Kotireddy

Towards Robust Low-Energy Houses
A Computational Approach for Performance Robustness

Assessment using Scenario Analysis
Rajesh Kotireddy
/ Department of the Built Environment
bouwstenen 245
A Computational Approach for Performance Robustness Assessment
using Scenario Analysis
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven,
op gezag van de rector magnificus prof.dr.ir F.P.T. Baaijens, voor een commissie
aangewezen door het College voor Promoties, in het openbaar te verdedigen op
dinsdag 4 september 2018 om 13:30 uur
door
Rajesh Reddy Kotireddy
geboren te Kotireddygaripalli, Andhra Pradesh, India

Dit proefschrift is goedgekeurd door de promotoren en de samenstelling van de
promotiecommissie is als volgt:
voorzitter : prof.ir. E.S.M. Nelissen

e
1 promotor : prof.dr.ir. J.L.M. Hensen
copromotor : dr.ir. P. Hoes
leden : prof.dr.ir.arch. S. Roels (KU Leuven)

prof.dr. P. Heiselberg (Aalborg University)
prof.dr. L.C.M. Itard (TU Delft)
dr. R. Choudhary (Cambridge University)
prof.ir. J.D. Bekkering
Het onderzoek of ontwerp dat in dit proefschrift wordt beschreven is uitgevoerd in

overeenstemming met de TU/e Gedragscode Wetenschapsbeoefening.
A Computational Approach for Performance Robustness Assessment
using Scenario Analysis
This research has received funding from the Green Tech Initiative of the Eurotech
Universities Alliance and support by the VABI-PIT Fund.
© Rajesh Kotireddy, 2018
All rights reserved. No part of this publication may be reproduced, distributed, or

transmitted in any form or by any means, including photocopying, recording, or
other electronic or mechanical methods, without the prior written permission of the
author.
Cover image: Licensed from IStock (Getty Images)
A catalogue record is available from the Eindhoven University of Technology Library

ISBN: 978-90-386-4497-4
NUR: 955
Published under Bouwstenen series 245 of the Department of the Built

Environment, Eindhoven University of Technology.
Printed by TU/e Print service (Dereumaux), The Netherlands.

To Ma & Pa
Acknowledgements
In my view, a PhD is a “process of learning”.
This process of learning has been successful thanks to several people who
contributed one way or another in helping me to complete this dissertation. It has
been a very special learning process thanks to all the wonderful people in the
computational building performance research group.
Firstly, I am grateful to Prof. Jan Hensen for giving me the opportunity to pursue
this process of learning in his computational building performance research group.
Thank you for your guidance and support in all aspects of this process of learning. I
feel very privileged to have completed this process of learning under your
supervision. I respect and admire your method of supervision. I particularly
appreciated the many astute insights and constructive feedback provided in progress
meetings, your encouragement of researchers to always consider the practitioners’
perspective and to understand the importance of balancing professional and personal
life. I have learned a lot from you in the past four years. In my culture, one should
give reverence in due order to “Mother, Father, Teacher, and God”. In your case, this
is very true. Thank you very much for being a great teacher.
I am very thankful to Dr. Pieter-Jan, who was my daily supervisor. Pieter-Jan is a

great supervisor and always provided constructive and critical feedback. Thank you
for being patient with my so-called two-minute discussions that often lasts for an
hour. I have really enjoyed our weekly meetings and learned a lot from you, especially
about robustness. Your feedback has helped greatly to enhance the overall quality
and structure of this dissertation.
I would like to express my gratitude to the doctoral committee members Prof. Staf
Roels, Prof. Per Heiselberg, Prof. Laure Itard, Dr. Ruchi Choudhary and Prof. Juliette
Bekkering for their time and constructive and valuable feedback. Your comments on
this dissertation have helped me immensely to improve the overall quality of this
thesis. My sincere thanks to user group members Ed Rooijakker, Hennekeij Ronald,
Paul Korthof, Steven Mast and Wim Plokker for their valuable time and feedback in
the suitability and usability assessment of the developed research approach in
practice. Their practical insights helped greatly to enhance suitability and usability of
the developed research approach in practice. I would like to thank Sanket Puranik
VII
Acknowledgements
and Sergio Costa for testing the developed approach using practical case studies in
their final graduation projects. I am very thankful to Jesse Plas for his help with the
collection and translation of reports related to statistics of Dutch dwellings. I would
like to thank Benjamin Manrique for co-writing a journal paper. My sincere thanks
to Dr. Mohamed Hamdy for his valuable input on optimization techniques. Many
thanks to Rohith for his help with the cover design. I would like to express my
gratitude to Duncan Harkness for his advice on academic writing and for
proofreading this dissertation.
Many thanks to all researchers in the building performance group for their valuable
feedback during progress meetings, which enhanced learning process. It has been a
great learning experience and a pleasure discussing robustness and so-called multi-
dimensional bubble charts with you all. Special thanks to Roel, who always had time
and answers for my questions. I really admire your expertise on a wide range of
research topics. Thank you Roel for being the Wikipedia of our research group. My
sincere thanks go to Ignacio and Raul for their valuable suggestions in both
professional and personal life. I learned a lot from you both and enjoyed our
conversations on various topics over drinks and dinners. Many thanks to all
administration staff of the Built Environment department, especially Building
Physics and Services (BPS) unit secretariat Leontine for ensuring a smooth learning
process.
This process of learning has been very enjoyable thanks to the amazing friends I
made here, especially Vojta, Rebeca, Parisa and Petr, who gave me a great start to the
social aspects of this process. Thanks to their wonderful company, I felt like home
far away from home. Isa, Olga, Zahra, Sanket and Christos have also been great
friends and their support has been incredible in this process of learning. I really
enjoyed coffee breaks at work with you all. My sincere thanks to fellow lunch club
members Luyi, Marie, Shikha and others for cooking all those delicious meals and
making lunch time memorable. I would also like to thank all BB&B club members
for memorable Friday evenings and nice conversations, which were a great source of
cultural learning. I am very thankful to Rebeca, Petr, Vojta, Eki, Tomaz, Ignacio,
Christos, John and Rubina for being great hosts and wonderful company during my
trips to your countries. Thank you to all my friends and colleagues who visited my
remote village in India. You have left there great memories which will be cherished
forever.
VIII
During this process of learning, I met many wonderful people from all over the world
who came to work at the BPS unit. Many thanks to Adam Bognar, Adam Wills,
Alessio, Antia, Asit, Azee, Bashar, Benedetto, Bruno, Chul-Sung, Christina, Dmitry,
Emy, Evangelos, Fotis, Gerardo, Hamid, Indra, Johann, Juliette, Katarina, Kennedy,
Marcel, Martina, Massimo, Munish, Qin, Raffaelle, Rizki, Sai, Sam, Sammy, Sanja,
Stephan, Yasin and many others for creating a very nice working environment. I
would like to thank my team mates of the TUe cricket team for successfully winning
the league a couple of times in the last four years. It has been a pleasure representing
the team and special thanks to Roshan for his support. I would like to thank my
fellow members at the Indian student association of TUe for welcoming me in
organizing Indian cultural events like Diwali and Holi. My sincere thanks to friends
at the Telugu association of the Netherlands, especially Siva Ram for organizing
various Telugu cultural events that made me feel at home.
I would also like to express my gratitude to many people who helped and motivated
me to start this process of learning. Special thanks to Dr. Subash Chandra Bose, who
is a constant source of inspiration, for his support throughout my studies. My sincere
thanks to Prof. Venkatarathnam for motivating me to pursue PhD during my MS. I
would like to extend thanks to my school teachers; (the late) Subramanyam and
Venugopal for their encouragement to always aim high. Very special thanks to my
cousin Madhava Reddy for his continuous support in all aspects of my life.
My family’s endless support ensured the successful completion of this process of

learning. Thank you, Ma and Pa, for your love, care and support. Many thanks to my
sister Nivedita, brother-in-law Ramana Reddy, brother Ramesh and sister-in-law
Lavanya for their unconditional love and support. I am forever indebted to my
brother Ramesh for his great support, and without his support, I would not have
been here. Very special thanks to my nephews Gokul, Rahul and little one Charith
for funny Skype conversations on weekends. You guys are my best buddies and sorry
for being a virtual uncle. Many thanks to my friend Monika for her continuous
support and nice talks whenever I am down. Special thanks to my fiancé, Gowthami,
for her patience during my PhD, otherwise a ‘dual PhD’ would not have been such
an easy process of learning.
Rajesh Kotireddy
IX
Summary
The built environment is under an obligation to move towards low-energy buildings

that provide high indoor environmental quality at low or no CO 2 emissions to reduce
the man-made impacts on the environment. Reductions of CO 2 emissions from
buildings are typically achieved by improving building insulation levels, using
highly efficient glazing, using energy efficient equipment and integrating renewable
energy technologies into buildings and into the wider built environment.
Considering the substantial cost required for the implementation of these
measures, future operational performance is a significant criterion in the design
process of low-energy houses.
The operational performance of low-energy houses (e.g. energy bills and comfort
conditions) is influenced by a multitude of dynamic factors, including occupant
behaviour, future climate conditions and economic factors. At the time of designing
a house, it is largely unknown how these influences will unfold over its life-span.
Typically, designers make use of set of assumptions to describe various operational
and external aspects such as occupancy patterns, temperature setpoints, plug loads
and weather conditions. Many of these assumptions are generally based on
empirically-derived educated guesses, and often need to rely on incomplete
knowledge and approximations. As such, there is a reasonable chance that the actual
conditions will differ from the assumptions.
In low-energy houses, these deviations are a serious concern, as their performance

is very sensitive to uncertainties. In addition, multiple low-energy building design
configurations lead to similar performances under deterministic conditions, but
these configurations can have significantly different magnitudes of performance
variation under uncertainties. Despite these concerns, uncertainties are rarely
considered in the design of low-energy houses, and hence the decision-making
processes may inadvertently result in designs that are sensitive to uncertainties and
might not perform as intended. To cope with these issues, it is important to analyze
the propagation of uncertainties and their impact on building performance. This is
also essential for the purpose of risk assessment in the early design phase for design
decision support. Therefore, performance assessment of low-energy houses taking
into account uncertainties should be assessed in the design phase to identify robust
designs that are capable of delivering the desired performance over their life-span.
XI
Summary
However, in many cases, the probability of occurrence of these uncertainties is

unknown, and as such it is hard for designers to quantify their impact. The approach
taken here to capture and quantify as many of these uncertainties as is feasible is to
describe and test future ‘scenarios’ that are likely to have an impact on building
performance. The integration of these scenarios into the performance robustness
assessment allows for a thorough investigation of these uncertainties. Accordingly,
a non-probabilistic robustness assessment based on scenario analysis is essential to
the goal of identifying robust designs. In practice, the design decision-making
process is complicated as it routinely involves multiple decision makers, with
multiple and conflicting performance requirements. As such, a multi-criteria
assessment is required to enable various decision makers to choose their preferred
robust designs from a large design space. Therefore, the aim of this research is to
develop a computational performance robustness assessment (CPRA) approach to
assess the performance robustness of low-energy houses considering future
scenarios to identify robust designs for various decision makers that have the
potential to deliver the desired performance over building’s life-span.
The computational performance robustness assessment (CPRA) approach

comprises multi-criteria assessment and multi-criteria decision making considering
multiple performance indicators and their corresponding robustness. In this
approach, taking into account the decision maker’s preferences, building design
space, future scenarios and performance indicators are defined. The performance of
the design space for future scenarios is assessed using building performance
simulations with multiple performance indicators and their corresponding
robustness. Different robustness assessment methods from various fields such as
statistics, operations research, structural design etc. are reviewed, and a few methods
that are used for robustness assessment using scenario analysis, namely the max-
min method, the best-case and worst-case method, and the minimax regret method,
are applied in the building performance context.
The developed CPRA approach is implemented using a simulation framework, and

to make this simulation framework more efficient and relevant in practice, scenario
sampling using the uniform Latin hypercube sampling method and a genetic
algorithm based multi-objective optimization method are implemented. While
identifying the same robust designs, the implemented simulation framework can
save up to 94–99.9% of computational time compared to a full factorial approach,
which require millions of simulations.
XII
The CPRA approach is demonstrated using residential house case studies for both
renovations and new houses with the policymaker and the homeowner as decision
makers. The developed CPRA approach is assessed for its usability and suitability in
practice with a user group comprising of representatives from leading building
consultant and services companies in the Netherlands. This CPRA approach is
presented to the user group at various stages and the feedback from the user group
is incorporated to improve this approach.
The case study results revealed that the designs with moderate to high insulation
levels and large renewable energy and storage (RES) systems were the most preferred
robust designs for the policymaker. In contrast, for the homeowner, designs with
relatively low insulation levels and small RES systems were the most preferred robust
designs. Designs with low to moderate insulation levels and large RES system were
the most preferred robust designs for both decision makers combined, which is a
compromise between the robust designs that were selected for the homeowner and
the policymaker separately.
In summary, using the developed CPRA approach, a decision maker can select a
robust design from the large design space based on optimal performance and
performance robustness or can trade off the selected robust designs with required
additional investment cost. Furthermore, the decision maker can choose a robust
design by prioritizing a performance indicator and carrying out trade-off with the
performance and robustness of other performance indicators and required
additional investment cost. In addition, each decision maker can choose design
options separately, which are more robust to their preferred performance indicators.
In conclusion, the developed CPRA approach could be useful when various decision
makers with multiple performance requirements are involved in a project, and it can
be effective in identifying a robust design from a large design space. This CPRA
approach can be used by designers and consultants, as a first step, to aid decision
makers in the design phase to identify robust low-energy building designs that have
the potential to deliver the desired performance in future operation.
XIII
Table of contents
Acknowledgements ................................................................................................... VII
Summary ..................................................................................................................... XI
Nomenclature .......................................................................................................... XVII
1. Introduction ...........................................................................................................1
1.1 Need for low-energy houses ................................................................................. 1
1.2 Problem definition................................................................................................ 3
1.3 Research aim and objectives .............................................................................. 10
1.4 Research methodology ........................................................................................ 11
1.5 The scope of this research ...................................................................................13
1.6 Thesis outline...................................................................................................... 14
2. State of the art: performance robustness assessment .......................................17

2.1 Uncertainty sources ............................................................................................ 17
2.2 Performance assessment methods to quantify the impact of uncertainties .... 21
2.3 Robustness assessment in literature .................................................................24
2.4 Selected robustness assessment methods ......................................................... 33
2.5 Robust design selection ...................................................................................... 38
2.6 Concluding remarks ...........................................................................................42
3. Methodology to develop and test the CPRA approach ...................................... 45

3.1 Overview of methodology ................................................................................... 45
3.2 The developed and tested CPRA approach ....................................................... 48
3.3 Practical use of the CPRA approach .................................................................. 65
3.4 Concluding remarks ........................................................................................... 65
4. Implementation of the CPRA approach in the simulation framework ........... 67

4.1 Overview of the simulation framework ............................................................ 67
4.2 Selected tools in the simulation tool chain ....................................................... 70
4.3 Improving simulation framework efficiency .................................................... 71
4.4 Improving simulation framework efficiency using scenario sampling ........... 73
4.5 Improving simulation framework efficiency using multi-objective
optimization ....................................................................................................... 80
4.6 Concluding remarks ......................................................................................... 104
5. Description of case studies and decision makers’ preferences ...................... 107

5.1 Decision makers ............................................................................................... 107
5.2 Case studies ...................................................................................................... 107
5.3 Design option space of the case studies ........................................................... 111
5.4 Scenarios ............................................................................................................ 117
XV
Table of contents
5.5 Performance indicators ..................................................................................... 125

5.6 Case study simulation model ........................................................................... 132
6. Demonstration of the developed CPRA approach using case studies ............ 137
6.1 Overview ............................................................................................................ 137
6.2 Demonstration of the CPRA approach using renovation house case study with
the policymaker as a decision maker ............................................................... 139
6.3 Demonstration of the CPRA approach using the renovation house case study
with the homeowner as a decision maker ...................................................... 164
6.4 Demonstration of the CPRA approach using the renovation house case study
with both policymaker and homeowner as decision makers .......................... 177
6.5 Demonstration of the CPRA approach using the new house case study ...... 190
6.6 Comparison of robust design options for both case studies .......................... 197
6.7 Concluding remarks ........................................................................................ 201
7. Suitability and usability assessment of the CPRA approach.......................... 205

7.1 Overview ........................................................................................................... 205
7.2 Evaluation of CPRA approaches with the end users ...................................... 205
7.3 Feedback from user group meetings ............................................................. 206
7.4 Improvement of CPRA approach using iterative process .............................. 208
7.5 Practical use of the developed CPRA approach ............................................. 209
7.6 Concluding remarks .........................................................................................214
8. Conclusions, original contributions, limitations and future work ................. 215

8.1 Summary ........................................................................................................... 215
8.2 Main conclusions ..............................................................................................218
8.3 Original contributions of this research ............................................................ 221
8.4 Limitations and future research ...................................................................... 223
Appendices ................................................................................................................ 225

Appendix A. Glossary: high-performance buildings ................................................... 225
Appendix B. Thermo-physical properties of building envelope materials .................. 231
Appendix C. Investment and replacement costs of design options ............................ 233
Appendix D. Demonstration of the CPRA approach using renovation house case
study...................................................................................................................237
Appendix E. Demonstration of the CPRA approach using new house case study .... 267
Appendix F. User group meetings: response to the main feedback .......................... 287
References ................................................................................................................. 293
Publications list ...........................................................................................................311
Curriculum Vitae ....................................................................................................... 313
XVI
Nomenclature
Abbreviations
AHP : Analytical Hierarchy Process
ASHP : Air-Source Heat Pump
Ause : Appliance use
BENG : Bijna Energie Neutrale Gebouwen (in English: Nearly
Zero Energy Buildings)
CF : Crossover Fraction
COP : Coefficient of Performance
CO2 : Carbon dioxide
CPRA : Computational Performance Robustness Assessment
CS : Climate scenario
DHPA : Dutch Heat Pump Association
DHW : Domestic Hot Water
ELECTRE : Elimination and Choice Translating Reality
EPBD : Energy Performance of Buildings Directive
EPC : Energy Performance Coefficient
FF : Full Factorial
GA : Genetic Algorithm
GC : Global Cost
GSHP : Ground-Source Heat Pump
HR107 : High Efficient gas boiler
HVAC : Heating, Ventilation and Air-Conditioning
ICa : Additional Investment Cost
IHG : Internal Heat Gains
LHS : Latin Hypercube Sampling
LPD : Liters Per Day
Luse : Lighting use
MCDM : Multi-Criteria Decision Making
MC : Maintenance Cost
NM : Net-Metering
NSGA : Non-Sorted Genetic Algorithm
nZEB : Nearly Zero Energy Buildings
NZEB : Net-Zero Energy Buildings
OP : Occupancy Profile
OS : Occupant Scenario
OC : Operational Costs
P : Building Envelope Package
PEB : Plus-Energy Buildings
PF : Pareto Fraction
PI : Performance Indicator
PROMETHE : Preference Ranking Organization Method
PV : Photo-Voltaic
RC : Replacement Cost
XVII
Nomenclature
RES : Renewable Energy and Storage System

RI : Robustness Indicator
RP : Renovation Package
SDHW : Solar Domestic Hot Water System
SPF : Seasonal Performance Factor
TS : Tournament Size
TOPSIS : Technique for Order Preference by Similarity to Ideal
Solutions
ULH : Uniform Latin Hypercube Sampling
Vent : Ventilation
Wopen : Window opening
WWR : Window to Wall Ratio
ZCB : Zero Carbon Buildings
Symbols
α : Coefficient of realism
ΔR : Relative Deviation
d : design
e : Escalation factor
Elimp : Imported electricity
Elexp : Exported electricity
fCO2,El : CO2 emission factor for electricity
fCO2,NG : CO2 emission factor for natural gas
fd,e : Discount factor
k : Life-span of a component
n : Calculation period of global cost
PInorm : Normalized performance indicator
PEl : Electricity price
PNG : Natural gas price
r : Interest rate
re : Real interest rates
R : Regret
RInorm : Normalized robustness indicator
Ta : Ambient temperature
Te,ref : Reference outdoor temperature
Ti : Indoor temperature
Thsp : Heating setpoint temperature
Tn : Neutral temperature
Tlower : Lower temperature limit of comfort band
Tupper : Upper temperature limit of comfort band
XVIII
1.Introduction
First, this chapter explains the need for low-energy houses and discusses the issues
concerned with the current design practice of these houses. Then, this chapter emphasize
the importance of performance assessment under uncertainties to address these concerns.
The research aims and objectives within the scope of this research are formulated in this
chapter. The research methodology developed to meet these research objectives is
summarized and the chapter concludes by providing an outline of this dissertation.
1.1 Need for low-energy houses
The built environment is among the largest energy consumers and contributors to
man-made climate change. To reduce this impact, the built environment is under an
obligation to move towards low-energy buildings. For instance, to address increasing
environmental concerns in the European Union, the energy performance of
buildings directive (EPBD) recast states that all new buildings should be nearly zero
energy from 2020 [EPBD, 2010]. Based on the EPBD directive [European
Commision, 2002; EPBD, 2010], the Netherlands has set national targets to achieve
45-80% energy saving in the built environment and an energy performance
coefficient (EPC) of zero for new buildings from 2020 [Hermelink et al., 2013]. EPC
is an index to calculate the energy efficiency in new buildings [Betlem et al., 2010];
the evolution of EPC requirements for new buildings for the past few decades and
near future is shown in Figure 1.1.
Similarly, for existing buildings, the Netherlands has set a target of achieving energy
label B averaged over the housing stock by the end of 2020 [Hermelink et al., 2013].
It can be seen from Figure 1.2 that the majority of the housing stock is still below
energy label B and requires major renovation [Meijer et al., 2009] to meet the 2020
targets. Furthermore, all buildings should achieve a 90% reduction in CO2
emissions by 2050 [EFFRA, 2013] compared to 1990 and preferably be
energy/carbon neutral buildings as indicated in the energy-efficient buildings multi-
annual roadmap by 2050 [European Commission, 2009] in order to move towards a
de-carbonized economy as directed by the European Union [European Commission,
2011]. To meet these stringent targets, buildings are currently designed based on
1
Chapter 1. Introduction
many sustainability frameworks, standards, codes and regulations such as Trias

Energetica [Lysen, 1996; VROM, 2010; RVO, 2013], Passivehaus [PHPP, 1998; PEP,
2008] etc. so that buildings deliver optimum indoor environmental quality (IEQ) at
minimum environmental impact, i.e. low-energy buildings.
1.4
Energy Performance
Coeffecient (EPC)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
Figure 1.1 Minimum EPC requirements for new buildings across different periods
[Agentschap-NL, 2012].
A
6% 4% 11%
B
11%
16% C
21% E
F
31%
G
Figure 1.2 Distribution of energy labels among labelled dwellings as of January 1, 2017
[Compendium voor de Leefomgeving, 2017].
A low-energy house is defined as a building “built according to special design criteria

aimed at minimizing the building’s operating energy” [Sartori and Hestnes, 2007]. In
the EPBD concerted action, which aimed to identify terms and definitions for high-
performance buildings [Erhorn and Erhorn-Kluttig, 2011], it was found that low
energy strategies underpinned all of the identified terms, such as low-energy house,
2
Need for low-energy houses
passive house, zero-energy house, plus-energy house, and zero-carbon house [Kibert
and Fard, 2012]. These low energy strategies are typically implemented by improving
building insulation levels, using highly efficient glazing, using energy efficient
equipment and integrating renewable energy systems into the built environment
[Gram-Hanssen, 2013; Butera, 2013; Becchio et al., 2015; Voss et al., 2011].
Considering the high economic costs required for the implementation of these
measures in the built environment, it is important to ensure that these measures
deliver the desired performance over the building’s life-span.
1.2 Problem definition
1.2.1 Issues in current design practice
New houses and renovations are currently designed using sustainability frameworks
and building codes and regulations to meet the aforementioned multi-annual targets
directed by the European Union. The main design strategy is to reduce the energy
demand for space heating and cooling as much as possible and to cover this demand
to a significant extent or fully through the use of renewable energy generation
systems. For instance, based on Passivehaus standards, the annual heating energy
demand for a conditioned building space should be less than 15 kWhth/m2a [PHPP,
1998; PHI, 1990]. Similarly, in the Netherlands, based on “Bijna Energie Neutrale
Gebouwen” (BENG1 – translates to English as nZEB, nearly zero energy building)
standards, the annual heating and cooling demand should be less than 25 kWhth/m2a
[RVO, 2015b]. This low energy demand is typically achieved by minimizing
transmission and ventilation losses and by maximizing passive solar gains. Hence,
low-energy houses, designed based on sustainability frameworks and standards,
have very highly insulated and air tight building envelopes and consequently low
heating and cooling energy demands.
In low-energy houses, due to their low energy demands, the variations in building
operation compared to design assumptions (e.g. occupant behavior) could
significantly influence the energy performance of these houses [Maier et al., 2009;
Martinaitis et al., 2015], resulting in a large deviation between measured and
1
BENG means “Bijna Energie Neutrale Gebouwen” in Dutch, which is roughly
translated as nearly zero energy building (nZEB). Hereafter, in this dissertation,
nZEB is used in place of BENG.
3
predicted energy use [de Wilde, 2014; Majcen et al., 2013]. For instance, large
variations are observed in the measured heating energy use across several identical
low-energy houses (see Figure 1.3). It can be observed that different heating setpoints
are preferred in different houses. For the same heating setpoint temperature (house
number 02 & 04), there is a huge difference in heating energy consumption; the
maximum difference in heating energy consumption among all houses is about
65%. This performance difference in identical houses is probably due to variations
in occupant behavior, which indicates the importance of the integration of these
variations in the design phase to reduce the performance deviation during operation
compared to predicted performance.
Heat Temperature
Figure 1.3 Measured indoor temperature and heating energy consumption in several
identical low-energy houses [Maier et al., 2009].
Annual percentage of overheating based on
Passivehaus standard
Annual overheating percentage based on the actual number of hours with elevated temperature during the monitoring period
Annual overheating percentage based on the anticipated number of hours with elevated temperature during the rest of the cooling season
Allowable annual overheating percentage based on the Passivehaus standard
Figure 1.4 Overheating hours in social housing flats built to the Passivehaus standards,
across different monitored periods (A: 17 Aug – 30 Sep, B: 3 July – 5 August, C: 1 May –
30 August) [Sameni et al., 2015].
4
Problem definition
Low-energy houses can also lead to indoor environmental quality issues such as
overheating. This overheating issue is increasingly evident in the houses with highly
insulated and airtight building envelopes [McLeod et al., 2013; Sameni et al., 2015;
Rodrigues et al., 2013; Zero Carbon Hub, 2015; de Wilde et al., 2008]. An example
of this overheating issue observed in social housing flats that are built to Passivehaus
standards is shown in Figure 1.4 [Sameni et al., 2015]. It can be observed that the
annual percentage of overheating is way above the acceptable limits based on the
Passivehaus standard. This study shows that houses built to the Passivehaus
standard are at considerable risk of overheating and that 72% of monitored flats
failed to operate as designed.
Overheating issues in low-energy houses are likely due to a number of reasons,

including heat being trapped in highly insulated air-tight envelopes and variations in
occupant behaviour such as internal heat gains, appliance use, window opening and
control of ventilation systems. This risk of overheating may significantly increase in
the future due to increases in outdoor temperature as a result of climate change.
The literature reveals that variations in occupant behavior and weather are among
the major factors that influence building performance [Hoes et al., 2009; Guerra-
Santin and Itard, 2010; Yan et al., 2015; de Wilde and Tian, 2009; de Wilde and
Coley, 2012]. In addition, policy changes also influence building performance and
future policies will have a drastic effect on the way these low-energy houses are
designed. For instance, in the current net-metering policy in the Netherlands, the
energy imported and exported on an annual basis are equally priced if the annual
net-exports is ≤3500 kWhe/a. If the annual net-exported electricity exceeds this limit,
then the surplus energy is paid at low prices [RVO, 2015c; E.ON, 2017b; Nuon, 2017].
Therefore, the grid is used as a virtual energy storage in the design of low-energy
houses, which results in the grid often being oversupplied and under great stress in
summer months. It seems improbable that this current net-metering policy will
continue in the future [KEMA, 2016; RVO, 2015c] and thus, these buildings should
maximize the self-consumption of energy generation as exporting energy to the grid
will likely be unprofitable in the future. Ultimately, achieving a balance between
energy demand and renewable energy generation is essential [Kotireddy et al., 2015]
to deliver the desired performance over a building’s life-span. To do so, variations in
building operation and external factors should be considered in the design process.
5
However, in the conventional design approach, building performance is predicted

based on a set of assumptions about a building’s operational conditions and other
external factors. In reality, as noted in the previous examples, the operational
performance of buildings (e.g. heating demand and comfort conditions) is
influenced by a multitude of dynamic factors, including occupant behaviour [Yan et
al., 2015], future climate [de Wilde and Coley, 2012] and economic factors [Rysanek
and Choudhary, 2013]. At the time of designing a building, it is usually unknown to
designers how these influences will unfold over a building’s life-span. Therefore, it
is necessary to make assumptions regarding such scenarios in order to communicate
building performance targets during the design phase. The use of building
performance simulation (BPS) can be a valuable tool in this context, as it can assist
decision making by enabling quantitative comparisons between design intent and
the predicted performance of various design options [Attia et al., 2012; Clarke and
Hensen, 2015].
Typically, BPS users make use of an average scenario to describe various operational
and external aspects such as occupancy patterns, temperature setpoints, plug loads,
and weather conditions. Many of these assumptions are usually based on
empirically-derived educated guesses, but often need to rely on incomplete
knowledge and approximations. As such, there is a reasonable chance that the actual
conditions will differ from the assumptions made in that average scenario. These
deviations are a serious concern for low-energy houses, as their performance is very
sensitive to such uncertainties [Hoes et al., 2009; Van Gelder et al., 2014; McLeod et
al., 2013; Rysanek and Choudhary, 2013]. Despite this importance, uncertainties are
rarely considered in the design of these buildings, and hence the decision-making
processes may inadvertently result in designs that are sensitive to uncertainties and
might not perform as intended [Mavrotas et al., 2015; Hopfe et al., 2013].
To ensure intended performance, buildings can also be designed to have a relatively

high tolerance to uncertainties. Buildings with this characteristic are said to be robust,
but tend to rely on oversized systems that require high investment and operating
costs [Gang, Wang, Xiao, et al., 2015; Lu et al., 2017] and invariably lead to
environmental impacts. In addition, multiple building design configurations lead to
similar performances under deterministic conditions, but these configurations can
have significantly different magnitudes of performance variation under uncertainties
[Kotireddy et al., 2017a]. Therefore, it is essential to achieve robust designs
considering uncertainties that have potential to deliver desired performance [Gang,
6
Problem definition
Wang, Xiao, et al., 2015; Østergård et al., 2017; Lu et al., 2017] at low costs. To cope
with these issues, it is important to analyze the propagation of uncertainties and their
impact on building performance in a systematic manner. This is also essential for
the purpose of risk assessment in the early design phase for design decision support.
For instance, conducting performance robustness assessment is necessary to
address one of the main issues that the building industry is currently facing in the
development of low-energy houses; performance deviation during operation
compared to the predicted performance [Kotireddy et al., 2018].
Therefore, performance assessment of low-energy houses taking into account these

uncertainties should be assessed in the design phase to identify robust designs that
have the potential to deliver the desired performance over the building’s life-span
[Leyten and Kurvers, 2006].
1.2.2 Quantification of impact of uncertainties

It is important to quantify the impact of uncertainties on building performance
[Woloszyn and Beausoleil-Morrison, 2017] to help decision makers to make
informed design decisions considering uncertainties. Depending on the
requirements of decision makers and the stage in the building life-cycle, different
sources of uncertainty play a role, and different uncertainty analysis techniques can
be used to quantify their impact for the task at hand. Many different terms are used
in literature to describe the ability of a system to deliver performance in an uncertain
environment [Ross et al., 2008; Chalupnik et al., 2013; Mekdeci et al., 2015].
Commonly-encountered terms in statistics, systems engineering, manufacturing
analysis, aerospace engineering and operations research include reliability,
robustness, resilience, flexibility and adaptability among others. Finding an
appropriate method based on relevance and the requirements of a project is a tedious
task for a designer. This issue is seldom addressed in the building performance
context and there is a growing need for integration of such methods [Tuohy, 2009;
de Wilde and Coley, 2012; Østergård et al., 2017]. Therefore, these methods from
different fields are reviewed and the selected methods in the present context are
discussed in detail in the next chapter.
It is found that robustness assessment is widely used if the assessment period spans
years to a few decades. In addition, the nature of uncertainties and their range are
precisely defined in robustness assessment [Anderies et al., 2013; Chalupnik et al.,
2013; Mekdeci et al., 2015]. In the present context, the life-span of energy efficiency
7
measures and renewable energy technologies considered for low-energy house

designs is about 25-50 years and the range of uncertainties for assessment is also
defined. Therefore, the robustness assessment approach is used in this research.
Robustness, in this research, is defined as the ability of a building to maintain the desired
performance under uncertainties in building operation and external factors.
1.2.3 Performance robustness assessment based on scenario analysis

In many cases, the designer has limited or no information about the probability of
the occurrence of uncertain situations, and it is thus difficult to quantify the
associated risks. For instance, in most cases it is unknown during the design phase
what type of households will occupy the building over its life-span and what their
corresponding behaviors will be. Similarly, large uncertainties are associated with
climate change projections [van den Hurk et al., 2006; KNMI, 2014]. In addition, it
is difficult to probabilistically define uncertainties in the future economy such as
electricity prices and policy changes [Rysanek and Choudhary, 2013]. As such, one
way to proceed is to use ‘scenarios’, which can be understood as formulated
alternatives when probabilities of uncertainties are unknown [Struck and Hensen,
2013; Kim, 2013] and can be used to integrate uncertainties into the performance
robustness assessment [Struck and Hensen, 2013; Hopfe et al., 2013].
Scenarios are used to present a range of possible alternatives so that the robustness
of designs can be assessed based on how different designs perform in each of these
alternatives [Polasky et al., 2011]. For instance, using scenario analysis, the risk can
be quantified based on an optimistic or a pessimistic approach using the best-case
and worst-case scenarios. Thus, scenarios aid the better understanding of
uncertainties and help to determine designs that are robust across range of possible
futures [Moss et al., 2010].
The non-probabilistic decision rules have been implemented to identify robust

building retrofits under technical and economic uncertainties by [Rysanek and
Choudhary, 2013], and this research demonstrated that the approach was useful for
scenario modelling and that it allowed for easier identification of robust designs
among design alternatives. Similarly, Hoes et al., [2011] carried out building
performance robustness assessment considering scenarios dealing with
uncertainties in occupant behavior. The preferred robust design using this method
is more robust to occupant behavior but could result in very uncomfortable indoor
temperatures (e.g. overheating), as observed in their previous study [Hoes et al.,
8
Problem definition
2009]. This overheating risk will be even higher in the future due to climate change
[McLeod et al., 2013; Rodrigues et al., 2013; de Wilde et al., 2008], and hence it is
important to include uncertainties in climate change in the design process [de Wilde
and Coley, 2012; Coley et al., 2012]. Climate change scenarios are included in
performance robustness assessment by [Nik et al., 2015; de Wilde et al., 2008;
Kotireddy et al., 2015; Kotireddy et al., 2018].
In the reported research, robustness assessment is carried out separately for user
scenarios [Hoes et al., 2011; Kotireddy et al., 2015], technical and economic scenarios
[Rysanek and Choudhary, 2013] and climate scenarios [Nik et al., 2015; de Wilde et
al., 2008; Kotireddy et al., 2015]. In such cases, a design that is robust to one scenario
could be sensitive to other scenarios. As such, a performance robustness assessment
considering all scenarios is essential in order to provide decision makers with
insights into how a design performs in each scenario in order to enhance the design
decision-making process.
1.2.4 Multi-criteria assessment and multi-criteria decision making

In a typical building design decision-making process, various decision makers such
as architects, designers, engineers, owners etc. are involved. The goal of the process
is to select their preferred designs from a design space. In practice, this process is a
complicated and difficult task, especially when it involves decision makers with
multiple and conflicting performance requirements [Mela et al., 2012]. For example,
if the decision maker is a homeowner, then his/her design selection criteria will
probably depend on indoor environment quality and operating and investment costs.
These priorities can be contrasted with, for example, those of a policymaker, who is
more focused on CO2 emissions and investment costs required for the
implementation of CO2 reduction measures. Similarly, for both decision makers the
design selection criteria may depend on indoor environment quality, operational and
investment costs, and CO2 emissions. In such cases, a multi-criteria performance
assessment considering two or more performance indicators is typically used [Perera
et al., 2013; Hamdy et al., 2013].
The difficulty of the decision making task increases significantly if uncertainties are
also included, and this issue is rarely addressed in the building performance context
[Hopfe et al., 2013]. In the literature, multi-criteria performance assessment
considering uncertainties has been carried out by [Y. Sun, Huang, et al., 2015; Zhang
et al., 2016]. However, they use predefined weights for each performance indicator;
9
prescribing weighting factors often requires knowledge and expertise that may often
be unavailable. Furthermore, assigning weights is difficult when two or more
performance indicators are considered [Mela et al., 2012]. To overcome these issues,
a multi-criteria assessment method based on a trade-off approach to identify robust
designs for various decision makers is desirable. In addition, to facilitate the selection
of the most robust design from a large design space for various decision makers, a
multi-criteria decision-making method is essential.
1.3 Research aim and objectives
It is clear from literature that there is a lack of a holistic design decision support
methodology for performance robustness assessment of low-energy houses
considering future scenarios. To bridge this methodological gap, this research aims
to develop a computational approach to assess the performance robustness of low-
energy houses for future scenarios that considers uncertainties in building operation
and external factors in order to identify robust designs for various decision makers.
To meet this research goal, the following objectives are formulated.
The objectives of this research are:
1. Develop a computational performance robustness assessment (CPRA)

approach that integrates uncertainties in multi-criteria assessment using
scenario analysis to quantify robustness and facilitate the selection of robust
designs for various decision makers.
2. Develop a simulation framework to implement the CPRA approach in order

to carry out performance robustness assessment of low-energy houses
across the considered scenarios.
3. Demonstrate, using case studies, how the CPRA approach can aid various
decision makers to identify robust designs based on their preferences.
The specific objectives of this research are:
4. Identify uncertainties in building operation and external conditions such as

occupant behavior, climate change, policy changes and integrate identified
uncertainties in the building performance assessment through scenario
analysis.
10
Research aim and objectives
5. Identify appropriate robustness assessment methods for scenario analysis

in building performance and adjacent fields, and implement them in the
present context.
6. Carry out multi-criteria performance assessment of various low-energy

building designs for future scenarios using multiple performance and
robustness indicators.
7. Implement multi-criteria decision making using the trade-off approach on

the Pareto front and multi-criteria decision-making method (MCDM) to
identify robust designs for different decision makers based on their
preferred performance and robustness indicators.
8. Carry out sensitivity analysis to identify the most influential scenarios on

performance and robustness indicators.
9. Assess the suitability and usability assessment of the developed CPRA

approach in practice with the end users.
1.4 Research methodology
The first step of this research is to review different uncertainty sources and identify
uncertainties arising from building operation and from other external factors.
Similarly, different robustness assessment methods from other fields, including the
building performance context, are reviewed. Based on this review, appropriate
robustness assessment methods based on decision makers’ attitude towards the risk
acceptance in the decision-making process are selected. In addition, different multi-
criteria decision-making methods are compared to find the suitable method in the
present context.
The next step is to develop the computational performance robustness assessment

(CPRA) approach. In this approach, multi-criteria performance assessment and
multi-criteria decision making considering multiple performance indicators are
implemented to identify robust designs for various decision makers. The developed
approach is tested using case studies and these results are presented to the end users.
The feedback from the end users on the CPRA approach is then incorporated, as
shown in Figure 1.5, through an iterative process to enhance its suitability and
usability in practice. After several iterations, the final CPRA approach is developed.
11
The developed approach is implemented using a simulation framework. Conducting

a performance and robustness assessment of the design space across the considered
scenarios with multiple performance and robustness indicators is computationally
expensive. Therefore, a multi-objective optimization method and a scenario
sampling strategy is used to reduce this computational cost. Robustness indicators
calculated using three robustness assessment methods are integrated in the multi-
objective optimization method. This simulation framework is used to test the CPRA
approach using case studies to identify robust designs for different decision makers.
Selection of robust designs is demonstrated using a trade-off approach and a multi-
criteria decision-making method through various visualization methods to enhance
the design decision-making process. Sensitivity analysis is carried out to identify the
most influential scenarios on the preferred performance and robustness indicators.
Develop the
CPRA
approach
Improve the
CPRA Test the
approach CPRA
through an approach in
iterative case studies
process
Receive the Present the

feedback CPRA
from end approach
users on the using case
CPRA study results
approach to end users
Figure 1.5 The research methodology implemented in this work to develop and test the
computational performance robustness assessment (CPRA) approach.
12
The scope of this research
1.5 The scope of this research

The developed CPRA approach is generic and can be used for performance
robustness assessment of new houses and renovated houses in a holistic way and
can also be used for robustness assessment of individual energy systems. However,
there are several areas that fall outside the scope of this work, which are elaborated
below.
i. Uncertainties
It is important to consider all uncertain parameters that can influence the
building performance over the building’s life-span. In literature, it is found
that uncertainties in occupant behavior, climate change and policy changes are
major factors influencing building performance. Therefore, these factors are
the main focus of the research. However, there are some uncertainties that
influence building performance which are not included in this research, such
as technology innovation and changes in energy markets, uncertainties in
thermo-physical properties of materials and performance degradation of
building components and energy systems. Since the purpose of this study is to
develop and demonstrate the CPRA approach, including these uncertainties
would significantly increase the complexity of the case studies and would result
in distraction from the main focus of the research. It is worth noting that
including these uncertainties would not change the CPRA approach.
ii. Low-energy houses

Low-energy house is the most widely used term among 23 different terms
identified to define high-performance buildings across 14 European Union
member states in the EPBD concerted action [Erhorn and Erhorn-Kluttig,
2011]. In addition, as noted earlier, all these identified terms such as passive
house, very low-energy house, zero-energy house, plus-energy house, zero-
carbon house follow low energy strategies. The design space considered in this
research constitutes the aforementioned houses, but the focus is neither
restricted to a particular type of building design nor is it to distinguish between
these building designs. The glossary of definitions of these high-performance
buildings is summarized in Appendix A [Mlecnik, 2012; Voss et al., 2012a].
iii. Case studies

The developed CPRA approach is tested using case studies of Dutch residential
houses. These case studies include various energy efficiency measures and
13
renewable energy systems for new houses and renovations. Robust design
options of these measures are identified for different decision makers. It is
reported in literature that novel building components such as climate adaptive
building shells can enhance building robustness [Loonen et al., 2013; Loonen
et al., 2017], however, they are not considered in this research. Similarly, for
reasons of scope, this research does not include districts/neighborhoods or
commercial buildings.
iv. Decision makers

The developed CPRA approach is demonstrated using case studies for the
policymaker and the homeowner, two key decision makers in the building
industry, who represent different interests in the performance of buildings.
The same approach can be used by other actors involved in the design decision-
making process, such as energy performance contractors, grid operators and
service providers. However, to limit the scope of this research, only the
policymaker and the homeowner are considered here.
1.6 Thesis outline

The thesis outline is presented in Table 1.1 and the organization of thesis is
summarized below.
Table 1.1 Details of the thesis chapters and the respective objectives addressed in each
chapter.
Addressed
Chapter Title research
objectives
2 State of the art: performance robustness assessment 4-5
Methodology to develop and test the CPRA

3 1
approach
Implementation of the CPRA approach in the
4 2, 4-5
simulation framework
Description of case studies and decision makers’
5 3,4
preferences
Demonstration of the developed CPRA approach
6 3, 6-8
using case studies
Suitability and usability assessment of the CPRA
7 9
approach
Conclusions, original contributions, limitations and
8 -
future work
14
Thesis outline
⎯ Chapter 2 provides an overview of performance robustness assessment

methods in different fields. This chapter concludes with a description of the
selected performance robustness assessment methods and multi-criteria
decision-making methods.
⎯ Chapter 3 presents a methodology to develop and test the computational

performance robustness assessment (CPRA) approach. The suitability and
usability assessment of the CPRA approach for end users is also discussed
in this chapter. An iterative process to improve the CPRA approach by
incorporating the feedback from end users is presented. The final CPRA
approach is described in the second part of this chapter.
⎯ Chapter 4 describes the simulation framework developed to implement the

CPRA approach. This chapter describes the simulation tools used in the
simulation framework. Furthermore, it describes the methods to make this
framework computationally efficient and, thus, to enhance its usability in
practice.
⎯ Chapter 5 describes the case studies used to test the CPRA approach using
the simulation framework. The design option space, future scenarios and
performance indicators are also described. Finally, the simulation models of
case studies are presented.
⎯ Chapter 6 demonstrates the CPRA approach using case studies. Robust

designs are identified for different decision makers using three robustness
assessment methods. A comparison is made among robust designs for
different decision makers for the tested case studies.
⎯ Chapter 7 assesses the suitability and usability of the developed CPRA

approach for end users. The practical use of the CPRA approach is
demonstrated using real case studies through PDEng and MSc graduation
projects.
⎯ Chapter 8 summarizes and concludes this research and highlights the

original contributions that it makes. The chapter ends by discussing
limitations of this research and plausible future work.
15
2. State of the art: performance
robustness assessment
This chapter reviews the uncertainty sources that can influence building performance and
methods to quantify the impact of these uncertainties. Robustness assessment methods
from different fields are reviewed and methods used in the present context are discussed.
Similarly, different approaches used to identify robust designs are reviewed, the selected
methods are presented, and the visualization methods used to present these results are
reviewed in this chapter.
2.1 Uncertainty sources

In conventional building design practice, performance assessment of building
designs is carried out using a set of fixed boundary conditions/assumptions and the
results of this assessment are used as criteria in the design decision-making process
[Attia et al., 2012]. In reality, as noted in Chapter 1, uncertainties abound in building
design and operation, and these uncertainties could affect the ability of designs to
realise the desired performance. Therefore, different uncertainty sources are
reviewed to find the relevant uncertainty sources in the present context. The different
purposes for analysing the propagation of uncertainties require different sources of
uncertainty to be considered. The main sources of uncertainties in building
performance predictions [De Wit and Augenbroe, 2002] are:
i. Modelling uncertainties, which arise due to simplification of complex

physical processes.
ii. Numerical uncertainties, which arise due to considered numerical methods

and model discretization.
iii. Input uncertainties, which arise due to unknown or uncertain parameter

values [Hoes, 2014].
It is acknowledged in the literature that fit-for-purpose modelling is essential to

reduce modelling uncertainties, for instance, in the case of systems modelling [Trcka
and Hensen, 2010] and occupant behaviour modelling [Gaetani et al., 2016a]. In the
17
Chapter 2. State of the art: performance robustness assessment
majority of studies, the influence of modelling and numerical uncertainties is

considered to be negligible as the physical model, spatial discretization and time
steps for assessment are assumed to be correctly chosen [Hoes, 2014]. Because of its
significant influence on building performance, the primary focus of this research is
also on the third category: input uncertainties.
There are many sources of input uncertainties that can influence building
performance, and these input uncertainty sources are broadly categorized in two
types as shown in Figure 2.1 [Francis and Bekera, 2014; Hopfe and Hensen, 2011; de
Wilde et al., 2008; Ramallo-González et al., 2015; Kiureghian and Ditlevsen, 2009]:
i. Epistemic uncertainties, which arise due to a lack of knowledge about a

variable/system property.
ii. Aleatory uncertainties, which arise due to the intrinsic randomness of a

variable/system property.
Uncertainty sources
(input uncertainties)
Epistemic Aleatory
Building specific Internal External

properties e.g. Uncertainties in
e.g. Uncertainties in
e.g. Uncertainties in building operation
weather data, climate
thermo-physical properties, such as occupant
change etc.
thickness, infiltration etc. behavior
Figure 2.1 Classification of input uncertainty sources that are typically integrated in
building performance simulations.
In practice, both epistemic and aleatory uncertainties are simultaneously present [Li
et al., 2012] and the choice of categorizing uncertainties as epistemic or aleatory
depends on the modeller [Kiureghian and Ditlevsen, 2009]. The epistemic
uncertainties can be reduced by acquiring further information about variable/system
18
Uncertainty sources
properties [Kiureghian and Ditlevsen, 2009]. These epistemic uncertainties include

uncertainties in building specific parameters such as thermo-physical properties
(e.g. thermal conductivity, density, specific heat capacity), thickness of building
materials, infiltration rates [Hopfe and Hensen, 2011; Chinazzo et al., 2015a].
Aleatory uncertainties are stochastic in nature and are irreducible [de Wilde and
Tian, 2009]. The aleatory uncertainties are grouped into internal and external
uncertainty sources (Figure 2.1). Internal uncertainty sources include the
uncertainties that arise in the operation of a building over its life-span, such as
occupant behaviour [Hopfe and Hensen, 2011]. External uncertainty sources include
uncertainties that arise in external factors [De Wit and Augenbroe, 2002] such as
weather conditions and the economy.
Extensive studies have been carried out by integrating epistemic uncertainties in

building performance simulations, notably on thermo-physical parameters such as
thermal conductivity, density and thermal capacity of building materials [Macdonald
and Strachan, 2001; De Wit and Augenbroe, 2002; Struck et al., 2009b; de Wilde
and Tian, 2009; Hopfe and Hensen, 2011; Eisenhower, Neill, Narayanan, et al., 2011;
Silva and Ghisi, 2014; Van Gelder et al., 2014] and infiltration [Macdonald and
Strachan, 2001; de Wilde and Tian, 2009; Hopfe and Hensen, 2011; Heo et al., 2012].
Likewise, extensive studies have been carried out on aleatory uncertainty sources and
particularly on occupant behaviour, which is considered a major factor causing
variations in building performance predictions [Yan et al., 2015]. This is evident as
there has been a huge surge in recent studies addressing occupant behaviour
uncertainties in building performance assessment [Hoes et al., 2009; Mavrogianni
et al., 2014; Hong et al., 2016; K. Sun and Hong, 2017; Van Gelder et al., 2014; Silva
and Ghisi, 2014; L. Wang et al., 2012; de Wilde, 2014; Struck et al., 2009a; Clevenger
and Haymaker, 2006; Macdonald and Strachan, 2001]. Similarly, much research
effort has been directed at uncertainties in weather data and climate change
[Clevenger and Haymaker, 2006; L. Wang et al., 2012; O’Neill and Eisenhower,
2013b; Nik and Kalagasidis, 2013; de Wilde and Coley, 2012; Holmes and Hacker,
2007; Leichenko, 2011; Wan et al., 2011; Mulville and Stravoravdis, 2016] and also
economic uncertainties [Rysanek and Choudhary, 2013; Rasouli et al., 2013; Fawcett
et al., 2012; Burhenne et al., 2013; Hamdy et al., 2013].
19
In general, uncertainties are integrated in building performance assessment through

uncertainty analysis for different purposes. Some purposes for quantifying
uncertainties using BPS are:
i. Improving models and model calibration [O’Neill and Eisenhower, 2013b;

Heo et al., 2012; De Wit and Augenbroe, 2002; Coakley et al., 2014].
ii. Overheating risk assessment [McLeod et al., 2013; Jenkins et al., 2014;
Hamdy et al., 2017].
iii. Enhancing the design decision-making process and early-phase design

support [Rysanek and Choudhary, 2013; Hopfe and Hensen, 2011; Rezaee et
al., 2015; Hopfe et al., 2013; Y. Sun, Huang, et al., 2015; Jafari and Valentin,
2017].
These purposes consider different sources of uncertainties. For instance,

uncertainties in thermo-physical parameters are typically used for quality assurance.
However, the impact of these epistemic uncertainties can be reduced by gathering
more detailed information, or by using complex and refined models. On the other
hand, aleatory uncertainties cannot be reduced or avoided and must be considered
in the performance assessment [Mullins et al., 2016]. Furthermore, scenarios
(aleatory uncertainties) are essential for long-term performance assessment of
buildings, especially performance robustness [Hopfe and Hensen, 2011]. The
purpose of using scenarios is to gain an improved understanding of the impact of
uncertainties to facilitate decision making during the design selection process with
the goal of choosing a design that has ability to perform under a variety of possible
future situations [Moss et al., 2010]. Therefore, in this work, scenarios (aleatory
uncertainties) are used for performance assessment.
In summary, it is evident from the literature that occupant behavior is a major factor
that influences building performance [Hoes et al., 2009; Guerra-Santin et al., 2010;
Yan et al., 2015; Gram-hanssen and Georg, 2017]. In addition, there is a growing
need for the integration of uncertainties in climate change in building performance
assessment to ensure the long-term performance and the adaptability of buildings to
climate change [de Wilde and Coley, 2012]. Similarly, integration of economic
uncertainties is essential to ensure the cost-optimality of designs [BPIE, 2010] both
in the near future and over the building’s life-span [Fawcett et al., 2012]. Therefore,
in this work, performance assessment is carried out considering uncertainties in
occupant behaviour, climate change and policy changes.
20
Performance assessment methods to quantify the impact of uncertainties
2.2 Performance assessment methods to quantify the impact of

uncertainties
2.2.1 Overview of methods

Typically, reliability, robustness, resilience, flexibility and adaptability (see Table 2.1)
are used across different disciplines to quantify the impact of uncertainties and to
determine the ability of a system/design to perform under uncertainties [Ross et al.,
2008; Chalupnik et al., 2013; Mekdeci et al., 2015]. However, some of these terms
are used interchangeably in literature. For example, robustness and resilience are
often used interchangeably to assess long-term performance under uncertainties
[Bankes, 2010; Anderies et al., 2013]. This is also the case with flexibility and
adaptability [Gosling et al., 2007; Geraedts, 2008; Gosling et al., 2013]; and with
robustness and reliability [Huang and Du, 2007; Nik and Kalagasidis, 2013; Baker et
al., 2008]. As such, it is generally difficult and confusing for a designer to choose an
appropriate assessment method. Therefore, these methods are reviewed and
compared below to find an appropriate method for the present context.
⎯ Reliability is defined as the ability of a system/design to deliver the required

performance without failure over a specified time, in line with the definition
of IEA EBC Annex 55 [IEA, 2013]. Furthermore, a reliable system/design
performs as stated under specified [IEEE, 1994] and uncertain conditions
[Chalupnik et al., 2013].
⎯ Robustness is defined as the ability of a system/design to deliver the

required performance [McManus and Hastings, 2006] under uncertainties
in operating conditions [Olewnik et al., 2004] and in specified
environmental conditions [Andersson, 1997]. It is also defined as the ability
of system/design to perform under a range of uncertainties [Chalupnik et
al., 2013] and to succeed across future scenarios [Mekdeci et al., 2015; Bettis
and Hitt, 1995].
⎯ Resilience is defined as the “ability of a system, as built/designed, to do its basic

job or jobs not originally included in the definition of the system’s requirements in
uncertain or changing environments” [Chalupnik et al., 2013; Hollnagel, 2014].
It is also defined as the ability of a system to withstand/recover/
mitigate/cope with foreseen and unforeseen uncertain conditions [Hassler
and Kohler, 2014; Nicol and Knoepfel, 2014; Hollnagel, 2014].
21
⎯ Adaptability is defined as the ability of a system/design to deliver the

required performance in uncertain conditions by responding to or changing
its function or structure [Gosling et al., 2013; Mekdeci et al., 2015].
⎯ Flexibility is defined as the ability of a system/design to deliver the required

performance in uncertain conditions by adapting to or restructuring its
function or structure [Geraedts, 2008; Mekdeci et al., 2015].
2.2.2 Comparison of methods and selection of appropriate method in the

present context
It is evident from the above comparison that the various uncertainty mitigation
methods differ greatly in the range of uncertainties considered and in their ability to
perform under these uncertainties. The range of uncertainties that can be considered
in the assessment of these methods is shown in Table 2.1. For instance, reliability is
used to assess whether the system consistently performs in accordance with the
designed conditions. Since reliability does not take into account uncertainties in
external factors, which are crucial in the performance assessment of buildings over
their life-span, it is not considered in this work. Similarly, flexibility and adaptability
are not considered suitable assessment methods as adaptive (i.e., changeable or
reconfigurable) building elements are not commonly encountered in the dwellings.
The relevant assessment methods in the present context are robustness and
resilience as both of these assessment methods consider uncertainties in building
operation and uncertainties arising from external factors.
Table 2.1 Various concepts considered for assessment under different uncertainties (adopted
from [Chalupnik et al., 2013]).
Uncertainties
Uncertainties Capable of adapting
in system/ Uncertainties
in system/ system/building
Concept building in external
building structure to cope
function factors
operation with uncertainties
Reliability X
Robustness X X
Resilience X X X
Adaptability X X X
Flexibility X X X X
22
Performance assessment methods to quantify the impact of uncertainties
In the present context, adopted from [Anderies et al., 2013], robustness differs from
resilience in the following aspects:
• Robustness focuses on “designing fail-safe systems within a defined range of

uncertainty” [Anderies, 2014; Anderies et al., 2013], whereas resilience tries
to “build fail-safe systems capable of learning, self-organizing, and adapting to
change” [Anderies et al., 2013].
• In robustness assessment, the nature of uncertainties (e.g. occupant

behavior) and their range is defined. In resilience assessment, uncertainties
even include unexpected conditions such as natural disasters.
• A precise definition of performance indicators (e.g. cost, comfort etc.) is

essential in robustness assessment. The function of a building and its
corresponding performance indicators can change over time in the resilience
assessment.
• Robustness assessment focuses on short to intermediate periods and ranges

from the building scale to regional levels. Resilience assessment focuses on
intermediate to long-term periods from cities to national level (see Table 2.2).
Table 2.2 Comparison of robustness and resilience based on design challenge, time period
and scale [Anderies, 2014].
Design
Time-period Scale Consideration
challenge
Micro-scale Building –
Short term Robustness
(Months-years) neighborhood level
Intermediate Meso-scale Cities – regional Robustness and
term (Years-decades) level Resilience
Macro-scale Regional – national
Long-term Resilience
(decades-centuries) level
In this research, performance assessment is carried out on residential houses over

their life-span, which is a few decades. To clarify, while the structure of a building
should have a longer life-span, the insulation and energy systems within the building
typically need to be upgraded or replaced after a few decades. Furthermore, the
nature and range of uncertainties is defined and unexpected uncertainties such as
natural disasters are not considered. The performance indicators on which
assessment under uncertainties has to be carried out are known before the
assessment. Therefore, the robustness assessment approach is used in this research
and robustness is redefined as follows:
23
Robustness is the ability of a building to maintain the desired performance under

uncertainties in building operation such as occupant behavior and in external conditions
such as weather conditions, climate change and policy changes.
It is worth noting that “robustness” in this study is the “performance robustness”.
2.3 Robustness assessment in literature
2.3.1 Robustness in adjacent fields
Robustness is a statistical approach that can be used to reduce errors due to

uncertainties [Bickel, 1976]. Genichi Taguchi, who is considered the father of robust
design [Beyer and Sendhoff, 2007], proposed a robust design approach using a
signal-to-noise ratio measure in order to reduce variations in output (signal) due to
uncertainties (noise). The Taguchi method, developed around 1950, and further
improved methods [Taguchi, 1987; Taguchi et al., 2000] are used to ensure a
designed product performs as intended, irrespective of uncertainties that can occur
over its life-span such as uncertainties in design, manufacturing and operation
[Huang and Du, 2007].
According to the Taguchi method, a robust design is one with minimum variation
around the target value, where the target value can be a specific nominal value, or
zero or infinite. In summary, mean and variance are indicators for robustness. These
indicators may not be preferable in all cases, especially when the likelihood of
occurrence of scenarios is equally probable or the probability of occurrence is
unknown. In such cases, taking the mean across scenarios nullifies the concept of
formulating scenarios as alternatives since it flattens out the results.
To overcome this difficulty, different methods for robustness assessment using

scenario analysis are implemented [Aissi et al., 2009; Averbakh, 2000]. In these
methods, the performance of a design is assessed for each scenario and robust
designs are those with the best possible performance even in the worst-case scenarios
or those that perform well in each scenario. In addition, Beyer and Sendhoff [2007]
reviewed different robustness methods and compared them with the Taguchi
method. An example of the robustness concept proposed in that study is shown in
Figure 2.2, where they stress the need to take into account different uncertainty
sources that may arise during the design process, such as uncertainties in
environmental and operating conditions, and design parameters.
24
Robustness assessment in literature
Robust design
performance evaluation
operating conditions
environmental and
system/building
Uncertainties in
Uncertainties in
System/building Objective
Uncertainties in
design parameters
Optimization
strategy
Figure 2.2 Robust design optimization considering different uncertainties such as uncertain
environmental and operating conditions, design parameter tolerances, and uncertainties
concerning the observed system performance (adopted from [Beyer and Sendhoff, 2007]).
This method has been recently adopted in building design optimization by [Ramallo-
González et al., 2015].
Large variations in performance,

1
PI2
not robust
PI2
2 2
4
3 3
5 5
Robust Pareto solutions
PI1 PI1
Actual (nominal) performance Variation in performance due to uncertainties
Figure 2.3 Robust Pareto solutions calculated considering both actual performance
(bubbles) and robustness (dotted circles) of performance indicators (PI) as the primary
criteria (adopted from [Gunawan and Azarm, 2005]). The left graph shows considered
design alternatives and the right graph presents robust designs among considered design
alternatives.
25
Similarly, Wang et al., [2014] proposed three different approaches to achieve robust
designs; by giving equal priority to actual performance and robustness, or by
prioritizing robustness as a primary criterion, or by prioritizing robustness as a
secondary criterion. In their study, robust designs are achieved considering actual
performance as the primary criterion and robustness as the secondary criterion.
However, to achieve realistic robust designs, both actual performance and
robustness should be treated equally, or a trade-off between these two can be
implemented, as presented in [Gunawan and Azarm, 2005]. In this method, robust
Pareto solutions are obtained, as shown in Figure 2.3, by considering both robustness
(dashed circles) and actual performance (bubbles) as primary criteria. The design
that has least variations in performance due to uncertainties is considered to be the
most robust. It can be observed from Figure 2.3 that design 2 is robust but not
optimal, whereas design 1 is optimal but not robust and design 4 is neither optimal
nor robust; thus, all of these designs are not robust Pareto solutions. This method
allows end users to trade off between actual performance and robustness.
With the advent of the robust design concept, the robustness approach has become
a powerful tool in many fields including physics [Lesne, 2008]; analytical chemistry
[Heyden and Massart, 1996]; biology [Kitano, 2004]; manufacturing engineering
[Mondal et al., 2014]; software engineering [Shahrokni and Feldt, 2013]; networks
[Larhlimi et al., 2011]; grids [Solé et al., 2008] and other fields [Walsh et al., 2013;
Lusby et al., 2017]. While robustness has been used in a wide variety of fields, its use
is most commonly found in structural and manufacturing engineering [Huang and
Du, 2007; Beyer and Sendhoff, 2007; Mondal et al., 2014].
2.3.2 Robustness in building performance context

In the building performance context, the use of robustness assessment has increased
in the past decade. However, there is also a growing need for robustness assessment
to address various issues such as:
⎯ To propose robust energy concepts that can withstand uncertainties in the

future; e.g. robust net-zero emission building concepts [Wiberg et al., 2014];
and robust passive cooling concepts [Voss et al., 2007].
⎯ To address the shortcomings of sustainability frameworks regarding

robustness of buildings arising from the fact that these frameworks do not
consider uncertainties in occupant behavior or climate change [Tuohy,
2009].
26
⎯ To reduce the performance gap between measured and predicted

performance as suggested by [de Wilde, 2014], and to ensure the desired
performance not only in the near future but over the building’s life-span
[Fawcett et al., 2012].
⎯ To ensure the long-term performance and future adaptability of buildings

[Hopfe et al., 2013]. For instance, long term performance and adaption of
buildings with respect to climate change [de Wilde and Coley, 2012].
⎯ To evaluate the robustness of innovative building elements such as adaptive

building facades with respect to uncertainties in occupant behavior and
weather conditions [Loonen et al., 2017].
⎯ To assess robustness of design decisions considering multiple performance

criteria under uncertainties in occupant behavior and weather conditions
and, thus, to enhance confidence in design decisions [Østergård et al., 2017].
Table 2.3 provides an overview of different studies that have addressed some of the
aforementioned issues in the building performance context. For instance, Hoes et
al., [2009] was the first study to implement the Taguchi method in the building
performance context. This research used relative standard deviation (ratio of mean
to standard deviation), which is similar to signal to noise ratio, as the robustness
indicator. Using this method, a design with robust energy and comfort performance
was identified among six designs with respect to uncertain user behavior. The
identified robust design resulted in very high indoor temperatures. Therefore, the
authors concluded that it is important to consider absolute performance in addition
to relative robustness. However, in most studies on this topic [Parys et al., 2012;
Gang, Wang, Yan, et al., 2015; Lee and Hensen, 2015; Karjalainen, 2016; Leyten and
Kurvers, 2006; Hopfe et al., 2013], the trade-off between robustness and actual
performance in the design selection process is not made explicit and robustness is
often prioritized (see Table 2.3). As discussed earlier, it is important to consider both
actual performance and robustness in the design decision-making process
[Gunawan and Azarm, 2005], otherwise this process may result in unrealistic
designs such as glass houses or concrete bunkers, as pointed out by [Van Gelder,
2014].
Chinazzo et al., [2015b] and Van Gelder et al., [2014] used both actual performance
and robustness in identifying robust designs. The energy saving index and
effectiveness indicators were used to assess actual performance in these two studies
27
respectively. In addition, these two studies used different robustness indicators for
robustness assessment. However, in the former study, predefined weights were used
for standard deviation, and for the data interquartile region, which were part of the
robustness indicator. These predefined weights will vary based on performance
requirements and the preferences of the decision makers, and as such require
knowledge from decision makers. In the latter study, the robustness indicator
considering percentile distributions was used. This indicator can overcome the
unreliability issues with data outliers as reported in [Hopfe and Hensen, 2011]. In
the latter study, even though actual performance was considered in the design
decision-making process, details of actual performance are not given. Therefore, in
this approach, it is difficult to distinguish between similar performing designs,
especially when a large design space is considered, as indicators range only from 0-
1. Furthermore, in this study, the scenarios were considered in a probabilistic
assessment, in which the probabilities of the occurrence of the considered scenarios
are unknown.
Typically, scenarios are used as an alternate approach when probabilities of

uncertainties are unknown [Polasky et al., 2011; Hopfe et al., 2013]. Even for such
cases, mean (and variance) across scenarios has been considered, as observed in the
robustness assessment of energy retrofits for future climate scenarios [Nik et al.,
2015] and as the robustness indicator used in design robustness optimization [Hoes
et al., 2011]. However, this approach is debatable as the probability of occurrence of
a scenario is usually unknown. So, in effect, taking the mean across all scenarios
does not capture the impact of each scenario since taking the mean effectively
flattens out the results. This issue can be avoided by using a non-probabilistic
approach, for instance as implemented in [Rysanek and Choudhary, 2013] and [Hoes,
2014] to identify robust designs using scenario analysis (Table 2.3). In the former
study, Wald, Hurwicz and Savage criteria are used to identify robust designs, and it
was found that using these methods made it is easy to identify robust designs from
within a large design space. However, as in many cases, some of the methods
implemented in this study also only consider robustness in the identification of
robust designs; trade-offs between robustness and actual performance are not
explicit. In the latter study, the best-case and the worst-case method is used to find
robust designs using scenario analysis.
28
Table 2.3 Review of different studies on robustness assessment / robust designs considering
uncertainties.
Robustness
Considered
Author Purpose indicator/ Remarks
uncertainties
measure
Robustness
Easy to identify robust
assessment of
Ranking based options. Actual
heating, Design
[Leyten and on robustness performance is not
ventilation and assumptions,
Kurvers, hypothesis included. Ranking system
air-conditioning maintenance
2006] using a penalty is hypothetical, and it is
(HVAC) and controls
system unsure if this method is
systems and
adaptable for other cases.
buildings
Implemented Taguchi
Relative method in the building
standard performance context.
Robust design deviation Designs robust to occupant
[Hoes et al., Occupant
with respect to (RSD). Ratio of behavior could be very
2009] behavior
user behavior mean to sensitive to other
standard uncertainties and also
deviation resulted in very high
indoor temperatures.
Works only for normal
Comparison of
distribution and becomes
robust
[Hopfe and Robustness Physical, design unreliable if data has
regression and
Hensen, analysis for and scenario outliers. No trade-off
ordinary least
2011] design support uncertainties between actual
square
performance and
regression
robustness.
Robustness is also a
primary criterion in
Optimization of
Scenario addition to actual
building RSD (ratio of
uncertainties in performance. The obtained
[Hoes et al., performance mean to
user behavior Pareto front includes
2011] using a standard
(usage robustness.
robustness deviation)
scenarios) Mean and variance are
indicator
used in non-probabilistic
robustness assessment.
Comparison of
Probability of
Occupant deterministic and
achieving the
behavior, uncertain conditions
same output
setpoints of Only robustness is
Robust passive with
[Parys et al., natural considered, and actual
cooling uncertainties as
2012] ventilation, Air performance is not
concepts that of average
flow rates and, included in the design
uncertainty
weather data decision making; this
inputs (Monte
etc. might lead to oversized
Carlo analysis)
systems.
This approach considers
uncertainties and decision
makers’ attitude towards
Difference
Multi-criteria risk in the decision-
between the
[Hopfe et decision Physical making process.
best-case and
al., 2013] making under parameters Only the best and worst
worst-case
uncertainty performances are included
performance
for robustness assessment
and actual performance is
ignored.
29
This study showed that

non-probabilistic
Non- assessment is suitable
probabilistic when using scenarios.
[Rysanek
assessment to Technical- Wald, Hurwicz Easy to identify robust
and
find robust economic and Savage designs and enhances
Choudhary,
optimal uncertainties methods design decision-making
2013]
retrofitting process
designs
Robustness is prioritized
in design selection.
Design optimized
considering robustness.
The Pareto front with
Best-case and robust optimal solutions is
Robust designs
Occupant worst-case obtained. Trade-off
with respect to
[Hoes 2014] behavior method between robustness and
occupant
scenarios RSD for optimal actual performance is
behavior
control strategy present. The visualization
of results is enhanced by
embedding robustness as
an extra dimension.
Both actual performance
and robustness are
considered. Percentile
distributions are used to
remove outliers.
Probabilistic
Uncertainties in
methodology to Effectiveness Difficult to distinguish
[Van Gelder occupant
design robust and robustness between similar
et al., 2014] behavior and
low-energy indicators performing designs based
energy prices
buildings on these two indicators
whose values range from
0-1, making the decision-
making process difficult.
Scenarios are used in
probabilistic assessment.
Physical
parameters, This method allows system
HVAC sizing design sizing based on trade-off
[Y. Sun et
under parameters, Sensitivity index with risk that can be
al., 2014]
uncertainties weather and accepted by the decision
occupant maker.
behavior
Robustness of Modified robustness
Relative
building design indicator to overcome
standard
[Buso et al., with respect to Occupant issue reported in [Hoes et
deviation
2015] variations in behavior al., 2009]. However, trade-
referred to the
occupant off is not present in this
basic model
behavior study.
Similar to previous studies,
robustness is prioritized in
design selection and
Robustness
[Chinazzo therefore trade-off between
assessment of The spread of
et al., Weather robustness and actual
energy box plot
2015a] performance is not
refurbishments
present. A design that is
not optimal may have the
least spread.
30
Robustness
index based on Comparison between
variance, data actual performance and
Proposed
[Chinazzo distribution in robustness. Predetermined
robustness
et al., Weather box plots and weight is assigned for
assessment
2015b] comparison interquartile and standard
methodology
with base case; deviation in the robustness
Energy saving index.
index.
Self-induced
uncertainties in
cooling load by Easy to identify robust
Robust optimal
multiplying Maximum designs. Only robustness
[Gang, design of
with a factor regret using is considered in design
Wang, Yan, cooling systems
and minimax regret selection. Tradeoff is
et al., 2015] concerning
uncertainties in method ignored in design
uncertainties
resistance of selection.
chilled water
pipes
Monte Carlo
method to
assess the Robust optimal design is
impact of compared with that of
Robust optimal
uncertainties conventional design
design of
and Markov approach and reliability-
[Gang, cooling systems
Uncertainties in method to based design approach.
Wang, considering
cooling load quantify Trade-off is present and
Xiao, et al., cooling load
calculations reliability are the design with the lowest
2015] uncertainty and
used. cost is the most preferred.
equipment
Availability risk It is not clear how
reliability
cost is also used robustness is quantified in
as a criterion in this study.
design
selection.
This method is helpful in
identifying robust designs
Risk indicator
based on risk if few
that quantifies
Risk indicator scenarios are considered.
[Lee and surge in energy
to quantify Operational This could become
Hensen, consumption
robust building scenarios complex if a large number
2015] compared to
design of scenarios are
reference
considered. In addition,
scenario
actual performance is not
considered.
Impact of input
Input Robust designs are
uncertainties of
uncertainties in identified using multi-way
renewable
[Lu et al., building Sensitivity sensitivity analysis. Actual
energy systems
2015] electrical load, analysis performance and
on performance
wind velocity, performance robustness
robustness of
cooling load etc. are compared.
NZEB
Mean across scenarios is
considered in this study
Mean and
Robustness and this approach is
[Nik et al., variance of
assessment of Climate change questionable as the
2015] relative
energy retrofits probability of occurrence
deviation
of each scenario is usually
unknown.
31
Risk cost that

includes mean Risk is quantified in terms
Robust optimal Physical, design repair time and of cost. Only mean values
[Gang et al., design of and occupant mean failure are considered. The
2016] district cooling behavior time of selected design is robust to
systems uncertainties components cost but could be sensitive
under to comfort.
uncertainties
Robust design options save
up to 79% of energy.
Difference Robust design is done by
Comparison of
between improving design options
[Karjalaine robust designs Occupant
maximum and that are sensitive to
n, 2016] with an behavior
minimum occupant behavior. Trade-
ordinary design
performance off between actual
performance and
robustness is not explicit.
Robust cost-optimal
Resilience solutions are used to
Relative
assessment of assess their resilience for
[Ascione et difference
robust retrofit Climate change climate change. Smart
al., 2017] compared to
energy exhaustive search is used
base line.
measures to identify robust cost-
optimal solutions.
Comparison of different
Comparative
Households, Spread, robustness assessment
[Kotireddy assessment of
occupant deviation and methods using scenario
et al., different
behavior, maximum analysis. Trade-off between
2017b] robustness
climate change regret predicted performance and
indicators
robustness is present.
Households,
Robust optimal designs
occupant
[Kotireddy Methodology to considering multi-criteria
behavior, Maximum
et al., identify robust assessment and multi-
climate change. regret using
2017a; low-energy and criteria decision making.
In addition, net- minimax regret
Kotireddy NZEB building Trade-off between
metering method
et al., 2018] designs predicted performance and
considered in
robustness is explicit.
the first study.
2.3.3 Choosing appropriate robustness assessment methods
The choice of a robustness assessment method depends on the purpose of the study
and on the decision makers’ attitude towards the risk acceptance in decision-making
process [Hopfe et al., 2013; Polasky et al., 2011]. In practice, various decision makers
with different attitudes towards risk acceptance are involved in a project, and as such
it is important to quantify robustness/risk for each scenario to make informed design
decisions. Therefore, it is important to select appropriate methods that considers
different risk-taking approaches of decision makers. It is evident from the literature
review (see Table 2.3) that robustness assessment using scenario analysis is rarely
addressed, and often mean and variance are used for robustness assessment across
scenarios.
32
To fill this research gap, robustness assessment methods that are adopted for
scenario analysis are reviewed from different fields such as operations research [Aissi
et al., 2009; Averbakh, 2000; Ehrgott et al., 2014; Xidonas et al., 2017],
manufacturing [Chien and Zheng, 2012], economics [Sautua, 2017] and ecology
[Polasky et al., 2011].
It is found that the max-min method, a conservative approach, and minimax regret
method, a less conservative approach [Polasky et al., 2011; Aissi et al., 2009] are the
most commonly used methods for robustness assessment using scenario analysis
[Averbakh, 2000]. Similarly, following this logic, the best-case and worst-case
method, the more conservative approach [Walsh et al., 2013] based on the pessimistic
approach by [Hurwicz, 1952], has previously been implemented in the building
performance context [Hopfe et al., 2013; Hoes, 2014]. These three methods are
selected for robustness assessment, which take scenarios into account for robustness
assessment and also represent different attitudes towards risk acceptance by decision
makers in the decision-making process. These methods are discussed in detail in the
next section.
In addition, most of the reported work considered only robustness in the design
selection process, which resulted in designs with unacceptable ‘actual’ performance.
Therefore, in the present research, actual performance and robustness are
considered as primary criteria, and trade-off between actual performance and
robustness is implemented based on the decision maker’s preferences in the
selection process of robust design.
2.4 Selected robustness assessment methods
The max-min method, the best-case and worst-case method, and the minimax regret
method are used for robustness assessment in the present context. These methods
are summarized in Table 2.4 and illustrated using an example here. In this example,
the objective is to minimize the performance indicator, and robustness is calculated
accordingly. As such, a ’low value of a performance indicator’ is desirable and a ‘high
value of a performance indicator’ is undesirable.
Adoption of these methods in the building performance context is discussed in detail

in the next chapter.
33
2.4.1 The max-min method
This method aims at finding robust solutions that have the least variations even in
extreme scenarios [Wald, 1945]. In this method, the performance spread is used as
the robustness indicator of a design [Kotireddy et al., 2015], and is defined as the
difference between maximum performance and minimum performance across all
considered scenarios. This is illustrated using an example, as shown in Figure 2.4,
by comparing the performance of three designs across three scenarios.
For design A, scenario 1 results in maximum performance and scenario 3 yields

minimum performance, resulting in a spread of 2. Similarly, design B and design C
have spreads of 2 and 3. Comparing these spreads, design A and design B are equally
robust as they have the least spread. But, design A results in very high actual
performance and as such is not the preferred robust design. Therefore, design B is
the most preferred robust design.
It is worth noting that in this method, robustness of a design is calculated without

any inter-comparison between designs, and scenarios causing maximum and
minimum performance for a design are considered for robustness assessment.
However, to find the maximum and minimum performance for a design across the
considered scenarios, the performance assessment of a design across all considered
scenarios may be essential.
10
Performance indicator (-)
9 2
8
7
6
5
4
3 2 3
2
1
0
Design A Design B Design C
Scenario 1 Scenario 2 Scenario 3
Figure 2.4 Calculation of spread of a performance indicator for three imaginary designs for
the considered scenarios.
34
Selected robustness assessment methods
2.4.2 The best-case and worst-case method
This method aims at finding robust solutions that have the best performance even
in the worst-case scenario [Walsh et al., 2013], which is similar to the pessimistic
approach in the Hurwicz method [Hurwicz, 1952]. In this method, performance
deviation between the worst-case performance of a design and the best-case
performance of all designs across all scenarios is used as a measure of robustness. A
similar method has been applied in [Hopfe et al., 2013; Hoes, 2014] and is improved
here by considering the performance of all designs across all scenarios to find the
best-case performance, unlike the predefined best-case performance as in [Hoes,
2014]. For instance, to find the best-case performance, the performance of three
designs across three scenarios is compared in the example shown in Figure 2.5; here,
the best performance is achieved by design B for scenario 2. The worst-case
performance for design A is caused by scenario 1 and results in a deviation of 8.
Similarly, for design C, the deviation is 3, which is caused by scenario 2. Comparing
these deviations, it is clear that design B is the most robust as it has the least
deviation. Furthermore, design B is also the preferred robust design as it has better
actual performance.
In contrast to the max-min method, this method considers all scenarios for
performance robustness assessment. In addition, inter-comparison of designs across
all scenarios is made to find the best performing design.
10
9
8
7
6 8
5
4
3 2 3
2
1
0
Figure 2.5 Calculation of deviation of a performance indicator for three imaginary designs
for the considered scenarios.
35
2.4.3 The minimax regret method

The minimax regret method aims at finding a robust solution that performs close to
optimal performance for all scenarios [Averbakh, 2000]. This method [Savage, 1951]
is a combination of the minimax [Wald, 1945] and regret methods. This method has
been widely used for robustness assessment in various fields [Chien and Zheng,
2012; Ehrgott et al., 2014; Averbakh, 2000] and has recently been used in the
building performance context [Gang, Wang, Yan, et al., 2015; Rysanek and
Choudhary, 2013; Kotireddy et al., 2017b; Kotireddy et al., 2018].
In this method, for a given scenario, regret is expressed as the performance

difference between a design and the best performing design in that scenario. The
maximum regret across all scenarios is used as the measure of robustness. For
instance, for scenario 1, design C is the best performing design and thus results in
zero regrets (see Figure 2.6). For this scenario, design A and design B have regrets
of 8 and 2 respectively. Similarly, for scenario 2, design B is the best performing and
thus has zero regrets. Design A and design C have regrets of 7 and 1 respectively. For
scenario 3, design B has zero regrets and design A and design C have regrets of 4
and 1 respectively. Comparing the regrets of these designs across the considered
scenarios, it can be observed that design C has the lowest maximum regret and is
thus the most robust design. It is worth noting that design B performance is optimal
for most of the scenarios, and that design B is not far from the most robust design.
10
9
8
7
6 8 7 4
5
4
0 1
3 2 0 0 1
2
1
0
Figure 2.6 Calculation of regrets of a performance indicator for three imaginary designs for
36
Selected robustness assessment methods
In contrast to previously mentioned methods, the performance of the robust design

in this method is close to optimal performance for every scenario. Similar to the best-
case and worst-case method, inter-comparison of designs is considered in the
robustness assessment.
Table 2.4 Comparison of selected robustness assessment methods in the present context.
Best-case and
Minimax regret
Parameter Max-min method worst-case
method
method
What is the
Performance Performance Maximum
robustness
spread deviation performance regret
indicator?
Difference
Difference between
Difference between the best
the performance of
between performance of
a design and the
maximum and the entire design
What is the best performing
minimum space and the
calculation design for that
performance of a worst
method? scenario and the
design across performance of a
maximum
considered design across
difference across all
scenarios considered
scenarios
scenarios
Which
scenarios are
Extreme All All
used for
calculation?
Minimum or Minimum or Minimum or
What is the
ideally zero ideally zero ideally zero
most robust
performance performance maximum
design?
spread deviation performance regret
Risk can be
Risk is high; accepted as trade-
Risk is high;
Design should off; Design should
Design should
When to use? deliver the best work well (close to
work even in
performance in optimal
extreme scenarios
all scenarios performance) in all
scenarios
e.g. Designing e.g. Designing
e.g. Designing
Where can it be HVAC system HVAC system for
HVAC system for
used? for hospitals, residential
data centers
clean rooms buildings
37
2.5 Robust design selection
Design decision making is the process of selecting the best design from the available
alternative solutions, which are ranked based on evaluation criteria [Dey et al., 2016].
Finding the “best/optimal design” may be not be a feasible option in all situations in
practice as the decision-making process involves various decision makers with
multiple performance requirements. This complexity is even higher when
robustness (uncertainties) is considered in the decision-making process. Therefore,
different decision-making methods are reviewed below to select appropriate methods
that enhance the design decision-making process.
2.5.1 Multi-criteria decision making

Multi-criteria assessment is carried out to obtain a Pareto front that has a set of Pareto
optimal solutions [Medineckiene et al., 2015; Y. Sun, Huang, et al., 2015]. This Pareto
front enables decision makers to trade-off between alternative design solutions based
on their preferred performance indicators. This assessment is commonly
implemented for two indicators but rarely for three or more indicators. Thus, it could
be a challenging task for multiple performance indicators under uncertainty
[Østergård et al., 2017; Mela et al., 2012]. To address this issue, a multi-criteria
assessment considering multiple objectives representing a typical case of a building
design project is employed in this research. This is elaborated further in the next
chapter.
Additionally, to find one design among a set of alternatives on the Pareto front, a
multi-criteria decision making (MCDM) method is typically used [Kumar et al., 2017;
Mulliner et al., 2016; Medineckiene et al., 2015; Mela et al., 2012]. The MCDM
method supports the decision maker in choosing one preferred design from among
a set of available alternative designs. It ranks a design with respect to other designs
by meeting predefined criteria. There are several MCDM methods available in the
literature (see review studies [Polatidis et al., 2006; Kumar et al., 2017]) such as
outranking methods (e.g. TOPSIS, PROMETHE, ELECTRE) and utility based
methods (e.g. weighted sum method, weighted product method). However, there are
limited MCDM methods available for decision making under uncertainty [Hopfe et
al., 2013; Polasky et al., 2011; Rysanek and Choudhary, 2013]. It was found that the
Analytic Hierarchy Process (AHP), Wald’s criterion, the Hurwicz criterion and the
Savage criterion are used in the building performance context for decision making
under uncertainty [Rysanek and Choudhary, 2013; Hopfe et al., 2013].
38
Robust design selection
⎯ The AHP method [Saaty, 1987] is one of the most widely used MCDM
methods [Polatidis et al., 2006]. AHP allows decision makers to rank
performance indicators with respect to decision criteria and design
alternatives through pairwise comparison.
⎯ The Wald’s maximin criterion is a pessimistic approach that

maximizes/minimizes the worst-case scenario outcome. Using this method,
the design alternatives are ranked based on the worst-case performance and
the design with the minimum worst-case performance is the most preferred.
⎯ The Hurwicz criterion [Hurwicz, 1952] is commonly used for decision

making under uncertainty [Rysanek and Choudhary, 2013; Polasky et al.,
2011; Pavzek and Rozman, 2009]. A Hurwicz weight (0-1, called a
coefficient of pessimism), is used based on a decision maker’s approach
towards risk i.e. an optimistic or pessimistic approach. Maximax criterion
(weight=1) is an optimistic approach that maximizes/minimizes the
outcome for the best-case scenario and maximin criterion (weight=0) is a
pessimistic approach that maximizes/minimizes the outcome for the worst-
case scenario, which is similar to Wald’s maximin criterion. Using the
Hurwicz method, a decision maker can take the weight of 1 and assign a
certain fraction of this weight to optimism and assign the remaining weight
to pessimism.
⎯ The Savage minimax regret criterion [Savage, 1951] is also commonly used
for making decisions under uncertainty [Rysanek and Choudhary, 2013;
Polasky et al., 2011]. This method is a pessimistic approach that focuses on
minimizing the maximum opportunity loss (regret).
In the AHP method, a pairwise comparison is conducted, and the final rank of
designs is calculated on the basis of the average weights of performance indicators
and a design’s rank for each performance indicator. Hence, decision makers should
know the relative importance of all performance indicators. In practical situations,
various decision makers with multiple performance criteria are involved, thus
making the assigning of weights more complicated in the AHP method [Kumar et
al., 2017]. In the Wald criterion, the design is ranked based on the worst outcome
and is thus a pessimistic approach. The Hurwicz criterion represents a compromise
between the optimistic and pessimistic approach (equal weights) and also allows a
decision maker to be cautious (risk free) or adventurous (optimistic/risk based) in
the design decision-making process. The Savage criterion allows decision makers to
39
choose a design that has the least risk among alternatives that are ranked based on
the regret.
The current research involves decision making under uncertainty considering

decision makers’ attitudes towards risk. Therefore, the Hurwicz criterion is the
preferred MCDM method in the present context as it allows decision makers to
choose the most robust design based on their attitude towards risk in the decision-
making process. Furthermore, the Hurwicz criterion is often used for realistic
decision problems [Gaspars-Wieloch, 2014; Pavzek and Rozman, 2009].
2.5.2 Visualisation
Visualisation methods play an important role in aiding decision makers by providing
a meaningful platform to analyse the outcome of a design space in order to identify
and select preferred designs [Blasco et al., 2008]. Generally, it is difficult to visualize
two or more performance indicators and their corresponding robustness because
doing so results in a multi-dimensional Pareto front. Furthermore, in practice it is
often the case that a large design space is considered in the design decision-making
process to find preferred designs. Therefore, to facilitate decision makers in
identifying their preferred designs, different visualization methods are reviewed to
select appropriate methods that can enhance the decision-making process.
The most commonly used multi-dimensional visualisation methods are parallel

coordinate plots and scatter plots [Blasco et al., 2008; Østergård et al., 2017]. An
example of a parallel coordinates plots used to represent a multi-dimensional design
space and corresponding outputs is shown in Figure 2.7. In a parallel coordinates
plot, the multi-dimensional parameters are transformed into a two-dimensional plot,
making the plot very compact. However, for a large design space with multiple
performance indicators, the clarity diminishes and becomes difficult to visualize.
On the other hand, scatter plots can represent multi-dimensional parameters in an

array of scatter plots [Blasco et al., 2008]. In their study, they reduced the complexity
in visualization of multi-dimensional Pareto fronts using so-called level diagrams. In
these level diagrams, different outcomes are normalized and presented on three x-
axes and are compared with the same y-axis. However, this normalization may not
be preferred in all cases. This issue can be overcome by representing the third
dimension as an additional parameter in two-axes (x-y) by using scatter plots such as
bubble size or colour [Hoes, 2014; Kotireddy et al., 2015]. An example of this
40
Robust design selection
representation is shown in Figure 2.8, and in this visualization method implemented

by [Mavrotas et al., 2015], robustness is represented as an extra dimension using
bubble size in the scatter plot.
Box-plots are typically used to represent a range of data [Parys et al., 2012; de Wilde
and Tian, 2009; K. Sun and Hong, 2017; Burhenne et al., 2013] and histograms/bar
plots are often used to represent the variation of output parameters with input
parameters [Eisenhower, Neill, Fonoberov, et al., 2011; Mavrogianni et al., 2014;
Berger et al., 2014; Hopfe and Hensen, 2011; Østergård et al., 2017].
In this research, scatter plots are used to visualise multi-dimensional Pareto fronts
with robustness being included as bubble size in the scatter plot. Similarly, box-plots
are used to visualise the range of performance variation and histograms are used to
represent range of robustness indicators, among other purposes.
Figure 2.7 An example of parallel coordinate plot with a multi-dimensional design space
and performance indicators [Østergård et al., 2017].
Figure 2.8 Visualization of robustness as an extra dimension using scatter plots without
robustness (left figure) and with robustness indicated by bubble size (right figure) (adopted
from [Mavrotas et al., 2015]).
41
2.6 Concluding remarks

It can be concluded from the literature review that there is a methodological gap that
can be filled by a holistic approach that integrates the following:
i. Uncertainty sources in performance assessment.

ii. Appropriate methods to quantify robustness under these uncertainties.
iii. Suitable methods to enhance the design decision-making process by aiding
various decision makers in the selection of robust designs.
This gap leads to the research objectives as discussed in Chapter 1.
As a first step to meet these objectives, a literature review was conducted to identify
different uncertainty sources that can impact building performance over a building’s
life-span and corresponding methods to quantify the impact of these uncertainties
were analysed. The merits of using scenarios instead of probabilistic approaches
were highlighted, indicating that scenarios can be used as formulated alternatives in
cases when probabilities of uncertainties are unknown. In addition, different
robustness assessment methods were reviewed from the general literature and the
building performance context to identify appropriate methods.
The following conclusions can be drawn from this literature review
⎯ The integration of scenarios in building performance predictions can provide

a better understanding of the impact of uncertainties and also facilitate
decision making during the design selection process with the goal of
choosing a design that is robust to a variety of possible future situations. In
literature, it is recommended to use scenario analysis for long-term
performance assessment of buildings, especially performance robustness.
⎯ The review showed that mean and variance, the widely used robustness
indicators based on the Taguchi method, are of limited use in robustness
assessment using scenario analysis. The likelihood of the occurrence of any
scenario is usually unknown and taking the mean across scenarios nullifies
the concept of formulating scenarios as alternatives since it flattens out the
results. In addition, it was found that robustness can be used for performance
assessment under uncertainties if the nature and range of uncertainties and
performance indicators are precisely defined for assessment, and it is
necessary to include both actual performance and performance robustness
in the selection process of robust designs.
42
Concluding remarks
⎯ The maximum and minimum performance are considered for robustness

assessment in the max-min method. The performance of all designs across
all scenarios is compared to calculate performance robustness in the best-
case and worst-case method. The maximum difference, across all scenarios,
between the performance of a design and the best performance of the
corresponding scenario is compared to calculate the robustness using the
minimax regret method.
⎯ The max-min method can be used when a design should deliver the desired
performance even in extreme scenarios, whereas the minimax regret method
can be used when a design has to deliver optimal or close to optimal
performance for each scenario.
⎯ The best-case and worst-case method is a conservative approach as it yields a

robust design that has the best possible performance even in the extreme
case (worst-case scenario). Conversely, the minimax regret method is a less
conservative approach as it yields a robust design that performs as closely as
possible to the optimal performance for every scenario.
⎯ The max-min method and the best-case and worst-case method can be used
when the cost/risk associated with the failure of design is very high. The
minimax regret method can be used when a decision maker can accept a
certain range of performance variation; for instance, a homeowner can accept
designs with certain overheating hours as a trade-off with global costs and
required additional investment cost.
Based on relevance and applicability, the max-min method, the best-case and worst-
case method and the minimax regret method are used in this research. Similarly, the
Hurwicz criterion based MCDM method is used to find a robust design from a large
design space. The development of the computational performance robustness
assessment (CPRA) approach that integrates the aforementioned methods is
described in the next chapter. The adoption of the aforementioned robustness
assessment methods in the building performance context is also discussed in detail
in the next chapter.
43
3. Methodology to develop and
test the CPRA approach
A methodology to develop and test the computational performance robustness assessment
(CPRA) approach is presented in this chapter. The methodology is iteratively developed
through suitability and usability assessment with end users.
The final CPRA approach, the outcome of an iterative process, is described in detail in the
second part of this chapter. This approach comprises multi-criteria performance
assessment and multi-criteria decision making considering performance robustness.
Different robustness assessment methods are presented to aid decision makers to identify
robust designs. Robust design selection using trade-off approach and multi-criteria
decision-making methods is discussed. Various visualization methods to enhance the
decision-making process are presented. This chapter ends by describing practical uses of
the developed CPRA approach before offering conclusions.
3.1 Overview of methodology
Figure 3.1 depicts the methodology used to develop and test the computational
performance robustness assessment (CPRA) approach. The first step of this
methodology is to identify end users of the CPRA approach. The next step is to
develop a computational approach for performance robustness assessment. The
developed CPRA approach is assessed with the help of end users for its suitability
and usability in practice. The developed approach is tested using case studies and the
results are presented to end users through mock-up presentations. The feedback
from the end users is implemented in the iterative process to improve the CPRA in
order to enhance its suitability and usability in practice.
45
Chapter 3. Methodology to develop and test the CPRA approach
Start
Identify end users of the CPRA

approach
Develop multi-criteria performance

Develop the computational
assessment and multi-criteria decision
performance robustness assessment
making approach considering
(CPRA) approach
performance and robustness indicators
Apply the CPRA approach in a case Test the developed CPRA using case
study studies to identify robust designs
Improve the
CPRA approach
Present the CPRA approach and case
Evaluate the CPRA approach with study results to the end users to evaluate
end users suitability and usability of the CPRA
approach in practice
Receive the feedback and improve the

Feedback from end users CPRA approach to enhance its suitability
and usability in practice
End
Figure 3.1 A methodology to develop and test the computational performance robustness
assessment (CPRA) approach.
3.1.1 Suitability and usability assessment of the CPRA approach

Delivering mock-up presentations to focus groups is a promising method for
usability testing [Hopfe, 2009], among different methods [Preston, 2015]. Therefore,
a user group with potential end users of the CPRA approach was formed. The user
group comprised of representatives from some of the leading building consultancy
and services companies and a building performance simulation tool development
company in the Netherlands. The CPRA approach was presented at various stages to
the user group members in user group meetings. A total of four meetings were
conducted for this purpose, at approximately six-month to one-year intervals between
the meetings. The feedback from the user group was incorporated to improve the
46
Overview of methodology
CPRA approach. The discussion of suitability and usability assessment is presented

in the following subsections.
⎯ Applying the CPRA approach in a case study

The developed CPRA approach was iteratively tested and demonstrated using a case
study. Each iteration represents a different version of the CPRA approach.
⎯ Evaluation of the CPRA approach with end users

Different versions of CPRA were evaluated with the user group through mock-up
presentations. The purpose of these presentations was to evaluate if the proposed
CPRA approach and methods implemented in the approach can be used in practice.
Therefore, different aspects of the CPRA approach were presented in different user
group meetings.
⎯ Feedback from users group meeting

Feedback was received from different user group meetings on various aspects of the
CPRA approach, such as the considered design space and performance indicators,
implemented simulation and visualization methods and the applicability of the
CPRA approach. The response to the main feedback from all user group meetings is
summarized in Appendix F. It is worth noting that this feedback was received for
different iterations of the CPRA approach and the feedback was incorporated in the
CPRA approach after every user group meeting.
3.1.2 Improvement of CPRA approach through iterative process

The feedback of each meeting was incorporated in an iterative process, as shown in
Figure 3.2, and the updated CPRA approach was presented in the subsequent
meeting. After several iterations, the final CPRA approach was developed. The
results of the suitability and usability assessment including the feedback from user
group meetings and improvement of the CPRA approach is discussed in detail in
Chapter 7. The developed and tested CPRA approach based on this iterative feedback
process is described in the next section.
47
Suitability and usability

CPRA0
Case studies Robust designs assessment with users
(Initial)
group
CPRA1
CPRA2
CPRAn
CPRAn+1
(Final)
n+1
n
2
1
Feedback on previous version leads to next version (e.g. 0→ 1)
Figure 3.2 Improvement of the CPRA approach through an iterative process based on
feedback from user group meeting to reach the final version.
3.2 The developed and tested CPRA approach

The final CPRA approach is shown in Figure 3.3. Each step is described below and
in further detail in the following subsections.
Step 1: Identify decision makers, and based on decision maker’s preferences define
the following:
1a. Building design space
1b. Future scenarios
1c. Performance and robustness indicators
Step 2: Set up a building performance simulation model and simulate the

performance of the design space for future scenarios with defined
performance indicators.
Step 3: Multi-criteria assessment: carry out performance robustness assessment

considering multiple performance and robustness indicators.
Step 4: Multi-criteria decision making: provide design decision support to the

decision makers in identifying the preferred robust designs by prioritizing
the performance indicators based on their preferences. Carry out sensitivity
analysis to identify the most influential scenarios. Present these results via
different visualization methods to decision makers to enhance the design
48
The developed and tested CPRA approach
Start
Identify decision makers and their

preferences
Who? What? Why?
1b. Define the future 1c. Define performance and

1a. Define the design space
scenarios robustness indicators
Set up the computational building

performance simulation model
Simulate the performance of the design

space for all future scenarios
Multiple
Robustness
Carry out multi-criteria assessment performance
indicators
indicators
Present the results of multi-criteria

assessment using different visualization
methods to decision makers
Trade-off
solutions using Multi-criteria
Design decision support to decision
Pareto front and decision making
makers to select robust designs
using robust method
design options
Carry out sensitivity analysis and identify

the most influential scenarios on
preferred performance and robustness
indicators
End
Figure 3.3 The final computational performance robustness assessment (CPRA) approach
implemented in this study.
49
3.2.1 Identify decision makers and their preferences

The first step is to identify the decision makers and their preferences. For instance,
policymakers can use performance robustness as a requirement in future building
regulations to safeguard intended policy targets. They can also define policies
considering performance robustness to support adaptations of current buildings to
improve their performance in order to extend their life-span. Similarly, performance
robustness is a relevant concern for homeowners, since they wish to ensure their
preferred building performance over the building’s life-span.
Energy performance contractors can also benefit from performance robustness

assessment by reducing the performance gap between predicted and actual
operation. Similarly, by considering performance robustness, building designers and
consultants can design and deliver more robust buildings, thus improving the
satisfaction of their customers.
3.2.1.1 Define the building design space

The design space needs to be defined based on the requirements of decision makers
and on current and future building regulations such that the preferred design of a
decision maker will also meet the criteria of building codes and regulations [RVO,
2016b; RVO, 2015b; BPIE, 2010]. In practice, it is generally the case that several
design configurations lead to similar optimal performance under deterministic
conditions, but these configurations have significantly different magnitudes of
performance variation for future scenarios.
For instance, all designs shown in Figure 3.4 could be net-zer0 energy building
(NZEB) solutions under deterministic conditions. For example, a NZEB solution (D1)
can be achieved by combining very high insulation levels (P1) and a small renewable
energy generation and storage system (RES1). In contrast, another NZEB solution
(Dn) can be realised by combining relatively lower insulation levels (Pn) and a larger
renewable energy generation and storage system (RESn).
50
Renewable energy generation, kWhe/m2a Dn
RESn
D4
RES4
D3
RES3
D2
RES2
D1
RES1
P1 P2 P3 P4 Pn
Total energy demand, kWhe/m2a
Figure 3.4 Designs with different insulation levels (energy demand) and corresponding
onsite renewable energy generation systems (energy generation) to reach NZEB.
However, when uncertainties arise, these designs can have different magnitudes of
deviation in performance during operation compared to the predicted performance
in the design phase. Hence, the preferred design is based on predicted performance
and performance robustness, and as discussed in Chapter 1, a balance between
energy demand and energy generation may be required to achieve this preferred
NZEB design [Kotireddy et al., 2015].
51
3.2.1.2 Define the scenarios

Scenarios need to be defined that consider all uncertain and influential parameters
that can cause variations in the building’s performance over its life-span. Figure 3.5
provides an overview of scenarios that could be considered, e.g. different household
sizes (referred as occupant scenarios) and their corresponding behavior (referred as
usage scenarios) over the building’s life-span, external factors such as climate change
(referred as climate scenarios) and policy changes such as feed-in-tariff prices
(referred as policy scenarios). The combination of all these scenarios must be used
in the performance robustness assessment as the likelihood of the occurrence of any
of these scenarios is unknown. However, considering all scenario combinations
results in very high computational cost, and thus a sampling strategy is desirable to
find the smallest scenario sample that represents all scenario combinations, which
is discussed in detail in the next chapter.
Occupant Usage Policy Climate

scenarios scenarios scenarios scenarios
• One person • Occupancy patterns • Net-metering as • Reference climate
household • Temperature set usual • Change in air
• Two person points • Cap on exported temperature
household • Domestic hot water energy • Change in air
• Three person • Appliance and • Reduced price for circulation patterns
household lighting use feed-in-tariff • .....
• Four person • Internal heat gains • ......
household • .......
• .......
Scenarios
Figure 3.5 Scenarios formulated based on uncertainties in (future) household size,

occupant behavior, policy change and climate change.
3.2.1.3 Define performance indicators

The performance indicators need to be defined in accordance with the preferences
of the decision makers. For instance, a policymaker prioritizes low or no CO2
emissions associated with a building design, but not at the expense of high
investment costs. In contrast, a homeowner prioritizes designs with comfortable
52
indoor environment at low costs. Hence, CO2 emissions and investment costs are
the preferred performance indicators for the policymaker, while thermal comfort and
costs such as investment and operating costs are the preferred performance
indicators for the homeowner. Detailed description of these performance indicators
is presented in Chapter 5 (Section 5.5).
As noted earlier, in addition to predicted performance, performance robustness is

also a primary criterion in the decision-making process in this approach. To this end,
it is necessary to define robustness indicators calculated using selected robustness
assessment methods (Chapter 2). Further description of this area is included in the
next section (Section 3.2.3).
3.2.2 Computational performance prediction using building performance

simulations
The performance of the design space is predicted for future scenarios by using a
building performance simulation model. This performance assessment of the entire
design space is computationally very expensive. In the next chapter, different
simulation methods are discussed in detail to determine how best to reduce this
computational time. The case study models for building and energy system
simulation are described in Chapter 5 (Section 5.6).
3.2.3 Multi-criteria performance assessment using multiple performance and

robustness indicators
In practice, since each decision maker has multiple performance requirements, they
will generally be prepared to accept a trade-off solution [J. J. Wang et al., 2009; Deb
et al., 2014; Hamdy et al., 2013]. The decision maker will prioritize different
performance indicators in the decision-making process as noted earlier in Section
3.2.1.3. All these performance indicators are compared against additional investment
cost (design), which enables each decision maker to select a cost-optimal robust
design or to accept a trade-off with respect to the other performance indicators and
to the robustness of these performance indicators. As discussed earlier in Chapter 2,
three methods, namely the max-min method, the best-case and worst-case method,
and the minimax regret method are selected to assess the performance robustness
of building designs. Each method uses different means (indicator) to evaluate
robustness, as described in the following subsections.
53
3.2.3.1 The max-min method

In this method, the spread of a performance indicator, the difference between
maximum performance and minimum performance across all scenarios, is used as
a robustness indicator of a design. Using this method, the following steps are
implemented in the present research to select the most robust design of a design
space across the considered scenarios.
1. Assess the performance of designs (dm) for all scenarios (Sn) using a
performance indicator (PI).
2. Find the maximum and minimum performance of a design across all

scenarios, as shown in Table 3.1 .
3. Calculate the performance spread of a design across all scenarios. The

performance spread is the performance difference between the maximum
and minimum performance, as shown in Table 3.1.
4. Repeat steps 1-3 for other performance indicators.
In summary, the performance spread is used as a measure of robustness, and the

design that has the smallest performance spread is the most robust solution. Ideally,
the design with zero performance spread is the most robust solution of a design
space.
Table 3.1 Performance robustness (spread) calculations using the max-min method.
Scenarios Maximum Minimum Performance
performance performance spread
Designs S1 S2 … Sn (PImax) (PImin) (PImax-PImin)
max(PI11, min(PI11,
d1 PI11 PI12 … PI1n PImax1-PImin1
PI12,…PI1n) PI12,…PI1n)
max(PI21, min(PI21,
d2 PI21 PI22 … PI2n PImax2-PImin2
PI22,…PI2n) PI12,…PI1n)
… … … … … … … …
max(PIm1, min(PIm1,
dm PIm1 PIm2 … PImn PImaxm-PIminm
PIm2,…PImn) PIm2,…PImn)
The most robust design min(PImax-PImin)
54
3.2.3.2 The best-case and worst-case method

In this method, performance deviation between the worst-case performance of a
design and the best-case performance of all designs across all scenarios is used as a
measure of robustness. The following steps are implemented in the present research,
based on this method, to select the most robust design within a design space across
2. Find the minimum performance of a design across all scenarios.
3. Compare the minimum performance of all designs and find the best-case
performance of the entire design space i.e. minimum performance of all
designs across all scenarios, as shown in Table 3.2.
4. Find the worst-case (maximum) performance of a design across all

scenarios.
5. Calculate the performance deviation of a design, as shown in Table 3.2. The

performance deviation is the performance difference between the worst-
case performance of a design and the best-case performance.
In summary, performance deviation is used as the measure of robustness and design

with the smallest performance deviation is the most robust solution. Ideally, the
design with zero performance deviation is the most robust solution of a design space.
Table 3.2 Performance robustness (deviation) calculations using the best-case and worst-
case method.
Scenarios Worst-case Best-case Performance
performance performance deviation
Designs S1 S2 … Sn
(WC) (BC) (WC-BC)
max(PI11,
d1 PI11 PI12 … PI1n WC1-BC1
PI12,…PI1n) min(PI11,
PI12,…PI1n,
max(PI21,
d2 PI21 PI22 … PI2n PI21, PI12,…PI1n, WC2-BC2
PI22,…PI2n)
PIm1,
… … … … … … …
PIm2,…PImn)
max(PIm1,
dm PIm1 PIm2 … PImn WCm-BCm
PIm2,…PImn)
The most robust design min(WC-BC)
55
3.2.3.3 The minimax regret method

In this method, for a given scenario, performance regret is the performance
difference between a design and the best performing design in that scenario [Aissi
et al., 2009]. This is elaborated below. The maximum performance regret of a design
across all scenarios is the measure of its robustness. Using this method, the
following steps are implemented in the present research to select the most robust
design of a design space across the considered scenarios.
2. Find the best performing design for each scenario by comparing the
performance of all designs. In this work, we assume that the best
performing (optimal) design is the one with the minimum performance
for a scenario. For instance, as shown in Figure 3.6, designs dm and d2 are
the best performing (optimal) designs among four designs for scenarios S 1
and Sn respectively.
3. Calculate the regret (R) of a design for each scenario, as shown in Table 3.3.
The regret is the performance difference between the design and the best
performing design for a scenario. For instance, as shown in Figure 3.6, R11,
and R21 represent the performance regrets of designs d1, and d2 respectively
for scenario S1. Similarly, R1n and Rmn represent the performance regret of
designs d1 and dm respectively for scenario Sn. It is worth noting that
designs dm and d2 have zero regret for scenarios S1 and Sn respectively.
4. Find the maximum performance regret for each design across all
scenarios. For instance, when considering design dm for scenarios S1 and
Sn (Figure 3.7), the maximum performance regret of design, dm is Rmn.
In summary, the maximum performance regret is the measure of robustness; the

lower the maximum performance regret, the higher the robustness. Therefore, the
most robust design is the design with the lowest maximum performance regret, as
shown in Table 3.3
56
d1 d2 dm
Performance indicator
S1 Sn
Scenarios
Figure 3.6 Performance of various designs for scenarios S1 and Sn (optimal designs are
indicated in dotted lines).
d1 d2 dm
and performance regrets
Performance indicator
R1n
R2n=0
R11
Rmn
Rm1=0
R21
S1 Sn
Scenarios
Figure 3.7 Performance regret of designs for scenarios S1 and Sn.
57
Table 3.3 Performance robustness (maximum regret) calculations using the minimax regret method.
Scenarios
Designs S1 S2 … Sn
d1 PI11 PI12 … PI1n
d2 PI21 PI22 … PI2n
… … … … …
dm PIm1 PIm2 … PImn
Minimum
A1 = min (PI11,PI21, A2 = min (PI12, An = min (PI1n,
performance for …
… PIi1,PIm1) PI22, … PIi2,PIm2) PI2n, … PIin, PImn)
each scenario (A)
Performance regrets (R)

S1 S2 … Sn Maximum performance regret (Rmax)
d1 R11=PI11-A1 R12=PI12-A2 … R1n=PI1n-An Rmax1= max (R11, R12,…R1n)
d2 R21=PI21-A1 R22=PI22-A2 … R2n=PI2n-An Rmax2= max (R21, R22,…R2n)
… … … … … …
dm Rm1=PIm1-A1 Rm2=PIm2-A2 … Rmn=PImn-An Rmaxm=max (Rm1, Rm2,…Rmn)
The most robust design min(Rmax)
58
3.2.4 Multi-criteria decision making
3.2.4.1 Trade-off solutions

In multi-criteria performance assessment considering multiple performance and
robustness indicators, a multi-dimensional Pareto front containing a set of Pareto
solutions is obtained, thus enabling decision makers to trade off among alternative
robust solutions based on their preferred choice of performance indicators and their
corresponding robustness. This multi-criteria assessment facilitates different
decision makers to choose robust designs from a large design space. Each decision
maker could choose different robust designs from the same design space. In general,
it is difficult to visualize two or more performance indicators and their corresponding
robustness and hence, as discussed in Chapter 2, different visualization methods are
used to enhance the decision-making process. These visualization methods are
described in the next section.
3.2.4.2 Robust design options

Design options of all Pareto solutions are compared to provide an overview of
different design options of the entire Pareto front. This is done to aid decision
makers in the selection of different robust design options based on their preferred
performance. Each decision maker can apply a trade-off approach to determine
which design options will lead to optimal performance and robust design.
3.2.4.3 Multi-criteria decision-making (MCDM) method

In order to choose a robust design from among a set of available designs for all
decision makers, the MCDM method, based on the Hurwicz criterion [Hurwicz,
1952], is used to calculate the design score. As discussed in Chapter 2, the Hurwicz
criterion is selected in the present context based on its relevance and wide use in
decision making under uncertainty. The design that has the highest score is the most
robust design in a design space.
Using the Hurwicz criterion, a design score is calculated [ESTECO, 2015] by

normalizing the preferred performance and robustness indicators, as shown in
below equations:
(𝑃𝐼𝑖 − min(𝑃𝐼𝑖𝑗 ))
𝑃𝐼𝑖,𝑛𝑜𝑟𝑚 = 1 − ⁄ (3.1)
(max⁡(𝑃𝐼𝑖𝑗 ) − min⁡(𝑃𝐼𝑖𝑗 ))
59
(𝑅𝐼𝑖 − min(𝑅𝐼𝑖𝑗 ))
𝑅𝐼𝑖,𝑛𝑜𝑟𝑚 = 1 − ⁄ (3.2)
(max⁡(𝑅𝐼𝑖𝑗 ) − min⁡(𝑅𝐼𝑖𝑗 ))
where 𝑃𝐼⁡ and 𝑅𝐼 are performance and robustness indicators of a design i,

respectively; j represents design space.
The most robust design (RD) based on the Hurwicz criterion is
𝑅𝐷(𝑃𝐼𝑖,𝑛𝑜𝑟𝑚 , 𝑅𝐼𝑖,𝑛𝑜𝑟𝑚 ) = 𝑚𝑎𝑥𝑗 [∝ ⁡ ×(min(𝑃𝐼𝑖,𝑛𝑜𝑟𝑚 , 𝑅𝐼𝑖,𝑛𝑜𝑟𝑚 )) +

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(1−∝) × (max(𝑃𝐼𝑖,𝑛𝑜𝑟𝑚 , 𝑅𝐼𝑖,𝑛𝑜𝑟𝑚 ))] (3.3)
where ∝ is coefficient of pessimism, ∝⁡= 1 for the pessimistic approach, ∝⁡= 0 for
the optimistic approach and ∝⁡= 0.5 for the neutral approach. It is worth noting that
choosing a value of ∝⁡ depends on whether the decision maker adopts a conservative,
non-conservative or a neutral approach. In this study, a neutral approach is used,
since the risk-free and the risk-taking approaches are already implemented using
robustness assessment methods.
3.2.4.4 Sensitivity analysis

In order to provide additional information to decision makers about the influence of
scenarios on Pareto solutions, sensitivity analysis is carried out using the statistical
Mann-Whitney U test to identify the most influential scenarios. The Mann-Whitney
U test is preferred in this research among alternative sensitivity analysis methods
used in building performance simulations [Tian, 2013] because of its ability to
process random variables (scenarios in this research) that do not have any uniform
or normal distribution. This statistical test, also called the Wilcoxon rank-sum test
[Wilcoxon, 1945], is a non-parametric test, most commonly used to systematically
quantify the magnitude of difference between two samples (low-high scenario
samples in this case) [Y. Sun, Su, et al., 2015; Fadeyi, 2014; Mlecnik et al., 2012]. In
other words, the sensitivity index (p) determines whether the influence of two
samples of a scenario, e.g. low and high heating setpoints, on a performance
indicator differ significantly from each other or not. In this research, scenarios where
p (p = probability) <0.05 are assumed to be sensitive [Mlecnik et al., 2012; Gaetani et
al., 2016b].
The purpose of this sensitivity analysis is to facilitate decision makers to take extra
measures to reduce the influence of scenarios.
60
3.2.4.5 Visualization methods

The results of the performance robustness assessment and sensitivity analysis are
presented in different visualization methods to enhance the design decision-making
process. Several visualization methods such as parallel coordinates plots,
histograms, scatter plots and box plots were reviewed in the second chapter and the
selected methods are used in this research. The difficulty in visualization lies in
identifying a robust design from a large design space considering multiple
performance requirements. Scatter plots are commonly used to represent multi-
dimensional plots [Blasco et al., 2008; Østergård et al., 2017]. This is illustrated using
case study results from the study by [Kotireddy et al., 2017b]. The Pareto fronts for
considered decision makers are shown as 3D scatter plots (see Figure 3.8 and Figure
3.9). Figure 3.8 shows the visualization of a 3D Pareto front using the 3D scatter plot.
This will enable a decision maker to choose a preferred robust design, for instance
by using a trade-off approach considering all three indicators simultaneously.
However, it is hard to visualise using 3D scatter plots, especially when a large design
space is considered. Furthermore, the complexity of design decision-making process
using these multi-dimensional scatter plots increases if a decision maker has more
than three preferred performance indicators. For instance, Figure 3.9 shows a
visualization of a 5D Pareto front using two 3D scatter plots for a decision maker who
has five preferred performance and robustness indicators. Using these two 3D scatter
plots leads to a laborious and difficult process to select the preferred design by a
decision maker with five preferred performance and robustness indicators.
Therefore, to reduce this difficulty and enhance the design decision-making process,
robustness is introduced as an extra dimension in the scatter plots (bottom figure)
as shown in Figure 3.10. The left figure shows the predicted performance and the
right figure shows the performance robustness. In the bottom figure, robustness is
represented by bubble size. The smaller the bubble size, the more robust is the
design. In the scatter plots presented in Figure 3.8 - Figure 3.10, the additional
investment cost required for a design, shown on the X-axis, allows the decision
maker to trade off additional investment cost with predicted performance and
performance robustness of the design. Each bubble represents the predicted
performance (median value) of CO2 emissions of a design across the considered
scenarios, and the bubble size depicts the robustness of CO 2 emissions.
61
Figure 3.8 A 3D scatter plot of a Pareto front considering additional investment, CO2
emissions and corresponding robustness (maximum regret).
Figure 3.9 Representation of a 5D Pareto front considering additional investment cost,

global cost, overheating hours and their corresponding robustness (maximum regret) using
two 3D scatter plots.
Using this information, a decision maker can choose a robust design based on their
preferred performance range and can trade off with required additional investment
cost. For instance, if a decision maker prioritizes costs over performance robustness,
then the preferred robust design lies in the least robust designs region (see Figure
3.10). Contrariwise, if predicted performance and performance robustness are
prioritized, then the preferred robust design is in the most robust designs region. If
a decision maker chooses to trade off among predicted performance, performance
robustness and required additional investment costs, then the preferred robust
design is in the optimal robust designs region.
In the scatter plot (see Figure 3.10) the performance and robustness of a design is
included in a bubble, which makes it difficult to distinguish between designs with
similar performance. To provide better insights into the predicted performance and
performance robustness of a design, a few selected designs, marked in color in the
62
scatter plot (top figure), are compared in box-plots (middle figure) and bar plots
(bottom figure) in Figure 3.11. Box-plots show the variation of predicted performance
of selected designs across the considered scenarios and bar plots show the
performance robustness of designs. A decision maker can choose a robust design
from the box plots, i.e. a design with better predicted performance and low variations.
However, it is hard to distinguish between the robustness of these designs as a
design can have very high variations for an extreme scenario but can be optimal for
the remaining scenarios. Therefore, the absolute values of performance robustness
are shown in the bar plots. The decision maker can choose a robust design from the
bar plots based on performance robustness. In both cases, the preferred robust
design depends on the required additional investment cost, and this information can
be obtained from scatter plots.
The same method has been implemented for visualization in cases where a decision
maker has more than three preferred performance and robustness indicators.
Predicted (actual) performance Performance robustness

3500 3500
Performance robustness of
CO2 emissions, kgCO2/aCO2 emissions, kgCO2/a
CO2 emissions, kgCO2/a
3000 3000
2500 2500
2000 2000
1500 1500
1000 1000
500 500
3500
0 0
3000
-500 -500
15 20 25 30 35 40 45 15 20 25 30 35 40 45
2500
Additional investment cost, k€ Additional investment cost, k€
2000
1500
Predicted (actual) performance and performance robustness
1000 (robustness embedded as bubble size, 183-3385 kgCO2/a)
3500
500 3500 Least robust designs 3000 = Least robust design
0 3000
2500 = Most robust design

2500
-500
2000
15 20 2000 25 30 Optimal
35 robust
40 designs45
1500 based€1500
Additional investment cost, on trade-off
1000 1000
500 500 Most robust designs
0 0
-500 -500
15 20 25 30 15 3520 4025 4530 35 40 45
Additional investment cost, k€
Additional investment cost, €
Figure 3.10 An example of a visualization method by including performance robustness

(from Figure 3.8) as the third dimension in a scatter plot.
63
3500
3000
2500

2000
1500
1000
500
-500
15 20 25 30 35 40 45
Additional investment cost, K€
3500
Performance robustness (maximum
regret) of CO2 emissions, kgCO2/a
3000
2500
2000
1500
1000
500
Design-1 Design-2 Design-3 Design-4 Design-5
Figure 3.11 An example of different visualization methods used in this research to aid
decision makers in selecting preferred robust designs.
64
Practical use of the CPRA approach
3.3 Practical use of the CPRA approach

The developed CPRA approach has been tested using real case studies for its
suitability and usability with the help of one PDEng and two MSc graduates. The
final CPRA approach was implemented by these graduate students for
robustness/risk assessment in case studies for a leading building consultant and
services company (Homij), a social housing company (Woonbedrijf) and a housing
corporation (Mitros Housing Corporation). The outcomes of the CPRA adaptation in
these projects are summarized in Chapter 7. It is worth noting that these case studies
are not part of this research. Separate case studies are used to test the methodology
in this research, which are described in Chapter 5, and case study results are
discussed in Chapter 6.
• Risk-averse’ design solutions for Homij using building performance

simulations - A case study of MorgenWonen by Sanket Puranik for his
PDEng Dissertation at the Eindhoven University of Technology [Puranik,
2017].
• Planning for the future: developing a risk-averse strategy of future-proof

nearly zero energy building retrofits for Woonbedrijf considering multiple
KPIs and market scenarios, by Sergio Costa for his MSc Thesis at the
Eindhoven University of Technology [Costa, 2017].
• Future-proof residential flats: towards the development of robust energy

concepts considering uncertainties in climate change and in the energy
market to achieve ‘nul-op-de-meter’ 2 for clustered residential flats for the
Mitros Housing Corporation. This project is currently being investigated by
Gerton van Middendorp as part of a MSc graduation program at the
Eindhoven University of Technology.

In this chapter, a methodology to develop and test the computational performance
robustness assessment (CPRA) approach is proposed. The suitability and usability of
the developed CPRA approach is assessed with end users through mock-up
2
Nul-op-de-meter is roughly translated as zero on the meter. It means that energy on the meter
at the end of year is same as at the beginning of year, indicating that total energy consumed
by a house is supplied by onsite energy generation by RES system.
65
presentations. The developed CPRA approach is improved through an iterative

process considering feedback from user group meetings.
The final CPRA approach comprises multi-criteria performance assessment and

multi-criteria decision making, taking into account performance robustness among
other performance indicators. In this approach, by prioritizing the decision maker’s
preferences, building design space, future scenarios and performance indicators are
defined. The performance of the design space for future scenarios is assessed using
building performance simulations with multiple performance indicators and their
corresponding performance robustness.
Different robustness assessment methods such as the max-min method, the best-
case and worst-case method and the minimax regret method are used to evaluate
performance robustness in this approach. These methods can be selected by decision
makers based on their attitude towards risk in the decision-making process. The
Hurwicz criterion based MCDM method is used to rank designs for different
decision makers by prioritizing the performance indicators based on their
preferences. Finally, sensitivity analysis is carried out to identify the scenarios with
the greatest influence on performance and performance robustness. All of these
methods and analyses are presented using various visualization methods to enhance
the decision-making process.
The CPRA approach is generic and it can be used for performance robustness
assessment in a holistic approach for both new houses and renovations.
Furthermore, it can be used for performance robustness assessment of individual
energy systems such as HVAC, PV, and SDHW systems.
66
4. Implementation of the
CPRA approach in the
simulation framework
This chapter describes the simulation framework developed to implement the CPRA
approach described in the previous chapter. The simulation tools used in the simulation
framework are described. Furthermore, this chapter describes the methods to make this
framework computationally efficient and thus to enhance its usability in practice.
Integration of robustness assessment methods in the optimization process is a novel
approach and is described in detail in this chapter.
4.1 Overview of the simulation framework

In this chapter, the CPRA approach developed in Chapter 3 is translated into a
simulation framework, as shown in Figure 4.1. This framework comprises three
stages: pre-processing, simulation and post-processing, which are described below.
i. Pre-processing
In pre-processing, the design of experiments, which constitutes establishing the

design space, and the formulation of scenarios and their combination (simulations)
with the design space are developed. In this stage, the input files of the design space
and scenarios are created. Performance indicators that are relevant to the decision
makers such as comfort, cost etc. are defined and these multiple performance
indicators are used to assess the performance of the design space across the
considered scenarios.
ii. Simulation (experiment)
The performance of the design space across formulated scenarios using the defined
performance indicators is predicted using a simulation model developed in this stage
of the computational framework. Robustness of the design space is assessed using
different robustness assessment methods that were selected from a literature review
(Chapter 2).
67
Chapter 4. Implementation of the CPRA approach in the simulation framework
iii. Post-processing and analysis
The results of performance and robustness assessments are analyzed, and robust
designs are identified based on decision makers’ preferences in the post-processing
and analysis stage. These results are presented with different visualization methods
to enhance the design decision-making process.
In a typical design process, multiple configurations regarding building envelope,

HVAC systems and energy systems are considered to find an optimal design. These
multiple configurations typically result in a large design space. In addition, to find a
robust design, the performance and robustness of these large design spaces is
assessed considering a large number of scenarios that can influence a building’s
performance over its life-span. Therefore, the performance robustness assessment
of the design space across all formulated scenarios could demand millions of
simulations, which is computationally very expensive and might not be feasible for
use in practice. The computational time should be reduced in order to make this
simulation framework useful in practice. The computational cost of an exhaustive
search of large design space can be reduced notably by the following approaches:
• Using a white box (high-fidelity) model in combination with a smart

algorithm to reduce the number of simulations.
• Constructing a statistical model by sampling the original white-box model

to make it possible to do many more simulations at low computational costs
[Van Gelder, 2014].
Generally, in a building design project, the design options space and the scenarios
are defined based on the preferences of the stakeholders. These preferences will
likely vary from case to case. When using a meta-model, the model would need to be
developed for every specific case. Using high-fidelity model, the model user can
immediately get started. Similarly, in order to repeat the study for a similar climate,
it is not essential to develop a new meta-model. In addition, the physical integrity
within the models can be preserved as it is useful to have the possibility to look back
at which physical phenomena cause certain types of building designs to have high
performance. Therefore, a white box model in combination with smart algorithm is
used in this CPRA approach. Genetic algorithm (GA) based optimization method is
preferred in this research due to its fast convergence [Deb et al., 2002].
68
Overview of the simulation framework
Identify decision makers and their

preferences
Define Define
Define
design space performance
scenarios (Sn)
(dm) indicators (PI)
Create design of experiments (DOE)

(mxn)
Building and energy systems (BES)

simulation model
Performance assessment of design space

(dm) across scenarios (Sn) with
performance indicators (PI) using BES
model
Performance robustness assessment of

design space (dm) across scenarios (Sn)
using a robustness assessment method
Analyze results and identify robust

designs based on predicted performance
and performance robustness
Present these results in relevant

visualization methods to enhance
decision-making process
3500
3500
Performance robustness of CO2
3000 3000
emissions, kgCO2/a
2500 2500
2000 2000
1500
1500
1000
1000
500
500
0
-500 0
15 20 25 30 35 40 45
Additional investment cost, K€ Design-1 Design-2 Design-3 Design-4 Design-5
Figure 4.1 Simulation framework developed for the implementation of the CPRA approach.
69
However, using GA, there is a risk of missing certain interesting design solutions as
the whole design space is not covered by GA. In this research, extensive studies have
been carried out to find the optimal settings of GA that result in similar Pareto fronts
as an exhaustive search. Furthermore, the performance and robustness of a design
space must be assessed for the same sample of scenarios to find a robust design
within a design space. In this research, scenario analysis is carried out independently
to find the smallest scenario sample size that yields similar performance as that of
all scenario combinations. Therefore, two methods are implemented in this
framework to reduce computational costs:
i. Scenario sampling (Section 4.4)

ii. Genetic algorithm (GA) based multi-objective optimization (Section 4.5).
4.2 Selected tools in the simulation tool chain
4.2.1 Selection of building performance simulation (BPS) tool

Several tools have been developed in the past few decades to address the different
needs of building performance simulation. Therefore, the capabilities of the available
simulation tools vary greatly, as reported in [Crawley et al., 2006]. The choice of tool
depends on the research and on design objectives that need to be met and on the
relevant performance indicators that need to be calculated. This study focuses on
low-energy houses; therefore, the selected BPS tool should be able to model energy
efficiency measures and renewable energy technologies. Furthermore, the selected
simulation tool should be able to predict the dynamic behaviour and interaction of
these measures and energy systems while taking into account occupant behaviour
and external conditions. All energy and thermal interactions among the building,
energy systems, occupants and external conditions should be predicted dynamically.
In addition, robustness assessment methods are implemented in an external
programming platform and this external platform should be connected to the
simulation model in order to simultaneously evaluate robustness and actual
performance. As such, the selected simulation tool should be able to run in batch
mode.
Of the limited number of such available tools with the aforementioned capabilities,
TRNSYS, a transient system simulation program [Solar Energy Laboratory
University of Wisconsin-Madison et al., 2009], is a commonly used simulation tool
[Nguyen et al., 2014; Sousa, 2012] and is suitable for the present context.
70
Selected tools in the simulation tool chain
4.2.2 Selection of other tools

Mode Frontier [ESTECO, 2015], a tool for process integration and design
optimization, is used in pre-processing for scenario sampling strategies and in post-
processing for MCDM analysis. MATLAB, known for its extensive programming
capabilities and plotting functions, is used in the pre-processing stage to create the
design of the experiments (simulations) and in post-processing to visualize results,
as shown in Figure 4.2. In addition, all three robustness assessment methods are
implemented using MATLAB [MathWorks, 2016].
MATLAB is used as a process integrator that couples all building and energy system
models and is used as a platform to carry out multi-objective optimization of the
design space for the considered scenarios using a multi-objective optimization
genetic algorithm from the MATLAB optimization tool box. It is worth noting that
this simulation framework could have been implemented using other simulation
and optimization tools.
Figure 4.2 Simulation tools used in the different stages of the simulation framework.
4.3 Improving simulation framework efficiency

Ideally, the performance and robustness of the design space should be assessed for
all scenario combinations as the likelihood of any scenario combination is not
known, but this assessment is computationally very expensive. To reduce this
computational cost, the computational efficiency of the simulation framework is
enhanced by using a scenario sampling strategy and multi-objective optimization.
The improvement in computational efficiency using these methods is presented in
the next sections and the computational cost savings from these methods are
tabulated at the end of this section. The implementation of sampling strategy and
multi-objective optimization are carried out through a case study. The case study is
chosen from [Kotireddy et al., 2017b], and the case study design space and scenarios
considered for sampling strategy and multi-objective optimization are shown in
Table 4.1 and Table 4.2 respectively.
71
Table 4.1 Design options of the case study considered to implement scenario sampling and
multi-objective optimization in order to enhance the computational efficiency of the
simulation framework.
Design variant Options
Building envelope properties
[4.5/6/3.5, 6/7/5, 7/8/6, 9/9/7, 10/10/10]
(Rc-wall/roof/floor), m2k/W
WWR (%) [20, 40, 60]

Thermal mass [Light-weight, Medium-weight, Heavy-weight]
Infiltration, ach [0.12, 0.24, 0.36, 0.48]
PV system, m2 [5, 10, 15, 20, 25, 30]
2
Solar DHW system, m [0, 2.5, 5]
Table 4.2 Scenarios of the case study considered to implement the scenario sampling and
multi-objective optimization to enhance the computational efficiency of the simulation
framework.
Parameter Options
Occupant scenarios
Household size [1, 2, 3, 4]
Usage (Occupant behavior) scenarios
Heating setpoint (occupied), °C [18, 20, 22]

Heating setpoint (un-occupied), °C [14, 16, 18]
Occupancy profile [Evening, All-day]
Average electricity use for lighting, W/m2 [1,2,3]
Average electricity use for appliances, W/m2 [1,2,3]
Internal heat gains due to lighting and [2, 3, 4, 5, 6]
appliances, W/m2
Domestic hot water consumption, LPD/person [40, 60, 100]
Ventilation, ach [0.9, 1.2, 1.5]
Shading control ON if radiation is above, W/m2 [250, 300, 350]
and if Tindoor >24°C
Shading control OFF if radiation is below, W/m2 [200, 250, 300]
and if Tindoor <24°C
Climate scenarios
Reference climate and climate change scenario [NEN5060-2008, G, W, G+, W+]
72
Improving simulation framework efficiency
It is worth noting that separate case studies are used to demonstrate the CPRA
approach, which are described in Chapter 5. Four performance indicators, based on
the preferences of a policymaker (CO2 emissions, additional investment cost) and a
homeowner (overheating hours, additional investment cost and global cost) are
considered. These performance indicators are described in detail in Chapter 5. It is
noteworthy that some of the scenarios are varied together as they are inter-
dependent. For instance, internal heat gains due to appliances and lighting depends
on the usage of lighting and appliances. The total number of design combinations
and scenario combinations are 3240 and 29160 respectively. Performance
assessment of this design space (Table 4.1) across all scenario combinations (Table
4.2) requires 94.478 million simulations.
Full factorial simulations (design options × all scenario combinations) are used as a
reference to calculate savings in computational costs by using the scenario sampling
strategy and optimization method.
4.4 Improving simulation framework efficiency using scenario sampling

Sampling strategies are widely reported in the literature [Janssen, 2013] and also in
the building performance context [Macdonald, 2009; Burhenne et al., 2011; Hu and
Augenbroe, 2012; O’Neill and Eisenhower, 2013a] to reduce computational costs
associated with running large sets of simulations. This reduction is often achieved
by using a suitable sampling technique to find the smallest sample size that can
predict similar performance as that of a full sample. Therefore, a sampling strategy
is implemented in this research to find the smallest scenario sample that can predict
similar performance and robustness for a design as that of all scenario combinations.
In a conventional approach, the sampling strategy is selected based on convergence
i.e., mean performance and variance [Janssen, 2013]. However, for performance
robustness assessment, the performance range or distribution is also crucial in
selecting a sampling strategy and determining its smallest sample size.
Low-high scenario combinations are generally sufficient for performance robustness

assessment [Kotireddy et al., 2018] because the low-high scenario combinations
typically result in a performance range, and performance with the remaining
scenario combinations fall within this range. Therefore, for low-high scenario
combinations the sampling strategy is used instead of investigating all scenario
combinations. To justify this selection, the performance robustness of three designs,
selected from the design space presented in Table 4.1, is assessed with all scenario
73
combinations and low-high scenario combinations (Table 4.2), to evaluate if low-high

scenario combinations are sufficient for the performance robustness assessment for
multiple performance indicators.
Typically, sensitivity analysis is a more suitable method to find the scenarios causing
this performance range in the whole design space, since the scenario combinations
can interact differently with each design. However, sensitivity analysis for the entire
design space across all scenarios requires an exhaustive search and is
computationally expensive, making it infeasible in practice. Therefore, a crude
method is implemented by selecting three different designs from the design space.
The notable difference among the three designs is that design 1 has low insulation
levels (Rc =4.5/6/3.5 m2K/W for walls/roof/floor) and large RES systems (30 m2 PV
system and 2.5 m2 SDHW system), whereas design 3 has very high insulation levels
(Rc =10/10/10 m2K/W for walls/roof/floor) and smaller RES systems (15 m2 PV
system and 5 m2 SDHW system). Design 2 has intermediate insulation levels (Rc
=6/7/5 m2K/W for walls/roof/floor) and intermediate RES systems (25 m2 PV system
and 2.5 m2 SDHW system).
4.4.1 All scenario combinations vs low-high scenario combinations

Figure 4.3 compares the robustness of CO2 emissions, overheating hours and global
cost, which are the considered performance indicators in the case study [Kotireddy
et al., 2017b]. Here, robustness of these performance indicators is calculated using
three robustness assessment methods, namely the max-min method, the best-case
and worst-case method and the minimax regret method. It can be observed that both
the low-high scenario and all scenario combinations resulted in similar robustness
for each respective robustness assessment method. This observation is valid for all
three performance indicators. Furthermore, for a particular robustness assessment
method, both scenario combinations resulted in the same robust design for a
preferred performance indicator. For instance, design 3 is robust to CO2 emissions
using spread as a robustness indicator for both low-high and all scenario
combinations. Similarly, design 1 is the most robust design to overheating in all three
robustness assessment methods for both low-high and all scenario combinations.
It is worth noting that there is no difference in a particular robustness indicator for

all designs in both low-high and all scenario combinations.
74
Improving simulation framework efficiency using scenario sampling
5000
CO2 emisisons, kgCO2/a

4000
3000
2000
1000
0
Design 1
Design 2
Design 1
Design 2
Design 3
Design 1
Design 2
Design 3
Design 3
Spread of CO2 Deviation of CO2 Maximum regret of
emissions emissions CO2 emissions
1000
Overheating hours, h/a
800
600
400
200
0
Design 1
Design 2
Design 3
Design 1
Design 2
Design 3
Design 1
Design 2
Design 3
Spread of Deviation of Maximum regret of

overheating hours overheating hours overheating hours
70
60
Global cost, k€
50
40
30
20
10
0
Design 3
Design 1
Design 2
Design 3
Design 1
Design 2
Design 3
Design 1
Design 2
Spread of global Deviation of global Maximum regret of

cost cost global cost
Low-high scenarios (512) All scenarios (29160)
Figure 4.3 Comparison of performance robustness of CO2 emissions, overheating hours and
global cost of three designs calculated using three robustness assessment methods for all
scenario combinations (29160) and low-high scenario combinations (512).
75
Therefore, it can be concluded that low-high scenario combinations (512) result in

similar performance as that of all scenario combinations (29160), and thus low-high
scenario combinations are sufficient for the performance robustness assessment.
This approach would itself save about 98% of computational costs but would still
require a total of 1.6 million simulations (design options × low-high scenario
combinations). Hence, a sampling strategy, based on Monte Carlo sampling, is used
for low-high scenario combinations to find the smallest sample size of scenario
combinations that predicts similar performance.
4.4.2 Sampling strategy

Commonly used Monte Carlo sampling strategies in building performance
simulations are random sampling, Sobol sampling and Latin hypercube sampling
(LHS) [Macdonald, 2009; Burhenne et al., 2011]. In random sampling the sample is
generated according to a random distribution, which might easily result in clusters
and gaps, as shown in Figure 4.4. These clusters and gaps are avoided in the SOBOL
method as samples are generated as uniformly as possible (see Figure 4.4)
[Burhenne et al., 2011]. LHS has fast convergence [Janssen, 2013; Helton et al., 2006]
compared to the other two methods. The sample is generated by stratification and
input is divided into sub-intervals. Each interval has the same number of samples,
which are randomly generated. Due to efficient stratification of the LHS sampling
method, small sample sizes are sufficient to achieve desired outcome/accuracy levels
[Helton et al., 2006]. This sampling efficiency can be further enhanced by using the
uniform Latin hypercube (ULH) sampling method, because the ULH method
provides desired accuracy levels even at smaller sample sizes [Janssen, 2013]. In
addition, ULH sampling has the least variance of statistical distance between
generated samples. Thus, ULH sampling is the preferred sampling method in this
study.
The performance of three designs is assessed with different ULH samples of

scenario combinations ranging from 25-500, and the robustness of three designs for
these ULH samples is compared with low-high scenario combinations. Due to the
stochastic nature of the ULH sampling method, the performance assessment of
these samples is carried out multiple times to reduce stochasticity in sample
generation. The smallest sample size is selected based on mean and standard
deviation of the robustness indicator across multiple runs. The smallest sample that
has a similar (mean) robustness indicator as that of low-high scenarios combinations
and that is close to zero standard deviation across multiple runs is preferred.
76
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6
Variant1
0.6
Variant1
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Variant2 Variant2
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
Variant1
Variant1
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Variant2 Variant2
Figure 4.4 Distribution of 100 samples of two variants with different sampling strategies
(top left-Random; top right-Sobol; bottom left-LHS; bottom right-ULH). Each variant has
100 data points ranging from 0 to 1.
Figure 4.5 compares the mean and standard deviation of robustness of CO 2

emissions calculated using three robustness assessment methods for different ULH
sample sizes across multiple runs (in this case it is 10) with that of low-high scenario
combinations. For other performance indicators, only maximum regret is compared
in Figure 4.6 to avoid repetition and to enhance readability. It is worth noting that
these experiments are carried out for different robustness indicators. It can be
observed from Figure 4.5 that all ULH samples, except 25 and 50, for three designs
have similar robustness for multiple runs as that of low-high scenario combinations.
In addition, the standard deviation is close to zero (see Figure 4.5). It is worth
mentioning that discrepancies are observed in the case of overheating hours of
design 3 with the ULH sample of 300 (see Figure 4.6), and these discrepancies can
be avoided with more runs.
77
78
Maximum regret of
CO2 emissions, kgCO2/a Deviation of CO2 emissions, kgCO2/a Spread of CO2 emissions, kgCO2/a
1000
2000
3000
4000
5000
0
2000
4000
5000
1000
3000
0
1000
3000
4000
5000
2000
0
Design 1 Design 1 Design 1
Design 2
25
Design 2
25
Design 2
25
combinations (512).
Design 2
50
Design 2
50
Design 2
50
Mean
100
100
100
150
150
Design 3 Design 3 150 Design 3
200
200
200

250
250
250

300
300
300

350
350
350

Design 1 Design 1
Scenario sample size for each run

Design 1
400
400
400
Standard deviation
450
450
450

500
500
500


high
high
Low-
high
Low-
Low-
ULH scenario samples across multiple runs (10) compared to low-high scenario
Figure 4.5 Variation of mean and standard deviation of performance robustness of CO 2
emissions of three designs calculated using three robustness assessment methods for different
Maximum regret of Maximum regret of
Maximum regret of
overheating hours, h/a CO2 emissions, kgCO2/a
global cost, k€
0
5
10
15
20
25
30
35
0
100
200
300
400
500
1000
1400
1200
0
200
400
600
800
Design 2
25
Design 2
25
Design 2
25
Design 2
50
Design 2
50
Design 2
50
scenario combinations.
100
100
100
150
Mean
150
150

200
200
200

250
250
250

300
300
300

350
350
350

Design 1 Design 1

Design 1
400
400
400
Standard deviation
450
450
450

500
500
500

high
high
high
Low-
Low-
Low-
Figure 4.6 Variation of mean and standard deviation of performance robustness

(maximum regret) of CO2 emissions, overheating hours and global cost of three designs for
different ULH scenario samples across multiple runs (10) compared to low-high (512)
79
It can be noted that the relative deviation (∆R) of robustness with ULH samples in
comparison to low-high scenario combinations is close to 0-5% for the ULH samples
≥ 100 for all performance indicators of the three designs. The acceptable range of
relative deviation depends on the preferred performance indicators and their
consequent impact on decision making. Moreover, it also depends on the field of
application of this CPRA approach as no relative deviation is preferred in buildings
with critical operating conditions, such as data centers and clean rooms, because the
risk/costs associated with failure of a design are very high. However, in residential
buildings, certain ranges of relative deviation can be accepted as they may not have a
considerable impact on design decision making. Moreover, standard deviation of the
ULH sample of 100 scenarios is close to zero for all performance indicators for all
three designs. Relative deviation is not improved drastically (between 0-5%) by an
increase in sample size.
Hence, the ULH sample of 100 scenario combinations is chosen for the performance
robustness assessment in this research.
4.5 Improving simulation framework efficiency using multi-objective

optimization
Multi-objective optimization is used to find trade-off solutions using a Pareto front
when multiple and often conflicting performance requirements are involved in a
project. A genetic algorithm (GA) based (notably the non-dominated sorting genetic
algorithm NSGA-II [Deb et al., 2002]) multi-objective optimization is the most
widely used in building design optimization [Evins, 2013; Machairas et al., 2014] as
shown in Figure 4.7. The wide use of NSGA-II is due to: its ability to find multiple
Pareto solutions in a single run; a better distribution of solutions in the Pareto front;
and better convergence near the true Pareto front [Deb et al., 2002].
Therefore, a NSGA-II based multi-objective optimization is implemented in this

study to find robust optimal solutions in a design space across formulated scenarios
considering multiple performance and robustness indicators.
80
Improving simulation framework efficiency using multi-objective optimization
Figure 4.7 Different optimization methods used in building research field [Evins, 2015]
Due to the inherent stochastic nature of the GA algorithm, it is important to test the
performance of the algorithm. In order to assess the GA’s performance, the true
Pareto front of a design space is calculated using a full factorial approach. For
instance, the true Pareto front of CO2 emissions calculated by a full factorial approach
using three robustness assessment methods is shown in Figure 4.8. In this section,
only CO2 emissions are compared, unlike in the previous section; the three
robustness indictors are retained as they differ in calculation approach, and thus,
also in the corresponding optimization process. The objective of this experiment is
to find a similar Pareto front to that of the true Pareto front using the GA-based
optimization method in the least possible number of iterations.
The implementation of optimization using three robustness assessment methods

and a method to find the optimal settings to match the true Pareto front are
elaborated in the next sections.
81
Figure 4.8 True Pareto front (blue bubbles) of CO2 emissions of the design space (gray
bubbles) calculated using three robustness assessment methods by the full factorial (FF)
approach. Bubble size represents corresponding robustness in each method. The smaller the
bubble size, the more robust is the design.
82
4.5.1 Implementation of robustness indicators in GA
In this research, a typical GA-based optimization cannot be implemented in a

straightforward way. In a typical GA-based optimization process, the GA creates the
design population for a generation and then the objective functions of this design
population are calculated. For instance, as shown in the dotted lines of Figure 4.9,
for each considered scenario, the performance of a design population for a
generation is assessed using building and energy system simulation (BES) model,
and then performance robustness of the design population is evaluated. The design
population for the next generation is defined by the GA, based on the calculated
objectives (actual performance and robustness) of the previous generation, and this
process continues until the optimization criterion is met. However, in this research,
to include performance robustness as an objective in the multi-objective
optimization, the robustness indicators such as maximum regret and deviation can
only be calculated after the performance assessment is conducted for the entire
design population for all scenarios [Kotireddy et al., 2017b].
Therefore, the fitness function (objective function) has to be defined in such a way
that the optimization process halts after every generation until the calculation of the
performance robustness is finished. Hence, the optimization process in the current
study is nested across three loops (see Figure 4.9) as discussed below, to ease
calculation of robustness indicators that require pausing of the GA algorithm:
1. Main loop – In this loop, population of design alternatives is updated for

different generations based on objectives. Robustness indicators are also
calculated in this loop.
2. Designs loop – This is a sub loop of the main loop, where the performance of
all design populations is calculated and the performance indicators matrix
from this loop is returned to the main loop.
3. Scenarios loop – This loop is a sub loop of the designs loop, where the
performance of each design is assessed for each scenario and the
performance indicator vector of a design across the considered scenarios is
returned to the designs loop.
In this optimization process, for a particular generation, the performance of a design

is calculated for the considered scenarios in the scenario loop and the performance
indicator vector of a design is returned to the designs loop.
83
YES
Evaluate performance Evaluate Select the parents
Create new generation
Create initial of generation for performance based on objective Optimization
Start g=0 based on crossover, Stop
population considered scenarios robustness of function to create criteria is met?
mutation etc.
using BES model design population new generation
d=1
NO
Performance
d=m indicators matrix
For design,
of generation, g
d =1:m across considered
d=d+1 scenarios, PI mXn
S=1
Performance Scenario
indicators For sample (n)
vector for a
design across
scenarios, using a
scenarios,
S=n S =1:n sampling
PI1xn S=S+1 method
Simulate the
performance of
design, d for
scenario, S
g=g+1
g = generation
d = design
m = number of designs in generation
S = scenarios
n = number of scenarios
Figure 4.9 The extended GA-based optimization implemented in this study compared to a typical GA-based optimization (indicated in dotted
line) for performance robustness optimization using scenario analysis.
84
In order to evaluate the performance of all designs across the considered scenarios,
the scenario loop needs to be nested within the designs loop. As a result of this
nesting, the performance indicator’s matrix of all designs across all scenarios is
returned to the main loop, where the robustness assessment method is applied.
Based on the predicted performance and performance robustness, the design space
for the new generation is updated by the genetic algorithm. This process continues
until the optimization criterion is met. In this work, the optimization process stops
if the average relative change in the best fitness function value over 20 generations
is less than 0.001. This stopping criterion is tested with the true Pareto front
resulting from full factorial analysis. It is worth noting that there are many alternate
stopping criteria, such as generational distance [Deb et al., 2002; Deb and Jain, 2013;
Hamdy et al., 2016]. This option is not considered in this study as it is beyond the
scope of this research.
4.5.2 Modifying the fitness function of the GA by storing design archives

The calculation of robustness indicators differs for the three robustness assessment
methods, as described in the previous chapter. The main difference is that the spread
is calculated for each design of the population without any inter-comparison of
performance of other designs of the population. On the other hand, maximum regret
and deviation are calculated with inter-comparison of performance of other designs.
Hence, these two indicators are calculated after assessing the performance of the
entire population. Therefore, implementation of maximum regret and deviation in
the optimization process requires modifications to the standard GA fitness
(objective) function. Otherwise, optimization yields completely different Pareto
fronts as illustrated with an example in Figure 4.10, which compares the Pareto front
of an optimization run with maximum regret as one of the objectives in the standard
fitness function and in the modified fitness function.
In the case of the standard fitness function, maximum regret is calculated for each
generation without storing any design archive of previous generations, which thus
results in zero maximum regret for at least one design in each generation. This zero
maximum regret is because the regret is calculated based on the best performing
design for a scenario, and for the first design in each generation the regret is always
zero, as seen in Figure 4.10(a), as there is no other design to compare with. Generally,
this design may not be the most robust, despite having zero maximum regret, when
compared to other designs in each generation and also with the entire design archive,
as seen in the Pareto front with the modified fitness function (see Figure 4.10(b)).
85
Therefore, the design archive of previous generations must be stored by the GA and
the objectives of each design should be updated before proceeding to the next
generation. This update of objectives at the end of each generation cannot be done
in a straightforward approach by typical optimization software tools. Therefore, the
fitness function is defined in such a way that the GA halts and updates objectives at
the end of every generation by retrieving the design archive. In the modified fitness
function, the GA pauses after every generation to update objectives as well as store
the design archive of previous generations. This is necessary to calculate objectives
in each generation. This calculation takes into account the current design population
and the design archive of previous generations, as shown in Figure 4.11, in order to
enable the inter-comparison of the performance of all designs in these two methods.
a) A standard fitness function

4500 4500
3500 3500
Maximum regret of
2500 2500
1500 1500
500 500
-500 -500
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
All solutions of an optimization run Pareto solutions All solutions of an optimization run Pareto solutions
b) The modified fitness function

4500 4500
3500 3500
Maximum regret of
2500 2500
1500 1500
500 500
-500 -500
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
All solutions of an optimization run Pareto solutions All solutions of an optimization run Pareto solutions
Figure 4.10 The Pareto front of an optimization run with maximum regret as an objective
in the fitness function. a) A standard fitness function b) The modified fitness function.
86
Max-min method
Calculate Select the parents

performance based on objective
spread of design function to create
population new generation
Scenario
sample using
a sampling
Best-case and worst-case method
method
Calculate YES
Evaluate performance of Choose a Select the parents
performance
Create initial generation for robustness based on objective Optimization
Start g=0 deviation for new Stop
population considered scenarios assessment function to create criteria is met?
population and
using BES model method new generation
design archive
Design archive = [ ] NO
Create new generation Minimax regret method

based on crossover,
mutation etc.
Calculate
Select the parents
performance
g= g+1 based on objective
regrets for new
function to create
population and
new generation
design archive
Update design archive for all generations
Figure 4.11 A multi-objective optimization approach considering multiple performance indicators and robustness indicators calculated using
three robustness assessment methods.
87
4.5.3 Optimal settings of GA for different robustness assessment methods

By definition, the full factorial approach results in a true Pareto front for a design
space, while the GA-based optimization typically results in an approximation of the
true Pareto front, which is inherent to the stochastic nature of the GA algorithm. GA
parameter settings have a strong influence on the performance of the GA, and hence,
it is important to determine the optimal settings to enhance computational efficiency
in the process of converging to the Pareto front. In this research, the calculated
Pareto front is converged by meeting the optimization criterion for a given GA
parameter’s settings.
The main parameters of the GA (NSGA-II) [Deb et al., 2002] are:
• Population size (PS) determines the number of individuals (design

alternatives) in a population at each generation.
• Generations (g) determine the number of evaluations in an optimization run.
• Crossover fraction (CF) determines the fraction of population at the next

generation.
• Pareto fraction (PF) controls the elite members of the population for every
generation to maintain the diversity of the population for convergence to an
optimal Pareto front.
• Selection function, which is tournament size (TS) in the case of multi-

objective optimization, determines how the GA selects the parents of the
crossover members and selects the mutation members for the next
generation.
The default values of these GA parameters in MATLAB are CF=0.8, g=100*number

of design variables, PF=0.35 and TS=4. Typically, a large population size enhances
the probability of the GA returning the global optimum since it has conducted a
thorough search of the design space, but this thoroughness slows down the
algorithm and therefore increases the computational time. Hence, an optimal
population size is essential for achieving close to the true Pareto front with low
computational time. Similarly, increasing the number of generations often improves
the final solution space. In this study, a criterion based on the average relative change
in the fitness function over a set of generations is used to stop the optimization
process, and hence a predefined number of generations is not essential and is
therefore not considered in the evaluation of optimal settings. However, number of
88
iterations, which is the product of generations and population size, is used as the
criterion to select the optimal settings of GA parameters. A higher CF often results
in a local optimum, and a lower CF requires more iterations to reach the optimization
criterion as new designs are added to the population at every generation.
These settings generally depend on the design space and fitness function. The
optimal settings of the GA parameter could be different for the three robustness
assessment methods as the methods of evaluation of objectives (fitness function) are
different. Therefore, optimal settings of GA parameters are determined using the
aforementioned case study for three methods. It is worth mentioning that uniform
creation and mutation functions are considered in this study to avoid non-integer
values of design variants. In addition, a uniformly distributed initial population
which covers the design space uniformly is provided for all optimization runs.
The optimal settings of GA parameters are determined based on the following

aspects:
i. Fast convergence: Minimum number of iterations (defined as the product

of generations and population size) required to meet the optimization
criterion is used to test the convergence of an optimization run.
ii. Reaching the true Pareto front: A high matching index (defined as the
percentage of Pareto solutions with a GA parameter setting that matches the
true Pareto solutions).
4.5.3.1 Default settings
The Pareto fronts of three robustness assessment methods with default MATLAB
values of the GA parameters (CF=0.8; PF=0.35; TS=4) over multiple runs (5 in this
case) are compared with their corresponding true Pareto fronts as shown in Figure
4.12. The blue bubbles represent the true Pareto front and the red bubbles represent
the calculated Pareto front with default GA parameters settings, while the bubble
size represents the robustness of the corresponding method.
It can be observed that there is good agreement between these two Pareto fronts. A
matching index of 66.6%, 71.4% and 69.5% is achieved with an average of 876, 744
and 792 iterations over 5 runs using spread, deviation and maximum regret as
robustness indicators, respectively.
89
Bubble size = Spread of CO2 emissions (523-1121 kgCO2/a)

3500
3000
CO2 emissions, kgCO2/a 2500
2000
1500
1000
500
-500
15 20 25 30 35 40 45 50
Bubble size = Deviation of CO2 emissions (971-3785 kgCO2/a)

3500
3000
2500
2000
1500
1000
500
-500
15 20 25 30 35 40 45 50
Bubble size = Maximum regret of CO2 emissions (182-3740

kgCO2/a)
3500 Rc= 10/10/10 m2K/W for
3000 wall/roof/floor
U = 0.4 W/m2K
2500 WWR =20%
Thermal mass= light-weight
2000 Infiltration = 0.12ach
PV system = 30m2
1500 SDHW system = 5m2
CO2 emissions = -192 kgCO2/a
1000 Maximum regret of CO2
emissions = 418 kgCO2/a
500
0
-500
15 20 25 30 35 40 45 50
True Pareto front calculated using full factorial approach

Pareto front calculated uisng default GA parameters settings
Figure 4.12 Comparison of the true Pareto front and a Pareto front with default GA
parameter values provided in MATLAB for three robustness assessment methods. Blue
bubbles represent the true Pareto front and red bubbles represent a Pareto front for the
default GA parameter’s settings. It can be noted that Pareto solutions of true Pareto front
are overlapped by the Pareto front calculated using default settings of GA parameters.
90
Using default settings there is risk of losing about 30% of robust designs compared
to the full factorial approach, but these settings save a considerable amount of
computational cost. Generally, finding the true Pareto front may not be essential and
may sometimes be unfeasible. Therefore, it is up to the end user to select default
settings by accepting the risk of losing 30% of the robust designs. The calculated
Pareto front using default GA parameter values deviates from the true Pareto front
of the full factorial approach and thus proves that the default MATLAB values of the
GA’s parameters are not optimal in the present context. Hence, further investigation
of different values of GA parameters is essential to find optimal settings and to avoid
overlooking any robust designs.
4.5.3.2 Optimal settings

To find the optimal settings, the following steps are executed, as shown in Figure
4.13.
1. Define the range of different GA parameters (p= a total of 192 parameter

combinations).
2. Run the optimization by choosing a robustness assessment method for every
combination of GA parameter values.
3. Repeat the optimization process multiple times (5) for every combination of
GA parameter values to reduce the stochasticity effect of some of the GA
parameters, such as CF and PF.
4. Repeat steps 2-3 for other robustness assessment methods.
5. Calculate the number of iterations required to meet the optimization criteria
and matching index of GA parameter combinations for three robustness
assessment methods.
6. Find the optimal GA parameter values based on iterations and the matching
index for each robustness assessment method. The combination that has the
highest matching index and the least number of iterations is the most
optimal.
The same stopping criterion is used in the optimization process for the three
robustness assessment methods with all GA parameter values. For instance, for
different population size, different generations are used to measure the average
relative change in the fitness function such that the minimum number of iterations
for the stopping criterion of the optimization process is the same in all cases.
91
Start
Max-min method
GA Parameter settings
(p)
CF = [0.5, 0.6, 0.7, 0.8]
PF =[0.2, 0.3, 0.4, 0.5] Calculate Select the parents
TS =[2, 3, 4, 5] performance based on objective
PS = [20, 30, 40] spread of design function to create
population new generation
p =1 Scenario
sample using
a sampling
Run optimization multiple times (r) r=1
Best-case and worst-case method
method
No Calculate YES
Evaluate performance of Choose a Select the parents
p=p+1 r=r+1 performance
If r=5 Create initial generation for robustness based on objective Optimization
g=0 deviation for new
population considered scenarios assessment function to create criteria is met?
population and
using BES model method new generation
Yes design archive
Design archive = [ ] NO
No
If p=192
Create new generation Minimax regret method
based on crossover,
Yes
mutation etc.
Evaluate convergence and matching
index for a setting for all robustness Calculate
Select the parents
assessment methods performance
g= g+1 based on objective
regrets for new
function to create
population and
Find the optimal setting for all new generation
design archive
robustness assessment methods
Update design archive for all generations

Stop
p = GA parameter combinations
r = multiple runs
g = generation
Figure 4.13 A method to find the optimal settings for GA parameters for different robustness assessment methods.
92
The number of iterations required to meet the optimization criterion with different
GA parameters for three robustness assessment methods is shown in Figure 4.14.
The range of the box plot for a GA parameter value is caused by other GA parameters
and stochasticity (5 runs). It is evident from Figure 4.14 that the large population size
(PS=40) requires more iterations to meet the optimization criterion compared to
other population sizes. A population size of 20 requires least iterations to meet the
optimization criterion. Conversely, the matching index with the lower population
size of 20 is significantly lower compared to that of higher population size, as shown
in Figure 4.15. A population size of 30 is an optimal trade-off between iterations and
the matching index.
Similarly, a low Pareto fraction requires high iterations for all three methods. A high
Pareto fraction leads to faster convergence because optimization reaches a local
optimum with high Pareto fractions. Contrariwise, a low PF requires more iterations
as it tries to reach the global optimum. High Pareto fractions are optimal if the design
archive is considered when evaluating objectives, which can be justified by a higher
matching index (see Figure 4.15) for deviation and regrets with a PF of 0.5.
Furthermore, in the case of high PF, other parameters have a limited effect (small
range of boxplot as seen in Figure 4.14 and Figure 4.15) on the number of iterations
required and on the matching index. It can be concluded that higher Pareto fractions
and lower crossover fractions are optimal values for GA parameters for optimization
using the best-case and worst-case method and the minimax regret method.
The matching index is improved up to 90% on average for all methods when the
optimal values for the GA are used. A matching index as high as 100% is achieved
in the case of the best-case and worst-case method. The optimal settings and
corresponding computational cost reductions for both decision makers are shown in
Table 4.3. It can be concluded from Figure 4.14 and Figure 4.15 that higher Pareto
fractions and lower crossover fractions are optimal values for GA parameters for
optimization using the best-case and worst-case method and the minimax regret
method. The optimal values for GA parameters for the optimization using the max-
min method are close to the MATLAB default values. The optimal settings for GA
parameter values are determined for the same case study with a homeowner as a
decision maker. The matching index is improved up to 90% on average for all
methods, except for the max-min method in the case of the homeowner, which is
slightly above that of the default values. The optimal settings and corresponding
computational cost reductions for both decision makers are shown in Table 4.3.
93
Spread Deviation Maximum regret
Figure 4.14 Number of iterations required to meet the optimization criteria for different GA
parameter values for three robustness assessment methods. Each box consists of all values
of other parameters and multiple runs.
94
Figure 4.15 Matching index of different GA parameters for three robustness assessment
methods. Each box consists of all values of other parameters and multiple runs.
95
Among the three robustness assessment methods, optimization using the max-min
method requires more iterations than the other two methods to meet the
optimization criteria, as depicted in Figure 4.14. This difference in convergence can
be attributed to the calculation approach of the objectives in these methods. For
instance, when comparing an optimization run with default settings using spread
(Figure 4.16) and maximum regret (Figure 4.17) as robustness indicators, it can be
observed that optimization using spread took about 62 (1860 iterations) generations
to converge, whereas optimization using maximum regret took only 23 (690
iterations) to converge. In these figures, the Pareto front grouped for five generations
is shown separately for actual performance and performance robustness, and also
different y-axis scales are used for visualization purpose.
The difference in convergence rates for these two methods is because in the max-
min method, robustness (spread) is optimized with respect to the best performing
scenario of a design and there is no inter-comparison of designs. Therefore, actual
performance and robustness do not necessarily follow the same trend, but often
conflict as observed in Figure 4.16. For instance, a design with very high CO2
emissions can have the least spread across scenarios. In addition, the spread of a
design population of a particular generation is quite scattered. Therefore, the max-
min method requires a higher number of generations to converge.
In contrast, in the minimax regret method, robustness (maximum regret) of a design

is optimized with respect to the best performing design in that particular generation.
The design with the optimal performance will have the least maximum regret, and
thus both the actual performance and performance robustness follow the same
trend, as demonstrated in Figure 4.17. Furthermore, for each generation, objectives
are calculated considering the current population and design archive of previous
generations due to the inter-comparison of the performance of all designs. Since the
GA takes design history into account, the population at the next generation depends
on the entire design archive instead of on the previous generation. Thus, the relative
change in the fitness function reduces as the Pareto front converges more quickly
with an increase in generations, resulting in less iterations required to meet the
optimization criteria. Therefore, the minimax regret method requires a lower
number of iterations to converge compared to the max-min method.
Similar observations can be made for the best-case and worst-case method as the
design archive is considered in this method as well in the calculation of robustness.
96
3500
3000
CO2 emissions, kgCO2/a 2500
2000
1500
1000
500
-500
15 20 25 30 35 40 45 50
Pareto solutions g1-g5 g6-g10
g11-g15 g16-g20 g21-g25
g26-g30 g31-g35 g36-g40
g41-g45 g46-g50 g51-g55
g56-g60 g61-g62
2000
1500
Spread of
1000
500
0
15 20 25 30 35 40 45 50
Pareto solutions g1-g5 g6-g10

g11-g15 g16-g20 g21-g25
g26-g30 g31-g35 g36-g40
g41-g45 g46-g50 g51-g55
g56-g60 g61-g62
Figure 4.16 Variation of the Pareto front across different generations of an optimization
run with default settings using the max-min method. The top graph shows actual
performance and the bottom graph shows robustness (spread).
97
3500
3000
2500
2000
1500
1000
500
-500
15 20 25 30 35 40 45 50
Pareto solutions g1-g5 g6-g10 g11-g15 g16-g20 g21-g23
4000
3500
3000
Maximum regret of
2500
2000
1500
1000
500
0
15 20 25 30 35 40 45 50
Pareto solutions g1-g5 g6-g10 g11-g15 g16-g20 g21-g23
Figure 4.17 Variation of the Pareto front across different generations of an optimization
run with default settings using the minimax regret method. The top graph shows actual
performance and the bottom graph shows robustness (maximum regret).
98
4.5.4 Computational cost savings with sampling strategies and multi-objective

optimization
The reduction in computational costs is calculated by comparing the computational

costs arising from using the optimal settings with the computational costs of using
the full factorial approach. It can be observed from Table 4.3 that, for all three
methods on average, when using optimal settings, a matching index of 90% is
achieved at less than 0.1 % and 4-5% of the computational cost of a full factorial
assessment with all scenario combinations and low-high scenario combinations,
respectively. This is valid for all methods except for the max-min method with
homeowner as the decision maker. Similarly, the matching index is improved by up
to 22-28% at more or less same number of iterations by using optimal settings
compared to using default MATLAB values.
4.5.5 Validation
The improved Pareto fronts with optimal settings are compared in Figure 4.18 with
those of the default values in reference to true Pareto fronts. The blue bubbles
represent true Pareto fronts and red bubbles represent Pareto fronts for the
corresponding optimal and default settings. It is noteworthy that the range of
maximum regret of CO2 emissions is slightly different for Pareto fronts optimized
with optimal settings of GA parameters, as seen in Figure 4.18. This difference is
attributed to the optimal settings yielding more Pareto solutions compared to the
Pareto solutions calculated using the full factorial approach. Accordingly, the
different values of maximum regret for the same design are due to the inter-
comparison of designs in the corresponding Pareto front for the maximum regret
calculations (see Table 4.4).
It can be seen from Figure 4.18 that when using default values, there is a risk of
losing some robust designs. Conversely, with the optimal settings this risk is reduced
to a significant extent, especially with the best-case and worst-case method and
minimax regret method. However, the selection of optimal GA parameter settings
depends on whether these parameters lead to the same robust design as the full
factorial approach. For instance, this can be validated by comparing the robust design
obtained using the optimal settings with the equivalent design obtained using the
full factorial approach. However, the choice of robust design depends on the
preference of decision makers and their approach towards risk in the decision-
making process. To simplify the selection process, the Hurwicz criterion is used to
99
identify the most robust design using a full factorial approach and a GA-based
optimization for three robustness assessment methods with their corresponding
optimal settings.
Robust designs for the policymaker using the full factorial (FF) approach and GA-
based optimization for three methods with their corresponding optimal settings are
tabulated in Table 4.4. It can be observed that both optimization methods (FF and
GA-based optimization) result in the same robust design for the corresponding
robustness assessment method, indicating that GA-based optimization with optimal
settings is valid and can be used to reduce computational time without
compromising the outcome. Therefore, these optimal settings are used in this
research for optimization.
It is noteworthy that the optimal settings largely depend on the considered design
space and the objectives of optimization as the settings are different for the three
robustness assessment methods and also the two decision makers (see Table 4.3).
Therefore, this optimization method is case study dependent. In practice, it may not
be feasible to determine optimal settings for each case study, thus alternative
optimization methods such as hyper-heuristic based GA optimization, which
optimizes GA parameters in addition to design space during an optimization
process, may be desirable. These hyper-heuristic optimization methods are well
implemented in other fields [Kumari et al., 2013; Cowling et al., 2002], but rarely
exploited in the building performance context. However, this implementation
deviates from the scope of this research and could be a useful area of investigation
in future work.
100
Table 4.3 Optimal settings of GA parameter values selected based on the matching index and iterations for this case study with policymaker
and homeowner as decision makers. Henceforth, these settings are used for optimization in this research.
Optimal settings
Default
Parameter Policymaker Homeowner
settings
Spread Deviation Maximum Spread Deviation Maximum
regret regret
Crossover fraction (CF) 0.8 0.7 0.5 0.5 0.8 0.8 0.8
Pareto fraction (PF) 0.35 0.5 0.3 0.5 0.5 0.5 0.4
Selection function (TS) 4 4 4 4 5 2 4
Population size 30 40 30 30 40 30 40
ULH scenario sample 100 100 100 100 100 100 100
Matching index (%) 66 – 71.5 88.8 100 91.3 71 92 91.6
Iterations required to meet
optimization criteria 744 – 876 945 744 798 810 786 760
(averaged over multiple runs)
Computational cost reduction
compared to FF with low- 94.72-
94.30 95.52 95.19 95.12 95.26 95.42
high scenario combinations 95.52
(%)
Computational cost reduction
compared to FF with all 99.91 99.90 99.92 99.92 99.91 99.92 99.92
scenario combinations (%)
101
Table 4.4 Robust designs selected using the Hurwicz MCDM method for three robustness assessment methods using the full factorial approach
(FF) and GA-based optimization with corresponding optimal settings.

Design variants Optimal Optimal Optimal
FF FF FF
GA GA GA
Rc-Wall, m2K/W 4.5 4.5 4.5 4.5 4.5 4.5
Rc-Roof, m2K/W 6 6 6 6 6 6
Rc-floor, m2K/w 3.5 3.5 3.5 3.5 3.5 3.5
Windows U value, W/ m2k 1.43 1.43 1.43 1.43 1.43 1.43
WWR (%) 20 20 20 20 20 20
Light- Light- Light- Light- Light- Light-
Thermal mass
weight weight weight weight weight weight
Infiltration, ach 0.12 0.12 0.12 0.12 0.12 0.12
PV size, m2 20 20 30 30 30 30
SDHW, m2 0 0 0 0 0 0
Additional investment cost, k€ 21.6 21.6 24.7 24.7 24.7 24.7

CO2 emissions, kgCO2/a (median value across
1506 1506 522 522 522 522
considered scenarios)
Performance robustness of CO2 emissions,
843 843 1669 1669 1274 976
kgCO2/a
102
Default settings Optimal settings

i. Spread
Bubble size = Spread of CO2 emissions (523-1121 kgCO2/a) Bubble size = Spread of CO2 emisisons (523-1121 kgCO2/a)
3500 3500
3000 3000

2500 2500
2000 2000
1500 1500
1000 1000
500 500
0 0
-500 -500
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
ii. Deviation
Bubble size = Deviation of CO2 emissions (971-3785 kgCO2/a) Bubble size = Deviation of CO2 emissions (971-3785 kgCO2/a)
3500 3500
3000 3000

2500 2500
2000 2000
1500 1500
1000 1000
500 500
0 0
-500 -500
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
iii. Maximum regret

Bubble size = Maximum regret of CO2 emissions (182-3740 Bubble size = Maximum regret of CO2 emissions (256-3740kgCO2/a)
kgCO2/a)
3500 3500
3000 3000
2500 2500
2000 2000
1500 1500
1000 1000
500 500
0 0
-500 -500
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
True Pareto front calculated using full factorial approach True Pareto front calculated using full factorial approach
Pareto front calculated uisng default GA parameters settings Pareto front calculated using corresponding optimal settings of GA parameters
Figure 4.18 Comparison of the optimal Pareto front using default MATLAB values and
corresponding optimal settings and a ULH sample of 100 scenarios with the true Pareto
front obtained using the full factorial approach for three robustness assessment methods.
Blue bubbles represent the true Pareto front and red bubbles represent an optimal Pareto
front with corresponding settings. It can be noted that Pareto solutions of true Pareto front
are overlapped by Pareto front calculated using default settings of GA parameters.
103

A simulation framework was developed to test the CPRA approach. GA Multi-
objective optimization in combination with the ULH scenario sampling method
were implemented in this framework to enhance its usability in practice. The fitness
function of the GA was modified to implement robustness indicators as one of the
objectives of the optimization problem. The optimal settings for GA parameters were
determined for each robustness assessment method.
The following conclusions are drawn from the developed simulation framework and
the results of case study considered in this chapter:
⎯ The method of calculation of objective function differs for different

robustness indicators. For instance, spread is calculated for each design of
the population without any inter-comparison of the performance of other
designs of the population, and thus one design at a time is considered when
calculating the objectives. Contrariwise, maximum regret and deviation are
calculated with inter-comparison of the performance of other designs, and
thus objectives are calculated after the calculation of the performance of the
entire population. Furthermore, for each generation, objectives are
calculated considering the current design population and design archive of
previous generations because of the need for inter-comparison of the
performance of all designs. In such cases, the typical GA fitness function
may not be suitable; instead, a modified fitness function as presented in this
research may be used.
⎯ In the max-min method, robustness (spread) is optimized with respect to

the best performing scenario of a design, whereas in the best-case and worst-
case method robustness (deviation) is optimized with respect to the best
performing case of all designs and scenarios. In the minimax regret method,
robustness (maximum regret) is optimized by minimizing the maximum
performance difference over all scenarios between the performance of a
design and the best performance of the corresponding scenario.
⎯ Low-high scenario combinations were sufficient for performance

robustness assessment, which can reduce computational time for this case
study by about 98% compared to assessment with all scenario
combinations.
104
Concluding remarks
⎯ The ULH sampling strategy was chosen for scenario sampling on low-high
scenarios due to its fast convergence. Using the ULH sampling strategy, a
sample of 100 scenario combinations was the smallest sample size that
yields similar performance robustness as that of low-high scenario
combinations for the considered performance indicators of this case study.
⎯ The matching index of a Pareto front can be improved up to 90% on average

with optimal settings compared to default values for different robustness
assessment methods. A matching index of 100% was achieved for the best-
case and worst-case method.
⎯ The simulation framework implemented in this research using scenario

sampling and multi-objective optimization methods could save up to 94-
99.9% of computational costs for the considered case study compared to a
full factorial approach with all scenario combinations, which require
millions of simulations.
The developed simulation framework was used to test the CPRA approach using
case studies that are described in the next chapter. The test results are described
in Chapter 6. A ULH sample of 100 scenarios and optimal settings of GA
parameters for corresponding robustness assessment methods are used to test
the CPRA approach using case studies.
105
5. Description of case studies
and decision makers’
preferences
This chapter describes the case studies used to test the CPRA approach developed in
Chapter 3 using the simulation framework described in Chapter 4. Based on decision
makers’ preferences, the design option space, future scenarios and performance indicators
are described in this chapter. The simulation models of the case study buildings are also
presented.
5.1 Decision makers

The developed CPRA approach is demonstrated using case studies for the following
two key decision makers:
• Policymakers – prefer robust designs with low CO2 emissions and low
investment costs to enable the policy of providing subsidies for the
implementation of CO2 reduction measures for end users.
• Homeowners – prefer robust designs that have the potential to deliver a

comfortable indoor environment with low operational costs and investment
costs.
5.2 Case studies

In the Netherlands, the housing stock comprises 7.6 million dwellings as of 2016
[CBS, 2016c]. This stock constitutes different dwelling types; Figure 5.1 shows the
dwelling types as of 2005, which is the most recent data available [Koninkrijksrelaties
Ministerie van Binnenlandse Zaken, 2016]. It can be seen from Figure 5.1 that the
terraced house is the most common dwelling type in the Netherlands; it represents
43% of the total housing stock. As discussed in Chapter 1, based on the EPBD
directive [EPBD, 2010], the Netherlands has set a national target of achieving energy
label B, averaged over the housing stock, by the end of 2020 [Hermelink et al., 2013]
and an energy neutral housing stock by 2045. Around 77% of the labelled terraced
houses currently have energy label C or lower (see Figure 5.2).
107
Chapter 5. Description of case studies and decision makers’ preferences
x 100000 10
9
8
7
Number of dwellings
6
5
4
3
2
1
0
Detached Semi-detached Terraced Apartment Others (Deck
(Townhouse) access,
Maisonette and
flats)
until 1945 1945-64 1965-74 1975-91 1992-2005
Figure 5.1 Different dwelling types in the Netherlands as of 2005 (Source: ABF research
[Koninkrijksrelaties Ministerie van Binnenlandse Zaken, 2016]).
100% G
with difefrent energy labels
Percentage of dwellings
80% F
E
60%
D
40%
C
20%
B
0% A
Terraced house
(Townhouse)
Figure 5.2 Percentage of labelled terraced houses with different energy labels as of 1 January,
2017 [Compendium voor de Leefomgeving, 2017].
Figure 5.3 Examples of a terraced house built in 1975-91 (left image) and newly built
terraced house (right image) (Source: Agentschap-NL [Agentschap NL, 2011; Agentschap
NL, 2013a]).
108
Case studies
Therefore, most of these terraced houses need renovation to meet the Dutch national
policy targets. Since the majority of these terraced houses were built during 1975-91,
terraced houses built from this period were selected for a renovation case study
building for reasons of representativeness.
In addition, around 45,000-50,000 new houses are being built every year
[Koninkrijksrelaties Ministerie van Binnenlandse Zaken, 2016], which represents
approximately 0.6% of the existing building stock [Filippidou et al., 2016]. Therefore,
determining the type of new house to use as the case study building had the potential
to turn into a complex task. To simplify this selection problem, the current research
turned to six reference buildings proposed by Agentschap-NL [Agentschap NL,
2013a]. Similar to the renovation case study building, the terraced house was chosen
for the new house case study from among these six reference buildings for reasons
of comparability. These buildings are described below, and simulation models of
these buildings are described in the next sections. Examples of these two buildings
are shown in Figure 5.3.
5.2.1 Renovation house case study

The renovation house case study is a three-story building with a gross surface area of
153 m2 and a treated floor area of 123 m2(see Table 5.1). This house is constituted by
heavy-weight floor, wall and roof constructions. The walls are made of bricks, prefab
concrete and insulation. The ground floor is made of concrete with insulation. The
insulation levels for the façade and the roof were about Rc = 1.3 m 2K/W in 1975 and
were increased up to 2.0 m2K/W in 1988. This house has identically sized windows
on the north and south façades and a small window on the west façade. The total
window area is about 27.4 m2, which is mostly double glazing and a small portion of
single glazing.
The house is heated by a central heating system using a high efficiency (HR107)
natural gas boiler. Domestic hot water needs are also met by the HR107 boiler. The
house is ventilated with mechanical extraction and also natural ventilation. There is
no heat recovery system coupled with the mechanical ventilation system. Active
cooling systems are not present in this house. There are no renewable energy
generation and storage systems installed in the house.
109
5.2.2 New house case study

The new house case study is also a three-story building with a gross surface area of
124 m2 and a treated floor area of 104 m2. Similar to the renovation house, this house
is also constituted by heavy-weight floor, wall and roof constructions. The layout of
this house is shown in Figure 5.4 and its dimensions are tabulated in Table 5.1. The
external walls consist of brick, air cavity, insulation and brick. The ground floor is
made of wood, concrete and insulation. The roof is comprised of wood, insulation,
prefab concrete, and roof tiles. The insulation levels for the façade and ground floor
are approximately Rc= 3.5 m2K/W and for the roof it is approximately 4 m2K/W.
Internal walls are made of gypsum board (12mm), insulation (75mm) and gypsum
board (12mm). The north and south walls have identically sized windows of about 12
m2 on each façade. In addition, a window of 1.4 m2 is located on the roof, facing north.
All these windows are of the highly efficient HR++ glazing type with a U value of
1.65 W/m2K.
This house is centrally heated by a HR107 natural gas boiler and ventilated using
balanced mechanical ventilation with a heat recovery unit with an efficiency of 95%.
A combi-boiler is used for domestic hot water needs. In addition, a solar thermal
collector system of 2.3 m2 is used to provide hot tap water. There are no active cooling
systems, photovoltaic (PV) systems or energy storage systems installed in this
building.
Table 5.1 Dimensions of each zone of the renovation house and the new house case studies.
Parameter Renovation house New house
Length, m 6 5.1
Width, m 11 8.9
Height, m 2.6 2.86
Ground floor (Zone-1), m2 66 46.2
First floor (Zone-2), m2 66 45.5
Attic (Zone-3), m2 21 32.6
Glazing, m2 27.4 25.4
Gross area, m2 153 124
Treated floor area, m2 123 104
110
Case studies
Figure 5.4 Layout of the reference Dutch terraced house, showing different floors, and front
and back view of the building. All dimensions are in mm.
5.3 Design option space of the case studies

In this section, design options for the building, energy and storage systems are
described. An overview of the design space for both case studies is tabulated at the
end of this section.
111
5.3.1 Building envelope

Different building design options, as shown in Table 5.2 and Table 5.3, are varied to
form the building envelope packages for both the renovation and the new houses.
The insulation level required to achieve energy label B for the selected terraced house
is 2.5 m2K/W for the building envelope. Similarly, for new buildings, the minimum
insulation levels of floor, walls and roof required to meet the energy performance
coefficient of 0.4, as stated in the Dutch current building codes are 3.5, 4 and 6
m2K/W respectively [RVO, 2016a]. These insulation levels are further increased to 6,
7 and 7 m2K/W for nearly zero energy buildings (nZEB=BENG) and up to 6, 8.5 and
10 m2K/W for zero energy buildings (ZEN=NZEB) [RVO, 2016b; RVO, 2015b].
Similarly, the use of double glazing windows became more common in the period
between 1975-91 [Agentschap NL, 2011], and since 2015, double glazing windows
have been overtaken by HR++ glazing windows to meet the Dutch building code
requirements [Agentschap NL, 2013a; RVO, 2016a]. Furthermore, to achieve low-
energy houses, the use of triple glazing is essential to meet the set targets [RVO,
2016b]. Building airtightness (infiltration rates) has improved considerably in recent
years, which has significantly reduced heat loses in buildings [RVO, 2016a; RVO,
2016b; RVO, 2015b].
In order to satisfy different building codes and standards, building envelope

properties such as window type, infiltration rates and insulation levels for walls, floor
and roof are varied at the same time to form building envelope packages for the
renovation and the new houses (see Table 5.2 and Table 5.3). For instance, RP1 can
fulfill the energy label B requirements and RP3 can meet the current Dutch building
codes [RVO, 2016a]. Similarly, RP4 and RP5 can meet the Dutch nearly zero and
zero-energy building standards [RVO, 2016b; RVO, 2015b] respectively, and RP6-7
can meet Passivehaus standards. Similar observations can be made for new building
envelopes P1-P5. It is worth noting that RPo represents the reference renovation
house and Po represents the reference new house, which are described in section
5.2.1 and 5.2.2, respectively.
Building orientation is varied from 0-180 degrees for the new house, however, the
orientation is fixed for the renovation house, which is N-S in this case study. Window
to wall ratio is varied from 20-80% at the step size of 20% for both buildings.
Similarly, light-weight and heavy-weight constructions are considered for the new
house, whereas only a heavy-weight construction is chosen for the renovation house.
Thermo-physical properties of different building envelope materials considered in
this study are described in appendix B.
112
Design option space of the case studies
5.3.2 Heating and ventilation systems

In the Netherlands, around 97% of buildings are heated using natural gas as a
heating source [CBS, 2016b], as shown in Figure 5.5. Most of these buildings use gas
boilers for heating [Agentschap NL, 2011]. Very few buildings are equipped with
district heating systems and heat pumps. However, these heating systems are being
encouraged in future Dutch building regulations [RVO, 2016b]. According to the
Dutch heat pump association, the heat pump market is gradually increasing and it
is expected that around 500,000 heat pumps will be installed by 2020 and as many
as 1.8 million by 2030 [DHPA, 2015]. Among these heat pumps, air source and
ground source heat pumps are most widely used in the Netherlands [DHPA, 2015].
0.82% 0.34% 1.68%

0.37%
Natural gas
Oil
Electricity
Wood
96.80%
Others
High efficiency (HR 107) boiler

3% (η=107%)
High efficiency (HR 100) boiler
8% (η=100%)
Improved efficiency (VR) boiler
(η=75%-85%)
14% Conventional (CR) boilers
44%
(η=70%-80%)
Local (individul) heating (gas or
oil)
District heating
31%
Block heaters (heating for an
apartment block)
Others
Figure 5.5 Different types of heating systems (top figure) and their heating sources (bottom
figure) commonly used in the Netherlands [CBS, 2016b; Agentschap NL, 2011].
113
Therefore, a highly efficient gas boiler (HR107), an air source heat pump (ASHP)
and the ground source heat pump (GSHP) are used as heating system options (see
Table 5.2 and Table 5.3) for both the renovation and the new house case studies.
These heating systems are used to convert the dynamic heating demand in to
equivalent gas and electricity. The seasonal performance factors (SPF) of ASHP and
GSHP are assumed to be 2.7 and 3.5 respectively [ISSO, 2011a; Hermelink et al.,
2013; Fleiter et al., 2016]. The SPF is the ratio of delivered useful heat to the total
supplied energy over the seasons. SPF is used instead of coefficient of performance
for these heat pumps to assess overall annual performance rather than nominal
performance [DHPA, 2015]. District heating is not considered as it falls outside the
scope of this study. To reduce the heating demand of different design options, a
balanced mechanical ventilation system with a heat recovery unit is used. However,
the reference renovation house is ventilated solely by a mechanical ventilation system
as it was not equipped with a heat recovery unit.
5.3.3 Renewable energy and storage systems

Both reference houses have no photovoltaic (PV) system, however, the new house
has a solar thermal system of 2.3 m2 installed for domestic hot water applications.
Installation of PV systems is being highly encouraged and becoming customary to
achieve low-energy houses based on current and future Dutch building regulations.
In these regulations, a PV system of 5-25.5 m2 is recommended, depending on the
regulations [RVO, 2015b; RVO, 2016b]. Therefore, renewable energy systems such
as solar domestic hot water systems and PV systems are considered in this study. A
standalone solar thermal collector system with an auxiliary heater is used to meet
domestic hot water (DHW) needs. A storage tank of 200 liters with an auxiliary
immersion heater of 2KW capacity is used in this study. Different sizes of solar
thermal and PV collectors considered in the design space are shown in Table 5.2 and
Table 5.3. The size of these collectors is varied based on the typical size of each panel.
The total electricity consumption for heating, ventilation, DHW system pump and
auxiliary heater, lighting and appliances of the building is met by an onsite PV
system. A photovoltaic system with a module efficiency of 18.3% and an inverter with
a conversion efficiency of 97.5% were chosen in this study [E.ON, 2016]. Each panel
has a peak capacity of 300 Wp. Different capacities of PV systems, as shown in Table
5.2 and Table 5.3, are considered for the design option space. It is worth noting that
the maximum size of PV systems is limited by the available effective roof area.
Generally, large PV systems increase the building’s interaction with the grid due to
114
Design option space of the case studies
the excesses of energy that they produce, which can cause great stress on the grid
during summer months. Therefore, to reduce grid interaction by improving self-
consumption of electricity by the building, an electric battery storage system is used.
Different capacities of electric battery storage are considered for both the new house
and the renovation case studies. The battery capacities are chosen based on typical
capacities available in the market.
5.3.4 Overview of design option space
Table 5.2 and Table 5.3 provide a summary of the defined design space considered
for the renovation case study and the new house case study. It is noteworthy that RP0
and P0 are used as reference for the renovation and the new house case studies
respectively to find alternative robust design options.
Table 5.2 Overview of the design option space considered for the renovation house case study.
Design Options
parameter
Renovation RP0
RP1 RP2 RP3 RP4 RP5 RP6 RP7
package (Reference)
Rc-floor, m2k/W 1.3 2.53 3.5 3.5 6 5 6 10
Rc-wall, m2k/W 1.3 2.53 3.5 4.5 7 7 8.5 10
Rc-roof, m2k/W 1.3 2.53 4 6 7 8 10 10
Window U value,
5.2, 2.9 1.8 1.65 1.43 1.01 1.01 0.86 0.52
W/m2K
Window g value 0.81, 0.75 0.61 0.62 0.6 0.38 0.38 0.59 0.58
Infiltration,
1 [1, 0.625, 0.5, 0.4, 0.10]
dm3/sm2
Orientation N-S N-S
WWR (%) - [20- 80] at a step size of 20
Heating and ventilation system
Ventilation [Mechanical extraction, Balanced mechanical ventilation system with

system heat recovery]
Heating system [HR107, ASHP, GSHP]
115
Renewable energy and storage systems
PV system, m2 [0-32] at a step size of 3.2
SDHW system,
[0-10] at a step size of 2.5
m2
Orientation, (°) [0-180] at a step size of 90
Tilt angle, (°) [0-90] at a step size of 45
Electric battery
[0-20] at a step size of 4
storage, kWh
Table 5.3 Overview of the design option space considered for the new house case study.
Design parameter Options

P0
Renovation package P1 P2 P3 P4 P5
(Reference)
Rc-floor, m2k/W 3.5 3.5 6 5 6 10
Rc-Wall, m2k/W 3.5 4.5 7 7 8.5 10
Rc-roof, m2k/W 4 6 7 8 10 10
Window U value, W/m2K 1.65 1.43 1.01 1.01 0.86 0.52
Window g value 0.62 0.60 0.38 0.38 0.59 0.58
Infiltration, dm3/sm2 0.625 [0.625, 0.5, 0.4, 0.15, 0.10]
Orientation N-S [0-180] at a step size of 90
WWR (%) - [20, 40, 60, 80]
Heating and ventilation system
[Balanced mechanical ventilation system with heat

Ventilation system
recovery]
Heating system [HR107, ASHP, GSHP]
Renewable energy and storage systems

PV system, m2 [0-32] at a step size of 3.2
SDHW system, m2 [0-10] at a step size of 2.5
Orientation, (°) [0-180] at a step size of 90
Tilt angle, (°) [0-90] at a step size of 45
Electric battery storage, kWh [0-20] at a step size of 4
116
Scenarios
5.4 Scenarios
Scenarios are defined considering uncertain and influential parameters that can
impact the preferred performance indicators of the decision makers over the
building’s life-span. The following occupant, usage, policy and climate scenarios are
considered in this study.
5.4.1 Occupant scenarios

Four occupant scenarios are formulated based on Dutch household statistics [CBS,
2016c]. The first scenario, a single person, represents 37% of the Dutch households.
The second scenario, a two-person household, accounts for 33% of the Dutch
households, as shown in Table 5.4. Similarly, for occupant scenarios 3 and 4,
households of three and four persons respectively occupy the building. The main
difference between these scenarios is the heat gain due to the number of occupants
and their corresponding behavior in the building. Heat gains due to occupants are
defined based on their activity level. For instance, based on ASHRAE 55 standards
[ASHRAE, 2010] 120W of heat gain is assumed for light work in the living room,
and 100W is assumed for resting in a bedroom. Convective and radiative factors for
these heat gains are assumed to be 0.5 for each.
5%
13% 1
37% 2
12% 3
4
≥5
33%
Figure 5.6 Percentage of dwellings occupied by different household types.
5.4.2 Occupant behaviour (usage) scenarios

For each of the occupant scenarios, usage scenarios are formulated based on
occupant behavior with respect to energy use in the building. These usage scenarios
span very careful energy users to energy-wasting users; well-informed to poorly
informed users and also cover different types of equipment with low to very high
117
efficiencies. For instance, 1W/m2 average electricity use for appliances represents
very careful energy users and also highly efficient equipment. Occupancy patterns,
heating setpoint temperatures, lighting and appliance use, ventilation rates,
domestic hot water consumption and shading control are varied for usage scenarios,
as shown in Table 5.4. These occupant behavior scenarios are different for living
room and bedroom, and are hence modelled with different profiles for each zone.
5.4.2.1 Occupancy patterns and heating setpoints

Occupancy patterns and the corresponding heating setpoints are chosen based on
[Ministerie van VROM, 2009]. The evening occupancy profile represents 19% and
the all-day occupancy profile accounts for 48% of Dutch households, respectively
[Ministerie van VROM, 2009]. It is assumed that working people represent the
evening profile and retired people stay at home all-day. The heating setpoint
temperature is varied form 18-22°C during occupied hours and reduced to 14-18°C
during unoccupied hours, as shown in Table 5.4. These heating setpoint preferences
are based on careful energy users, energy wasting users and also based on occupant
age. The hourly profiles of occupancy patterns and their corresponding heating
setpoints, for both ground and first floor, are shown in Figure 5.7. It is worth noting
that the attic (second floor), which is seldom used by occupants, is not conditioned.
100% 100%
Occupancy presence
Occupancy presence
0% 0%
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time, h Time, h
Ground floor (Zone1) First floor (Zone2) Ground floor (Zone1) First floor (Zone2)
24 24
Heating setpoint
temperatture, °C
Heating setpoint
22
temperature, °C
22
20 20
18 18
16 16
14 14
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time, h Time, h
Ground floor (Zone1) First floor (Zone2) Ground floor (Zone1) First floor (Zone2)
Figure 5.7 Occupancy patterns and their corresponding preferred heating setpoint range for
an usage scenario. Left figures represent people being home all-day (so-called all-day
profiles) and right figures represent evening profiles.
118
Scenarios
5.4.2.2 Appliance use, lighting use and corresponding internal heat gains
Three scenarios are considered for average electricity use for lighting and appliances,
as shown in Table 5.4. Each scenario has a similar usage profile for occupancy pattern
but differs in peak loads, resulting in different average electricity consumption. For
the average scenario, electricity consumption for lighting [Agentschap NL, 2013c;
RVO, 2015a] and appliances [Papachristos, 2015] is in line with the average electricity
consumption of Dutch households of about 3500kWh for lighting and appliances
[CBS, 2016a]. Internal heat gains due to lighting and appliances is varied in
combination with appliance use and lighting use, from 2 to 6 W/m2 based on
[NEN7120, 2011; Hoes et al., 2011].
14
Appliance use and correposning
12
internal heat gains, W/m2
10
8
6
4
2
0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time, h
Ground floor (Zone1) First floor (Zone2) Average
14
Lighting use and corresponding
12
internal heat gains, W/m2
10
8
6
4
2
0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time, h
Ground floor (Zone1) First floor (Zone2) Average
Figure 5.8 Appliance use, lighting use and their corresponding internal heat gains (IHG)
for an evening occupancy profile with an average energy usage behavior.
119
Lighting, appliance use, and their corresponding internal heat gains are triggered in
proportion to hourly occupancy profiles and reduced to base load (standby mode)
when idle. An example of this assumption is given for the average usage scenario for
evening occupancy in Figure 5.8. For instance, internal heat gains (IHG) due to
appliance use and lighting use is generally higher in the morning and evening due
to occupants’ activity. The IHG profiles for each zone are based on [ISSO, 2011b].
The convective factor of all internal heat gains is 0.5 and for lighting it is about 0.6.
The remainder of the internal heat gains in all cases is transmitted by long-wave
radiation.
5.4.2.3 Domestic hot water use

Domestic hot water (DHW) consumption is varied from 40L/day to 100L/day per
occupant for different usage activities based on [NEN7120, 2011; NEN7120+C2, 2012]
and [Guerra-Santin and Silvester, 2016]. DHW usage for different purposes for an
average user and the profile of DHW usage for three scenarios are shown in Figure
5.9.
60
DHW usage, LPD per person
50
40
30
20
10
0
Bath Shower Sink Clothing Dish wash
(hand wash) (hand)
60
DHW usage, LPD per person
50
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time, h
40 60 100
Figure 5.9 Domestic hot water usage for different purposes (average usage scenario) and
DHW profiles for three usage scenarios.
120
Scenarios
The DHW usage profiles are based on different usage patterns used in the
Netherlands surveyed by [NEN7120, 2011] to determine the generation efficiency of
domestic hot water devices. The major difference between these DHW usage profiles
is the duration of shower time. It is worth noting that the hot water for the
dishwasher and washing machine is assumed to be supplied by the in-built heating
systems of these machines.
5.4.2.4 Ventilation by window opening and mechanical system

A minimum ventilation rate of 0.9 ach, as decreed by Dutch building regulations, is
maintained in the building to improve the indoor air quality, regardless of infiltration
rates. The ventilation rate is increased up to 1.5 ach for the high usage scenario during
occupied hours [Hoes et al., 2011].
To reduce overheating during summer, the building is cooled through natural

ventilation by opening windows. The maximum achievable air change rates through
windows largely depends on occupant behavior [Yun et al., 2009], among other
factors. Therefore, two window opening scenarios based on the occupant’s level of
activity are considered [Plas, 2017]. In the fully opened scenario, a maximum
ventilation rate of 5 ach [Hamdy and Hensen, 2015] is assumed, which is reduced to
1 ach in the case of the partial opening scenario [Mulville and Stravoravdis, 2016] (see
Table 5.4). These air change rates through window opening are in addition to
mechanical ventilation rates and infiltration rates in the building.
5.4.2.5 Shading control

The windows have external shading devices to reduce glare and overheating in
summer. However, shading devices should not reduce passive solar gains during
cold days. Therefore, shading control (by occupants) of external shading devices of
windows is implemented based on radiation levels on the façade and on indoor
temperature [Hoes, 2014], as shown in Table 5.4. For instance, for an average
scenario, if the indoor temperature is lower than 24°C, solar shading is lowered if
the radiation on the façade is more than 300 W/m2 and is raised as soon as the
radiation falls below 250 W/m2. The solar shading is lowered if the indoor
temperature is higher than 24°C, irrespective of radiation levels, to avoid overheating
in the building.
121
5.4.3 Policy scenarios

A net-metering system is currently used in the Netherlands to measure net annual
energy consumption, i.e., the difference between the energy imported by a building
from the grid and the energy exported from a building’s onsite energy generation
system to the grid. In the current policy, the energy imported and exported are
equally priced up to 3500 kWhe/a delivered to the grid. Beyond this limit, the price
is 0.07 €/kWhe. However, since so many buildings have taken advantage of this
option to sell excess energy to the grid, the grid is often oversupplied and under great
stress in summer months. Therefore, it seems probable that this current net-
metering pricing model will be terminated in the future [KEMA, 2016; RVO, 2015c]
and hence, net-metering scenarios are considered that represent business as usual
and the termination of the current net-metering policy.
5.4.4 Climate scenarios

Four scenarios for future climate change in the Netherlands [van den Hurk et al.,
2006] are used in this study. Climate change scenarios are based on global mean
temperature rise and changes in atmospheric air circulation patterns in comparison
to values in 1990. In these four climate change scenarios, scenario G represents a
moderate increase of global temperature of +1°C in 2050, whereas scenario W
represents an extreme case of an increase of +2°C in 2050 relative to 1990. Scenarios
G and W do not consider changes in air circulation patterns. In contrast, scenarios
G+ and W+ include changes in air circulation patterns along with a rise in global
mean temperature, as shown in Figure 5.10. These climate change scenarios were
updated in 2014 [KNMI, 2014] based on two emission scenarios – RCP 4.5
representing stabilization, and RCP 8.5 representing a high emission scenario
(Figure 5.10). These updated scenarios include changes in sea level, precipitation and
radiation among other parameters, in addition to changes in global temperature and
air circulation patterns. However, the air circulation patterns and global temperature
in 2050 for the scenarios proposed in 2014 are the same as in the scenarios proposed
in 2006, as seen in Figure 5.10.
Hence, the weather files that are developed based on the climate change scenarios
proposed in 2006 are used in the current study. It is noteworthy that performance
prediction with 2014 scenarios could differ with that of 2006 due to changes in the
radiation parameter in 2014 scenarios.
122
Scenarios
Figure 5.10 Change in global temperature predicted for four climate change scenarios in
2050, compared to 1990 (left) [van den Hurk et al., 2006] and 1981-2010 (right) [KNMI,
2014].
In addition to climate change scenarios, a typical climate reference year, NEN 5060
[NEN, 2008], is considered (Table 5.4), which is based on average months of 20 years
of historical weather data.
5.4.5 Overview of scenarios

Table 5.4 shows the summary of occupant, usage, policy and climate scenarios
considered in this research. It is worth noting that some of the scenarios marked
with * are varied together. For instance, internal heat gains due to lighting and
appliances are varied in proportion with electricity use for lighting and appliances.
An ULH sample of 100 low-high scenario combinations is used for performance
robustness assessment of the design space.
123
Table 5.4 Summary of future scenarios that consider uncertainties in households, occupant
behavior, policy changes and climate change.
Parameter Options References
Occupant scenarios
Household size [1, 2, 3, 4] [CBS, 2016c]
Occupant behavior (usage) scenarios

Occupancy profile (OP) [Evening, All-day] [Ministerie van VROM,
2009]
Heating setpoint (Thsp) (occupied), °C [18, 20, 22] [Ministerie van VROM,
2009]
Heating setpoint (Thsp) (un-occupied) *, °C [14, 16, 18] [Ministerie van VROM,
2009]
Average electricity use for lighting (Luse), [CBS, 2016a;
[1,2,3] Agentschap NL, 2013c;
W/m2 RVO, 2015a]
Average electricity use for appliances (Ause), [CBS, 2016a;

[1,2,3] Papachristos, 2015]
W/m2
[NEN7120, 2011; RVO,
Internal heat gains due to lighting and 2015a; Agentschap NL,
[2, 3, 4, 5, 6] 2013c; Papachristos,
appliances (IHG)*, W/m2 2015; CBS, 2016a;
Hoes et al., 2011]
[NEN7120, 2011;
Domestic hot water consumption (DHW), NEN7120+C2, 2012;
[40, 60, 100] Guerra-Santin and
LPD per occupant Silvester, 2016]
Ventilation, ach [0.9, 1.2, 1.5] [Hoes et al., 2011]
[5 ach at full opening, 1 [Hamdy and Hensen,

Window opening, ach 2015; Mulville and
ach at partial opening] Stravoravdis, 2016]
Shading control ON if radiation is above

[250, 300, 350] [Hoes, 2014]
W/m2 and if Tindoor >24°C
Shading control OFF if radiation is below*,
[200, 250, 300] [Hoes, 2014]
W/m2 and if Tindoor <24°C
Policy scenarios
[KEMA, 2016; RVO,
Net-metering (NM) [Yes, No] 2015c]
Climate scenarios
[NEN5060, G, W, G+, [NEN, 2008; van den
Reference climate and climate change Hurk et al., 2006]
W+]
* This scenario is coupled with the previous scenario.
124
Performance indicators
5.5 Performance indicators

The following performance indicators are defined based on the decision makers’
preferences. For instance, a policymaker prioritizes low or no CO2 emissions
associated with a building design, but not at the expense of high investment costs.
In contrast, a homeowner prioritizes designs with comfortable indoor environment
at low costs.
5.5.1 CO2 emissions

CO2 emissions are calculated based on energy imports and exports by the building
to and from the grid. An emission factor of electricity (𝑓𝐶𝑂2,𝐸𝑙 ) of 0.540 kgCO2 per
kWh of electricity [Papachristos, 2015] is used to calculate CO2 emissions due to
electricity imports (𝐸𝑙𝑖𝑚𝑝 ) and avoided CO2 emissions due to electricity exports
(𝐸𝑙𝑒𝑥𝑝 ). In the case of the natural gas boiler as the heating system, an emission factor
( 𝑓𝐶𝑂2,𝑁𝐺 ) of 0.203 kgCO2/kWh is used to calculate CO2 emissions due to gas
consumption (𝐸𝑁𝐺 )⁡[Maas, W., Zijlema, 2017]. There is no distinction between the
emission factors for imported and exported electricity, as the exported electricity is
assumed to replace equivalent electricity production by the grid. However, the effect
of the net-metering scenario is considered in the CO2 emission calculations.
If net-metering is available, both imported and exported electricity is considered in

the CO2 emission calculations (see equation 5.1) and if net-metering is terminated,
exported electricity is discarded in the CO2 emission calculations (see equation 5.2).
Net-metering is available:
𝐶𝑂2 = (𝐸𝑙𝑖𝑚𝑝 × 𝑓𝐶𝑂2,𝐸𝑙 + 𝐸𝑁𝐺 × 𝑓𝐶𝑂2,𝑁𝐺 ) − (⁡𝐸𝑙𝑒𝑥𝑝 × 𝑓𝐶𝑂2,𝐸𝑙 )⁡⁡⁡⁡⁡⁡⁡(5.1)
Net-metering is terminated:
𝐶𝑂2 = 𝐸𝑙𝑖𝑚𝑝 × 𝑓𝐶𝑂2,𝐸𝑙 + 𝐸𝑁𝐺 × 𝑓𝐶𝑂2,𝑁𝐺 (5.2)
It is worth noting that the embodied emissions are not included in CO2 emission
calculations and thus CO2 emissions in this study are only operational CO 2
emissions.
125
5.5.2 Additional investment cost (ICa)

The investment cost of a design is the sum of the investment costs of the design
options such as insulation packages, windows for different WWRs, SDHW system,
PV system and heating and ventilation systems. Fixed costs for all designs e.g. land,
labor etc. are not considered. In the case of the renovation house case study, the
investment cost of a design is the additional amount required by a design compared
to the reference renovation house (RP0). Similarly, in the case of the new house case
study, the investment cost of a design is the additional amount required by a design
compared to the reference new house (P0). Hence, in this work investment cost is
referred to as additional investment. The details of the investment cost of all design
options are presented in Appendix C.
𝑗
𝐴𝑑𝑑𝑖𝑡𝑖𝑜𝑛𝑎𝑙⁡𝑖𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡⁡𝑐𝑜𝑠𝑡⁡(𝐼𝐶𝑎 ) = ∑ 𝐼𝐶𝑗 − 𝐼𝐶𝑅 ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.3)

𝑖=1
where 𝐼𝐶 is the investment cost of the different design options, 𝐼𝐶𝑅 is the investment
cost of the reference house (i.e. for the new house it is P0 and for the renovation
house it is RP0). The index i represents a design option and j represents total number
of design options.
5.5.3 Global cost (GC)

Global cost, which comprises investment, replacement, maintenance and operating
costs, is calculated to predict the future financial implications of designs and, thus,
to find the cost-optimal solutions. These costs are calculated for a 30-year period
[EPBD, 2010], because assumptions on interest rates and energy prices are difficult
to forecast beyond this period [BPIE, 2010].
Global cost is calculated by the following equation [Hamdy et al., 2013]:

𝑗 𝑗
𝐺𝑙𝑜𝑏𝑎𝑙⁡𝑐𝑜𝑠𝑡⁡(𝐺𝐶) = ∑ 𝐼𝐶𝑎,𝑗 + ∑ 𝑅𝐶𝑗 + 𝑂𝐶 + 𝑀𝐶⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.4)

𝑖=1 𝑖=1
where⁡𝐼𝐶𝑎 is the additional investment cost of the different design options, 𝑅𝐶 is the
replacement cost of the building components and energy systems that have a life-
span of less than 30 years, 𝑂𝐶 is the operating cost, and MC is the maintenance costs.
Fixed costs are excluded in this study since they are the same for all designs. The
index i represents a design option and j represents total number of design options.
126
To calculate the net present value of these future costs, replacement and operating
costs are discounted using real interest rates and escalation in energy prices. A real
interest rate⁡(𝑟) of 2.8% is assumed based on the average real interest rate in the
Netherlands over the past 20 years. Similarly, an energy price escalation rate (𝑒) of
1.1% is assumed based on the average rise in electricity price in the Netherlands for
the past 10 years [CBS, 2015]. It is worth noting that additional investment cost is
used in global cost calculations instead of initial investment cost for comparison with
the reference houses.
5.5.3.1 Replacement costs (RC)

The replacement costs are calculated for the building and energy systems
components that are replaced within the GC calculation period. For instance, a PV
system has a typical life-span of about 25 years and hence needs to be replaced.
Similarly, the inverter has to be replaced after every 10 years of use. On the other
hand, the building envelope has a life-span of about 40-50 years and is therefore not
considered in the replacement cost calculations. The replacement costs are
discounted to get the net present value of these costs based on the following equation
[Hamdy et al., 2013].
𝑅𝐶 = 𝐼𝐶𝑖 × (1 + 𝑟)−𝑘 (5.5)
where IC is the initial investment cost, the index i is the component that is being
replaced, r is the real interest rate and k is the life-span of the component being
replaced.
5.5.3.2 Operating costs (OC)

The operating costs are calculated based on annual energy costs for gas and electricity
consumption. Exported electricity is also considered in the calculation of operating
cost, depending on the net-metering scenario. If net-metering is available, both
imported and exported electricity are considered in operating cost calculations, as
shown in the equation below:
𝑂𝐶 = (𝐸𝑙𝑖𝑚𝑝 × 𝑃𝐸𝑙 + 𝐸𝑁𝐺 × 𝑃𝑁𝐺 ) − (⁡𝐸𝑙𝑒𝑥𝑝 × 𝑃𝐸𝑙 )⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡ (5.6)
If net-metering is terminated, exported energy is not considered in operating cost,

which is calculated as shown in the equation below:
𝑂𝐶 = 𝐸𝑙𝑖𝑚𝑝 × 𝑃𝐸𝑙 + 𝐸𝑁𝐺 × 𝑃𝑁𝐺 (5.7)
where 𝑃𝐸𝑙 ⁡is price of the electricity (€/kWhe) and 𝑃𝑁𝐺 is price of natural gas (€/m3).
127
A single tariff rate of 0.190 €/kWhe is used to calculate the electricity costs, and a
rate of 0.657 €/m3 for gas is used to calculate the natural gas consumption costs
[CBS, 2016a; RVO, 2017b]. In addition, in the case of net-metering scenario, the
surplus exported electricity is priced at 0.07 €/kWhe if the annual net exported energy
is more than 3500 kWhe/a [RVO, 2015c; E.ON, 2017b; Nuon, 2017]. Similar to the
replacement costs, the operating costs are discounted to get the net present value of
these costs.
OC⁡ = fd,e × OC⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.8)
where OC is the operating costs (€), fd,e is the discount factor considering real
interest rates and the escalation rate of energy prices. This discount factor fd,e is
calculated by the following equation:
1−(1+re )−n
fd,e = re
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.9)
where n is the calculation period, re is the real interest rate (r) that takes in to
account effect of the escalation rate (e) of energy prices, which is calculated by the
following equation
r−e
re = ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.10)
1+e
5.5.4 Thermal comfort (overheating hours)

Thermal comfort in a residential building largely depends on the occupant(s) ability
to adapt to the outdoor environment. Therefore, the thermal comfort model based
on adaptive temperature limits proposed by [Peeters et al., 2009] is implemented in
this work. These temperature limits are calculated based on reference outdoor
temperatures and neutral comfort temperatures. The neutral comfort temperature is
the temperature at which the majority of a group of people are comfortable.
Reference outdoor temperature ( 𝑇𝑒,𝑟𝑒𝑓 ) is the arithmetic average of today’s
temperature and temperatures of the previous three days, as shown in equation 5.11.
This approach is used as the weather of recent days has an effect on the clothing
choice of occupants and their perception of what constitutes a comfortable
temperature range [Morgan and de Dear, 2003].
(𝑇𝑡𝑜𝑑𝑎𝑦 +0.8𝑇𝑡𝑜𝑑𝑎𝑦−1 +0.4𝑇𝑡𝑜𝑑𝑎𝑦−2 +0.2𝑇𝑡𝑜𝑑𝑎𝑦−3 )

𝑇𝑒,𝑟𝑒𝑓 = 2.4
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.11)
128
where 𝑇𝑒,𝑟𝑒𝑓 is the reference outdoor temperature (°C), 𝑇𝑡𝑜𝑑𝑎𝑦 is the arithmetic
average of today’s maximum and minimum outdoor temperature (°C). 𝑇𝑡𝑜𝑑𝑎𝑦−1 ,
𝑇𝑡𝑜𝑑𝑎𝑦−2 and 𝑇𝑡𝑜𝑑𝑎𝑦−3 are the arithmetic average of the maximum and minimum
outdoor temperature of yesterday, 2 days ago and 3 days ago respectively (°C).
In this model, the neutral comfort temperatures are different for different room
functions such as living room, bedroom and bathrooms. In this work, bathrooms
and bedrooms are combined into a single zone (Zone 2) and therefore, neutral
temperatures and acceptable comfort limits of bedrooms are used for this zone since
occupancy presence in bedrooms is much higher than in the bathrooms. For
bedrooms, a minimum temperature of 16°C is assumed for a neutral comfort
temperature in winter conditions (𝑇𝑒,𝑟𝑒𝑓 < 0⁡°𝐶) and a maximum temperature of
26°C is assumed [CIBSE, 2006] for temperature limits [Peeters et al., 2009].
The neutral temperature for bedroom is calculated as shown in equation 5.12.
𝑇𝑛 = 16⁡°𝐶⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡𝑇𝑒,𝑟𝑒𝑓 < 0⁡°𝐶⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
⁡𝑇𝑛 = 0.23 × 𝑇𝑒,𝑟𝑒𝑓 + 16⁡°𝐶⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡0⁡°𝐶 ≤ 𝑇𝑒,𝑟𝑒𝑓 < 12.6⁡°𝐶⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
𝑇𝑛 = 0.77 × 𝑇𝑒,𝑟𝑒𝑓 + 9.18⁡°𝐶⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡12.6⁡°𝐶 ≤ 𝑇𝑒,𝑟𝑒𝑓 < 21.8⁡°𝐶⁡⁡⁡⁡⁡
⁡⁡⁡⁡⁡⁡⁡⁡𝑇𝑛 = 26⁡°𝐶⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡𝑇𝑒,𝑟𝑒𝑓 ≥ 21.8⁡°𝐶⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.12)
where 𝑇𝑛 is the neutral comfort temperature (°C) and 𝑇𝑒,𝑟𝑒𝑓 is the reference outdoor
temperature (°C).
Similarly, the neutral comfort temperature for living room, is shown in equation 5.13.
𝑇𝑛 = 0.06 × 𝑇𝑒,𝑟𝑒𝑓 + 20.4⁡°𝐶 ⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡𝑇𝑒,𝑟𝑒𝑓 < 12.5⁡°𝐶 ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
⁡⁡⁡⁡⁡⁡⁡⁡⁡𝑇𝑛 = 0.36 × 𝑇𝑒,𝑟𝑒𝑓 + 16.63⁡°𝐶 ⁡⁡⁡𝑓𝑜𝑟⁡⁡⁡𝑇𝑒,𝑟𝑒𝑓 ≥ 12.5⁡°𝐶 ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.13)
Since indoor temperatures are very unstable due to various heat gains and losses in
the building, neutral temperatures may not be met all the time in the building.
Therefore, the acceptable temperature are used [Peeters et al., 2009], which
determines the comfort bands (or in other words how much above or below the
neutral temperature) that the occupant(s) found acceptable. These bands are
calculated in equations 5.14 and 5.15 for the bedrooms and the living room
respectively.
129
30
Daily mean ambient temperature, °C

25
20
15
10
5
0
-10 -5 0 5 10 15 20 25 30 35
-5
-10
Reference outdoor temperature (Te, ref) °C
Reference climate Climate change scenario

Figure 5.11 Variation of reference outdoor temperature (equation 5.11) with daily mean
outdoor temperature.
Thermal comfort adaptive temperature limits for reference climate
32
Indoor operative tempearture, °C
30
28
26
24
22
20
18
16
14
-10 -5 0 5 10 15 20 25 30
Reference outdoor temperature (Te,ref), °C
Tn_Livingroom Tn_Bedroom
Tupper limit_bedroom Tlower limit_Bedroom
Tupper limit_Living room Tlower limit_Living room
Thermal comfort adaptive temperature limits for a climate change scenario

32
Indoor operative tempearture, °C
30
28
26
24
22
20
18
16
14
0 5 10 15 20 25 30
Reference outdoor temperature (Te,ref), °C
Tn_Livingroom Tn_Bedroom
Tupper limit_bedroom Tlower limit_Bedroom
Tupper limit_Living room Tlower limit_Living room
Figure 5.12 Adaptive temperature limits of comfort bands based on reference outdoor
temperatures of reference climate and a climate change scenario.
130
For bedrooms:
𝑇𝑢𝑝𝑝𝑒𝑟 = 𝑚𝑖𝑛(26⁡°𝐶, 𝑇𝑛 + 𝑤 × 𝛼)⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
⁡⁡⁡𝑇𝑙𝑜𝑤𝑒𝑟 = 𝑚𝑎𝑥(16⁡°𝐶, 𝑇𝑛 − (1 − 𝑤) × 𝛼)⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.14)
For living room:
⁡⁡⁡⁡𝑇𝑢𝑝𝑝𝑒𝑟 = 𝑇𝑛 + 𝑤 × 𝛼⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
⁡⁡⁡𝑇𝑙𝑜𝑤𝑒𝑟 = 𝑚𝑎𝑥(18⁡°𝐶, 𝑇𝑛 − 𝑤 × (1 − 𝛼))⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.15)⁡
where 𝑇𝑢𝑝𝑝𝑒𝑟 is the upper limit of the comfort band (°C), 𝑇𝑙𝑜𝑤𝑒𝑟 is the lower limit of
the comfort band (°C), 𝑤 is the width of the comfort band (°C) and 𝛼 is constant,
which ranges from 0-1. The widths of the comfort band for 80% and 90%
acceptability levels are 7°C and 5°C respectively. In the current research, a 90%
acceptability level is used.
In the present research, the thermal comfort model is adapted for both reference
climate and climate change scenarios. For instance, reference outdoor temperature
for a typical reference climate scenario (NEN 5060) and a climate change scenario
are calculated (see Figure 5.11). These reference outdoor temperatures are used to
calculate the corresponding neutral temperatures for both climates. The adaptive
temperature comfort band is calculated based on reference outdoor temperatures for
both reference climate and climate change scenarios, as shown in Figure 5.12. Since
the ideal heating system is used for heating in these case studies, it is assumed that
there are no underheating hours in the building.
Overheating hours are the total number of hours exceeding the allowable maximum
indoor temperatures during occupancy in a year. However, this does not quantify the
magnitude of overheating. The magnitude of overheating (∆T*h) is quantified by
multiplying the degree of temperature excess (∆T) by the number of hours (h) that
the excess exists. In this work, overheating hours are the weighted overheating hours.
5.5.5 Robustness indicators
The following robustness indicators, described in detail in Chapter 3, are used to

calculate the performance robustness of a performance indicator.
i. Spread; using the max-min method.

ii. Deviation; using the best-case and the worst-case method.
iii. Maximum regret; using the minimax regret method.
131
5.6 Case study simulation model

A detailed building and energy systems simulation model is developed to predict the
thermal, energy and economic performance of the design space across formulated
scenarios. The schematic of this simulation model is shown in in Figure 5.13. This
simulation model has input files, building model, HVAC system model, energy and
storage system models and comfort model, which are described below.
5.6.1 Input files: weather and occupant behavior profiles

A user defined weather file format (Type 99) is used in TRNSYS instead of a standard
typical meteorological year format (Type 15-tmy format) to model the impact of
outdoor environment. Type 99 is preferred because the hourly weather data
generated for the climate change scenarios is not available in standard formats. The
weather data includes diffuse and direct radiation on the horizontal surface, dry bulb
temperature, relative humidity, wind speed and wind direction. This component
calculates radiation parameters for different orientations and inclination angles. The
output from the weather component is connected to all components that are
influenced by outdoor environment, such as building, PV system, SDHW system,
and ventilation system.
Occupant behavior is typically modelled either in a simplified manner with

predefined values for different occupant behavior input parameters, or in a more
complex manner with stochastic behavior models or multi-agent models [Hoes et al.,
2011]. Since all possible scenarios are considered in the present work, and to avoid
extra complexity and computational resources, the simplified approach is used in
this study. The hourly profiles are developed for all occupant behavior scenarios,
such as heating setpoint profiles, IHG profiles, DHW profiles etc. Different occupant
behavior profiles are generated for different zones. In addition, some of the occupant
behavior profiles are modelled together. For instance, window opening profile is
modelled based on occupants’ presence in each zone and their activity. Occupant
behavior profiles are modelled using time dependent forcing functions (Type 14) in
TRNSYS. Multiple forcing functions are developed for different scenarios. For
instance, the manual control of the shading devices is modeled using hysteresis
controllers (Type 2) for each orientation. Input radiation from the weather file and
indoor temperature from the building model (Type 56) are used to control the
shading devices.
132
Case study simulation model
5.6.2 Building model

A multi-zone resistance capacitance (RC) building model is developed using the Type
56 component in the TRNSYS simulation tool to calculate the heating demand and
temperature. This building model is divided into three thermal zones to enable the
calculation of the temperature and energy demand of each zone. The living room
and kitchen on the ground floor form the first zone, three bedrooms and a bathroom
on the first floor constitute the second zone, and the attic on the second floor is the
third zone. This multi-zone building model in TRNSYS (Type 56) requires input
parameters such as geometries, building envelope properties, infiltration rates, and
window types. All of these parameters are connected to the building model through
an input file, as shown in Figure 5.13. Similarly, input parameters related to external
factors such as weather data are connected to the building.
5.6.3 HVAC system model

The heating demand calculated using an ideal heating system in the building model
is connected to a heating system (see Figure 5.13) to calculate the equivalent amount
of electricity/gas required for the heating by using a conversion factor based on the
type of heating system. In order to reduce the complexity of the simulation model
and to retain the main focus of this research, a detailed heating system model is not
considered in this study. The mechanical ventilation system is modelled through
hysteresis differential control. In the case of the mechanical ventilation with heat
recovery system, an air-air heat recovery device (Type 667) is coupled with this
differential controller. Heat recovery is bypassed when the room temperature (T i) is
greater than the heating setpoint and when the ambient temperature (Ta) is greater
than room temperature.
Natural ventilation, as a free cooling option to reduce overheating in the summer, is

modeled by using hysteresis (Type 2) differential controllers for window opening
scenarios. Windows are assumed to be opened by occupants if
𝑇𝑖 > ℎ𝑒𝑎𝑡𝑖𝑛𝑔⁡𝑠𝑒𝑡𝑝𝑜𝑖𝑛𝑡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
𝑇𝑖 > 𝑇𝑎 ⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡
𝑇𝑎 < 𝑚𝑎𝑥𝑖𝑚𝑢𝑚⁡𝑙𝑖𝑚𝑖𝑡⁡𝑜𝑓⁡𝑡ℎ𝑒𝑟𝑚𝑎𝑙⁡𝑐𝑜𝑚𝑓𝑜𝑟𝑡⁡𝑏𝑎𝑛𝑑⁡𝑤𝑖𝑑𝑡ℎ⁡⁡⁡⁡⁡⁡⁡⁡⁡(5.16)⁡
The maximum limit for thermal comfort is determined based on adaptive

temperature limits proposed by [Peeters et al., 2009].
133
5.6.4 Energy and storage systems model

Renewable energy and storage system models such as the PV system model, the
SDHW system model and the electric battery model are developed to calculate net-
energy imports form the grid and exports to the grid. The PV system is modeled
using a Type 562 simple photovoltaic system, which is connected to the weather
component for the input of the solar radiation on each panel. The outdoor
temperature is taken into account in the calculation of the efficiency of the solar
panels. This type models a photovoltaic array based on the user provided efficiency
for each module. The efficiencies of the panels depend on the surrounding
temperatures and incident radiation. Therefore, efficiency of the PV system is
determined by using the performance of a reference condition in combination with
linear modifiers for a specific cell temperature and incident radiation.
A simplified model with solar thermal collectors (type 73) and a storage tank (type 4)
is used to model the SDHW system. An auxiliary heater immersed in a storage tank
will function if the available solar radiation is not able to meet the DHW demand
profile for a particular scenario. An electric battery-based storage model (Type 47) is
used to store the surplus energy generated by the PV system. This model of a battery
operates in conjunction with solar cell array and power conditioning components.
The model specifies how the state of charge varies over time, given the rate of charge
or discharge. A minimum state of charge of 20% is required, and thus the maximum
allowed state of discharge is 80%.
5.6.5 Comfort model

The comfort model is also developed in TRNSYS using an equation editor, and all
the equations presented in section 5.5.4 were developed in the equation editor.
Weather related input parameters and building temperatures are connected to this
equation editor. In this model, the reference outdoor temperature ( 𝑇𝑒,𝑟𝑒𝑓 ) is
preprocessed using weather files for reference weather data and climate change
scenarios. This information is supplied through the data reader Type 9.
All performance indicators are calculated using the models described above, as
shown in Figure 5.13. This simulation model is connected to MATLAB to calculate
different robustness indicators of these performance indicators. It is worth noting
that the equations of these TRNSYS components can be found in mathematical
references in the TRNSYS manual.
134
Case study simulation model
Orientation
Energy prices
Size
Investment costs
Efficiency
Replacements costs
Tilt angle
Reference weather data Total costs

Electricity
Rc-Wall Future climate change PV system model generation
Rc-Floor weather data
Rc-Roof
U window
Thermal mass Net imports/
Orientation Net exports
Infiltration Electricity
Heating consumption
WWR
demand HVAC system model by HVAC To
system
Electric battery grid
Building model model
From
Number of occupants
Electricity grid
DHW consumption
Occupancy profile
demand SDHW system model by DHW
Setpoint temperatures CO2
system
Appliance use emissions
Lighting use
Internal heat gains
DHW use Thermal
Shading control Electricity comfort
IEQ
Window opening parameters Comfort model consumption (Overheating)
Ventilation rates by plug loads
Figure 5.13 Schematic overview of the building and energy system simulation model.
135
6. Demonstration of the
developed CPRA approach
using case studies
This chapter demonstrates the developed CPRA approach using the case studies of the
renovation house and the new house. In the first part of this chapter, the developed CPRA
approach is demonstrated in detail using the renovation house case study with the
policymaker, the homeowner and both as the decision makers. Similarly, in the second
part of this chapter, the demonstration of the CPRA approach using the new house case
study is summarized. This chapter concludes with a comparison of robust design options
of both case studies for all decision makers.
6.1 Overview
Firstly, the CPRA approach is demonstrated using the renovation house case study
that is described in Chapter 5. The aim of this study was to find robust renovation
measures for the policymaker and for the homeowner as decision makers, and also
for both decision makers combined. In practice, as discussed in Chapter 1 and
Chapter 3, the design decision-making process is a complex task, especially when
various decision makers with multiple performance requirements are involved.
Therefore, to facilitate end users with different decision-making options and to
enhance the design decision-making process, different robust design selection
approaches are presented. For instance, robust design options of all Pareto solutions
provide a complete overview of the design space and can aid decision makers in
choosing which design options lead to optimal performance and the most robust
designs. Similarly, using the MCDM method, a decision maker can easily select the
most robust design based on the highest design score. The selection of robust
designs is illustrated for three robustness assessment methods using the following
approaches:
i. Using trade-off solutions from the Pareto front.

ii. Selecting robust design options separately from Pareto solutions.
iii. Using the MCDM method based on the Hurwicz criterion.
137
Chapter 6. Demonstration of the CPRA approach using case studies
However, to enhance readability, the CPRA approach is demonstrated in detail using

three robustness assessment methods only for one decision maker i.e. policymaker.
For homeowners and both decision makers combined, the demonstration is carried
out in detail using the minimax regret method and demonstration with other
methods are summarized here. Nevertheless, the demonstration using the max-min
method and the best-case and worst-case methods are presented in detail in
Appendix D.
Secondly, the robust designs of the new house case study selected using the
aforementioned approaches for three robustness assessment methods are
summarized. The results of this case study are presented in Appendix E to avoid
repetition and to enhance readability of this chapter. In the last part of this chapter,
different robust design options are compared for both case studies to allow decision
makers to choose cost-optimal robust design options.
In this research, different visualization methods, which are reviewed in Chapter

2 and described in Chapter 3, are used to enhance the design decision-making
process.
In the scatter plots that are used for the trade-off approach, the additional
investment cost (ICa) required for a design is shown on the X-axis to allow the
decision maker to trade off ICa with predicted performance and robustness of the
design. The median value of a performance indicator across the considered
scenarios is used to represent the predicted performance. Here, each bubble
represents this median value of a design and bubble size depicts the robustness
of a design. To be consistent throughout this chapter, the maximum size of
bubble is fixed and the bubble size is varied in proportion to the range of
robustness indicator values. The smaller the bubble size, the more robust is the
design.
In the box plots that are used to capture variations in predicted performance, the
range of each box represents the variation of a performance indicator across the
considered scenarios. Similarly, in the case of scenario analysis, the range of a
box for a scenario represents the variation of a performance indicator across all
Pareto designs and remaining scenarios.
Only Pareto solutions, calculated considering ICa, predicted performance and

corresponding robustness, are analyzed. The reference building design (ICa = 0
k€) is marked in green for ease of comparison.
138
Renovation house case study - policymaker as a decision maker
6.2 Demonstration of the CPRA approach using renovation house case

study with the policymaker as a decision maker
6.2.1 Trade-off solutions using Pareto front

The Pareto fronts for the policymaker calculated using the three robustness
assessment methods are shown in Figure 6.1, Figure 6.3 and Figure 6.5. The Pareto
front for the policymaker was calculated considering his/her preferred performance
and robustness indicators; CO2 emissions and corresponding robustness, and
additional investment costs (ICa). This calculation resulted in a 3D Pareto front, and
to enhance visualization the third dimension is included as bubble size (robustness).
In these figures, the required ICa is shown on the x-axis to allow the policymaker to
trade-off CO2 emissions and their corresponding robustness with the required ICa.
The policymaker prioritizes a robust design that has low CO 2 emissions and high
robustness (small bubble size). They can also trade off the preferred robust design
with required ICa. The policymakers can choose one of the methods discussed below
in selecting the preferred robust design based on their attitude towards risk
acceptance in the decision-making process.
6.2.1.1 The max-min method

Figure 6.1 shows the Pareto front of the design space for the policymaker calculated
using the max-min method considering additional investment costs (ICa), CO2
emissions and the spread of CO2 emissions. Each bubble represents the median
value of CO2 emissions of a design across the considered scenarios, and the bubble
size depicts the spread (robustness) of CO2 emissions. The design with zero ICa is
the reference building (RP0). It can be observed from the Pareto front that the
existing building resulted in the highest CO2 emissions and also the largest spread
of CO2 emissions among all designs. This indicated that renovation of the existing
building was essential, and therefore, the policymaker would prefer to find robust
renovation measures to reduce these CO2 emissions and their corresponding spread.
In a typical building design decision-making process, decision makers are interested

in designs that are in different budget ranges (ICa). To find the preferred robust
renovation measures for the policymaker, designs in different ICa ranges were
compared (see Figure 6.1). For instance, in the ICa range of 0-15 k€, CO2 emissions
and their corresponding spread gradually decreased in proportion to increases in ICa.
In this ICa range, the preferred robust design for the policymaker would be the
139
design with low CO2 emissions and the lowest spread of CO2 emissions. Similarly,
in the ICa range of 15-30 k€, the predicted performance is similar, and the robustness
improved further (in line with) with an increase in ICa. Consequently, the
policymaker would prioritize spread of CO2 emissions and can trade off the predicted
performance with required ICa. In contrast, in the ICa range of 30-45 k€, the predicted
performance was further improved, whereas the spread of CO2 emissions was higher
than that of designs in the ICa range of 15-30 k€. Therefore, in this ICa range the
policymaker would prioritize predicted performance and can trade-off the required
ICa with the spread of CO2 emissions.
Bubble size = Spread of CO2 emissions (2268-3948 kgCO2/a)

4000
1
3500
3000
2500
2000
1500
2 3
1000
500
0
-5 5 15 25 35 45
Additional investment cost (ICa), k€
Figure 6.1 The Pareto front for the policymaker calculated using the max-min method. The
spread is included as bubble size. The green bubble represents the reference building design.
Comparing the entire Pareto front, the designs in the ICa range of 0-15 k€ resulted
in high CO2 emissions and corresponding spread (bigger bubble size), and thus may
not be preferred robust designs for the policymaker. Conversely, the designs in the
ICa range of 15-30 k€ resulted in low CO2 emissions and smaller spread, and
accordingly were more robust and more preferred designs compared to the first ICa
range. However, these designs incurred high ICa. Similarly, the designs in the ICa
range of 30-45 k€ resulted in further reductions of CO2 emissions, but the
improvement in robustness does not outweigh the required ICa. Therefore, the
preferred robust design of the policymaker depends on the required ICa of the
design. This is elaborated further by comparing a few designs selected from the
Pareto front in different ICa ranges. The details of these Pareto designs selected for
further analysis are tabulated in Table 6.1.
140
It is worth noting that the upper part of the Pareto front was largely dominated by
designs that have large renewable energy and storage (RES) systems, and the lower
part of Pareto front was dominated by designs with improved insulation levels (RP3-
RP5) and large RES systems (see Table 6.1). The selected Pareto designs were
grouped in three ranges of ICa, are shown in Figure 6.2. In this figure, the variation
of predicted performance across considered scenarios is shown in box plots and the
corresponding spread is shown in bar plots.
Firstly, a comparison is made between the four selected designs in the ICa range of
0-15 k€. All these designs have the same renovation package (RP0), but differ in RES
systems, as shown in Table 6.1. Among these designs, the reference building (ICa=0
k€) had very high CO2 emissions and also the largest spread. Comparing the next
two designs, it can be noted that the design with an ICa of 9.1 k€ had better predicted
performance but resulted in a larger spread compared to the design with an ICa of
4.1 k€. This larger spread of the design with an ICa of 9.1 k€ was due to higher
variations in CO2 emissions for extreme scenarios. In such cases, if the policymaker
prioritizes robustness, then the design with an ICa of 4.1 k€ would be the most
preferred robust design. In contrast, if the policymaker prioritizes predicted
performance, then the design with an ICa of 9.1 k€ would be preferable. However,
this improvement in predicted performance comes at the expense of 5 k€. Among
these four designs, the design with an ICa of 14.9 k€ was the most robust, because
this design had the least spread and better predicted performance. Within this budget
range, it was evident that investing is energy generation and storage systems resulted
in more robust designs than the designs with improved building insulation levels.
In fact, it can be inferred from Figure 6.2 that variations in CO2 emissions across
considered scenarios can be reduced to a large extent by adding PV and battery
storage systems to an existing building without improving insulation.
Similarly, comparing designs in the ICa range of 15-30 k€, it can be observed that the
design with an ICa of 23.6 k€ was the most preferred robust design for the
policymaker, because this design had low CO2 emissions and the smallest spread of
CO2 emissions among the four selected designs. Compared to designs in the ICa
range of 0-15 k€, these designs have larger PV and SDHW systems and also bigger
battery capacities (see Table 6.1). In addition, the design with an ICa of 29.3 k€ has
improved insulation; renovation package RP3 (Rc = 3.5/4.5/6 m2K/W for
floor/wall/roof; U = 1.43 W/m2K). Despite these improved insulation levels, higher
variations of CO2 emissions and a larger spread were observed with this design
141
compared to the design with an ICa of 23.6 k€. This large spread was attributed to
smaller battery capacity, which resulted in a reduction in self-consumption of
electricity, especially in the case of net-metering scenarios.
In the last ICa range, the policymaker would prioritize the design with an ICa of 38.6
k€, because this design had a low median value and the smallest spread of CO 2
emissions. However, the improvement in robustness does not outweigh the required
ICa, because it can be seen from Figure 6.2 that the designs in the ICa range of 15-45
k€ resulted in similar spread, except for designs with an ICa of 23.6 and 38.6 k€.
These two designs have similar energy and storage systems, but the design with an
ICa of 38.6 k€ has renovation package RP4 (Rc = 6/7/7 m2K for floor/wall/roof; U =
1.01 W/m2K). The improved insulation of this design resulted in better predicted
performance compared to the design with an ICa of 23.6 k€. However, the spread of
CO2 emissions was similar for both designs and indicates that in this method,
robustness was influenced more by internal load dominated parameters such as
lighting use and appliance use than the shell load dominated parameters such as
insulation levels. The influence of scenarios and design options on predicted
performance and robustness are discussed in detail in the next sections. The
improvement in predicted performance comes at the expense of an ICa of 15.8 k€.
Hence, to select a preferred robust design, the policymaker should compare
predicted performance and robustness, and trade these off with required IC a.
It is noteworthy that the designs (ICa of 23.6 and 38.6 k€) that had maximum size of
PV system, SDHW system and battery storage capacity were the most robust among
the selected designs, indicating that the max-min method resulted in conservative
robust designs. In this method, the predicted performance was optimized and
variations across scenarios were reduced. This reduction in variations, as noted
earlier, could also be achieved by including large RES systems. Among other design
options, it was found that the gas boiler (HR107) was the most robust heating option
for all selected designs. Similarly, south oriented PV and SDHW systems at a tilt
angle of 45° were the most robust options. Designs with a WWR of 40% and an
infiltration rate of 1 dm3/ds2, and designs with a WWR of 20% and an infiltration
rate of 0.1 dm3/ds2 were the most robust options in the ICa range of 0-30 k€ and 30-
45 k€, respectively. The robust designs options considering the entire Pareto front
are discussed in detail in the next section.
142
Table 6.1 Details of the selected Pareto designs, for further analysis, in different additional investment cost ranges from the Pareto front of the
policymaker calculated using the max-min method.
ICa, k€ 0 4.2 9.1 14.9 15.9 19.7 23.6 29.3 31.3 34.6 34.7 38.6
Renovation package (RP) RP0 RP0 RP0 RP0 RP0 RP0 RP0 RP3 RP3 RP4 RP5 RP4
Infiltration, dm3/ds2 1 1 1 1 1 1 1 0.1 0.1 0.1 0.1 0.1
WWR, % 40 40 40 40 40 40 40 20 20 20 20 20
HVAC system HR107
PV system, m2 0 19.2 25.6 25.6 32 32 32 32 32 32 32 32
SDHW system, m2 0 0 0 7.5 0 5 10 7.5 7.5 10 7.5 10
Orientation S
Tilt angle, ° 0 45 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 8 8 20 20 20 12 16 12 16 20
6000
Spread of CO2 emissions,

5000
4000
kgCO2/a
3000
2000
1000
0
0.0 4.2 9.1 14.9 15.9 19.7 23.6 29.3 31.3 34.6 34.7 38.6
Additional investment cost of selected designs (ICa), k€
Figure 6.2 Variation of CO2 emissions (box plot – left figure) and corresponding spreads (bar plot – right figure) across considered scenarios
of selected Pareto designs for further analysis.
143
6.2.1.2 The best-case and worst-case method

Figure 6.3 shows the Pareto front of the design space for the policymaker calculated
using the best-case and worst-case method considering additional investment costs
(ICa), CO2 emissions and the corresponding deviation. Like Figure 6.1, each bubble
represents the median value, whereas bubble size depicts the deviation (robustness)
of CO2 emissions of a design across the considered scenarios. Similarly, the
policymaker prioritizes a design with low CO2 emissions and the lowest deviation,
and can trade these off with required ICa. As observed in the case of the max-min
method, the existing building (green bubble) has the highest CO2 emissions and the
largest deviation of CO2 emissions among all Pareto solutions.
The CO2 emissions and their corresponding deviation gradually decreased in

proportion to increases in ICa. However, the improvement in robustness beyond 15
k€ ICa does not outweigh the required ICa, because deviations of CO2 emissions were
found to be similar for most of the designs beyond 15 k€ of IC a. In this method, the
predicted performance was optimized and the deviation of the worst-case
performance of a design from the best-case performance of the entire design space
was minimized. The same can be observed from Figure 6.3 that the designs beyond
the 15 k€ ICa had similar worst-case performances, but the predicted performance
was better for these designs, meaning that they might be preferred robust designs
for the policymaker. This improvement in predicted performance incurred high
additional investment costs.
Bubble size = Deviation of CO2 emissions (2336-5525 kgCO2/a)

4000
1
3500
3000
2500
2000
1500 2 3
1000
500
0
-5 5 15 25 35 45
Figure 6.3 The Pareto front for the policymaker calculated using the best-case and worst-
case method. The deviation is included as bubble size. The green bubble represents the
reference building design.
144
Using this information, a decision maker can choose the required ICa range and
choose a robust design from among the alternatives within that range. To provide
better insight into the selection of robust designs for the policymaker, a few designs
that were randomly selected on the Pareto front in different additional investment
cost ranges (Figure 6.3) were compared; see Figure 6.4. The details of these selected
designs are tabulated in Table 6.2. Like observations made with the max-min
method, the upper part of the Pareto front (see Figure 6.3) is dominated by designs
with large RES systems, whereas the lower part of the Pareto front is dominated by
designs with improved renovation packages (RP1-RP6) in combination with large
RES systems.
In the first ICa range (0-15 k€), the predicted performance and robustness gradually
increased in proportion to increases in ICa arising from the inclusion of large PV
systems. In this ICa range, the policymaker would prioritize a design with low CO2
emissions and the lowest deviation. For instance, by comparing the first four designs
in Figure 6.4, it can be observed that the design with an ICa of 14 k€ was the most
robust, because this design had better predicted performance and the lowest
deviation among the four selected designs. These designs have the same renovation
package and other design options except for PV system and battery capacity (see
Table 6.2). As observed for the max-min method, the design with the large PV system
and battery capacity yielded the most robust options using this method.
In the second ICa range (15-30 k€), it can be inferred that the designs with an ICa of
19.8, 21.7 and 27.8 k€ had similar deviations (2500-2595 kgCO2/a), but differed in
predicted performance. This difference in predicted performance was due to the
designs having different insulation levels, SDHW system sizes and battery capacities
(see Table 6.2). The design with an ICa of 21.7 k€ had the lowest deviation among the
selected designs. However, the design with an ICa of 27.8 k€ had the best predicted
performance among these designs, but a higher deviation compared to the design
with an ICa of 21.7 k€, because it had a slightly higher worst-case performance (see
Figure 6.4). In such cases, the decision maker would prioritize either predicted
performance or robustness and trade these off with the required ICa.
In the last ICa range, designs have different renovation packages and battery
capacities, but the same PV system of 32 m2 and SDHW system of 7.5 m2. It can be
observed that four designs had similar predicted performance and deviation, but the
design with an ICa of 39.8 k€ was more robust, because this design had better
145
predicted performance and lower deviation compared to other designs. This design
could be the preferred robust design if the policymaker prioritized predicted
performance and robustness. Otherwise, the design with the lowest ICa among the
four designs could be preferred as the deviation of CO2 emissions of these designs
was very similar. The design with an ICa of 21.7 k€ was the most preferred robust
design among all selected designs, because this design had better predicted
performance and robustness at lower ICa.
The robust design options among the selected designs were the designs with large
PV systems (19.2-32 m2) and bigger battery capacities (8-20 kWh), which were
similar to the robust design options using the max-min method. In contrast, here,
maximum size of SDHW system was limited to 7.5 m2 using the best-case and worst-
case method. Designs with renovation package RP6 (Rc = 6/8.5/10 m2K/W; U =0.86
W/m2K) were the most robust. Other design options such as WWR, HVAC system,
infiltration rates, tilt angle and orientation of energy generation systems followed a
similar trend as observed using the max-min method.
146
policymaker calculated using the best-case and worst-case method.
ICa, k€ 0 4.2 9.1 14.0 15.9 19.8 21.7 27.8 32.7 33.2 37.9 39.8
Infiltration, dm3/ds2 1 1 1 1 1 1 1 0.1 0.1 0.1 0.1 0.1
WWR, % 40 40 40 40 40 40 40 20 20 20 20 20
HVAC system HR107
PV system, m2 0 19.2 25.6 32 32 32 32 32 32 32 32 32
SDHW system, m2 0 0 0 0 0 5 7.5 7.5 7.5 7.5 7.5 7.5
Orientation, ° S
Tilt angle, ° 0 45 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 8 16 20 20 20 16 16 20 16 20
6000
Deviation of CO2 emissions,

5000
4000
kgCO2/a
3000
2000
1000
0
0.0 4.2 9.1 14.0 15.9 19.8 21.7 27.8 32.7 33.2 37.9 39.8
Figure 6.4 Variation of CO2 emissions (box plot – left figure) and corresponding deviation (bar plot – right figure) across the considered
scenarios of the selected Pareto designs for further analysis.
147
6.2.1.3 The minimax regret method

The Pareto front calculated using the minimax regret method considering ICa, CO2
emissions and the corresponding maximum regret of the design space for the
policymaker is shown in Figure 6.5. Similar to the Pareto fronts of the max-min and
the best-case and worst-case method, each bubble represents median value, while the
bubble size depicts the maximum regret (robustness) of CO2 emissions across the
considered scenarios. The smaller the bubble size, the more robust is the design. It
can be observed from Figure 6.5 that the robustness of the designs improved
gradually in line with an increase in ICa.
For instance, in the first ICa range (0-15 k€), the predicted performance and
robustness significantly improved with an increase in ICa. In this ICa range, the
policymaker would prioritize a design with low CO2 emissions and the lowest
maximum regret. In the second ICa range (15-30 k€), predicted performance of all
designs was similar. The improvement in robustness does not outweigh the required
ICa and accordingly, in this range of ICa, the policymaker would prioritize designs
with the lowest ICa. In contrast, in the last ICa range, the predicted performance and
robustness of designs improved further with an increase in ICa (30-45 k€). The
maximum regret is even close to zero for a few designs in this IC a range, indicating
that these designs were optimal for most of the scenarios. This optimal performance
was the result of improved insulation levels in combination with large RES systems.
In this ICa range, since the predicted performance of the designs was similar, the
policymaker would prioritize robustness and trade this off with the required IC a.
Bubble size = Maximum regret of CO2 emissions (56-3830 kgCO2/a)

4000
1
3500
3000
2500
2000
1500
2 3
1000
500
0
-5 5 15 25 35 45
Figure 6.5 The Pareto front for the policymaker calculated using the minimax regret
method. The maximum regret is included as bubble size. The green bubble represents the
148
A few selected designs in these ICa ranges were compared for further analysis to
provide better insights into robust design selection. The details of the selected
designs for further analysis are tabulated in Table 6.3, and these designs are
presented in Figure 6.6 for comparison. Similar to observations made with the
previous methods, the upper part of the Pareto front (see Figure 6.5) is largely
dominated by designs with large RES systems, whereas the lower part of the Pareto
front is dominated by designs with improved renovation packages (RP2-RP6) in
combination with large RES systems (see Table 6.3).
It can be seen from Figure 6.6 that the design with an ICa of 14 k€ was the most
robust among the selected designs within the ICa range of 0-15 k€. This design had
better predicted performance and the lowest maximum regret among the four
selected designs, which was attributed to larger PV systems and bigger battery
capacities. Similarly, in the second ICa range (15-30 k€), the design with an ICa of
22.7 k€ was the most robust. In the second ICa range, designs had different
renovation packages (RP0-RP3) and RES systems, but design with large RES systems
(PV = 25.6m2, SDHW= 10 m2 and Battery is 20 kWh) were the most robust despite
not having improved insulation. In the third ICa range of 30-45 k€, a few designs
achieved very low CO2 emissions and even close to zero maximum regret of CO 2
emissions, indicating that these designs were optimal for most of the scenarios since
this method optimizes robustness with respect to optimal performance. Therefore,
it can be inferred that the inclusion of RES systems improved the robustness of
designs and that higher insulation levels in combination with large RES systems
yielded designs with optimal and robust performance. Comparing the designs in this
ICa range, the design with an ICa of 40.9 k€ had the lowest maximum regret and also
better predicted performance, and accordingly, was the most robust design.
However, the design with an ICa of 35.2 k€ also achieved similar robustness but with
a reduction in cost of 5.7 k€ compared to the design with an ICa of 40.9 k€. In this
case, the policymaker would prefer a design with an ICa of 35.2 k€ compared to other
designs, because it resulted in similar robustness at lower ICa.
Among all selected designs, the design with an ICa of 35.2 k€ was the most preferred
robust design for the policymaker as it had better predicted performance and close
to zero maximum regret. Among the design options, designs with RP3-RP6 were the
most robust. Similarly, designs with a PV system of 25.6-32 m2, an SDHW system
of 7.5-10 m2 and battery capacity of 16-20 kWh were the most robust design options.
149
policymaker calculated using the minimax regret method.
ICa, k€ 0 4.2 9.1 14.0 18.8 22.7 28.9 30.3 32.2 33.7 35.2 40.9
Infiltration, dm3/ds2 1 1 1 1 1 1 0.1 0.1 0.1 0.1 0.625 1
WWR, % 40 40 40 40 40 40 20 20 20 20 20 20
HVAC system HR107
PV system, m2 0 19.2 25.6 32 25.6 25.6 32 25.6 32 25.6 32 32
SDHW system, m2 0 0 2.5 0 7.5 10 0 7.5 0 10 10 7.5
Orientation S
Tilt angle, ° 0 45 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 4 16 16 20 20 16 20 16 16 20
6000

5000
Maximum regret of
4000
3000
2000
1000
0
0.0 4.2 9.1 14.0 18.8 22.7 28.9 30.3 32.2 33.7 35.2 40.9
Figure 6.6 Variation of CO2 emissions (box plot – left figure) and corresponding maximum regret (bar plot – right figure) across considered
scenarios of the selected Pareto designs for further analysis.
150
It is noteworthy that the selection of robust design using the Pareto front is illustrated
for only a few randomly selected designs across different ICa ranges. Therefore, to
find the preferred robust design across the entire Pareto front, the robust design
options of all Pareto solutions are compared and alternative methods of selecting
robust designs are discussed in the next section.
Using the trade-off implemented in this CPRA approach, a decision maker can
prioritize predicted performance or robustness, and can trade this-off with required
additional investment costs to find preferred robust designs.
6.2.2 Robust design options
In the case of the trade-off approach using Pareto solutions, there is a possibility that
more preferred robust designs have been missed as the trade-off is illustrated using
only a few randomly selected designs. To ameliorate this problem, all design options
from the Pareto solutions are compared in Figure 6.7 to allow decision makers to
choose robust design options individually. This figure gives an indication of which
design options could lead to optimal performance and a robust design. All design
options from three different robustness assessment methods were compared. The
triangles represent spread, rectangles represent deviation and circles represent
maximum regret; the design options with the lowest values of these indicators are
deemed to be robust.
It is worth noting that for a particular design option, the variation of other design
options is shown to indicate the interaction among different design options. For
example, RP0 has multiple maximum regret values for CO2 emissions (circles), and
these were due to other design options such as infiltration rates, PV system and
WWR. The selection of robust designs for the policymaker based on robust design
options with three robustness assessment methods is discussed below.
151
6000 6000 6000

5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
Renovation package (RP) Infiltration rate, dm3/ds2 WWR, (%)
6000 6000 6000

1 = HR107

0 = N-S facing S
5000 2 = ASHP 5000 5000

90 = E-W facing E
4000 3 = GSHP 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
1 2 3 0 90 180 0 45 90
HVAC system type Orientation of RES, (°) Tilt angle of RES, (°)
6000 6000 6000 6000


5000 5000 5000

5000
4000 4000 4000
4000
3000 3000 3000
2000
3000 2000 2000
1000 2000 1000 1000
0 0 0
0
1000
6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
PV system size, m2 Solar DHW system size, m2 Battery capacity, kWh
0
20 40 60 80
Figure 6.7 Variation of robustness of CO2 emissions for different design options of all Pareto
solutions for the policymaker calculated using three robustness assessment methods
(Triangles represent spread, rectangles represent deviation and circles represent maximum
regret).
The max-min method: It can be observed from Figure 6.7 that designs with renovation
packages RP0 and RP4 resulted in the smallest spread, but RP0 has zero ICa.
Consequently, the policymaker would prefer RP0. Similarly, designs with an
infiltration rate of 0.1 dm3/ds2 were the most robust, however, designs with an
infiltration rate of 1 dm3/ds2 also yielded similar robustness. These different ranges
of infiltration rates among robust design options were attributed to the combination
of different insulation levels (RP0-RP4) with these infiltration rates. Designs with
WWRs of 20% were the most robust options. The HR107 gas boiler was the most
robust heating system because it had a similar emission factor of primary energy as
the ASHP and GSHP heating systems, but it required a lower ICa. This outcome is
in line with a study reported in literature [Blom et al., 2011]. South facing PV and
SDHW systems at a tilt angle of 45° were the most robust options, because these
design options maximized energy generation.
152
Designs with large RES systems were the most robust using the max-min method,
as noted earlier. Large PV and SDWH systems reduced CO2 emissions of a design,
and bigger battery capacity increased self-consumption of electricity and
consequently reduced the spread of CO2 emissions, especially in the case of the no
net-metering scenario. Therefore, a PV system of 32 m2, an SDHW system of 10 m2
and a battery capacity of 20 kWh were the most robust options. Thus, it can be
inferred that robustness of existing buildings can be enhanced, without improving
insulation, simply by adding large RES systems. These large RES systems are
deemed to be conservative design options, and also may not be preferred by
homeowners from an aesthetic point of view. The robust deigns for the homeowner
are discussed in the next section of this chapter.
The best-case and worst-case method: In this method, deviation is minimized with
respect to the best performance across the entire design space. Typically, higher
insulation levels in combination with large RES systems resulted in the best-case
performance. Designs with RP6 had the lowest deviation of CO2 emissions and
accordingly, RP6 was the most preferred renovation package. Designs with low
infiltration rates (0.1 dm3/ds2) and low WWRs (20%) were the most robust, because
these design options reduced heating demand and consequently reduced CO2
emissions. Similar to results from the max-min method, here, the HR107 gas boiler
was the most robust heating system, and energy generation systems oriented
towards the south at a tilt angle of 45° were the most robust. In addition, designs with
large PV system (32 m2) and bigger battery capacity (20 kWh) were the most robust
using this method. In contrast, the maximum size of SDHW system was limited to
7.5 m2 in this method as improving insulation levels yielded more robust design
options than opting for larger SDHW systems. This difference was because the
deviation with respect to the best-case performance cannot be minimized with the
inclusion of large RES systems alone.
The minimax regret method: Designs with renovation packages RP3 and RP6 resulted
in zero maximum regret of CO2 emissions (the most robust), but RP6 incurred
higher ICa compared to RP3. Therefore, the policymaker would prefer designs with
the RP3 renovation package. These zero maximum regrets for CO2 emissions can be
achieved for the same renovation package by including large RES systems, as can be
seen from Figure 6.7 and Table 6.3. In this method, since the robustness was
optimized with respect to the optimal performance, this optimal performance could
also be attained with different design options. It can be noted that designs with the
153
RP3 renovation package and PV system of 32 m2, SDHW system of 10 m2 and battery
capacity of 16 kWh resulted in zero maximum regret of CO2 emissions. Similarly,
the same result can also be achieved by designs with RP6 renovation package, PV
system of 32 m2, SDHW system of 7.5 m2 and battery capacity of 16 kWh. Similar to
observations made earlier for the other two robustness assessment methods, it was
found that HR107 was the most robust heating system, and that energy generation
systems oriented towards the south at a tilt of 45° were the most robust design
options.
The notable difference among robust design options for three robustness assessment
methods was that using the max-min method the designs with existing insulation
levels and large RES systems can reduce the spread of CO2 emissions significantly.
In contrast, the designs with enhanced insulation levels and large RES systems were
the most robust with the other two methods. For all three robustness assessment
methods, it was found south facing PV and SDHW systems at a tilt angle of 45° was
the most robust option. Similarly, the designs with low WWRs and low infiltration
rates were the most robust options.
Therefore, using the CPRA approach, a decision maker can choose different design
options based on their predicted performance (CO2 emissions) and robustness
(spread/deviation/maximum regret) and can trade these off with the required
additional investment costs.
6.2.3 The most robust design using the MCDM approach
In order to ease the design decision-making process further and to facilitate decision
makers in identifying the most robust design, the design score of Pareto solutions
was calculated using the Hurwicz criterion considering predicted performance and
robustness. The design with the highest score is the most preferred robust design.
The design scores are compared against required additional investment cost (ICa) in
Figure 6.8 to allow the decision maker either to choose the most robust design or
trade off design score with the required ICa. For instance, in the max-min method,
the design score did not improve significantly beyond 23.6 k€ of ICa. Therefore, the
policymaker could prefer the design with an ICa of 23.6 k€. Alternatively, the
policymaker could opt for the most robust design with the highest design score. It
can be observed that for all three methods, the design with the corresponding highest
ICa was the most robust design, except for the minimax regret method, where the
design with ICa of 35.2 k€ was also equally robust as the design with ICa of 40.8 k€.
154

1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Design score
Design score
Design score
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Additional investment cost (ICa), k€ Additional investment cost (ICa), k€ Additional investment cost (ICa), k€
Figure 6.8 The design scores of Pareto solutions for the policymaker calculated using the
Hurwicz criterion for three robustness assessment methods considering predicted
performance and corresponding robustness. Reference building design is indicated in green
(ICa = 0k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
Design options score

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9
1 = HR107 0.9
0 = N-S facing S 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 0 45 90
1 6000 1 1
0.9 0.9 0.9

0.8 5000 0.8 0.8
0.7 0.7 0.7

0.6 4000 0.6 0.6
0.5 0.5 0.5
0.4 3000 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
2000
0.1 0.1 0.1
0 0 0
0 1000
6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0
20 40 60 80
Figure 6.9 The design scores of different design options of Pareto solutions for the
policymaker calculated using the Hurwicz criterion for three robustness assessment methods
regret).
155
The design score of different design options of Pareto solutions were compared and
presented in Figure 6.9 to enable the policymaker to select design options separately
based on their corresponding highest design score. It can be inferred from Figure
6.7 and Figure 6.9 that both approaches of robust design selection, i.e. trade-off and
MCDM, yield similar robust design options. For instance, in both cases, large RES
systems were the most robust for all three robustness assessment methods. Similar
observations can be made for heating system, and orientation and tilt angle of energy
generation systems. The most robust design options selected based on the highest
design score considering the decision makers attitude towards risk acceptance in the
decision-making process are summarized below.
The max-min method: It can be observed from Figure 6.9 that designs with
renovation package RP4 had the highest design score, whereas RP0 also had a design
score close to 1. These similar designs score were the result of combinations of other
design options such as the inclusion of large RES systems. As expected, designs with
the large RES systems (PV system of 32 m2, SDHW system of 10 m2 and battery
capacity of 20 kWh) were the most robust. Similarly, the designs with an infiltration
rate of 0.1 dm3/ds2 and a WWR of 20% were the most robust. These observations
were similar to the robust design options based on trade-off approach considering
the entire Pareto front.
Therefore, if the policymaker opted for a conservative approach in the decision-

making process, then the most preferred robust design among all Pareto solutions
was the design with low to moderate insulation levels, airtight envelope, small
WWRs and large RES systems oriented towards the south at a tilt angle of 45°.
The best-case and worst-case method: Similar to the observations made for robust
designs using the max-min method, here designs with RP6 renovation package had
the highest design score, however, RP2 also had a design score close to 1. Other
robust design option such as WWRs, infiltration rates and RES systems were similar
to robust design options based on trade-off approach using the entire Pareto front.
Therefore, if the policymaker targeted the best performance in all scenarios, then the
most preferred robust design among all Pareto solutions was the design with very
high insulation levels, airtight envelope, small WWR and large RES systems oriented
towards the south at a tilt angle of 45°.
156
The minimax regret method: In contrast to observations made with the previous two
methods, designs with renovation packages RP3 and RP6 had the highest design
score, and in this case, the policymaker would prefer RP3 due to its low IC a. As
observed for the best-case and worst-case method, other robust design options were
similar to robust design options based on trade-off approach.
Therefore, if the policymaker chose a less conservative approach and accepted a

certain level of risk (maximum regret) in the decision-making process, then the most
preferred design among all Pareto solutions would be the design with moderate
insulation levels, airtight envelope, a small WWR and large RES systems oriented
towards the south at a tilt angle of 45°.
6.2.4 Scenario analysis

The performance variations of these Pareto solutions depend on the considered
scenarios. Sensitivity analysis was carried out using the Mann-Whitney U test. The
sensitivity index (p) was calculated for each scenario by pooling all low and high
values as two separate samples, and p determined the influence of these two samples
on predicted performance and robustness indicators. Robustness indicators for each
method were calculated by considering all Pareto solutions of a scenario
combination. For instance, the spread of a scenario combination was calculated
considering the maximum and minimum performance across all Pareto designs for
that particular scenario combination. This calculated spread of CO2 emissions, as
shown in Figure 6.10, is comparable with the spread of CO2 emissions of the Pareto
front (see Figure 6.1).
It is worth remembering that scenarios where p<0.05 are assumed to be sensitive.

In other words, the higher the p value, the more robust the design is to that particular
scenario. These values for predicted performance and for spread, deviation and
maximum regret are shown in the corresponding x-axis labels of each scenario in
Figure 6.10, Figure 6.11, and Figure 6.12, respectively. Scenarios that are sensitive to
either predicted performance or robustness are shown in these figures. It is worth
noting that all scenarios influence (p<0.05) either predicted performance or
robustness. The white box plots represent low values of scenarios and colored box
plots represent high values of corresponding scenarios. The range of the box of a
scenario includes all scenario combinations of Pareto solutions for that particular
scenario. For instance, in Figure 6.10, the range of the box for a low occupant
scenario includes all scenario combinations of Pareto solutions with low occupant
levels.
157
The max-min method: It can be seen from Figure 6.10 that all low-high scenarios
influenced the predicted performance, and that, as expected, among these scenarios
occupant behavior related scenarios had the greatest influence. Higher CO2
emissions and corresponding variations were observed with higher occupant levels,
which was due to increased electricity consumption. In contrast, there was a slight
reduction in CO2 emissions in the all-day occupancy profile (OP) compared to the
evening occupancy profile. This reduction, especially in the case of the net-metering
scenario, was due to increased self-consumption of electricity, as the electricity
demand is proportional to the electricity generation profile. The variations in CO2
emissions were higher with higher (red boxes) heating setpoints (Thsp), appliance use
(Ause), lighting use (Luse), IHG and DHW use. It is worth noting that in climate
scenarios (CS), CO2 emissions were lower in the climate change scenario (red box)
compared to the reference climate. This reduction in CO 2 emissions was attributed
to reduced heating demand because of increased outdoor temperatures due to
climate change.
In contrast, only a few low-high scenarios influenced the spread of CO2 emissions.
In particular, the scenarios with the greatest influence were number of occupants
and their corresponding usage of lighting and appliances. In contrast, the scenarios
that influence shell dominated loads (e.g. heating demand) such as heating setpoint
(Thsp) and climate scenarios (CS) did not influence the spread of CO2 emissions. This
difference in influence was attributed to renovation package RP0, which was
dominant in the Pareto solutions. Since the majority of Pareto designs have the RP0
renovation package, the spread of CO2 emissions was largely due to plug loads rather
than to shell loads.
It is intriguing that termination of net-metering (red box) resulted in higher

variations in CO2 emissions, but in a lower spread compared to the presence of net-
metering (white box). This difference in influence on predicted performance and
robustness was probably due to varied capacities of battery of the Pareto solutions.
Higher battery capacities were least influenced by net-metering, whereas smaller
battery capacities were greatly influenced by net-metering scenarios due to reduced
self-consumption of generated electricity. It is worth noting that interaction between
scenarios was present and the influence of each scenario on predicted performance
and robustness was also affected by other scenario combinations.
158
Low scenarios High scenarios
OS Occupant scenario
OP Occupancy profile
Thsp Heating setpoint temperature
Luse Lighting use
Ause Appliance use
IHG Internal heat gains
DHW Domestic hot water use
Vent Ventilation
Wopen Window opening
CS Climate scenarios
NM Net-metering
Figure 6.10 Variation of CO2 emissions and corresponding spread of Pareto solutions with
low and high scenarios for the policymaker. The white box plots represent low scenarios and
filled box plots represent high scenarios. The x-axis represents the sensitivity index (p) for
each scenario. If p<0.05, the scenario is sensitive; if p>0.05, it is less sensitive but more
robust.
159
The best-case and worst-case method: Similarly, in the case of the best-case and worst-
case method all low-high scenarios influenced the predicted performance, and
occupant behavior related scenarios were the most influential scenarios (see Figure
6.11). Among occupant behavior scenarios, number of occupants and their
corresponding usage of lighting and appliances were the most influential scenarios.
In addition, scenarios of high usage of appliance, lighting and corresponding IHG
resulted in the largest deviation. This deviation was due to the worst-case
performance, i.e. very high CO2 emissions of a design due to increased electricity
consumption in these scenarios. In contrast to the max-min method, the termination
of net-metering resulted in higher CO2 emissions and also to a larger deviation
compared to the net-metering policy. This influence was inevitable as termination of
net-metering policy does not consider exported energy in CO2 emission calculations.
Figure 6.11 Variation of CO2 emissions and corresponding deviation of Pareto solutions
with low and high scenarios for the policymaker. The white box plots represent low scenarios
and filled box plots represent high scenarios. The x-axis represents the sensitivity index (p)
for each scenario. If p<0.05, the scenario is sensitive; if p>0.05, it is less sensitive but more
robust.
160
The minimax regret method: In the case of the minimax regret method, similar trends
of influential scenarios can be observed as in the max-min method (see Figure 6.12).
As expected, number of occupants, appliance use, lighting use and corresponding
IHG were the most influential scenarios on CO2 emissions and corresponding
maximum regret.
Figure 6.12 Variation of CO2 emissions and corresponding maximum regret of Pareto
solutions with low and high scenarios for the policymaker. The white box plots represent
low scenarios and filled box plots represent high scenarios. The x-axis represents the
sensitivity index (p) for each scenario. If p<0.05, the scenario is sensitive; if p>0.05, it is
less sensitive but more robust.
Therefore, as demonstrated using the scenario analysis implemented in the CPRA

approach, decision makers can identify the scenarios with the most influence on the
preferred performance indicators and can adopt extra measures to reduce their
influence.
161
6.2.5 Summary
The CPRA approach was demonstrated using a renovation case study building with
the policymaker as a decision maker. The multi-criteria assessment and multi-
criteria decision making were carried out considering three preferred performance
and robustness indicators by the policymaker. In this demonstration, robust designs
were identified using the trade-off and the MCDM approaches. These robust designs
are tabulated for three robustness assessment methods in Table 6.4. Here, with the
exception of the max-min method, it can be observed that both robust design
selection approaches yielded the same robust designs. It is worth mentioning that in
the case of the max-min method, two designs with an ICa of 23.6 k€ and 38.5 k€
respectively had similar spread of CO2 emissions. Furthermore, the design with an
ICa of 38.5 k€ had better predicted performance, however, the improvement in this
predicted performance does not outweigh the required ICa. Hence, in the case of the
max-min method, the design with an ICa of 23.6 k€ was the preferred robust design
using the trade-off approach.
Table 6.4 Comparison of robust designs of the renovation house case study selected for the
policymaker using the trade-off and the MCDM approaches for three robustness assessment
methods.
The most The most The most
Design Robust Robust Robust
robust robust robust
design design design
options design design design
based on based on based on
trade-off trade-off trade-off
MCDM MCDM MCDM
ICa, k€ 23.6 38.5 39.8 39.8 35.2 35.2
RP0 (Rc= RP4 (Rc= RP6 (Rc= RP6 (Rc= RP3 (Rc= RP3 (Rc=
Renovation 1.3/1.3/1.3 6/7/7 6/8.5/10 6/8.5/10 3.5/4.5/6 3.5/4.5/6
package, m2K/W; U m2K/W; U m2K/W; U m2K/W; U m2K/W U m2K/W; U
(RP) = 2.9 = 1.01 = 0.81 = 0.81 = 1.43 = 1.65
W/m2K) W/m2K) W/m2K) W/m2K) W/m2K) W/m2K)
Infiltration,
1 0.1 0.1 0.1 0.625 0.625
dm3/ds2
WWR, % 40 20 20 20 20 20
HVAC
HR107 HR107 HR107 HR107 HR107 HR107
system
PV system,
32 32 32 32 32 32
m2
SDHW
10 10 7.5 7.5 10 10
system, m2
Orientation S S S S S S
Tilt angle, ° 45 45 45 45 45 45
Battery,
20 20 20 20 20 20
kWh
162
In practice, decision makers can choose a robustness assessment method based on

their approach towards risk in the decision-making process. For instance, if the
policymaker opted for a conservative approach to decision making, then the design
with an ICa of 23.6 k€ would be the most preferred robust design based on trade-off,
and the design with an ICa of 38.5 k€ would be the most preferred robust design
based on the MCDM method. Similarly, if the policymaker preferred to have the best
performing design even in the worst-case scenarios, then the design with an ICa of
39.8 k€ would be the most robust design based on the trade-off and the MCDM
approaches. This design has deep renovation package, RP6, and also very large RES
systems. In contrast, if the policymaker opted for a less conservative approach, then
the design with an ICa of 35.2 k€ would be the most preferred robust design. This
robust design represents a compromise between the robust designs selected using
the max-min and the best-case and worst-case methods.
In summary, for quicker identification of robust designs, the MCDM method may
be preferred as both robust design selection approaches implemented in this CPRA
approach yielded similar robust designs, except for the max-min method. The CPRA
approach also provides a decision maker with information to trade off investment in
improving building insulation levels with that of RES systems. In addition, decision
makers could choose design options that were the most robust to the preferred
performance indicators, such as insulation levels or energy generation systems,
individually. For instance, based on results of this case study, it may be wiser to invest
in RES systems rather than improve insulation levels to enhance building design
robustness with respect to CO2 emissions. Similarly, policymakers could use this
approach when defining building codes and regulations based on robust design
options. For example, based on the case study results presented here, building codes
could be upgraded by limiting insulation levels to a certain extent and opting for
larger RES systems. However, these robust design options might differ if the
homeowner is a decision maker. The CPRA approach is demonstrated using the
same case study with the homeowner as decision maker in the next section.
163
6.3 Demonstration of the CPRA approach using the renovation house

case study with the homeowner as a decision maker
The notable difference in the demonstration of the CPRA approach with the
homeowner as a decision maker is that the homeowner has different interests in
building performance and a greater number of preferred performance indicators
than the policymaker, namely ICa, GC and overheating hours. Therefore, multi-
criteria assessment and multi-criteria decision making was carried out considering
these preferred performance indicators and their corresponding robustness
calculated using three robustness assessment methods. However, to enhance
readability, the demonstration is presented only for the minimum regret method in
this section. The demonstration using the other two methods and comparison with
the minimax regret method are presented in Appendix D. Nevertheless, robust
design selected using three robustness assessment methods are summarized at the
end of this section.

The Pareto front for the homeowner calculated using the minimax regret method is
shown in Figure 6.13. The Pareto front for the homeowner was calculated
considering his/her preferred performance and robustness indicators; i.e. global
cost, overheating hours and corresponding robustness, and required additional
investment costs (ICa). This calculation results in a highly complex 5D Pareto front.
However, this 5D Pareto front is shown as two 3D graphs to enhance visualization,
with ICa being the common axis in both graphs. In these figures, each bubble
represents the median value of the predicted performance and the bubble size
depicts the corresponding robustness of a design across the considered scenarios.
For instance, the top figure presents overheating hours (each bubble) and the
maximum regret of overheating hours (bubble size). Similarly, the bottom figure
presents global cost (GC) and the maximum regret of GC. In both of these figures,
the required ICa is shown on the x-axis to facilitate clear visualization of the
alternative designs, which could be used by the homeowner to trade off among global
cost, overheating hours and their corresponding robustness.
The homeowner prioritizes a robust design that has low GC and the lowest
maximum regret of GC (smaller bubble size). In addition, the preferred design
should have less overheating hours and also the lowest maximum regret of
overheating across the considered scenarios. Furthermore, the homeowner can trade
off preferred robust design with required ICa.
164
Renovation house case study - homeowner as a decision maker
Bubble size = Maximum regret of overheating hours

(15-66 h/a)
60
50
40
30
20
10
0
-5 0 5 10 15 20
Bubble size =Maximum regret of global cost

(1.5-20.8 k€)
100
Global cost (GC), k€/30 years
80
60
40
20
0
-5 0 5 10 15 20
Figure 6.13 Pareto front for the homeowner optimized using the minimax regret method.
The top figure represents overheating hours and the bottom figure represents global costs. In
both figures, the corresponding maximum regret is included as bubble size. The green
bubble represents the reference building design.
It can be observed from Figure 6.13 that there are no Pareto solutions beyond an ICa
of 15 k€, indicating that designs with higher investment costs were not robust
solutions for the homeowner using this method. This observation was similar to that
of the best-case and worst-case method, and stands in contrast to results from the
max-min method (see Figure D.1 and Figure D.4 in Appendix D). Furthermore, all
Pareto solutions had low maximum regrets for overheating (<66 h/a). In these
Pareto solutions, the designs with low GC resulted in maximum regrets for
overheating, and the inverse was also true. The range of maximum regret for
overheating was between 15-66 h/a for all designs, and if the homeowner accepts
165
this range of overheating as a trade-off with global cost, then the preferred robust
designs were in the GC (median value) range of 47-60 k€. In this range, it can be
noticed that the designs with ICa ranging from 8-12 k€ are more robust compared to
the designs with ICa ranging from 0-8 k€. Even though these robust designs incurred
additional investment of up to 4 k€, they could reduce the maximum regret of global
costs by up to 19 k€.
Conversely, if the homeowner has low tolerance towards overheating and prefers to
bear more costs, then preferred robust designs that had the lowest maximum regrets
for overheating were in the GC range of 55-60 k€ with the corresponding ICa ranging
from 10-15 k€. In addition, the reference building also had less overheating hours
and relatively low maximum regret for overheating hours. However, this building
resulted in very high GC and corresponding maximum regret. Therefore, to find the
preferred robust design for the homeowner by prioritizing the performance
indicators and corresponding robustness, a few selected designs from different ICa
ranges are compared; see Figure 6.14. The details of these selected Pareto designs
are tabulated in Table 6.5. It can be observed from Figure 6.14 that designs with an
ICa of 0 and 2.6 k€ had the same maximum regret for overheating, which is because
these designs differ only in energy systems (see Table 6.5), which had no influence
on overheating. Similar observations can be made for designs with an ICa of 12.8 and
13.1 k€.
Table 6.5 Details of selected Pareto designs, for further analysis, in different additional
investment cost ranges from the Pareto front of the homeowner calculated using the
minimax regret method.
ICa, K€ 0 2.6 8.6 9.9 10.6 11.1 11.7 12.6 12.8 13.1 14 14.5
Renovation
0 0 1 1 1 1 1 1 1 1 1 1
package (RP)
Infiltration,
1 1 0.5 0.6 0.1 0.5 1 0.4 0.5 0.5 0.1 0.5
dm3/ds2
WWR, % 40 40 20 20 20 20 20 40 40 60 60 60
HVAC system HR107
PV system, m2 0 9.6 0 3.2 9.6 9.6 9.6 9.6 9.6 3.2 9.6 9.6
SDHW
0 0 0 0 0 0 0 0 0 0 0 0
system, m2
Orientation, ° S S E S S S S S S S S S
Tilt angle, ° 0 45 45 0 45 45 45 0 0 45 45 45
Battery, kWh 0 0 0 0 0 0 0 0 0 0 0 0
166
200
overheating hours, h/a
Maximum regret of
150
100
50
0
0.0 2.6 8.6 9.9 10.6 11.1 11.7 12.6 12.8 13.1 14.0 14.5
100
Maximum regret of global
cost (GC), k€/30 years
80
60
40
20
0
0.0 2.6 8.6 9.9 10.6 11.1 11.7 12.6 12.8 13.1 14.0 14.5
Figure 6.14 Variation of overheating hours, global costs (box plots) and corresponding
maximum regret (bar plots) across considered scenarios of selected Pareto designs for
further analysis.
167
By comparing the four designs in the first group of ICa range, it can be inferred that
overheating hours and corresponding maximum regrets increased in line with an
increase in ICa. The increase in overheating was due to improved insulation levels
and airtightness and a reduced WWR. This lack of robustness to overheating with
airtight and highly insulated building envelopes was because of the heat gains due
to IHG and solar gains being trapped in these buildings. Moreover, the potential of
natural cooling will be reduced in the future due to climate change. These
observations are similar to several studies reported in the literature [Zero Carbon
Hub, 2016; Mulville and Stravoravdis, 2016; Rodrigues et al., 2013; Sameni et al.,
2015]. In addition, a small WWR reduces the window opening area, consequently
reducing the potential of natural ventilation in summer, resulting in overheating. In
contrast, these improvements resulted in a reduction of GC (operational costs) and
corresponding maximum regrets. This reduction in operational costs was largely due
to lower heating demands with these designs. Among these designs, the design with
an ICa of 8.6 k€ resulted in the lowest maximum regret for GC and was hence the
most preferred robust design for the homeowner. However, this design resulted in
higher overheating regret hours of about 30 h/a compared to the reference building.
If the homeowner accepts this risk of overheating, the preferred robust design
resulted in a significant reduction of GC regrets of about 13 k€.
In the second group of ICa range, the overheating hours and corresponding
maximum regrets gradually decreased in line with an increase in ICa. In contrast,
the GC and corresponding maximum regrets gradually increased in proportion to
increases in ICa. This contrasting trend was probably due to improved infiltration
rates (up to 1 dm3/ds2) and a high WWR (40%), which reduced overheating, albeit at
high heating demands. In this ICa range, the homeowner would prefer the design
with an ICa of 10.6 k€ if global costs were prioritized. This design had the lowest
maximum regret of GC, which was about 1.56 k€, because this design was more
optimal than other designs for most of the scenarios, and was, hence, the most robust
to GC. Conversely, the design with an ICa of 12.6 k€ was the most preferred if
overheating was prioritized. However, this design would incur 7.7 k€ higher
maximum regret of global cost. To reach a compromise between these performance
indicators and corresponding robustness, the homeowner would prefer the design
with an ICa of 11.1 k€, which had lower maximum regret of overheating hours than
the design with an ICa of 10.6 k€ and also lower maximum regret of GC than the
design with an ICa of 12.6 k€. Therefore, in such cases, it can be concluded that the
168
CPRA approach enhances the design decision-making process by elucidating the

most preferred robust design, especially among similar performing designs.
Similarly, in the last group of ICa range, all designs had similar overheating hours
and corresponding robustness as these designs had the same renovation package
and similar airtightness, except for the design with an IC a of 14 k€. For this design,
the impact of low infiltration was nullified by a large WWR, which enhanced
ventilation due to its large window opening area. The preferred design in this IC a
range solely depends on the GC and corresponding robustness, which can be traded
off with required ICa. Among four designs, the design with an ICa of 12.8 k€ was the
most preferred robust design for the homeowner as it had better predicted
performance of GC and lower maximum regret of GC.
robust design for the homeowner as it had lower maximum regrets for GC and
overheating hours. This design had renovation package RP1 with a small PV system
of 9.6 m2. In addition, using this method, designs with small RES systems without
any battery were the most robust for homeowners.

In order to identify which design options could lead to optimal performance and to
a robust design, the influence of design options on all Pareto solutions for the
homeowner was analyzed, as presented in Figure 6.15 and Figure 6.16. The preferred
design option should have low global cost and the lowest maximum regret of GC
(Figure 6.16) and less overheating hours and the lowest maximum regret of
overheating (Figure 6.15). Only design options that influence overheating such as
renovation package, infiltration rate and WWR are shown in Figure 6.15.
As expected, designs with low insulation levels were less prone to overheating and
were thus the most robust. For instance, RP1 had close to zero maximum regrets for
overheating. In addition, RP1 had the lowest maximum regret of GC compared to
other renovation packages, and was accordingly the most preferred robust design
option for the homeowner. It is intriguing to note that the designs with low
infiltration rates were the most robust to GC. This finding was attributed to the
reduced heating demand with low infiltration rates. On the other hand, designs with
these infiltration rates were prone to overheating and were hence less robust to
overheating.
169
70 70 70

Overheating hours,h/a
60 60 60
50 50 50
40 40 40
30 30 30
20 20 20
10 10 10
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
Figure 6.15 Variation of maximum regret of overheating hours for different design options
of all Pareto solutions for the homeowner. Each bubble represents maximum regret of
overheating hours for a design option. Only design options that influence overheating are
shown here.
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

0 = N-S facing S
1 = HR107 boiler
80 2 = ASHP 80 90 = E-W facing E 80

3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 0 45 90
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
Figure 6.16 Variation of maximum regret of global cost for different design options of all
Pareto solutions for the homeowner. Each bubble represents maximum regret of global cost
for a design option.
It is worth noting from Figure 6.15 that different infiltration rates resulted in similar
maximum regret of overheating, which was attributed to the combination of
infiltration rates and different renovation packages. Designs with large WWRs were
the most robust to overheating because they provide enhanced natural ventilation
due to the larger opening area of windows. WWRs of 40 and 60 were the most robust
to overheating using the minimax regret method. In contrast, these designs with
170
large WWRs were less robust to GC due to their high IC a, despite the reduction in
operational costs with low heating demand. Similar to robust heating systems for the
policymaker, the HR107 gas boiler was the most robust heating option, which was
attributed to no ICa being necessary for HR107 and also to its relatively low
operational costs due to cheap gas prices in the Netherlands. It is worth noting that
renewable energy systems had no influence on overheating, therefore robust design
options for these systems were solely based on GC. Designs with small RES systems
were more robust design options for the homeowner. This finding was attributed to
low operational costs and low ICa, among other factors. Operational costs were less
dependent on the size of PV system, especially in the case of net-metering
termination, as the excess energy exported to the grid does not lower operational
costs.
It can be noted that the maximum regret of GC gradually decreased in proportion to

increases in the size of PV system. This reduction in maximum regrets of GC was
due to the presence of the net-metering policy. In the case of net-metering, the
designs with large PV systems were more optimal compared to designs with small
PV systems. Consequently, large PV systems resulted here in low maximum regrets
of GC. There was no significant benefit in installing solar DHW systems, because
similar variations were observed in GC for designs with and without solar DHW
systems. These comparable variations were probably due to the reduction in
operational costs of these systems being nullified by the required ICa. Similar to the
robust design options for the policymaker, south facing PV and SDHW systems at a
tilt angle of 45° were most robust. In summary, the most preferred robust design
options for the homeowner were renovation package RP1, high infiltration rates, a
large WWR and a small PV system of 9.6m2 at a tilt angle of 45° oriented towards
the south.
The design score of Pareto solutions calculated using the Hurwicz criterion
considering global cost, overheating hours and their corresponding maximum
regrets were compared; see Figure 6.17. The design with the highest score was the
most robust, which implies that this design performs better than all other designs
considering all performance indicators and their corresponding maximum regrets.
171
1
0.9
0.8
0.7
0.6
Design score
0.5
0.4
0.3
0.2
0.1
0
-5 0 5 10 15 20
Figure 6.17 The design scores of Pareto solutions for the homeowner calculated using the
Hurwicz criterion considering predicted performance and the corresponding maximum
regret. The green bubble represents the reference building design.
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9 1 = HR107 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 0 45 90
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
Figure 6.18 The design scores of different design options of Pareto solutions calculated using
the Hurwicz criterion for the homeowner considering preferred performance indicators and
their corresponding maximum regret.
In contrast to the design score of Pareto solutions for the policymaker, here, the
design score did not improve significantly with an increase in ICa, indicating that
172
these designs were not cost-optimal robust solutions. This difference in variation of
design score with ICa for the policymaker and the homeowner is attributed to GC.
However, for the homeowner, ICa is part of GC, which is indirectly considered in
design score calculations. It can be observed from the Figure 6.17 that the design
with an ICa of 11.1 k€ had the highest design score, and was accordingly the most
robust. However, there were several designs with similar scores and, therefore, the
design scores of different design options were compared to find more preferred
robust design options; see Figure 6.18. It is worth recalling that for each design
option, the variation of other design options is presented to show the interaction
among different design options.
The robust design options based on the trade-off (see Figure 6.15 and Figure 6.16)
and MCDM approaches (see Figure 6.17 and Figure 6.18) were found to be similar
for most of the design options, and these design options are summarized in Table
6.6. For instance, designs with renovation package RP1 were the most robust design
options for the homeowner using both robust design selection approaches. Similarly,
designs with low infiltration rates and high WWRs were the most robust design
options as they reduced GC and overheating, respectively, and also improved
corresponding robustness. In addition, designs with small RES systems were the
most robust. It is worth noting that different options of a design parameter resulted
in similar design scores, which is attributed to the interaction between design
parameters as observed in the case of the designs with PV systems of 0, 3.2 and 9.6
m2.
Figure 6.19 shows the variation of overheating hours, global cost and their
corresponding maximum regret for Pareto solutions across all low-high scenarios.
The sensitivity index (p) is shown on the x-axis labels for each scenario. The scenarios
that are not sensitive (p>0.05) to either overheating or maximum regret of
overheating such as heating setpoint, DHW use and net-metering are not shown in
overheating graphs. It can be observed that the variations in overheating increased
in all high scenarios, except for ventilation by means of a mechanical ventilation
system (Vent) and through window opening (Wopen), which is inevitable as higher
ventilation rates remove the heat gains in the buildings resulting in less overheating
hours. Similarly, better shading control can reduce overheating hours significantly,
consequently improving a design’s robustness to overheating (see Figure 6.19).
173
Figure 6.19 Variation of overheating hours, global cost and their corresponding maximum
regret of Pareto solutions with low and high scenarios for the homeowner. The white box
plots represent low scenarios and filled box plots represent high scenarios. The x-axis
represents the sensitivity index (p) for each scenario. If p<0.05, the scenario is sensitive; if
p>0.05, it is less sensitive but more robust.
174
The climate scenarios (CS) were the most influential on overheating, as the reference
climate scenario led to the least overheating and the future climate change scenario
led to the most overheating. These variations in overheating indicate the importance
of considering uncertainties in climate change while designing robust buildings. It
is inevitable that overheating increases with more occupants, as observed in the case
of the high occupant scenario. This increase in overheating was attributed to the rise
in heat gains due to the presence and activity of occupants. Internal heat gains due
to lighting and appliance use were particularly influential scenarios on overheating.
It can be noted that a few low-high scenarios influenced only actual performance; i.e.
variations in maximum regret were similar for both low and high values of a
scenario. For instance, low-high scenarios for appliance and lighting use and their
corresponding IHG influenced global cost, but did not influence the maximum
regrets of global cost. This is because the regrets are calculated based on inter
comparison of designs, and the usage profiles of these scenarios are the same for all
designs, resulting in the same net imported and exported energy for these scenarios,
especially when there was no battery system as in the current case. Even though both
GC and CO2 emissions are calculated based on net imported and exported energy,
CO2 emissions were largely influenced by Ause and Luse, because the Pareto solutions
for the policymaker had large battery systems, and therefore, net imports and exports
of these solutions were varied greatly with low and high Ause and Luse respectively.
In contrast, the temperature setpoints and climate scenarios influenced energy

consumption for different designs, and consequently resulted in variations in global
cost as well as corresponding maximum regret. Therefore, it can be inferred that the
scenarios that influenced shell dominated loads (e.g. heating demand) such as
heating setpoint (Thsp) and climate scenarios (CS) had a greater influence on the
maximum regret of GC than the plug load dominated scenarios did.
6.3.5 Summary
The CPRA approach was demonstrated with the homeowner as a decision maker. In
this demonstration, the multi-criteria assessment and multi-criteria decision making
were carried out with five preferred performance and robustness indictors by the
homeowner. Note that demonstration of the CPRA approach for the homeowner
using the max-min method and the best-case and worst-case method are presented
in Appendix D.
175
Robust designs for the homeowner were identified using the trade-off and the
MCDM approaches, and these designs were compared for three robustness
assessment methods and are presented in Table 6.6. In summary, using the CPRA
approach, the homeowner can opt for a conservative or less conservative approach
while choosing robust designs. As observed in the demonstration for the
policymaker, both robust design selection approaches implemented in this CPRA
approach yielded the same robust design, except for the max-min method. Therefore,
for quicker identification of robust designs, the MCDM method may be preferred
[Rysanek and Choudhary, 2013]. In addition, the homeowner could choose a robust
design by prioritizing either global cost or overheating hours, which can be traded
off with the robustness indicators and required ICa.
The CPRA approach also provides the homeowner with information to trade off
investment in improving building insulation levels with that of RES systems. For
instance, based on this case study’s results, using the best-case and worst-case
method (deviation) and the minimax regret method (maximum regret), the most
robust solution was to double the insulation levels of existing buildings and opt for
small RES systems. These robust design options can deliver the best performance or
close to optimal performance for most of the considered scenarios. Conversely, if the
homeowner opts for a conservative approach (the max-min method) and prefers to
have designs that deliver the desired performance even in extreme scenarios, then
the designs with relatively higher insulation levels, RP3 (Rc = 3.5/4.5/6 m 2K/W; U =
1.3 W/m2K) and large RES systems were the most robust (see Table 6.6).
Using the scenario analysis presented in this CPRA approach, the homeowner could
identify the most influential scenarios and opt for extra measures to reduce this
influence. For instance, in this case study it was evident that the occupant behavior
related scenarios were the most influential and that overheating risk could be
reduced to a large extent by better shading control and enhanced ventilation through
window openings, accordingly improving the design’s robustness to overheating. It
can be inferred from Table 6.4 and Table 6.6 that the robust designs for the
homeowner and the policymaker are very different. Therefore, to reach a
compromise between these two decision makers, the selection of a robust design
must consider the preferred performance and robustness indicators of both the
homeowner and the policymaker. Hence, the CPRA approach is demonstrated using
the same case study with both the homeowner and the policymaker as decision
makers in the next section.
176
homeowner using the trade-off approach and the MCDM method for three robustness
assessment methods.
MCDM MCDM MCDM
ICa, k€ 31.3 22.5 11.2 11.2 11.1 11.1
RP3 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc =
3.5/4.5/6 2.5/2.5/2.5 2.5/2.5/2.5 2.5/2.5/2. 2.5/2.5/2.5 2.5/2.5/2.5
RP m2K/W; U m2K/W; U m2K/W; U 5 m2K/W; m2K/W; U m2K/W; U
= 1.43 = 1.65 = 1.65 U = 1.65 = 1.65 = 1.65
Infiltration,
0.1 1 0.1 0.1 0.5 0.5
dm3/ds2
WWR, % 40 20 20 20 20 20
HVAC
GSHP HR107 HR107 HR107 HR107 HR107
system
PV system,
32 32 12.8 12.8 9.6 9.6
m2
SDHW
5 0 0 0 0 0
system, m2
Tilt angle, ° 45 45 0 0 45 45
Battery,
16 20 0 0 0 0
kWh
6.4 Demonstration of the CPRA approach using the renovation house case
study with both policymaker and homeowner as decision makers
In practice, the preferred robust design of a homeowner should also meet the
requirements of building codes and regulations, generally defined by policymakers.
Similarly, the building codes and regulations defined by policymakers should also
deliver preferred performance for homeowners to improve satisfaction as they are
the end users of these buildings. As noted earlier, the robust designs, selected
separately, differ a lot for these decision makers as they have different interests in
building performance. Therefore, to reach a compromise, the Pareto front for both
policymaker and homeowner combined is optimized considering the preferred
performance indicators and corresponding robustness of both decision makers. This
optimization resulted in a complex 7D Pareto front, which is shown as three 3D
177
graphs with ICa being the common x-axis in all graphs (see Figure 6.20). In all
graphs, the corresponding robustness is included as the bubble size. In contrast to
earlier demonstrations, the preferred robust design for both decision makers should
have low CO2 emissions, less overheating hours and GC, and also high robustness
to CO2 emissions, overheating hours and GC. As mentioned earlier, the results of
CPRA approach demonstration using the minimax regret method are presented here
and the results using the other two methods are presented in Appendix D.

Figure 6.20 shows the Pareto front for both decision makers, calculated using three
robustness assessment methods. In this figure, graphs from top to bottom represent
CO2 emissions, overheating hours and GC, respectively. The notable difference in
this Pareto front compared to the Pareto fronts for the policymaker and the
homeowner calculated separately is that the Pareto solutions can be found in the
entire ICa range of 0-50 k€ for all three methods. This observation is in contrast to
the Pareto fronts for individual decision makers, particularly the homeowner, where
there are no Pareto solutions beyond a certain level of ICa.
It is worth noting that there are different layers of Pareto fronts (Figure 6.20) and
some of these Pareto solutions have higher CO2 emissions and lower robustness to
CO2 emissions compared to the reference building. It is highly intriguing that these
solutions still feature in the Pareto front. Their presence is attributed to lower GC
and/or less overheating hours and/or higher corresponding robustness. The same
can be observed from the overheating and GC graphs (2nd and 3rd row), where these
Pareto solutions have high CO2 emissions, lower GC, less overheating hours and
higher corresponding robustness compared to the reference building. Similar to the
Pareto solutions for the policymaker, it can be observed that the predicted
performance and corresponding robustness of CO2 emissions improved gradually in
line with an increase in ICa. For instance, in the ICa range of 0-15 k€, the predicted
performance and robustness of CO2 emissions significantly improved proportionally
to an increase in ICa. It is worth noting that the designs with similar overheating
hours and corresponding robustness can be found across this entire ICa range. This
similar overheating and corresponding robustness was attributed to the designs
having the same renovation package and airtightness, because the difference among
these designs was mostly different RES systems. Therefore, in this ICa range, both
decision makers should be able to find a trade-off among CO2 emissions, GC and
corresponding robustness.
178
Renovation house case study - both policymaker and homeowner as decision makers
Bubble size =Maximum regret of CO2 emissions

(3.6-5063 kgCO2/a)
6000

5000
4000
3000
2000
1000
0
-5 5 15 25 35 45 55
Additional investment cost (IC a), k€
Bubble size = Maximum regret of overheating hours

(15-711 h/a)
60
50
40
30
20
10
0
-5 5 15 25 35 45 55
Bubble size =Maximum regret of global cost

(2.2-53.9 k€)
100
80
60
40
20
0
-5 5 15 25 35 45 55
Figure 6.20 Pareto front for the policymaker and the homeowner calculated using the
minimax regret method. The figures from top to bottom represent CO2 emissions,
overheating hours and global costs, respectively. In all figures, the maximum regret of CO2
emissions is included as bubble size. The green bubble represents the reference building
design.
179
In the ICa range of 15-30 k€, CO2 emissions further reduced and corresponding
robustness further improved in line with an increase in IC a. In contrast, GC
increased, and the corresponding robustness decreased in line with an increase in
ICa. However, the Pareto solutions with similar overheating hours and
corresponding robustness can be found across this entire range of ICa. Therefore, in
this ICa range, the decision makers can choose between robust designs by making a
trade-off among CO2 emissions, GC and corresponding robustness. In the ICa range
beyond 30 k€, the reduction in CO2 emissions does not outweigh the required ICa.
However, robustness improved greatly in line with an increase in IC a and even the
maximum regret was close to zero for a few Pareto solutions in this ICa range. This
optimal performance was attributed to the inclusion of very high insulation levels in
combination with large RES systems.
Conversely, the GC increased gradually, and the robustness of GC also reduced

considerably (bigger bubble size) in line with an increase in ICa. In addition, the
overheating risk of the Pareto solutions in this ICa range was unacceptably high. This
higher overheating risk was largely due to the very high insulation levels (RP4-RP7)
and airtightness (0.1 dm3/ds2) of these building designs. As noted earlier, the Pareto
solutions in the first ICa range were dominated by designs with large RES systems.
In contrast, the Pareto solutions in the ICa range of 15-30 k€ were dominated by
designs with low to moderate insulation levels in combination with large RES
systems. Similarly, in the ICa range beyond 30 k€, Pareto solutions were dominated
by designs with very high insulation levels and large RES systems.
To provide better insights into the selection of robust designs for both policymaker
and homeowner considering seven preferred performance and robustness
indicators, a few designs selected randomly from the Pareto front are compared in
Figure 6.21. The details of these designs are tabulated in Table 6.7. It can be observed
from Figure 6.21 that different designs were robust to different performance
indicators. For instance, the design with an ICa of 43.8 k€ had the lowest maximum
regret of CO2 emissions and was hence the most robust to CO2 emissions among all
selected designs. This design’s lowest maximum regret of CO2 emissions was
attributed to its optimal performance for most of the scenarios because of its very
high insulation levels and large RES system.
180
Table 6.7 Details of selected Pareto designs, for further analysis, in different additional
investment cost ranges from the Pareto front (see Figure 6.20) of both decision makers
calculated using the minimax regret method.
ICa, k€ 0 5.6 10.3 14.8 16.5 20.4 25.1 31.6 36.1 43.8 44.4 50.4
Renovation
package 0 0 1 1 1 1 1 6 4 7 5 7
(RP)
Infiltration,
1 1 0.4 0.625 0.4 0.4 1 0.1 0.1 0.1 0.1 1
dm3/ds2
WWR, % 40 40 20 20 20 20 20 20 20 20 80 80
HVAC HR107
system
PV system,
0 28.8 6.4 32 32 32 32 28.8 32 32 32 32
m2
SDHW
0 0 0 0 0 0 0 0 5 10 10 10
system, m2
Orientation S S E E S S
Tilt angle, ° 0 45 45 45 0 45 45 45 45 45 45 45
Battery,
0 0 0 0 4 12 20 16 20 20 20 20
kWh
6000
Maximum regret of
5000
4000
3000
2000
1000
0
0.0 5.6 10.3 14.8 16.5 20.4 25.1 31.6 36.1 43.8 44.4 50.4
800
Maximum regret of
600
400
200
0
0.0 5.6 10.3 14.8 16.5 20.4 25.1 31.6 36.1 43.8 44.4 50.4
120
Maximum regret of global
cost (GC), k€/30 years
100
80
60
40
20
0
0.0 5.6 10.3 14.8 16.5 20.4 25.1 31.6 36.1 43.8 44.4 50.4
Figure 6.21 Variation of CO2 emissions, overheating hours, global cost (box plots – left
figures) and their corresponding maximum regret (bar plots – right figures) of the selected
Pareto solutions (see Table 6.7).
However, designs with very high insulation levels in combination with very low
infiltration rates resulted in the highest maximum regret of overheating hours. As
181
such, it may not be the preferred robust design for both decision makers. If so, the
design that is the most acceptable based on the trade-off of all preferred performance
and robustness indicators would be the preferred robust design for both decision
makers. For instance, among the first four designs, the design with an ICa of 14.8 k€
resulted in lower maximum regrets for all preferred performance indicators.
Similarly, the design with an ICa of 20.4 k€ and the design with an ICa of 44 k€ were
the most preferred robust designs among the next four and the last four designs,
respectively.

The design options of the Pareto solutions calculated using the minimax regret
method for CO2 emissions, overheating hours and global costs are shown in Figure
6.22, Figure 6.23 and Figure 6.24 respectively. It is worth remembering that the
design options with the lowest maximum regrets of these indicators are deemed to
be the most robust and worth recalling that for a particular design option, the
variation of other design options is shown to indicate the interaction among different
design options.
It can be noted that the designs with renovation packages RP4, RP6 and RP7 resulted
in the lowest maximum regret for CO2 emissions. Similarly, designs with RP1, RP3
and RP5 also had close to zero maximum regrets for CO 2 emissions. This lowest
maximum regret for CO2 emissions with these different renovation packages was
due to large capacities of RES systems. Among these renovation packages, RP1 had
the lowest maximum regret for GC and overheating hours, and was accordingly the
most preferred robust design option for both decision makers. Similarly, different
infiltration rates such as 0.1, 0.5 and 1 dm3/ds2 resulted in similar maximum regrets
for overheating hours and CO2 emissions, whereas an infiltration rate of 0.1 dm3/ds2
resulted in the lowest maximum regret of GC and was thus the most preferred robust
design option for both decision makers.
It can also be observed that all WWRs resulted in the lowest maximum regret for
CO2 emissions and overheating hours with the exception of a WWR of 20% for
overheating hours. However, the maximum regret for overheating hours with a
WWR of 20% was closer to zero. Furthermore, this WWR had the lowest maximum
regret for GC. Hence, a WWR of 20% was the most preferred robust design option.
It is intriguing that different values of a design option resulted in the same maximum
regret for a performance indicator. For instance, different infiltration rates had
182
similar maximum regret for overheating hours. These comparable variations were
attributed to the interaction between different design options. For instance, the
renovation package with high insulation levels in combination with high infiltration
resulted in low maximum regrets for overheating. Similarly, the renovation package
with low insulation levels in combination with low infiltration rates also resulted in
low maximum regrets for overheating.
6000 6000 6000

5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
6000 6000 6000

1 = HR107

0 = N-S facing S
5000 2 = ASHP 5000 5000

90 = E-W facing E
4000 3 = GSHP 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
1 2 3 0 90 180 0 45 90
6000 6000 6000


5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
Figure 6.22 Variation of maximum regret of CO2 emissions for different design options of
all Pareto solutions for the policymaker and the homeowner. Each bubble represents
maximum regret of CO2 emissions for a design option.
800 800 800
700 700 700

600 600 600

500 500 500
400 400 400
300 300 300
200 200 200
100 100 100
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
Figure 6.23 Variation of maximum regret of overheating hours for different design options
of all Pareto solutions for the policymaker and the homeowner. Each bubble represents
maximum regret of overheating hours for a design option. Only design options that
influence overheating are shown here.
183
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

0 = N-S facing S

1 = HR107
80 2 = ASHP 80 90 = E-W facing E 80
3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 0 45 90
100 100 100


80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
Figure 6.24 Variation of maximum regret of global cost for different design options of all
Pareto solutions for the policymaker and the homeowner. Each bubble represents maximum
regret of global cost for a design option.
It was found that HR107 is the most robust heating system for both decision makers.
Similarly, south oriented PV and SDHW systems at a tilt of 45° were found to be the
most robust options. As expected, the large PV system of 32m2 resulted in close to
zero maximum regrets for CO2 emissions, however, a PV system of 12.8 m2 resulted
in the lowest maximum regret for GC. Hence, both decision makers could opt for a
trade-off between these two PV system sizes. It can be noted that the PV system of
12.8 m2 had very high maximum regrets for CO2 emissions, however maximum
regrets for GC with the PV system of 32 m2 were closer to that of PV system of 12.8
m2. Hence, the PV system of 32 m2 was preferable for both decision makers.
Similarly, designs with an SDHW system of 10 m2 were the most robust to CO2
emissions, whereas the design with no SDHW system was the most robust to GC.
To reach a compromise among these performance indicators, an intermediate
SDWH system size of 5 m2 may be preferred as it has comparatively low maximum
regrets for both CO2 emissions and GC. The same observations can be made for
battery capacity; a large battery of 20 kWh capacity was the most robust to CO2
184
emissions, and designs with no battery were most robust to GC. A battery capacity
of 12 kWh was found to be a reasonable compromise between these performance
indicators as it had comparatively low maximum regrets for both GC and CO 2
emissions. These design options are summarized in order to allow a comparison
with robust designs selected based on the MCDM method (see Table 6.8), which is
discussed in next section.
In summary, the robust design options for both decision makers were renovation
package RP1, low infiltration rates, small WWRs and large RES systems. These
robust design options represent a compromise between the robust design options
for the policymaker and the homeowner selected separately. For instance, designs
with renovation package RP1 were the most robust for both decision makers. This
finding was similar to what was found for the homeowner, but was in contrast to
what was found for the policymaker. Similarly, designs with large RES systems were
the most robust for both decision makers, and this finding was similar to the robust
design options found for the policymaker, but was in contrast to those found for the
homeowner.

The design scores of Pareto solutions calculated using the Hurwicz criterion
considering all preferred performance indicators and their corresponding maximum
regrets by both decision makers are presented in Figure 6.25. In contrast to the
design scores for the policymaker and similar to the design scores for the
homeowner, the design score gradually decreased beyond a certain ICa, indicating
that these designs were not cost-optimal robust solutions. In addition, a multi-layered
Pareto front can be seen, as observed earlier in Figure 6.20, because the design score
significantly increased until 8 k€ ICa and then gradually decreased in proportion to
increases in ICa. The most robust design based on the highest design score was the
design with an ICa of 20 k€. Alternatively, both decision makers could choose
different robust design options based on their highest score (see Figure 6.26) and
can trade off these options with the required ICa. It can be noted that the robust
design options using the MCDM approach were similar to the robust design options
using the trade-off approach, with the exception of SDHW system. These robust
design options were compared and are presented in Table 6.8.
185
1
0.9
0.8
0.7
Design score
0.6
0.5
0.4
0.3
0.2
0.1
0
-5 5 15 25 35 45 55
Figure 6.25 The design scores of Pareto solutions for both policymaker and homeowner
calculated using the Hurwicz criterion considering all preferred performance indicators and
their corresponding maximum regrets. The green bubble represents the reference building
design.
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9
1 = HR107 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 0 45 90
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
Figure 6.26 The design scores of different design options of Pareto solutions for the
policymaker and the homeowner calculated using the Hurwicz criterion considering all
preferred performance indicators and their corresponding maximum regrets.
186
The sensitivity indexes (p) of different low-high scenarios, which quantify the
influence of different scenarios on all preferred performance and robustness
indicators, were compared and are presented in Figure 6.27. In this figure, the
sensitivity index is shown for predicted performance and corresponding maximum
regrets of all preferred performance indicators by both decision makers. It is worth
noting that all scenarios influenced predicted performance of global cost and CO2
emissions (see Figure 6.27), where p=0 in all cases. In contrast, only scenarios that
were sensitive to either predicted performance or robustness of overheating are
shown here. For instance, the net-metering scenario had no influence on
overheating, and hence, is not shown here. In addition, to avoid repetition, variations
of predicted performance and corresponding robustness for different scenarios are
not shown here.
It can be inferred from Figure 6.27 that the number of occupants (OS) and their
corresponding behaviour, especially Ause, Luse and IHG, had a great influence on all
performance indicators and their corresponding maximum regrets. Similarly, CS
also influenced all performance indicators and their corresponding robustness.
However, there were a few scenarios that influenced predicted performance but not
robustness. For instance, all-day occupancy profiles slightly reduced both CO 2
emissions and GC compared to evening profiles, but did not have an impact on
robustness. This difference in influence on predicted performance and robustness
was due to the same average electricity consumption in both occupancy profiles
despite different magnitudes of usage at different times. Similarly, higher ventilation
rates increased CO2 emissions slightly and were thus sensitive to CO2 emissions.
However, these variations were too small to have an impact on the design’s
robustness to CO2 emissions.
The influence of scenarios on different performance indicators and their

corresponding maximum regrets for both decision makers was similar to
observations made for the homeowner and the policymaker separately. The only
exception was in the case of the homeowner for the maximum regret of GC, where
the Ause, Luse and corresponding IHG had no influence on robustness, which
contrasts with the influence of these scenarios for both decision makers combined,
as seen in Figure 6.27.
187
a) CO2 emissions
1.0
Sensitivity index (p) 0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
CO2 emissions Maximum regret of CO2 emissions
b) Overheating hours
1.0
0.9
Sensitivity index (p)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Overheating hours Maximum regret of overheating hours
c) Global cost
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Global cost Maximum regret of global cost
Figure 6.27 Sensitivity of various scenarios to the predicted performance and the
corresponding maximum regrets of preferred performance indicators of both policymaker
and homeowner calculated using the Mann-Whitney test. Scenarios where p<0.05 (dotted
line) are assumed to be sensitive.
188
The notable difference in both cases was the variable battery capacities among Pareto
solutions. For the homeowner, there was no battery system in the Pareto front. In
such cases, the net imported and exported energy of all designs were largely
influenced by shell load dominated scenarios such as T hsp and CS. In contrast, the
plug load dominated scenarios such as low-high Luse and Ause resulted in the same
net imported and exported energy as these profiles were the same for all designs and
therefore had little influence. In summary, occupant behaviour related scenarios and
climate scenarios were the most influential scenarios on all considered performance
indicators.
6.4.5 Summary
The CPRA approach was demonstrated for both decision makers, which resulted in
a multi-criteria assessment and multi-criteria decision making considering seven
preferred performance and robustness indicators. Robust designs selected using the
trade-off and the MCDM approaches for three robustness assessment methods were
compared and are presented in Table 6.8.
policymaker and the homeowner using the trade-off and the MCDM approaches for three
MCDM MCDM MCDM
ICa, k€ 39.1 22.02 29.8 22.05 23.94 20
RP3 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc = RP1 (Rc =
3.5/4.5/6 2.5/2.5/2.5 2.5/2.5/2.5 2.5/2.5/2.5 2.5/2.5/2.5 2.5/2.5/2.5
RP m2K/W; m2K/W; U m2K/W; U m2K/W; U m2K/W; U m2K/W; U
U = 1.43 = 1.65 = 1.65 = 1.65 = 1.65 = 1.65
Infiltration,
0.625 0.1 0.625 0.1 0.1 0.1
dm3/ds2
WWR, % 40 20 40 20 20 20
HVAC
system
PV system,
32 28.8 32 32 32 32
m2
SDHW
10 2.5 7.5 0 5 0
system, m2
Tilt angle, ° 45 45 45 45 45 45
Battery,
16 16 16 16 12 12
kWh
189
As noted earlier, the robust designs for both decision makers (see Table 6.8) were a
compromise between robust designs selected separately for the policymaker (see
Table 6.4) and the homeowner (see Table 6.6). It can be inferred from these tables
that the robust designs for both decision makers, for a particular robustness
assessment method, had the same renovation packages as those of the robust designs
for the homeowner. Similarly, the most robust heating system also follows the same
trend. Conversely, robust designs for both decision makers had similar RES systems
as robust designs for the policymaker for the corresponding robustness assessment
method.
Therefore, based on this case study’s results, it can be concluded that the most robust
renovation measure for an existing building was to improve the insulation level (2.5-
6 m2K/W) to a certain extent and then opt for large RES systems. These robust design
options are in contrast to sustainability frameworks and standards such as Trias
energetica and Passivhaus standards, where reducing the heating demand as much
as possible by including highly insulated and airtight building envelopes was the
main criterion. However, as stated earlier, many studies have reported high
overheating risks during summer in these buildings and these risks are bound to
increase in the future due to climate change, as observed in this research. These
overheating risks can cause serious health issues to tenants of all ages and especially
to elderly people. Instead, the environmental impact of the existing building stock
can be reduced to a large extent by the inclusion of large PV systems. The
corresponding grid stress caused by large PV systems can be reduced by including
batteries for energy storage. This comparison of different robust design options is
further elaborated in the next section.
In summary, using the CPRA approach, both decision makers can opt for a
conservative or less conservative approach while choosing robust designs. For
instance, if both decision makers opt for a less conservative approach, the most
robust option is to double the insulation levels of existing buildings and opt for large
RES systems to enhance existing building robustness based on this case study’s
results. Conversely, if both decision makers opt for a conservative approach and
prefer designs that deliver the required performance even in extreme cases, then the
designs with high insulation levels (Rc = 3.5/4.5/6 m2K/W) and large RES systems
were the most robust. Based on scenario analysis on this case study, it was evident
that the occupant behavior related scenarios and climate scenarios were the most
influential on the considered performance indicators and corresponding robustness.
190
New house case study
6.5 Demonstration of the CPRA approach using the new house case
study
The CPRA approach was demonstrated using the new house case study, described
in Chapter 5, to facilitate the policymaker, the homeowner and both decision makers
to find robust design options for the new buildings. However, this demonstration is
the same as that of renovation building and hence, only a summary of this
demonstration for all decision makers is presented in this section. The results of the
new house case study demonstration are presented in Appendix E.
6.5.1 Policymaker as a decision maker

The robust designs for the policymaker identified using the trade-off and the MCDM
approaches for three robustness assessment methods were compared and are
presented in Table 6.9. The most robust design options for the policymaker using
all three robustness assessment methods are summarized below.
The max-min method: If the policymaker opts for a conservative approach, then the
design with an ICa of 23 k€ that has building envelope package P1 and large RES
systems was the most preferred robust design based on trade-off (see Figure E.1 and
Figure E.2), and the design with an ICa of 28.19 k€ that has building envelope
package, P3 and large RES systems was the most preferred robust design based on
the MCDM approach (see Figure E.3 and Figure E.4). This contrast in the preferred
robust designs based on these two approaches was attributed to ICa, as noted earlier,
which was not considered in the design score calculations. This resulted in the
designs with high insulations and large RES systems as the most robust using the
MCDM approach, compared to that of trade-off approach.
The best-case and worst-case method: If the policymaker prefers to have the best
performing design even in the worst-case scenarios, then the design with an ICa of
33.6 k€ that has building envelope package P4 and large RES systems (see Figure E.1
and Figure E.2) and the design with an ICa of 40.9 k€ that has very high insulation
levels (P5) and large RES system (see Figure E.3 and Figure E.4) were the most
preferred robust designs based on the trade-off and the MCDM approaches,
respectively.
The minimax regret method: If the policymaker opts for a less conservative approach,
then the design with an ICa of 31.9 k€ that has building envelope package P3 and
large RES systems (see Figure E.1 and Figure E.2) and the design with an ICa of 39.1
191
k€ that has very high insulation levels (P5) and large RES systems (see Figure E.3 and
Figure E.4) were the most preferred robust designs based on the trade-off and the
MCDM approaches, respectively.
The notable difference between the Pareto solutions and corresponding robust
designs of the renovation and the new case studies was that for similar ICa, new
house robust designs have higher insulation levels compared to renovation house.
This difference was attributed to the higher initial investment cost of the reference
new house (P0) compared to the reference renovation house (RP0), which therefore
resulted in lower ICa of design alternatives for new building designs. Accordingly,
the robust designs of the new house have much higher insulation levels compared
to the renovation house.
Table 6.9 Comparison of robust designs of the new house case study with the policymaker
as decision maker, selected using the trade-off and the MCDM approaches for three
Robust The most The most The most
Design Robust Robust
design robust robust robust
design design
options based design design design
based on based on
on based on based on based on
trade-off trade-off
trade-off MCDM MCDM MCDM
ICa, k€ 23 28.19 33.6 40.9 31.9 39.1
P1 (Rc = P3 (Rc = P4 (Rc = P5 (Rc = P3 (Rc = P5 (Rc =
Building 3.5/4.5/6 5/7/8 6/8.5/10 10/10/10 5/7/8 10/10/10
envelope m2K/W; m2K/W; U m2K/W; U m2K/W; U = m2K/W; U m2K/W; U
package, P U = 1.43 = 1.01 = 0.81 0.55 = 1.01 = 0.55
Infiltration,
0.15 0.15 0.1 0.1 0.1 0.1
dm3/ds2
WWR, % 20 20 20 80 40 40
Building
S S S S S S
orientation
Thermal Light- Light- Heavy- Heavy- Heavy- Heavy-
mass weight weight weight weight weight weight
HVAC
HR107 HR107 HR107 HR107 HR107 HR107
system
PV system,
32 32 32 32 32 32
m2
SDHW
7.5 7.5 10 10 10 10
system, m2
PV and
SDHW S S S S S S
Orientation
Tilt angle, ° 45 45 45 45 45 45
Battery,
20 20 16 20 20 20
kWh
192
Like observations made for scenario analysis on the renovation house case study, it
was found that the number of occupants and their corresponding behavior were the
most influential scenarios on CO2 emissions and corresponding robustness (see
Figure E.5). In contrast to the renovation house case study, the heating setpoints had
little influence on CO2 emissions. This difference in influence was attributed to the
Pareto solutions that were dominated by building envelope packages with very high
insulation levels. These very high insulation levels resulted in very low heating
demands and consequently low CO2 emissions which were less influenced by
heating setpoints. Furthermore, the variation in this low heating demand had little
influence on CO2 emissions compared to the plug loads in these buildings.
6.5.2 Homeowner as a decision maker

The robust designs for the homeowner identified using the trade-off and the MCDM
approaches for three robustness assessment methods were compared and are
presented in Table 6.10. It is worth noting that both approaches to robust design
selection lead to the same robust designs with the best-case and worst-case method
and the minimax regret method, which is similar to the policymaker. However, using
the max-min method, robust designs selecting using the trade-off and the MCDM
approaches were slightly different in terms of battery capacity.
Like renovation house case study results, the designs with small RES systems and
low insulation levels were the most robust using the best-case and worst-case method
and the minimax regret method. These robust designs are in contrast with the robust
designs for the policymaker, which was attributed to their low operating and
investment costs and to better comfort; the main preferred performance indicators
of the homeowner. Similar to robust designs for the policymaker, designs with large
RES systems were the most robust with the max-min method, as this method leads
to conservative robust designs. The robust design options for the homeowner using
three robustness assessment methods are summarized below.
The max-min method: If the homeowner opts for a conservative approach, then the
design with an ICa of 17.2 k€ that has building envelope package P1 and large RES
systems was the most preferred robust design based on trade-off (see Figure E.6,
Figure E.7 and Figure E.8) and the design with an ICa of 15.2 k€ that has building
envelope package P1 and large RES systems was the most preferred robust design
based on the MCDM approach (see Figure E.9 and Figure E.10).
193
The best-case and worst-case method: If the homeowner prefers the best performing
design even in the worst-case scenarios, then the design with an ICa of 5.9 k€ that
has building envelope package P1 and small RES systems was the most preferred
robust design based on the trade-off and the MCDM approaches.
The minimax regret method: The most robust design using this method (see Figure
E.6 - Figure E.10) is the same as that of robust designs using the best-case and worst-
case method (see Table 6.10).
Table 6.10 Comparison of robust designs for the new house case study with the homeowner
as a decision maker, selected using the trade-off and the MCDM approaches for three
MCDM MCDM MCDM
ICa, k€ 17.2 15.2 5.9 5.9 5.9 5.9
P1 (Rc = P1 (Rc = P1 (Rc = P1 (Rc = P1 (Rc = P1 (Rc =
Building 3.5/4.5/6 3.5/4.5/6 3.5/4.5/6 3.5/4.5/6 3.5/4.5/6 3.5/4.5/6
envelope m2K/W; U m2K/W; U m2K/W; U m2K/W; U m2K/W; m2K/W; U
package, P = 1.43 = 1.43 = 1.43 = 1.43 U = 1.43 = 1.43
Infiltration,
0.1 0.1 0.4 0.4 0.4 0.4
dm3/ds2
WWR, % 20 20 20 20 20 20
Building
S S S E S E
orientation
Thermal Light- Heavy- Heavy- Heavy- Heavy- Heavy-
HVAC
HR107 HR107 HR107 HR107 HR107 HR107
system
PV system,
32 32 12.8 12.8 12.8 12.8
m2
SDHW
0 0 0 0 0 0
system, m2
PV and
SDHW S S S S S S
Orientation
Tilt angle, ° 45 45 0 0 0 0
Battery,
20 16 0 0 0 0
kWh
It can be noted that the designs with low insulation levels in combination with small
RES systems provide cost-optimal robust solutions for the homeowner and are
therefore the most preferred robust designs. Furthermore, using the best-case and
194
worst-case method and the minimax regret method, there were no designs including
a battery system among Pareto solutions, and this was attributed to high ICa of
batteries in addition to small RES systems, where most of the generated energy was
being utilized onsite. Similarly, designs with SDHW systems were not cost-optimal
robust solutions for the homeowner, which is similar to renovation house case study
results. The number of occupants and their corresponding behavior related scenarios
were the most influential on overheating and GC (see Figure E.11). In contrast to the
policymaker, the heating setpoints had a very large influence on GC and this
influence was attributed to low insulation levels of the Pareto solutions for the
homeowner and therefore the shell load dominated scenarios, such as heating
setpoints and CS, were the most influential. The same can be observed with climate
change because it had the least influence on CO2 emissions, but had very high
influence on GC as well as overheating hours. Furthermore, NM had no influence
on GC due to very small RES systems among Pareto solutions.
6.5.3 Both policymaker and homeowner as decision makers

Table 6.11 provides an overview of the robust designs selected based on the trade-off
(see Figure E.12-Figure E.15) and the MCDM (see Figure E.16 and Figure E.17)
approaches for both decision makers. These designs were compared for three
robustness assessment methods. It can be inferred from Table 6.11 that robust
designs were dominated by large RES systems, which was similar to the policymaker
but in contrast to the homeowner. Likewise, robust designs have low to moderate
insulation levels, which represents a compromise between the robust designs for the
homeowner and the policymaker selected separately. It can also be noted that robust
designs selected using the trade-off and the MCDM approaches are different, which
is in contrast with that of robust designs for the homeowner, but similar to that of
robust designs for the policymaker.
The max-min method: Similar to observations made with renovation case study
results, designs with low insulation levels (P1) and large RES systems were the most
robust for both decision makers.
The best-case and worst-case method: Designs with moderate insulation levels were the
most robust for both decision makers. Designs with large PV system (32m2), SDWH
systems (0-10 m2) and relatively small battery capacities (16 kWh) were the most
robust.
195
The minimax regret method: Designs with low to moderate insulation levels were the
most robust for both decision makers. Other robust design options are in line with
the robust design options using the best-case and worst-case method.
Similar to observations made earlier for the policymaker and the homeowner, the
number of occupants and corresponding behavior such as Ause, Luse and IHG, and
climate change were the most influential scenarios on all considered performance
and robustness indicators for the both decision makers combined (see Figure E.18).
The robust design options for different decision makers are compared for the new
house and the renovation house case studies in the next section.
Table 6.11 Comparison of robust designs for the new house case study with the policymaker
and the homeowner as decision makers, selected using the trade-off and the MCDM
approaches for three robustness assessment methods.
MCDM MCDM MCDM
ICa, k€ 25.2 17.2 29.2 21.5 26.9 19.1
P1 (Rc = P1 (Rc = P2 (Rc = P2 (Rc = P1 (Rc =
Building P3 (Rc =
3.5/4.5/6 3.5/4.5/6 6/7/7 6/7/7 3.5/4.5/6
5/7/8
envelope m2K/W; U m2K/W; U m2K/W; m2K/W; U m2K/W; U
m2K/W; U =
package, P = 1.43 = 1.43 U = 1.01 = 1.01 = 1.43
1.01W/m2K)
W/m2K) W/m2K) W/m2K) W/m2K) W/m2K)
Infiltration,
0.625 0.1 0.1 0.1 0.625 0.1
dm3/ds2
WWR, % 40 20 20 20 40 20
Building
S S S/E S S E
orientation
Thermal Light- Light- Heavy- Heavy- Heavy- Heavy-
HVAC
system
PV system,
32 32 32 32 32 28.8
m2
SDHW
7.5 0 10 0 10 0
system, m2
PV and
SDHW S S S S S S
Orientation
Tilt angle, ° 45 45 45 45 45 0
Battery,
20 20 16 16 20 12
kWh
196
Comparison of robust design options for both case studies
6.6 Comparison of robust design options for both case studies

In this section, the robust design options for different decision makers selected using
the three robustness assessment methods are compared for both case studies. The
purpose of this comparison is to provide information to the end users to choose
robust cost-optimal design options. This comparative assessment was an attempt to
provide informed design decision choices to the decision makers, which is similar to
the web-based tool known as the energy savings explorer developed by [RVO, 2017a].
6.6.1 Building envelope package

Designs with renovation package RP1 (Rc = 2.5/2.5/2.5 m2K/W; U=1.65 W/m2K)
were the most robust for the homeowner (see Table 6.6) and both decision makers
combined (see Table 6.8). This was probably due to the low ICa of RP1, consequently
resulting in cost-optimal robust solutions for the homeowner. In addition, designs
with this renovation packages resulted in less overheating hours. Furthermore, CO2
emissions associated with the designs with this renovation package could be reduced
to a significant extent by the inclusion of large RES systems and, therefore, this was
the most preferred robust renovation package for both decision makers. It is worth
noting that this renovation package [Agentschap NL, 2011] was sufficient to meet the
Dutch national targets of achieving energy label B by 2020 [Hermelink et al., 2013].
In addition, RP1 was the most robust renovation package for all robustness
assessment methods, with the exception of the max-min method, where designs with
RP3 were also found to be the most robust. This difference in robust renovation
package was attributed to reduction in GC in proportion to increases in ICa, using
the max-min method (see Figure D.2 in Appendix D), because the difference
between the maximum and minimum GC of a design was reduced due to low
operational costs. The low operating costs were due to improved self-consumption
of electricity, as these designs had GSHP as a heating system, which resulted in an
all-electric design.
In contrast to the most robust renovation packages for the homeowner and both
decision makers combined, the renovation packages with high insulation levels were
the most robust for the policymaker (see Table 6.4) using the best-case and worst-
case method and the minimax regret method. These renovation packages resulted in
optimal/best performance, which was crucial in optimizing the robustness with
these methods. However, using the max-min method, even the reference building
(RP0) with large RES systems also yielded the same robustness as that of RP4 (Rc =
197
6/7/7 m2K/W; U=1.01 W/m2K), as the variations in CO2 emissions can be reduced
to a significant extent by the inclusion of large RES systems.
In the case of the new house case study, the building envelope packages with very
high insulation levels (P3-P5) were the most robust options for the policymaker using
the best-case and worst-case method and the minimax regret method (see Table 6.9-
Table 6.11). These robust options were attributed to two factors; one was the optimal
performance with high insulation levels and the other was low ICa compared to
corresponding RP as noted earlier. In the case of the max-min method, as observed
for the renovation house case study results, the building envelope package P1 (Rc =
3.5/4.5/6 m2K/W; U=1.43 W/m2K) also resulted in the same robustness as that of P3
(Rc = 5/7/8 m2K/W; U=1.01 W/m2K). In contrast, the building envelope package P1
was the most robust for the homeowner for all three robustness assessment
methods.
Improving the insulation levels beyond the current Dutch standards [RVO, 2016a]
was not a cost-optimal robust option for the homeowner. However, when both
decision makers are considered, the building envelope packages P1-P3 were the most
robust for all three methods. As expected, the building envelope packages with low
insulation levels were the most robust using the max-min method. Conversely, since
the other two methods optimized robustness with respect to best
performance/optimal performance, building envelope packages with high insulation
levels (P2-P3) that meet the nZEB (BENG) and ZEN standards [RVO, 2016b; RVO,
2015b] were the most robust for both decision makers.
6.6.2 Infiltration (building airtightness)

There was no consistent trend among robust designs with infiltration rates for
different decision makers. Typically, designs with high infiltration rates were the
preferred robust design options for the homeowner as they were robust to
overheating. However, designs with low infiltration rates were the most robust to GC
due to low heating demand with these infiltration rates. In addition, the designs with
low infiltration rates in combination with low insulation levels were also the most
robust to overheating. Similar to several studies reported in the literature, designs
with very high insulation levels and low infiltration rates were prone to overheating
risk. For the policymaker, designs with low to moderate infiltration rates were the
most robust for both renovation and new buildings. This observation was valid for
the homeowner and also for both decision makers combined.
198
6.6.3 Window to wall ratio (WWR)

Designs with WWRs of 20-40% were the most robust options for all decision makers
in both case studies, with the exception of a few methods, as seen in Table 6.4, Table
6.6, and Table 6.8 for the renovation house and Table 6.9-Table 6.11 for the new
house. Typically, low WWRs were the most robust to GC due to their low ICa.
Interestingly, the reduction in heating demand with high WWRs did not outweigh
required ICa. Similar observations can be made for CO2 emissions. In contrast,
overheating hours were reduced significantly with high WWRs due to enhanced
natural ventilation, because of increased window opening area with these WWRs.
6.6.4 Building orientation and thermal mass

For the homeowner and both decision makers combined, buildings oriented towards
the south were the most robust option for all decision makers with the exception of
the best-case and worst-case method and the minimax regret method. This difference
was attributed to less overheating hours with the east oriented building despite the
fact that south orientation maximizes passive solar gains in the building and
consequently reduces heating demand. Since these two methods optimize
robustness with respect to optimal performance, the design options that yielded
optimal performance (east orientation for overheating in this case) might also result
in better robustness.
Heavy-weight buildings were the most robust for all decision makers using the best-
case and worst-case method and the minimax regret method. In contrast, light-
weight buildings were the most robust using the max-min method. It is worth
remembering that for the renovation house case study, the orientation (south) and
thermal mass (heavy-weight) were fixed.
6.6.5 HVAC system

In all cases, the HR 107 gas boiler was the most robust heating system, with the
exception of the max-min method when the homeowner was the decision maker. HR
107 gas boilers were more preferred due to the very low ICa of these boilers and to
the similarity of emission factors for the primary energy of all considered heating
systems. Furthermore, in the Netherlands, gas prices are cheaper than electricity
prices [CBS, 2016a; RVO, 2017b], resulting in low operational costs. In contrast, the
GSHP was the most robust heating option for the homeowner using the max-min
199
method. This robustness was a result of low variations in operational cost with
increased self-consumption of electricity with GSHP as a heating system.
To move towards all-electric dwellings, it may be required to provide subsidies for

ASHP and GSHP systems to encourage homeowners to prefer ASHP/GSHP over
HR107 gas boilers. Alternatively, the heat based grid system could be a feasible
option to prefer ASHP/GSHP over HR107 gas boilers [Manrique et al., 2018].
6.6.6 RES system

Typically, designs with large RES systems were the most robust for all decision
makers with the exception of the best-case and worst-case method and the minimax
regret method, when the homeowner was a decision maker. For instance, designs
with a PV system of 28.8-32 m2 were the most robust for all decision makers, except
for the homeowner using the best-case and worst-case method and the minimax
regret method (see Table 6.6 and Table 6.10). This difference for the homeowner
was probably due to designs with large PV systems not being cost-optimal robust
solutions. Designs with large SDHW systems (7.5-10 m2) were the most robust
options for the policymaker in both case studies. In contrast, designs with very small
or no SDHW were the most robust for the homeowner because these SDHW
systems do not outweigh benefits in reduction of GC compared to the required ICa.
Similarly, designs with large batteries were the most robust for the policymaker,
whereas battery systems do not feature on Pareto solutions for the homeowner except
for the Pareto solutions calculated using the max-min method.
In the case of the homeowner, designs with small PV systems of 9.6-12.8 m2 were
the most robust using the best-case and worst-case method and the minimax regret
method. Here, a large portion of this small amount of generated energy was utilized
onsite, and hence batteries for storage were not essential. Furthermore, designs with
batteries are not cost-optimal solutions. However, in the case of the max-min
method, designs with large PV systems were the most robust, and hence, large
batteries for storage were essential to maximise the utilisation of energy in the net-
metering termination scenario. Otherwise, the benefit of installing large PV systems
is nullified in case of net-metering termination.
Robust design options of RES systems for the both decision makers represent a
compromise between the robust designs of RES systems for the policymaker and the
homeowner selected separately. For instance, designs with battery capacities of 12-
200
16 kWh (see Table 6.8 and Table 6.11) were the most robust for both decision makers,
which represented a compromise between the preferred battery capacity for the
policymaker (20 kWh) and the homeowner (0 kWh except for the max-min method).
These observations were valid for both case studies.
Therefore, building designs with low-moderate insulation levels in combination with

large RES systems were the most robust for both decision makers combined and
could be sufficient to achieve a 90% reduction in CO 2 emissions by 2050 [EFFRA,
2013] and to move towards a de-carbonized economy [European Commission, 2011].
In addition, large RES systems are becoming more financially viable because the
price of PV has reduced three-fold in the last decade [E.ON, 2017a].
6.6.7 Orientation and tilt angle of PV and SDHW systems

South facing PV and SDHW systems at a tilt angle of 45° were the most robust option
in all cases, except for the best-case and worst-case method, where a tilt angle of 0°
was the most robust.
It is worth noting that the most robust design option for a decision maker was largely
influenced by required ICa. The ICa of these design options was calculated based on
investment cost of different design options collected from literature, websites and
reports [Fleiter et al., 2016; dubbelglasweetjes, 2016; Kingspan insulation, 2016].
However, there are uncertainties associated with these costs and these uncertainties
were not considered in this research to reduce the complexity and to focus on the
main aim of this dissertation i.e. to develop and demonstrate the CPRA approach.
Considering uncertainties in investment costs could lead to different robust designs,
but the CPRA approach remains the same and this could be part of future work.

The CPRA approach was demonstrated using case studies of renovation and new
houses with the policymaker, the homeowner and both as decision makers. The
multi-criteria assessment and multi-criteria decision making was carried out with
multiple (3-7) performance and robustness indicators based on preferences of both
decision makers. Robust designs selection was illustrated with different approaches
to facilitate decision makers with different decision-making options, and thus, to
enhance the design decision-making process. Scenario analysis was carried out to
identify the most influential scenarios on the preferred performance and robustness
indicators.
201
The following conclusions can be drawn from the demonstration of the CPRA
approach using both case studies:
⎯ The CPRA approach, as demonstrated, can be used by a decision maker to

select robust designs based on optimal (actual) performance and
performance robustness across considered scenarios. Using the multi-
criteria assessment implemented in this study, the decision maker can
choose a robust design by prioritizing a performance indicator and carry out
trade off with the other performance and robustness indicators and required
additional investment cost.
⎯ The CPRA approach also provides a decision maker with information to trade
off investment in improving the building envelope with that of RES systems.
For instance, policymakers can use this approach when defining building
codes and regulations based on robust design options such as limiting
insulation levels to a certain extent and opting for large RES systems as
observed from these case studies results.
⎯ In the case of a conservative approach in the decision-making process, the

designs with high insulation levels, RP4 for the renovation house and P3 for
the new house, and large RES systems were the most robust for the
policymaker. In contrast, the designs with low insulation levels, RP1 for the
renovation house and P1 for the new house, and the large PV system and
large battery were the most robust for the homeowner. A compromise
between these robust designs, i.e. designs with low insulation levels, RP1 for
the renovation house and P3 for the new house, and with large RES systems
(PV = 28.8-32 m2, SDHW= 0-2.5 m2 and battery capacity = 16-20 kWh), were
the most robust for both decision makers combined.
⎯ If the decision maker opts for the best performance in all scenarios, then the
designs with very high insulation levels, RP6 for the renovation house and
P5 for the new house, and large RES systems were the most robust for the
policymaker. In contrast, the designs with low insulation levels, RP1 for the
renovation house and P1 for the new house, and with small RES systems were
the most robust for the homeowner. The designs with low to moderate
insulation levels, RP1 for the renovation house and P2 for the new house, and
with large RES systems were the most robust for both decision makers.
202
Concluding remarks
⎯ If the decision-making process involves certain risk acceptance, then the

designs with moderate to very high insulation levels, RP3 for the renovation
house and P5 for the new house, and with large RES systems were the most
robust for the policymaker. In contrast, the designs with low insulation levels,
RP1 for renovation house and P1 for the new house, and with small RES
systems were the most robust for the homeowner. The designs with low to
moderate insulation levels, RP1 for renovation house and P3 for the new
house, and with large RES systems were the most robust for both decision
makers.
⎯ The max-min method yielded conservative robust designs and the minimax
regret method yielded optimal robust designs. In the case of the homeowner,
the preferred robust designs based on the max-min method have large RES
systems and, in contrast, the designs with small RES systems were the most
preferred robust designs with the minimax regret method. Similarly, in the
case of the policymaker, the preferred robust deigns with the minimax regret
method have very high insulation in combination with large RES systems,
because they lead to optimal and robust performance. Contrariwise, the
preferred robust designs based on the max-min method have very low
insulation and large RES systems because the variations in CO2 emissions
could be reduced to a significant extent by the inclusion of large RES systems.
⎯ In the case of the renovation house, doubling the insulation levels of the
existing house and opting for large RES systems is the most preferred robust
option for both decision makers based on the current case study results.
These insulation levels meet the requirements of energy label B, which is
also a requirement of the Dutch government for the renovation by the end of
2020. Similarly, for new houses, improving insulation levels beyond current
Dutch standards (P1) does not yield significant benefits. Airtight to
moderately airtight buildings with infiltration rates of 0.1-0.625 dm3/ds2
were the most robust for both decision makers. These robust design options
are in contrast to guidelines in sustainability frameworks and standards such
as Trias energetica or Passivhaus standards, where reducing the heating
demand as much as possible by including highly insulated and airtight
building envelopes was the main criterion. Similar to many studies in the
literature, buildings with highly insulated and airtight building envelopes
203
resulted in high overheating risks during summer and these risks are bound
to increase in the future due to climate change.
⎯ The HR107 gas boiler was the most preferred robust heating option for both
decision makers combined, compared to ASHP and GSHP, because of low
investment cost, and low operating costs owing to cheap gas prices. Also, the
HR 107 had a similar emission factor of primary energy as those of other
heating systems. Subsidies may be required for ASHP and GSHP
installations in order to move towards all electric dwellings.
⎯ Designs with large PV systems (28.8-32 m2) were the most robust for the
policymaker and also for both decision makers combined, because they
reduced CO2 emissions to a significant extent. In contrast, these large PV
systems were not robust for the homeowner as they were not cost-optimal.
Designs with PV system of 9.2-12.8 m2 were the most robust for the
homeowner in all cases. Batteries enhanced building robustness with respect
to net-metering scenarios for the policymaker and for both decision makers
combined. However, battery was not a cost-optimal robust option for the
homeowner, because the most robust designs for the homeowner has small
PV systems. SDHW was not a cost-optimal robust design option as these
systems are expensive despite the added advantage of reduction in
operational costs.
⎯ The designs with south facing RES systems at a tilt of 45° were the most
robust design options in all cases, as these design options maximized the
energy production. The advantage of improved self-consumption of
electricity with E-W orientation in the case of large PV systems was nullified
by the inclusion of battery storage systems in both case studies.
204
7. Suitability and usability
assessment of the CPRA
approach
This chapter demonstrates the suitability and usability assessment of the developed CPRA
approach in practice with the help of end users. The iterative process used to improve the
CPRA approach by incorporating the feedback from end users is discussed. Results of the
suitability and usability assessment using practical case studies are presented in this
chapter.
7.1 Overview
The suitability and usability of the developed CPRA approach was assessed with a
user group comprising of potential end users of this approach. A total of four
meetings were conducted for this purpose, as shown in Figure 7.1, to iteratively
improve the CPRA approach in order to arrive at the final approach. The outcome of
the suitability and usability assessment including the feedback from user group
meetings and improvement of the CPRA approach are discussed in the following
sections.
7.2 Evaluation of CPRA approaches with the end users
In the first user group meeting, an initial version of the CPRA approach was
presented. In addition, the need for performance robustness assessment in current
design practice was highlighted. The max-min method to evaluate performance
robustness (spread) of the design space across the considered future scenarios using
performance indicators based on decision makers’ preference was presented. The
selection of the most robust design for different decision makers based on their
preferred predicted performance and spread was discussed. In the next meeting, the
improved CPRA approach that included feedback from the first meeting was
presented to the user group. In this improved approach, the minimax regret method
to evaluate performance robustness was introduced to the user group.
205
Chapter 7. Suitability and usability assessment of the CPRA approach
Present initial CPRA Present improved Present improved Present improved

(CPAR0) using case CPRA (CPRA1) using CPRA (CPRA2) using CPRA (CPRA3)using
study results case study results case study results case study results
The final CPRA

Start User group meeting-1 User group meeting-2 User group meeting-3 User group meeting-4
approach
Receive feedback Receive feedback Receive feedback Receive feedback
Figure 7.1 Suitability and usability assessment of the CPRA approach at different user
group meetings.
The simulation methods used in the CPRA approach were also discussed in this
meeting. In the third meeting, it was demonstrated through a case study how
designers and consultants can use the developed approach in practice. In the last
meeting, the improved CPRA approach, which included feedback from all the
previous user group meetings, was demonstrated using the renovation house case
study.
7.3 Feedback from user group meetings

The feedback from user group meetings was generally positive. The feedback also
varied greatly due to the different backgrounds of user group members. However, all
user group members acknowledged the importance of performance robustness
assessment and the need for a holistic approach for performance robustness in
current design practice. In addition, user group members provided valuable feedback
on the developed CPRA approach from a practitioners’ point of view. Questions and
suggestions on various aspects of the CPRA approach by user group members in
different user group meetings are summarized below. The response to the main
feedback is documented in Appendix F.
⎯ Applicability of the CPRA approach
• Can the developed CPRA approach be generalized? Can it be tested using

new buildings as well as renovated buildings?
• Is the developed approach limited to only buildings? Can it be used to design
robust energy systems such as HVAC or renewable energy systems?
• What is the practical use of this approach?
206
Feedback from user group meetings
• Can this approach be used to improve energy performance predictions with

the Dutch EPC assessment to get closer to actual performance during
operation i.e. reducing the performance gap?
• How can outcomes of this research be transferred to the end user/customers?
For instance, how can this approach be used by a real estate company in
selling a house?
⎯ Design space, scenarios and performance indicators
• Suggestions on including different households in occupant scenarios.

• Relevance of performance indicators. For instance, choosing electricity
consumption as a performance indicator while not taking into account
electricity generation is not appropriate.
⎯ Performance robustness
• Does performance robustness consider risks associated with not meeting the
required design parameter in construction e.g. airtightness, which is very high
in practice?
• Does the robustness assessment also consider the risk that the building will
not function properly due to failure of energy systems? Robustness could also
mean the risk of setting the system incorrectly (higher risk for more
complicated systems, risk of failure of complex systems)?
⎯ Visualization methods
• How can a decision maker/ designer choose a robust design using the
developed CPRA approach?
• Are there any novel methods considered to compare multiple performance
indicators at once?
• How can results be demonstrated to clients in an easy way (e.g. one
sheet/slide) to enhance the design decision-making process?
⎯ Computational aspects
• What are the computational costs associated with the CPRA approach and
how can these computational costs be reduced?
It is worth noting that this feedback was received at different stages of the CPRA
approach and was incorporated into the CPRA approach after every user group
meeting. This process is explained in detail in the next section.
207
7.4 Improvement of CPRA approach using iterative process

The feedback of each meeting was incorporated in an iterative process and the
updated CPRA approach was presented in the next meeting (see Figure 7.1).
In the first user group meeting, feedback was mostly on robustness assessment.
User group members highlighted the importance of assessing the risk associated
with the failure of energy systems and malfunctions due to incorrect settings.
However, these uncertainties fall outside the scope of this research, as the focus is
on performance robustness assessment rather than system robustness. It is worth
noting that the CPRA approach is unaffected by including these uncertainties. The
focus of performance robustness in the present research was made clear in the
subsequent meeting. In addition, defined performance indicators based on the
decision makers’ preferences were updated based on the feedback from the previous
user group meeting. As such, costs are considered in the place of electricity
consumption. By doing so, it was possible to take into account cost savings due to
energy generation, which is one of the main criteria for end users in the design
Based on the feedback in the second user group meeting, more visualization
methods were included in the CPRA approach. Scenario sampling and multi-
objective optimization methods were implemented in the CPRA approach to reduce
computational costs associated with this approach. Similarly, implementation of this
approach in practice, which was a main discussion point in the second meeting, was
also addressed. The CPRA approach was generalized so that it can be used for
performance robustness assessment of both new and renovation houses.
Furthermore, it can be used for performance robustness assessment in a holistic
approach and for robustness assessment of individual systems such as HVAC, PV
and SDHW systems. These improvements were incorporated and the updated CPRA
approach was presented to the user group in the following meetings to demonstrate
its practical use.
In the last user group meeting discussion, the applications of this approach were
highlighted, such as reducing the performance gap, risk assessment and
performance guarantee. For instance, using this approach, performance range can
be given as a contracting option rather than nominal performance. In other words,
regret/spread/deviation can be added in contracting options in addition to actual
performance. Furthermore, the developed CPRA approach can be used to find risk
208
Improvement of CPRA approach using iterative process
averse solutions, which could be useful in energy performance contracting options.

Similarly, using the robustness indicators presented in this research, it is easy to
quantify the risk associated with preferred design options.
However, some suggestions from user group members, such as assessing the
usability of the approach in the EPC assessment to improve energy performance
predictions, were not implemented. This assessment falls outside the scope of this
research as the method implemented for energy performance predictions in this
approach is different from the method used in EPC assessment. The implementation
of the CPRA approach for EPC assessment requires further research. Similarly, to
reduce the complexity of the case studies, uncertainties in thermo-physical properties
and the uncertainties in not meeting the required design parameters in the
construction stage (e.g. infiltration rates) were not considered. These two suggestions
could inform future work.
7.5 Practical use of the developed CPRA approach
The suitability and usability of the final CPRA approach was assessed using practical
case studies with the help of PDEng and MSc thesis projects. This assessment is
illustrated below using relevant results from these studies. Detailed analysis can be
found in their full reports [Puranik, 2017; Costa, 2017].
7.5.1 Homij case study - risk-averse’ design solutions for Homij using building
performance simulations.
The purpose of this project [Puranik, 2017] was to assess the risks associated with
not achieving cost neutral (“nota-nul” in Dutch) bills for Homij buildings
[“MorgenWonen,” 2016] under uncertainties. The developed CPRA approach was
implemented in this project to reduce these risks by identifying robust designs
considering uncertainties in occupant behaviour and net-metering policies. The
results of this assessment were further used to investigate the business models for
Homij for energy performance contracting options under uncertainties.
The case study building is an all-electric nearly zero energy residential building. In
this project, similar uncertainties in occupant behaviour and net-metering scenarios
as in current research were used. The performance of the design space comprising
of multiple insulation levels and energy system capacities was assessed with costs,
CO2 emissions and energy matching indices under uncertainties to achieve a robust
design. The minimax regret method was used for the robustness assessment of these
209
performance indicators. An illustration of the implementation of the developed

CPRA approach in this project is shown in Figure 7.2 and Figure 7.3.
Figure 7.2 The Pareto front of the design space considering annual electricity costs, total cost
of ownership for energy measures (TCOe) and performance robustness (regrets). Numbers
show the designs chosen based on additional investment cost (AIC); for further analysis,
see [Puranik, 2017].
Figure 7.3 Variation of electricity costs for the selected Pareto solutions in the order of
increasing additional cost of investments for energy measures [Puranik, 2017].
It can be observed from these figures that similar visualization methods to those
implemented in the CPRA approach were used to present the results to stakeholders
of this project. In this study, it was found that designs with a Passive house standard
envelope and large PV systems were more robust to electricity costs. However, this
robustness comes at high total cost of ownership. The cost-optimal robust design is
210
Practical use of the developed CPRA approach
the Dutch current standard envelope with large PV systems. The results of other
performance indicators can be found in the full report [Puranik, 2017].
In summary, the developed approach was successfully implemented in this project

to identify robust design options to reduce the risk associated with energy
performance contracting. The outcome of this project is in line with the outcome of
the current research; that is, large energy systems are more robust options than
improved building insulation levels.
7.5.2 Woonberijf case study – planning for the future: developing a risk averse
strategy for future-proof nearly zero energy buildings
The purpose of this study [Costa, 2017] was to identify the most risk-averse first steps
to renovate the existing building stock of the social housing company, Woonbedrijf.
This study was carried out to achieve Woonbedrijf’s energy efficiency goals of
achieving Energy Label B by 2020 and a nearly zero energy building stock by 2045
[Oudenand and Gal, 2014]. To meet this purpose, the developed CPRA approach was
applied to a typical Woonbedrijf’ residential building (terraced house) to identify the
most robust roadmap. A roadmap is a combination of different renovation packages
that are applied at different points in time to achieve Woonbedrijf’s energy efficiency
goals by 2045. Further details of roadmaps can be found in the full report of this
project [Costa, 2017].
In this project, it was found that decision makers preferred the minimax regret
method among the three robustness assessment methods used in the CPRA
approach. As such, the minimax regret method was used to assess the robustness of
the roadmaps for future scenarios. These future scenarios considered uncertainties
in occupant behavior, emission factors and prices. It is worth noting that in this
project uncertainties in occupant behavior are grouped in two types of behavior:
energy saving and energy wasting behavior. Performance robustness assessment
was carried out with four key performance indicators; namely CO2 emissions,
internal rate of return, energy costs and comfort, which are relevant to both
Woonbedrijf and the residents.
The results of the robustness assessment were presented to the decision makers
using the visualization methods shown in Figure 7.4-Figure 7.6. Different
visualization methods were employed in this project as it was easy to represent all
considered scenarios due to their limited amount.
211
Annual CO2 emissions, 8000

7000
6000
kgCO2/a
5000
4000
3000
2000
1000
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Roadmap
Figure 7.4 CO2 emissions and performance regrets (represented as X) of different roadmaps
across the considered scenarios. Diamonds represent a constant gas emission factor and
triangles represent a switch after 15-years to natural gas grid scenarios [Costa, 2017].
3500
Annual energy costs, €/a
3000
2500
2000
1500
1000
500
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Roadmap
Figure 7.5 Annual energy costs and performance regrets (represented as X) of different
roadmaps across the considered scenarios (represented as diamonds) [Costa, 2017].
2500
2000
1500
1000
500
0
0 1 2 3 4 5 6 7 8 9
Renovation Package
Figure 7.6 Variation of overheating hours of each renovation package considering the best
occupant behavior (represented as blue diamonds), the worst occupant behavior
(represented as red diamonds) and the performance regrets of each renovation package
(represented as X) [Costa, 2017].
212
Practical use of the developed CPRA approach
In this study, it was found that the roadmaps with deep renovation are more robust
to internal rate of returns on investment and CO 2 emissions (see Figure 7.4), the
preferred performance indicators of Woonbedrijf. However, deep renovation
packages are at high risk of overheating for tenants, as seen in Figure 7.6.
Furthermore, the robustness of these roadmaps to annual energy costs is very similar
to that of roadmaps with low-medium insulation levels.
The developed CPRA approach was successfully implemented to identify the most
risk-averse first steps to renovate the existing building stock to meet Woonbedrijf’s
energy efficiency goals. Considering the preferences of Woondbedrijf and the
residents, it was found that the most risk-averse first step was to upgrade the
insulation of the building stock to Energy Label A (medium insulation levels, Rc of
1.47/6/3.5 m2K/W for wall, roof and floor) and to install large PV systems.
7.5.3 Mitros housing corporation case study – future-proof residential flats:

towards the development of robust energy measures to achieve ‘nul-op-de-
meter’ for clustered residential buildings
This project is currently being investigated [Middendorp, 2018] to identify robust
energy measures for the Mitros housing corporation that ensure “nul-op-de-meter”
(zero-on-the-meter) for clustered residential buildings and to achieve ‘Flat met
Toekomst’ (future-proof flat). The nul-op-de-meter [RVO, 2015c] concept for multi-
residential buildings has been initiated by the consortium ‘Nieuw Utrechts Peil’
[“Nieuw Utrechts Peil,” 2017], consisting of a construction company, an architect and
Nieman Consultancy. This concept is aimed at generating as much energy as is
consumed by a building and its occupants over a year.
The developed CPRA approach is used to assess the robustness of different measures
across future scenarios that consider uncertainties in climate change, occupant
behavior and energy prices to identify robust measures that meet zero-on-the-meter.
In this project, different measures such as insulation systems, efficient HVAC
systems and renewable energy systems are assessed for thermal comfort and annual
energy consumption and generation.
213

The suitability and usability of the developed CPRA approach was assessed with end
users through mock-up presentations. The feedback from the end users was
implemented through an iterative process to improve the developed CPRA approach.
The final CPRA approach has been successfully implemented in practical case
studies.
User group members recognized the need for a holistic approach for performance
robustness assessment in current design practice. It was found that low
computational costs are one of the main criteria for the implementation of the
developed CPRA approach in practice and visualisation methods are crucial in the
design decision-making process. The minimax regret method is the most preferred
robustness assessment method of the end users of the considered real case study
buildings. The results of the practical case studies are in line with the outcomes of
this research.
214
8. Conclusions, original
contributions, limitations and
future work
This chapter summarizes the CPRA approach and draws the main conclusions from the
developed CPRA approach, the implemented simulation framework, the case study
demonstrations and from the suitability and usability assessment of the CPRA approach.
In addition, this chapter describes the original contributions and limitations of this
research and concludes by indicating future research directions.
8.1 Summary
To address the increasing global concerns regarding climate change, low-energy
houses that provide high indoor environmental quality at low or no CO 2 emissions
are essential. These low-energy houses are typically achieved by integrating energy
efficiency measures and renewable energy technologies in the built environment,
which requires significant economic investment. Therefore, it is important to ensure
that these measures deliver the desired performance and thereby improve the
satisfaction of the end users. Uncertainties in occupant behavior, climate change and
policy changes influence the building performance greatly, causing variations in
energy, cost and comfort. Therefore, the aim of this dissertation was to develop a
computational approach that integrates uncertainties in building performance
assessment and quantifies the impact of these uncertainties to facilitate decision
makers in identifying preferred robust designs that are capable of delivering desired
performance in future operation.
As a first step to meet the aim of this dissertation, different uncertainty sources that
influence the building performance were reviewed in Chapter 2. It was found that
uncertainties in occupant behaviour and external factors such as climate change and
policy changes were major influencing parameters. Similarly, different methods to
quantify the impact of these uncertainties were reviewed and robustness assessment
was found to be the most appropriate method in the present context. In addition,
different robustness assessment methods were reviewed from general literature and
215
Chapter 8. Conclusions, original contributions, limitations and future work
in the building performance context to find the most appropriate method for
robustness assessment using scenario analysis. The max-min method, the best-case
and worst-case method and the minimax regret method were selected to address
different attitudes towards risk acceptance by decision makers. To facilitate the
selection of a robust design from a large design option space, different multi-criteria
decision-making methods used for decision making under uncertainties were
reviewed and the Hurwicz MCDM method was preferred as it allows a decision
maker to be cautious (risk free), optimistic (risk prone) or to opt for a compromise
between the optimistic and pessimistic approaches in the design decision-making
process.
The next steps taken to meet the aim of this dissertation can be summarized as:
1. Develop a computational performance robustness assessment (CPRA) approach that

integrates uncertainties in multi-criteria assessment using scenario analysis to quantify
robustness and facilitate robust design selection for various decision makers.
A methodology to develop and test the computational performance robustness

assessment (CPRA) approach was proposed in Chapter 3. The developed CPRA
approach was improved in an iterative process considering the feedback from
the end users through user group meetings. The final CPRA approach
comprised multi-criteria performance assessment and multi-criteria decision
making considering performance robustness, among other performance
indicators. In this approach, by prioritizing the decision maker’s preferences,
building design space, future scenarios, performance and robustness indicators
were defined.
The performance of the design space for future scenarios was assessed using
building performance simulations with multiple performance indicators and
corresponding robustness. The max-min method, the best-case and worst-case
method and the minimax regret method were used to evaluate robustness in this
approach. The Hurwicz multi-criteria decision-making method was used to rank
designs for different decision makers by prioritizing the performance and
robustness indicators based on their preferences. A sensitivity analysis was
carried out to identify the most influential scenarios on performance and
performance robustness. All of these results were presented using various
visualization methods to enhance the decision-making process.
216
Summary
2. Develop a simulation framework to implement the CPRA approach in order to carry

out performance robustness assessment of low-energy houses across future scenarios.
The developed CPRA approach was implemented using a simulation

framework, which was presented in Chapter 4. A GA-based multi-objective
optimization method and an ULH scenario sampling technique were
implemented to make this framework computationally efficient and, thus, to
enhance its usability in practice. Integration of deviation and maximum regret
into the optimization framework was not possible using a typical GA fitness
function. Therefore, the GA fitness function was modified to integrate
robustness indicators into the optimization process. It was found that the
optimal settings of GA parameters vary greatly depending on the objectives of
optimization. Therefore, optimal settings for each robustness assessment
method were determined.
3. Demonstrate how the CPRA approach can aid various decision makers to identify
robust designs based on their preferences using case studies.
The CPRA approach was demonstrated using the renovation and the new houses
case studies for policymaker, homeowners and both decision makers combined,
as presented in Chapter 6. The multi-criteria assessment and multi-criteria
decision making was carried out with multiple (3-7) performance and robustness
indicators based on the preferences of the considered decision makers. Robust
designs for both case studies were compared for all decision makers using three
robustness assessment methods, as presented in Chapter 6. The suitability and
usability of the developed CPRA approach was assessed with end users through
mock-up presentations, as presented in Chapter 7. The final CPRA approach was
successfully implemented in practical case studies and illustrated using selected
examples, as presented in Chapter 7.
This CPRA approach can be used by a decision maker to select robust designs
based on optimal (actual) performance and performance robustness across
considered scenarios. Using the multi-criteria assessment implemented in this
approach, the decision maker (e.g. homeowner) can choose a robust design by
prioritizing a performance indicator (e.g. GC) and carrying out trade-off with the
other performance indicators (e.g. overheating hours) and robustness indicators
and required additional investment cost.
217
The CPRA approach also provides a decision maker with information to trade
off investment in improving building envelope with that of RES systems. In
addition, decision makers can choose design options that are more robust to the
preferred performance indicators separately. Using the MCDM method, a
decision maker can easily select the most robust design from the design space
based on design score or can trade off the selected designs based on the design
score with required additional investment cost. This method is also useful for
quicker identification of robust designs form a large design space [Rysanek and
Choudhary, 2013]. Using scenario analysis, a decision maker can identify the
scenarios with the greatest influence on the performance and robustness
indicators and can adopt extra measures to reduce this influence.
The developed CPRA approach can be a first step for the designers and consultants
among other actors involved in the decision-making process to identify robust low-
energy building designs. The CPRA approach could be useful when various decision
makers are involved in a project with multiple performance requirements, and
effective in identifying a robust design from a large design space. This approach
could be an important step in many practical applications such as reducing the
performance gap, risk assessment, and performance guarantee. The developed
CPRA approach can be used to find risk averse solutions, which are prominent in
energy performance contracting options.
8.2 Main conclusions

The following main conclusions can be drawn from this research:
⎯ Scenarios can be used as formulated alternatives in cases when probabilities

of uncertainties are unknown. The integration of such approaches in
building performance predictions can provide a better understanding of the
impact of uncertainties and also facilitate decision making during the design
selection process with the goal of choosing a design that is robust to a variety
of possible future situations.
⎯ Literature review showed that mean and variance, the widely used
robustness indicators based on the Taguchi method, are of limited use in
robustness assessment using scenario analysis. The likelihood of the
occurrence of any scenario is usually unknown, and taking the mean across
scenarios nullifies the concept of formulating scenarios as alternatives since
it flattens out the results. In addition, it was found that it is necessary to
218
Main conclusions
include both actual performance and performance robustness in the

selection process of robust designs.
⎯ Low-high scenario combinations are sufficient for performance robustness

assessment, which can reduce computational time by about 98% compared
to assessment with all scenario combinations. In addition, using the ULH
sampling strategy on low-high scenarios, a sample of 100 scenarios is the
smallest sample size that yields similar performance robustness to low-high
scenario combinations for the considered performance indicators.
⎯ The choice of robustness indicator depends on the decision maker’s attitude

towards risk acceptance in the decision-making process. The results showed
that the max-min method can be used when a design has to deliver the
desired performance for all scenarios including extreme scenarios, whereas
the minimax regret method can be used when a design should deliver
optimal or close to optimal performance for each scenario. The max-min
method is considered to be a conservative approach as it yields a robust
design that has the least variations even in extreme scenarios. Conversely,
the minimax regret method is a less conservative approach as it yields a
robust design that performs as closely as possible to the optimal
performance for every scenario.
⎯ The max-min method and the best-case and the worst-case method can be
used when the cost/risk associated with the failure of a design is very high.
The minimax regret method is suitable when a decision maker can accept a
certain range of performance variation; for instance, a homeowner can
accept designs with certain overheating hours as a trade-off with global costs
and corresponding robustness.
⎯ The method of calculation of objective function of optimization process

differs for different robustness indicators. For instance, spread is calculated
for each design of the design population without any inter-comparison of
the performance of other designs of the population, and thus one design at
a time is considered for calculating the objectives. Contrariwise, maximum
regret and deviation are calculated with inter-comparison of the
performance of other designs. Therefore, objectives are calculated after
calculating the performance of the entire population. Furthermore, for each
generation, objectives are calculated considering the current design
population and design archive of previous generations because of the need
219
for inter-comparison of the performance of all designs. In such cases, the

typical GA fitness function may not be sufficient, so the modified fitness
function as presented in this research may be preferred.
⎯ The matching index of a Pareto front with optimal settings can be improved
by up to 90% on average in comparison to default MATLAB values for GA
parameters (around 66-71.5%) for different robustness assessment
methods. A matching index of 100% was achieved when using deviation as
a robustness indicator for this case study with the policymaker as a decision
maker.
⎯ The renovation case study demonstration results reveal that doubling the
insulation levels of the existing house and opting for large RES systems is
the most preferred robust design option for both decision makers. These
insulation levels meet the requirements of energy label B, which is also a
requirement for renovation buildings by the Dutch government by the end
of 2020. These robust design options are in contrast to guidelines in
sustainability frameworks and standards, where reducing the heating
demand as much as possible by including highly insulated and airtight
building envelopes was the main criterion. Similar to many studies in
literature, the buildings with highly insulated and airtight building
envelopes resulted in high overheating risks during summer and these risks
are bound to increase in the future due to climate change as observed in this
research.
⎯ The results of the new house case study demonstration reveal that improving
the insulation levels beyond the current Dutch standards (P1) [RVO, 2016a]
is not a cost-optimal robust option for the homeowner. However, when both
decision makers are considered, the designs with the building envelope
packages with similar and even higher insulation levels (P1-P3) are the most
robust with all three methods. Using the max-min method, designs with
building envelope packages with lower insulation levels (P1) are more robust
than higher insulation levels. In contrast, the other two methods, which
optimize robustness with respect to the best performance/optimal
performance building envelope packages with higher insulation levels (P2-
P3) are the most robust for both decision makers. These insulation levels
meet the nZEB (BENG) and NZEB (ZEN) standards [RVO, 2016b; RVO,
2015b].
220
Main conclusions
⎯ Designs with large RES systems were the most robust design options for the
policymaker and for both decision makers combined. Based on the case
study results, it may be wiser to subsidise the installation of RES systems
rather than improve insulation levels beyond a certain limit as robustness to
CO2 emissions and costs can be improved to a great extent by including large
RES systems. Furthermore, this would negate the overheating risks that are
intrinsic to very highly insulated and airtight building envelopes.
⎯ Scenarios that influence predicted performance do not necessarily influence

robustness. The occupant behavior related scenarios had the greatest
influence on all considered performance indicators. In addition, net-
metering and climate scenarios influenced CO2 emissions and global costs,
whereas the climate change scenario had the most influence on overheating
risks.
⎯ User group members recognized the need for a holistic approach for
performance robustness assessment in current design practice. It was
understood that computational costs are one of the main criteria for the
implementation of the developed CPRA approach in practice and that
visualisation methods are crucial in the design decision-making process.
The minimax regret method is the most preferred robustness assessment
method for the end users of the considered practical case studies. These case
studies’ results are in line with the outcomes of this research, such as
designs with low to moderate insulation levels and large energy systems are
found to be more robust than designs with very high insulation levels and
small RES systems.
8.3 Original contributions of this research
The novel premise of the proposed holistic CPRA approach lies in the integration of
uncertainties in multi-criteria assessment using scenario analysis and quantification
of robustness to facilitate the selection of robust designs for various decision makers.
For instance, adoption of robustness assessment methods in the building
performance context from other fields is a novel approach. In addition, integration
of robustness indicators by modifying the fitness function in multi-objective
optimization to enhance computational effectiveness is also a novel contribution in
this CPRA approach. Novel contributions of the proposed CPRA approach are
described below.
221
i. Multi-criteria assessment and multi-criteria decision making considering

robustness
In practice, various decision makers with multiple performance

requirements and different attitudes towards risk acceptance are involved in
a project. As such it is important to quantify robustness considering these
requirements to make informed design decisions, which is rarely addressed
in the building performance context. To address this issue, multi-criteria
performance assessment and multi-criteria decision making considering
performance robustness is implemented in the CPRA approach [Kotireddy
et al., 2018]. This multi-criteria assessment enables various decision makers
to choose robust designs from a large design space based on preferred
performance indicators and their corresponding robustness. In contrast to
multi-criteria robustness assessment implemented in literature, the current
assessment also accounts for optimistic and pessimistic approaches in
decision making.
ii. Integration of robustness assessment methods in the building

performance context
Adoption of different robustness assessment methods commonly used in

other fields such as statistics, operations research etc. in the building
performance context is a novel approach in this research [Kotireddy et al.,
2017b]. In addition, another novelty of this research is the use of a modified
fitness function in order to integrate robustness indicators as one of the
objectives of multi-objective optimization. The method of calculation of
objective function differs for different robustness indicators as some of the
robustness indicators has to be evaluated after the calculation of the
performance of the entire design population. Due to inter-comparison of
performance of designs to evaluate these robustness indicators, the design
population of each generation and the design archive of previous
generations are stored. In such cases, the typical GA fitness function is not
recommended; instead, a novel approach by using a modified fitness
function is implemented in this research to integrate robustness indicators
into the simulation framework to simultaneously optimize performance
robustness and actual performance.
222
Original contributions of this research
iii. Design approach considering balance between energy demand and energy
generation
In this research, a novel design approach considering balance between

energy demand (improving building envelope) and energy generation
(installing RES systems) is investigated to identify robust low-energy
building designs [Kotireddy et al., 2015; Kotireddy et al., 2017a]. This
approach is in contrast with sustainability frameworks and standards, which
recommend highly insulated and airtight building envelopes to reduce
heating demand as much as possible. The case study results revealed that
designs with high insulation levels and airtight envelopes are prone to
overheating as reported in literature and that achieving a balance between
insulation levels and size of energy systems is essential to achieve cost-
optimal robust designs.
8.4 Limitations and future research

⎯ In this research, despite considering many scenarios, uncertainties in
technology innovation and in the energy market are not included. In
addition, uncertainties in investment costs are also crucial as the preferred
robust design depends on additional investment costs. Considering these
uncertainties might enhance performance predictions, especially in long
term performance assessment, and therefore they could be included in
future work.
⎯ The implemented GA-based optimization can only provide an

approximation of the true Pareto front and therefore there is still a small risk
of missing some robust designs, as discussed in Chapter 4. This risk might
be reduced by implementing different optimization techniques [Nguyen et
al., 2014]. However, the CPRA approach remains the same. Determining the
optimal settings of GA parameters largely depends on the considered design
space and objective functions, consequently making them case study
dependent. In practice, it is a tedious task to determine the optimal settings
of GA parameters for every case study as well as for different decision
makers. Hence, the hyper-heuristic optimization method is recommended
[Cowling et al., 2002; Kumari et al., 2013] to optimize the optimization
process in addition to the design space. This method has been seldom used
in the building performance context and represents a good area for future
work.
223
⎯ It was noted that climate adaptive building shells can enhance building
robustness [Loonen et al., 2013; Loonen et al., 2017]. However, they are not
considered in this research for reasons of scope. Similarly, this research does
not include districts, neighborhoods or commercial buildings. It might be
interesting to assess if building robustness can be enhanced considering
low-energy communities rather than just standalone low-energy house.
Therefore, the developed CPRA approach can be used for performance
robustness assessment of novel building components, districts and
commercial buildings, but requires further research.
⎯ Suitability and usability assessment was carried out through mock-

presentations to a focus group. Online surveys and interviews might yield
better results. To enhance the suitability of this approach in practice, a good
graphical interface is required to aid the design decision-making process, as
pointed out by user group members. Similarly, the integration of robustness
assessment methods into the design phase of any building project will
provide added value to the end user. These aspects are interesting areas for
future work.
⎯ The CPRA approach can be integrated into a design tool to enhance

confidence in design decisions. This integration could be very useful in
practice and also a promising research area to explore. For instance, the
adaption of the CPRA approach for use in the Dutch EPC calculations could
enhance performance predictions and thus reduce the performance gap.
The assessment of current labelling systems considering robustness in
order to ensure required performance in the future could also be an
interesting research area to explore, as achieving high energy labels is not
necessarily a guarantee of the required performance [Majcen et al., 2013].
Similarly, integrating robustness (risk) into the energy contracting options
would yield better customer satisfaction. Using this approach, performance
range (regret/spread/deviation) can be given as a contracting option rather
than nominal/deterministic performance. However, this integration
requires further research.
224
Appendices
Appendix A. Glossary: high-performance buildings
This appendix lists and defines different high-performance buildings. These definitions are
derived from a review of relevant literature.
A.1 Overview
In the EPBD concerted action related to the terms and definition of high-
performance buildings, 23 identified terms relate to one of the following three main
aspects [Erhorn and Erhorn-Kluttig, 2011]:
i. low-energy consumption (energy-efficient house, low-energy house,

passive house, zero-energy house, plus-energy house etc.)
ii. low emissions (zero-carbon house, zero-emission house etc.)
iii. Sustainable or green aspects (Green buildings, bioclimatic house etc.)
Although these houses have many commonalities, they also differ in some respects.
Despite all these houses following low energy strategies [Kibert and Fard, 2012], the
definitions and the requirement of the so-called low energy varies greatly from
country to country. Therefore, this appendix aims to provide a comparison of these
definitions with a specific focus on the Netherlands. The glossary of definitions of
these buildings [Mlecnik, 2012; Voss et al., 2012a] is summarized below.
A.2 Energy-efficient building

An energy-efficient building is defined as a “building that has energy-efficient
technologies to provide high indoor environmental quality at low energy use, when operating
as designed, compared to similar buildings” [Meier, Alan; Olofsson, Thomas; Lamberts,
2002]. These buildings typically consume far less energy and produce much lower
CO2 emissions than traditional buildings, and require far less operating costs [Kibert
and Fard, 2012].
A.3 Low-energy building

A low-energy building is defined as a building “built according to special design criteria
aimed at minimizing the building’s operating energy” [Sartori and Hestnes, 2007]. Like
energy-efficient buildings, the energy consumption of low-energy buildings should
225
Appendices
be as low as possible. For instance, in the Netherlands it is about 60 kWh th/m2a for
space heating (and cooling) for low-energy buildings and this heating demand is
further lowered to about 30 kWhth/m2a for very low-energy buildings (see Figure A.1
and Table A.1). Similarly, primary energy use, which includes space heating and
cooling, domestic hot water needs and plug loads, should be less than or equal to 120
kWh/m2a for a low-energy building classification [Mlecnik, 2012]. It is worth noting
that the terms energy-efficient buildings and low-energy buildings are often used
interchangeability. In addition, in some cases, several other high-performance
buildings such as passive houses and nearly zero energy buildings are referred as
low-energy buildings [Kibert and Fard, 2012].
A.4 Passive building/Passive house

“A passive house is a building, for which thermal comfort (ISO 7730) can be achieved solely
by post-heating or post-cooling of the fresh air mass, which is required to achieve sufficient
indoor air quality conditions – without the need for additional recirculation of air.”
[Passipedia, 2017]. Passive houses are typically achieved by incorporating airtight and
very highly insulated building envelopes, resulting in ultra-low energy buildings that
require a very small amount of heating and/or cooling energy. Based on passive
house standards [PHI, 1990], the space heating demand of a passive house building,
calculated using the passive house planning package (PHPP), should be limited to
15 kwhth/m2a. Similarly, the primary energy consumption for space heating,
domestic hot water and auxiliary equipment without lighting and appliances is 45
kWh/m2a [Mlecnik, 2012], and this figure rises to about 120 kWh/m2a if it includes
lighting and appliances.
A.5 Nearly zero-energy buildings (nZEB)

A nearly zero-energy building is defined as a “building that has very high energy
performance resulting in nearly zero or a very low amount of energy demand, which is
covered to a large extent by renewable energy produced on-site or nearby” [EPBD, 2010;
Cole and Fedoruk, 2015]. However, there is little information regarding the so-called
“very low amount of energy demand” [Kurnitski et al., 2012]. In addition, it is not
clear what is included or excluded in this very low-energy demand [Voss et al., 2012a].
In the Netherlands, based on “Bijna Energie Neutrale Gebouwen” (BENG = nZEB)

standards, the annual heating and cooling demand should be less than or equal to 25
kWhth/m2a [RVO, 2015b]. The primary energy use for heating, ventilation, cooling
and domestic hot water should be less than or equal to 25 kWh/m2a and the share of
renewable energy should be at least 50%.
226
Glossary: high-performance buildings
A.6 Net zero-energy buildings (NZEB)

There are as many as 70 definitions and concepts of NZEB [Musall and Voss, 2013].
In addition, researchers have adopted different boundaries in the calculation of
NZEBs (e.g. energy flows from and to the grid), and have considered different types
of energy in the energy balance (e.g. heating, heating + DHW, heating + DHW +
plug loads etc.) and used different temporal energy matches (see Figure A.2).
NZEB is defined as a building that can meet all of its ‘energy requirements from low-
cost, locally available, non-polluting, renewable sources’ or, ‘more specifically a building[s]
that generates enough renewable energy on site to equal or exceed its annual energy use.’
[Torcellini et al., 2006; Cole and Fedoruk, 2015]. Similarly, annual balance can be
achieved by onsite energy generation or by importing energy from off-site energy
sources such as wind farms and PV farms [Marszal et al., 2011].
A.6 Plus-energy buildings (PEB)

At present, plus energy buildings are garnering a great deal of interest as they provide
added value to the end user by generating more energy than required [Cole and
Fedoruk, 2015]. A PEB is defined as a building that “produces more energy than is
needed, and … [exports] it to other buildings or systems, i.e. energy storage management or
feeding the extra energy produced to the grid” [Kolokotsa et al., 2011; Cole and Fedoruk,
2015]. In these buildings, unlike NZEB, total energy consumption, comprising space
heating, DHW, appliances, lighting etc. is considered in energy balance calculations.
A.7 Zero-carbon buildings (ZCB)

The zero-carbon building is popular in the UK. In these buildings, the space heating
demand is minimal and is met by renewable energy sources [Mlecnik, 2012]. The
UK sustainable code for homes defines a zero-carbon building as a building that has
the following properties: “net CO2 emissions resulting from total energy used in the
dwelling are zero or better. This includes the energy consumed in the operation of the space
heating/ cooling and hot-water systems, ventilation, all internal lighting, cooking and all
electrical appliances” [CLG, 2010].
A.8 Green buildings

As per US Green Building Council [USGBC, 2017], a green building is defined as
“the planning, design, construction, and operations of buildings with several central,
foremost considerations: energy use, water use, indoor environmental quality, material
section and the building's effects on its site”.
227
Appendices
A.9 Summary
The space heating demand requirements of a few building classifications are
presented in Figure A.1. This figure presents comparison of heating demand based
on the standards and simulated heating demand of similar buildings in the
considered design space of the case studies in this research. This heating demand is
calculated based on dynamic simulations for an average scenario of occupant
behavior and for a reference climate. In addition, this figure presents a comparison
of predicted total electricity demand of these buildings that include electricity
required for space heating by ASHP (SPF=2.7), DHW usage, appliances, lighting,
ventilation etc. for the same scenario.
Space heating demand, kWhth/m2a Total predicted electricity demand, kWhe/m2a
Existing building Existing building
Low-energy buildings Low-energy buildings
Very low-energy buildings Very low-energy buildings

Simulated Standards
BENG BENG
Passive buildings Passive buildings
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Figure A.1 Space heating demand limits for different types of buildings based on [Mlecnik,
2012; Agentschap NL, 2013b] and space heating demand and total predicted electricity
demand of the corresponding buildings from the design space of the current case studies.
It can be noted from Figure A.1 that the space heating demand of these buildings
differed considerably, ranging from 12 kWhth/m2a to 137 kWh kWhth/m2a. In
contrast, the difference in electricity demand among all buildings is between 39.2-
88 kWhe/m2a and is quite low compared to that of the space heating demand. An
existing reference building is also presented to demonstrate how these building
compare with each other and with a typical existing building. It is worth noting that
for the existing building (RP0), the heating demand is calculated based on the total
gas consumption of a reference building [Agentschap NL, 2011]. The purpose of this
comparison is not to validate the building models of these buildings as the heating
demand calculation methods are completely different, but to illustrate how different
buildings perform within the design space of the case studies. In the selected designs
from the design space of the renovation house case study, the very low-energy
buildings have renovation package RP3, nZEB (BENG) has RP4 and Passive house
has RP7. The equivalent building envelope packages can also be found in the design
space of the new house case study.
228
Glossary: high-performance buildings
Table A.1 Space heating demand, primary energy use and renewable energy requirements
for different buildings.
Space heating Primary
Renewable
Building type demand energy use Reference
energy
(kWhth/m2a) (kWh/m2a)
Low-energy 60 120 NA [Mlecnik, 2012]
Very low-
30 † NA [Mlecnik, 2012]
energy
1 2
Passive 15 45 -120 NA [PEP, 2008]
BENG (nearly
25 251 50% 3 [RVO, 2015b]
zero-energy)
[Torcellini et al.,
Zero-energy † † 100% 2006]
[Cole and
Plus-energy † † >100% Fedoruk, 2015]
1. Primary energy use for space heating, DHW and ventilation except for lighting
and appliances
2. Includes all energy use (lighting and appliances as well)
3. The renewable energy share calculations considers primary energy [RVO,
2015a].
† No clear definition of requirements
In most of the NZEB concepts implemented so far, energy efficiency has been given
priority in to achieve maximum reductions in energy demand. These low energy
demands are met by small renewable energy generation systems. This is in line with
the requirements set by frameworks and voluntary standards such as Trias
Energetica [Lysen, 1996; VROM, 2010] and Passivhaus [PHI, 1990; PEP, 2008]. It
is also widely reported in literature that buildings designed based on these
frameworks and standards are prone to overheating [Sameni et al., 2015; Zero
Carbon Hub, 2015]. Therefore, it is important to investigate whether NZEB concepts
can also be achieved by balancing energy demand and energy generation, without
necessarily focusing on maximizing the reduction of energy demand.
NZEB has become a popular ‘catchphrase’ that is used to describe the interaction
between energy-efficient measures and renewable energy integration in the building
[Voss et al., 2011]. Accordingly, several private firms in the Dutch market have
adopted this concept as part of their marketing strategies such as nota-nul (net-zero)
by Homij and ‘Flat met Toekomst’ (flat with future) by Nieman consultancy based on
the nul-op-de-meter (Zero-on-the-meter) concept introduced by the Dutch
government [RVO, 2015c].
229
Appendices
In the nota-nul concept by Homij, the target is to have net-zero costs. These buildings
are all electric and the total electricity consumption of the building is considered in
the annual balance to calculate net costs. Incentives such as tax exemptions are also
considered in the annual cost balance [Eck, 2016]. Similarly, in the ‘Flat met
Toekomst’ concept by Nieman consultancy, the total electricity consumed by a
building is accounted for in the annual energy balance. However, limits are placed
on how much energy can be used for different purposes, such as heating and DHW.
Specifically, space heating demand and DHW demand should be less than or equal
to 30 kWhth/m2a and 15 kWhth/m2a, respectively. Similarly, the electricity consumed
by appliance and lighting should be less than or equal to 26 kWhe/m2a [Middendorp,
2018]. If tenants exceed these limits, they will face additional charges.
Figure A.2 A path towards nZEB, NZEB and plus energy buildings (top figure) and
different balancing periods for calculation (bottom figure) [Voss et al., 2012b].
230
Appendix B. Thermo-physical properties of building envelope
materials
In this appendix, thermo-physical properties of different insulation materials and

construction materials are listed.
Thermo-physical properties of different construction materials used in the

renovation and new house case studies are presented in Table B.1. In addition to
these materials, an air gap of 10 cm is used in the walls, and floors are covered with
carpet. The insulation thickness of floor/wall/roof is varied to produce different types
of renovation packages and building envelope packages for the renovation house and
new house case studies respectively, as shown in Table 5.2 and Table 5.3. Different
types of insulation materials are used for floors, walls and roofs. These insulations
are selected from [Kingspan insulation, 2016].
The position of insulation is varied, as shown in Figure B.1, to form heavy-weight

and light-weight constructions similar to the study by [Roberz et al., 2017]. The
notable difference among these constructions is thermal inertia, which is very high
in the case of heavy-weight constructions due to high thermal capacity of
construction materials (see Table B.1).
Table B.1 Thermo-physical properties of different construction materials used in the

renovation and the new house case studies.
Construction Conductivity Density Specific heat capacity
material (W/m K) (kg/m3) (J/kg K)
Brick 1.30 2000 900
Insulation 0.03 50 840
Brick 1.30 2000 900
Cast concrete 0.39 2100 840
Concrete 2.10 2400 800
Roof tiles 0.84 1900 1000
Bitumen roof 0.17 1200 1000
Plywood 0.15 800 1200
231
Appendices
Air gap Insulation

Brick Brick
Outdoor (Ta) Indoor (Ti) Outdoor (Ta) Indoor (Ti)
Heavy-weight Light-weight
Figure B.1 Graphical representation of heavy-weight and light-weight wall constructions.
232
Appendix C. Investment and replacement costs of design options
In this appendix, the costs of different insulation materials and windows are tabulated.
The total cost of each renovation and building envelope package is calculated and a
comparison is made of the required ICa of implementing these packages in a reference
building. Similarly, the investment cost and replacement cost of different HVAC and RES
systems are compared.
The cost of insulation for floor, wall and roof for different thermal resistance levels
is shown in Table C.1. The prices for different windows are tabulated in Table C.2.
Based on these costs, the total investment cost for different renovation packages and
corresponding additional investment cost (ICa) of these renovation packages are
compared to the reference house, as shown in Table C.3. Similarly, for the new house
case study, these costs are shown in Table C.4.
Table C.1 Cost of insulation for floor, wall and roof for different thermal resistance values.
[Kingspan insulation, 2016].
Cost of insulation (€/m2)
Rc value (m2K/W)
Floor Wall Roof
10 55.1 55.1 55.1
9 49.9 49.9 49.9
8 44.5 46.0 44.5
7 39.0 39.8 39.0
6 33.5 35.0 33.5
5 28.1 32.0 28.1
4 22.6 26.3 22.6
3.5 20.4 24.6 20.4
2.5 16.2 19.3 16.2
2 13.8 17.1 13.8
1.3 11.4 11.4 11.4
Table C.2 Cost of windows [dubbelglasweetjes, 2016].
Window type U-value (W/m2K) Price (€/m2)

Regular double glazing 2.7 65
Double glazing HR 1.7-2.0 70
Double glazing HR+ 1.3.-1.6 75
Double glazing HR++ 1.2 80
Triple glazing HR+++ 0.5-0.9 120
233
Appendices
Table C.3 Cost of different renovation packages and corresponding ICa for the renovation
house case study.
Costs, € RP0 RP1 RP2 RP3 RP4 RP5 RP6 RP7
Floor 703 823 1037 1037 1708 1430 1708 2810
Wall 3375 3811 4870 5773 7870 7870 9489 10910
Roof 946 1107 1542 2291 2664 3040 3769 3769
Windows 1371 1477 1530 1582 1688 1688 2532 2532
Total cost of each
6397 7220 8981 10685 13931 14029 17498 20020
package (RP)
ICa 0 822 2582 4287 7533 7631 11100 13622
Table C.4 Cost of different building envelope packages and corresponding IC a for the new
house case study.
Costs, € P0 P1 P2 P3 P4 P5
Floor 1066 1066 1755 1469 1755 2886
Wall 5658 6707 9142 9142 11023 12673
Roof 1066 1755 2040 2328 2886 2886
Windows 1682 1740 1856 1856 2784 2784
Total cost of each package (P) 9472 11267 14793 14796 18447 21229
ICa 0 1796 5322 5324 8980 11757
It can be inferred form Table C.3 and Table C.4 that the ICa of new building envelope
packages is much lower than the corresponding renovation packages for the existing
building, which is attributed to the higher initial investment cost of the reference
new buildings’ envelope package (P0). In addition, the costs of these buildings differ
due to different areas of floor, walls, roof and windows.
It is worth noting that for higher WWRs, surface area of windows has been deducted
from the cost of the area of walls that they replace in the calculation of the IC a of
these packages. However, the higher cost of glazing required for higher WWRs is
included in the calculation of the ICa. The cost of infiltration ranges from 0-2314€,
depending on the building airtightness [Saint-Gobian Isover, 2016], and this cost
includes the costs required for taping around windows, wall-wall connections and
wall-floor connections [Plas, 2017].
The initial investment cost (ICi) and corresponding replacement costs of HVAC, PV,
SDHW and battery systems are tabulated in Table C.5. These costs are collected from
234
Investment and replacement costs of design options
different sources, notably from the RVO publication on the investment costs of PV
systems [RVO, 2014]. Similarly, the costs for heating and SDHW systems are taken
from the assessment report on solar thermal and heating system technologies by
[Fleiter et al., 2016]. Battery costs are based on Tesla Powerwall for residential
applications [Tesla, 2017].
It is worth noting that the ICi of some of these components is also considered as
additional investment cost since the reference houses does not contain such
components. For instance, in the case of the renovation house and the new house,
there are no PV and battery systems. Therefore, initial investment cost of these
components essentially becomes ICa. GC in this research is calculated for a 30-year
period, and building components and energy systems that have a life-span less than
30 year are replaced and are therefore considered in replacement cost (RC)
calculations. For instance, building insulation has a life-span longer than 30 years
and is accordingly not considered. In contrast, some systems such as battery, inverter
and gas boiler should be replaced twice during this period.
Table C.5 Initial investment cost (ICi) and corresponding replacement costs of different
HVAC and RES systems.
Life-span
System Unit ICi (€) Reference fd,e RC (€)
(n, years)
PV m2 185*A+680 [RVO, 2014] 25 0.60 111.4
Battery 12 0.78 352.7
kWh 450 [Tesla, 2017]
Battery2 24 0.61 276.5
Solar
m2 773 25 0.60 465.4
DHW
ASHP kWthp 1130 [Fleiter et 20 0.67 753.0
GSHP kWthp 1675 al., 2016] 20 0.67 1116.2
Gas boiler 15 0.74 412.3
kWthp 559
Gasboiler2 30 0.54 304.1
Inverter kW per 15 0.74 28.6
m2 of 38.7 [RVO, 2014]
Inverter2 30 0.54 21.1
PV
2Second time replacement of a component in the GC calculation period of 30 years .
235
Appendix D. Demonstration of the CPRA approach using the
renovation house case study
In this appendix, some of the results of the CPRA demonstration using the renovation
case study are presented to avoid repetition and improve the readability of Chapter 6. The
presented results include demonstration for the homeowner and both decision makers
combined using the max-min method and the best-case and worst-case method. These
demonstration results are compared with those of the minimax regret method.
D.1 Demonstration of the CPRA approach for the homeowner
D.1.1 Trade-off solutions using Pareto front

The Pareto fronts for the homeowner calculated using the max-min method and the
best-case and worst-case method are shown in Figure D.1 and Figure D.4.
D.1.1.1 The max-min method

It can be observed from Figure D.1 that the existing building (green bubble) had less
overheating hours and also a small spread across the considered scenarios. This
implies that the existing building was robust to overheating. However, this building
incurred large global costs (GC) and also resulted in the largest spread of GC.
Therefore, the homeowner would prefer to find robust renovation measures to
reduce this GC and its corresponding spread. The designs with similar or less
overheating hours and corresponding spread can be found across the entire IC a
range, except beyond 40 k€. The designs beyond this ICa range had a large spread of
overheating hours. On the other hand, the designs with lower GC compared to the
reference building can be found in the ICa range of 0-35 k€, as the corresponding
spread of GC of these designs gradually decreased in line with an increase in ICa. In
these Pareto solutions, the designs with low global costs resulted in a large spread of
overheating, and the inverse is also true.
It is intriguing that GC spread gradually decreased in proportional to increases in ICa

and GC, as observed in Figure D.2. The designs with high ICa resulted in high GC,
as seen in Figure D.2, but in low spread of GC as the difference between the
maximum and minimum performance was reduced due to low operational costs. It
is noteworthy that for a particular design, variation in GC across the considered
scenarios was largely caused by operational costs. It can be argued that these designs
237
Appendices
with high ICa had less variations and were more robust than designs with low IC a,
but this robustness comes at the expense of very high ICa and GC. Furthermore, the
overheating risk associated with these designs is very high.
The homeowner could either prioritize overheating hours or global costs and trade-
off these preferences with robustness and required ICa. For instance, if the
homeowner had low tolerance towards overheating and preferred to bear more costs,
then the preferred robust designs that had the lowest spread of overheating are in
the global cost range of 40-60 k€ with the corresponding ICa ranging from 3.8-42
k€. Conversely, if the homeowner accepts certain overheating risks as a trade-off with
the GC and corresponding robustness, then the preferred robust designs are in the
ICa range of 20-40 k€, because beyond the ICa of 40 k€, the improvement in
robustness does not outweigh required ICa and corresponding GC.
To provide better insights into prioritizing the performance and robustness

indicators and their consequent impact on design decision making for the
homeowner, a few selected Pareto designs in different ICa ranges are compared; see
Figure D.3. The box plots represent predicted performance and the bar plots indicate
the corresponding spread of the overheating hours and GC. The details of these
selected Pareto designs are tabulated in Table D.1.
The notable differences among the designs in the ICa range of 0-15 k€ are insulation
levels, RES systems and heating systems (see Table D.1). It can be observed from
Figure D.3 that designs with an ICa of 0 and 5.6 k€ had the same spread of
overheating, because these designs differ only in energy systems (see Table D.1),
which had no influence on overheating. Despite their very low overheating hours and
corresponding spread, these designs might not be preferred by the homeowner,
because these designs resulted in very high global costs and corresponding spread.
The next two designs had better predicted performance and a low spread of GC, but
resulted in a large spread of overheating hours. This large spread of overheating was
caused by extreme scenarios (e.g. climate change); it can be noted from the box plot
of overheating hours that four designs had similar overheating hours for most of the
scenarios. If the homeowner accepts this risk of overheating for extreme scenarios,
then the design with an ICa of 15.4 k€ was the most preferred robust design. This
design has renovation package RP1 with a GSHP heating system and a large PV
system of 32m2.
238
Demonstration of the CPRA approach using the renovation house case study
Bubble size = Spread of overheating hours (110-822 h/a)

60

50
40
30
20
10
0
-5 5 15 25 35 45 55
Bubble size = Spread of global cost (22-49.7 k€)

100
80
60
40
20
0
-5 5 15 25 35 45 55
Figure D.1 Pareto front for the homeowner optimized using the max-min method. The top
figure represents overheating hours and the bottom figure represents global costs. In both
figures, the corresponding spread is included as bubble size. The green bubble represents the
100
80
60
40
20
0
-5 5 15 25 35 45 55
Predicted performance Spread
Figure D.2 Variation of predicted performance and corresponding spread of global cost. The
green bubble represents the reference building design.
239
Appendices
Similarly, by comparing the designs in the second group of different ICa ranges, it
can be inferred that the design with an ICa of 30.7 k€ resulted in the lowest spread of
GC, but the largest spread for overheating hours. This large spread of overheating
was attributed to very high insulation levels (RP6; Rc = 6/8.5/10 m2K/W; U =0.86
W/m2K) and to the airtight building envelope of this design. In contrast, the design
with an ICa of 25.1 k€ had the lowest spread of overheating hours, but the largest
spread of GC. The design with an ICa of 30.7 k€ was the most robust to GC among
the four selected designs and hence, the most preferred design for the homeowner
if GC was prioritized. Similarly, the design with an ICa of 25.1 k€ was the most robust
to overheating and the most preferred for the homeowner if overheating hours were
prioritized. To reach a compromise between these performance indicators and
corresponding robustness, the design with an ICa of 16.5 k€ was the most preferred
robust design as it had lower spread of global cost compared to the design with an
ICa of 25.1 k€ and a lower spread of overheating compared to the design with an IC a
of 30.7 k€. Furthermore, the ICa of this preferred design is much lower than those of
the most robust designs for these two performance indicators. In addition, this
design had better predicted performance of GC.
Compared to designs in the previous ranges, it can be observed that all designs had
higher GC and overheating hours in the last range of ICa. This high GC was due to
the very high ICa of these designs, which was attributed to improved insulation levels
(RP3-RP7) and inclusion of large energy systems (see Table D.1). These upgrades
resulted in reduced operational costs, but not substantial compared to required
investment costs. Comparing the designs’ robustness to GC, it can be observed that
the designs with an ICa of 40 k€ and 44.5 k€ had the same spread of GC, but the
design with an ICa of 40 k€ had better predicted performance of GC and was hence
the most preferred robust design. Furthermore, this design had better predicted
performance and robustness to overheating hours compared to the design with an
ICa of 44.5 k€. Similar observations can be made with respect to GC for designs with
an ICa of 35.2 k€ and 51 k€. Among these two designs, the design with an ICa of 35.2
k€ had better predicted performance of GC and also lower additional investment
costs and was accordingly the most preferred design.
Overall, the design with an ICa of 35.2 k€, which has the RP3 renovation package,
GSHP heating system and large RES systems (PV= 32 m2, SDHW = 10 m2 and
battery =20 kWh) was the most preferred robust design for the homeowner because
it resulted in similar global costs as that of the design with an ICa of 51 k€, but at a
240
much lower ICa. It is worth noting that this design had a higher spread of about 50
h/a of overheating hours compared to the design that was the most robust (IC a =51
k€) to overheating in this ICa range, but the homeowner can trade off these
overheating hours to save 15.8 k€ of ICa.
Among all selected designs, it can be noted that the design with an ICa of 16.5 k€ was
the most preferred robust design due to its low ICa, GC and its less overheating
hours. Furthermore, the spread of GC and overheating hours was closer to the
designs that were the most robust to GC and overheating hours, respectively.
Using this method, it can be noted that the designs with very high insulation levels
(RP3-RP7) and large RES systems (PV = 32 m2, SDHW = 10 m2, battery =20 kWh)
with GSHP resulted in the lowest spread of GC because these design options reduced
operational costs by a significant amount across extreme scenarios. In contrast, the
designs with low insulation levels and high infiltration rates resulted in the least
spread of overheating due to improved ventilation and reduced heat trap of internal
heat gains and solar gains in the building.
Table D.1 Details of selected Pareto designs, for further analysis, in different additional
investment cost ranges from the Pareto front of the homeowner calculated using the max-
min method.
ICa, k€ 0 5.6 10.2 15.4 16.5 19.9 25.1 30.7 35.2 40 44.5 51
Renovation 0 0 1 1 1 1 1 6 3 5 7 7
package (RP)
Infiltration, 1 1 1 1 0.1 0.5 0.1 0.1 0.1 0.1 0.5 0.625
dm3/ds2
WWR, % 40 40 20 20 20 20 80 20 20 20 20 80
HR HR
HVAC system HR107 GSHP
107
GSHP
107
GSHP
PV system, m2 0 28.8 12.8 32 32 32 32 32 32 32 32 32
SDHW system, 0 0 0 0 5 5 5 0 10 10 10 10
m2
Orientation, ° S E E S E S S
Tilt angle, ° 0 45 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 0 0 0 0 0 16 16 20 20 20
241
Appendices
900
800
700
Spread of
600
500
400
300
200
100
0
0.0 5.6 10.2 15.4 16.5 19.9 25.1 30.7 35.2 40.0 44.5 51.0
100
Spread of global cost (GC),
80
k€/30 years
60
40
20
0
0.0 5.6 10.2 15.4 16.5 19.9 25.1 30.7 35.2 40.0 44.5 51.0
Figure D.3 Variation of overheating hours, global costs (box plots) and corresponding
spread (bar plots) across the considered scenarios of selected Pareto designs for further
analysis.
242
D.1.1.2 The best-case and worst-case method

It can be observed from Figure D.4 that there are no Pareto solutions beyond 20 k€,
indicating that designs with high investment costs are not cost-optimal solutions for
the homeowner using this method. Therefore, a different x-axis scale is used to
enhance visualization. The existing building had lower deviation of overheating and
higher deviation of GC compared to other Pareto solutions. The designs with
improved predicted performance of GC and lower deviation of GC compared to the
reference building can be found in all ranges of ICa. The deviation of GC gradually
decreased in proportion to increases in ICa.
In contrast to the max-min method, here, the deviation of GC is proportional to the

predicted performance of GC. The designs with similar or even less overheating and
the same deviation of overheating hours as those of the reference building can also
be found in the entire ICa range. However, in these Pareto solutions the designs with
low GC resulted in higher deviation of overheating hours, and the inverse is also
true. Therefore, to illustrate the impact of design decision making based on the
preferences of the homeowner, a few selected designs across different ICa ranges are
compared in Figure D.5. The details of these selected designs are tabulated in Table
D.2.
Firstly, by comparing the designs in the ICa range of 0-10 k€, it can be inferred from
Figure D.5 that the reference building and design with an ICa of 3.1 k€ had similar
overheating hours and corresponding deviation since these designs differ only in
RES systems. These designs resulted in high GC and corresponding deviation, and
hence might not be preferred by the homeowner. On the other hand, the designs
with an ICa of 8.7 k€ and 10.5 k€ that have improved insulation and airtightness
resulted in large deviations of overheating hours, but at low GC and corresponding
deviation. Among the four designs, it can be observed that the design with an ICa of
8.7 k€ had better predicted performance of GC and also the least deviation of GC.
This design resulted in a higher deviation of overheating hours of about 44 h/a
compared to the reference building, due to improved insulation and airtightness.
However, this improvement in insulation and airtightness had significant benefits
in the reduction of GC of about 10.2 k€. If the homeowner accepts this overheating
risk as a trade-off with the corresponding GC savings, then this design was the most
preferred robust design.
243
Appendices
Bubble size = Deviation of overheating hours

(112-343 h/a)
Overheating hours, h/a 60
50
40
30
20
10
0
-5 0 5 10 15 20
Bubble size = Deviation of global cost

(39.5-66.5 k€)
100
80
60
40
20
0
-5 0 5 10 15 20
Figure D.4 Pareto front for the homeowner optimized using the best-case and worst-case
method. The top figure represents overheating hours and the bottom figure represents global
costs. In both figures, the corresponding deviation is included as bubble size. The green
Similarly, by comparing the designs in the ICa range of 10-15 k€, it can be noted that
the predicted performance of overheating hours improved in line with an increase in
ICa, and the corresponding deviation gradually decreased in proportion to increases
in ICa. In contrast, the GC and corresponding deviation gradually increased in line
with an increase in ICa. In this ICa range, the design with an ICa of 13.9 k€ was the
preferred robust design if the homeowner prioritizes overheating at the expense of
ICa and extra GC. Similarly, if the homeowner prioritizes GC, then the preferred
design would be the design with an ICa of 11.2 k€, because it had better predicted
performance of GC and the lowest deviation of GC. To reach a compromise between
244
these performance indicators and corresponding robustness, the design with an ICa
of 13.2 k€ was more preferred as it had lower deviation of GC compared to the design
with an ICa of 13.9 k€ and lower deviation of overheating compared to the design
with an ICa of 11.2 k€.
In the last range of ICa, it can be observed that the design with an ICa of 17.1 k€ had
the lowest deviation of GC, but also had the highest deviation of overheating hours.
This large deviation of overheating hours was due to improved insulation levels, RP3
(Rc = 3.5/4.5/6 m2K/W; U =1.43 W/m2K), and airtightness. The same improvement
in combination with a PV system of 19.2 m2 resulted in significant benefits in
robustness to GC. Therefore, this design was the preferred robust design if the
homeowner prioritizes GC. Conversely, if the homeowner had low tolerance towards
overheating, then the design with an ICa of 16.2 k€ was the most preferred robust
design, because it had less overheating hours and corresponding deviation in
addition to relatively low GC and corresponding deviation. It is worth noting that the
design with an ICa of 17.2 k€ had higher deviation of GC compared to the reference
building, indicating that the designs beyond this ICa did not reduce GC (operational
cost) significantly compared to the required ICa and hence, there were no Pareto
solutions beyond this ICa with this method. It is also intriguing to observe that this
design had lower overheating hours compared to the reference building, which was
due to its large WWR, which resulted in a larger opening area of windows providing
more natural ventilation in summer.
Table D.2 Details of selected Pareto designs, for further analysis, in different additional
investment cost ranges from the Pareto front of the homeowner calculated using the best-
case and worst-case method.
ICa, K€ 0 3.1 8.7 10.5 11.2 12.3 13.2 13.9 16.2 16.8 17.1 17.2
Renovation
0 0 1 1 1 1 1 1 1 1 3 1
package (RP)
Infiltration, 1 1 0.625 0.5 0.1 1 0.4 1 0.1 0.625 0.1 1
dm3/ds2
WWR, % 40 40 20 20 20 20 40 40 80 80 20 80
HVAC system HR107
PV system, m2 0 12.8 0 0 12.8 12.8 12.8 12.8 12.8 12.8 19.2 12.8
SDHW system,
0 0 0 2.5 0 0 0 0 0 0 0 0
m2
Orientation, ° S E S
Tilt angle, ° 0 0 0 0 0 0 0 0 0 0 45 0
Battery, kWh 0 0 0 0 0 0 0 0 0 0 0 0
245
Appendices
900
800
700
Deviation of
600
500
400
300
200
100
0
0.0 3.1 8.7 10.5 11.2 12.3 13.2 13.9 16.2 16.8 17.1 17.2
100
Deviation of global cost (GC),
80
k€/30 years
60
40
20
0
0.0 3.1 8.7 10.5 11.2 12.3 13.2 13.9 16.2 16.8 17.1 17.2
Figure D.5 Variation of overheating hours, global costs (box plots) and corresponding
deviation (bar plots) across considered scenarios of selected Pareto designs for further
analysis.
246
robust design as it had relatively low GC, overheating hours and corresponding
deviations. In contrast to observations made for robust designs using the max-min
method, it was found that designs with small RES systems were more robust with
this method. Furthermore, designs with batteries for energy storage do not feature
in the Pareto front, which was probably due to the high ICa of batteries that might
not yield significant benefits in lowering operational costs in the case of Pareto
solution with small RES systems.
D.1.2 Robust design options

The influence of design options on Pareto solutions for the homeowner was
analyzed, as presented in Figure D.6 and Figure D.7. All design options are
compared for the max-min method and the best-case and worst-case method and the
results are presented in comparison with robust design options using the minimax
regret method.
The max-min method: As expected, designs with very high insulation levels (RP7; Rc
= 10/10/10 m2K/W) in combination with airtight building envelopes (0.1 dm3/ds2)
were the least robust to overheating. It can be concluded from Figure D.7 that the
reference building was the most robust to overheating. Similarly, as observed with
the robust design options for the minimax regret method, designs with low
insulation levels were less prone to overheating and the most robust to overheating.
For instance, designs with RP1 and RP3 had the least spread of overheating. In
contrast, designs with renovation packages of very high insulation levels had the least
spread of GC. However, the reduction in spread was not substantial with insulation
levels beyond RP3 as it can be noted that all these renovation packages (RP3-Rp7)
resulted in a spread of GC close to 20 k€. This indicates that improving insulation
levels beyond RP3 does not yield significant benefits in reduction of GC spread.
Therefore, designs with RP3 were the most preferred robust designs using the max-
min method.
Designs with large WWRs were the most robust to overheating, because they provide
enhanced natural ventilation due to the larger opening area of windows. In contrast,
these designs were less robust to GC due to their high IC a, despite the reduction in
operational costs arising from low heating demand.
247
Appendices
900 6000 900 900

800 CO2 emissions, kgCO2/a 800 800

700
5000 700 700
600 600 600
500
4000 500 500
400 400 400
3000
300 300 300
200 200 200
2000
100 100 100
0 0 0
0
10002
1 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
0
20 40 60 80
Figure D.6 Variation of robustness of overheating hours for different design options of all
Pareto solutions for the homeowner calculated using three robustness assessment methods
regret). Only design options that influence overheating are shown here.
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

0 = N-S facing S
1 = HR107 boiler
80 2 = ASHP 80 90 = E-W facing E 80

3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 0 45 90
6000
100 100 100

5000
80 80 80
4000 60
60 60
40
3000 40 40
20 2000 20 20
0 1000 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0PV system size, m2 Solar DHW system size, m2 Battery capacity, kWh
20 40 60 80

Figure D.7 Variation of robustness of global cost for different design options of all Pareto
solutions for the homeowner calculated using three robustness assessment methods
regret).
248
In contrast to robust design options using the minimax regret method, the GSHP
was the most robust heating option. Here, the reason that the GSHP was found to
be the most robust option was largely due to very low heating demands in highly
insulated buildings and because an all-electric option provided by the GSHP in
combination with the use of large PV systems increases self-consumption of
electricity. Similarly, designs with large RES systems, facing south at a tilt angle of
45°, were the most robust, because these systems reduced the GC spread drastically,
as demonstrated in Figure D.2, despite their high ICa and GC. These robust design
options again indicate the conservative approach of the max-min method.
The best-case and worst-case method: Designs with RP1 had the lowest maximum
deviation of GC and were hence the most preferred, which was same as the robust
design option using the minimax regret method. Likewise, the HR107 gas boiler was
the most robust heating option. In addition, designs with small RES systems were
the most robust design options for the homeowner. This finding is attributed to
operational costs and ICa, among other factors. Operational costs are less dependent
on the size of PV system as the excess energy exported to the grid does not lower
operational costs in the case of net-metering termination. South facing PV system at
a tilt angle of 0° were the most robust. This observation was in contrast to the robust
design options using the max-min method and the minimax regret method.
D.1.3 The most robust design using the MCDM approach

considering GC, overheating hours and corresponding robustness were compared
for the max-min method and the best-case and worst-case method; see Figure D.8.
As observed with design scores for the minimax regret method (see Figure 6.18), the
design scores using the other two methods gradually decreased for designs beyond a
certain ICa, indicating that these designs were not cost-optimal robust solutions. For
instance, in the case of the max-min method, the design scores beyond an ICa of 22.5
k€ gradually decreased with an increase in ICa. This observation stands in contrast
to the design scores for the policymaker. It can be inferred from Figure D.8 that the
designs with an ICa of 22.5 k€ and 11.2 k€ had the highest design scores using the
max-min method and the best-case and worst-case method, respectively. These
designs were the most robust for the respective methods. However, there are several
designs with similar scores. Therefore, the design scores of different design options
were compared to find the most preferred robust design options; see Figure D.9.
249
Appendices
Spread Deviation
1 1
0.9 0.9
0.8 0.8
0.7 0.7
Design score
Design score
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
-5 5 15 25 35 45 55 -5 5 15 25 35 45 55
Additional investment cost (IC a), k€ Additional investment cost (ICa), k€
Figure D.8 The design scores of Pareto solutions for the homeowner calculated using the
Hurwicz criterion for the max-min method and the best-case and worst-case methods
considering predicted performance and corresponding robustness. The reference building
design is indicated in green (ICa = 0 k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9 1 = HR107 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 0 45 90
1 6000 1 1
0.9 0.9 0.9

0.8 5000 0.8 0.8
0.7 0.7 0.7

0.6 4000 0.6 0.6
0.5 0.5 0.5
0.4 3000 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
2000
0.1 0.1 0.1
0 0 0
0 1000
6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0
20 40 60 80
Figure D.9 The design scores of different design options of Pareto solutions for the
homeowner calculated using the Hurwicz criterion for three robustness assessment methods
regret).
250
Designs with the renovation package RP1 were the most robust design options for
the homeowner with all three robustness assessment methods. For the max-min
method, RP3 was also found to be a robust renovation package. As observed earlier
using the trade-off approach, low infiltration rates and low WWRs were robust design
options as they reduce GC and improve corresponding robustness. Similarly, the
HR107 gas boiler was the most robust design option for the best-case and worst-case
method, and the GSHP was the most robust heating option for the max-min method.
In addition, designs with small RES systems were the most robust with the best-case
and worst-case method, whereas designs with large RES systems were the most
robust with the max-min method. It can be noted that the robust design options
based on trade-off (see Figure D.4, Figure D.6 and Figure D.7) and the MCDM (see
Figure D.8 and Figure D.9) approaches were found to be similar for most of the
design options. These design options are summarized and presented in Table 6.6 of
Chapter 6.
D.1.4 Scenario analysis
The sensitivity index (p) for different low-high scenarios, which quantifies the
influence of different scenarios on all preferred performance and robustness
indicators, was compared and is presented in Figure D.10. The scenarios that were
not sensitive (p>0.05) to either overheating or robustness of overheating, such as
heating setpoint, DHW use and net-metering, are not shown in overheating graphs.
As noted earlier with minimax regret method, overheating increases with more
occupants and their usage especially internal heat gains due to lighting and appliance
use. Similarly, ventilation rates by means of mechanical ventilation (Vent) and
window opening, and better shading control reduced overheating hours
significantly, thus improving a design’s robustness to overheating. Climate scenarios
are the most influencing scenarios on overheating and corresponding robustness.
The occupant behaviour related scenarios such as heating setpoints (Thsp), appliance
use (Ause), lighting use (Luse), IHG and DHW use were the most influential on GC.
Similar to observations made for CO2 emissions (in the case of policymaker),
variations in GC were high with high occupancy levels, which was due to increased
electricity consumption. In contrast, occupancy profile influenced only GC, but not
the robustness of GC. This was attributed to the same aggregated energy usage for a
particular usage scenario, despite different magnitudes of usage at different times
for all-day and evening occupancy profiles. Similarly, only a few low-high scenarios
251
Appendices
influenced the GC robustness; mostly the number of occupants and their

corresponding usage of lighting and appliances.
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Overheating hours Spread of overheating hours
Deviation of overheating hours Maximum regret of overheating hours
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Global cost Spread of global cost
Deviation of global cost Maximum regret of global cost
Figure D.10 Sensitivity of various scenarios to the predicted performance and corresponding
robustness of preferred performance indicators of the homeowner calculated using the
Mann-Whitney test. Scenarios where p<0.05 (dotted line) are assumed to be sensitive.
These values are shown for three robustness assessment methods (Stars represent predicted
performance, triangles represent spread, rectangles represent deviation and circles represent
maximum regret).
D.2 Demonstration of the CPRA approach for both decision makers
D.2.1 Trade-off solutions using Pareto front

Figure D.11 shows the Pareto front for both decision makers, calculated using the
max-min method and the best-case and worst-case method. In this figure, Pareto
fronts for these robustness assessment methods are shown in columns. The figures
row wise from top to bottom represent CO2 emissions, overheating hours and GC,
252
respectively. It can be noted that the Pareto solutions can be found in the entire IC a
range of 0-50 k€ for all three methods. This result is in contrast with the Pareto front
for the case of the homeowner calculated using the best-case and worst-case method,
because there were no Pareto solutions beyond a certain level of ICa. As noted earlier
using the minimax regret method (Chapter 6), there are different layers of Pareto
fronts calculated using these two methods (Figure D.11). Similar to the Pareto
solutions for the policymaker, it can be observed that the predicted performance and
corresponding robustness of CO2 emissions improved gradually in line with an
increase in ICa. For instance, in the ICa range of 0-15 k€, the predicted performance
and robustness of CO2 emissions significantly improved in proportion to increases
in ICa. It is worth noting that the designs with similar overheating hours and
corresponding robustness can be found across this entire ICa range. Therefore, in
this ICa range, both decision makers should be able to find a trade-off among CO2
emissions, GC and corresponding robustness.
In the ICa range of 15-30 k€, CO2 emissions further reduced and corresponding
robustness further improved in line with an increase in ICa. In contrast, GC gradually
increased and the corresponding robustness gradually decreased in proportion to
increases in ICa. As noted in the first ICa range, the Pareto solutions with similar
overheating hours and corresponding robustness can also be found across this entire
range of ICa. Similarly, in this ICa range, the decision makers can choose between
robust designs by making a trade-off among CO2 emissions, GC and corresponding
robustness. In the ICa range beyond 30 k€, the reduction in CO2 emissions did not
outweigh required ICa. Furthermore, robustness did not improve greatly in line with
an increase in ICa. Conversely, for the max-min method, the spread of GC gradually
reduced (smaller bubble size) in line with an increase in ICa as observed in the case
of Pareto solutions for the homeowner in this ICa range (see Figure D.1).
However, the overheating risk of these Pareto solutions is unacceptably high. For
instance, the spread of overheating of a few designs in this ICa range reached as high
as 822 h/a. This high overheating risk was largely due to the very high insulation
levels (RP4-RP7) and airtightness (0.1 dm3/ds2) of these designs. The notable
difference among these robustness assessment methods was that improvement in
robustness of CO2 emissions is not significant beyond 30 k€ of ICa using the max-
min and the best-case and worst-case method compared to the robustness calculated
using the minimax regret method (see Figure 6.20). This lack of improvement in
robustness for these Pareto solutions was due to very high worst-case performance.
253
Appendices
a) Spread b) Deviation
Bubble size = Spread of CO2 emissions Bubble size = Deviation of CO2 emissions
(2240-5337 kgCO2/a) (2252-5674 kgCO2/a)
6000 6000

5000 5000
4000 4000
3000 3000
2000 2000
1000 1000
0 0
-5 5 15 25 35 45 55 -5 5 15 25 35 45 55
Additional investment cost (ICa), k€ Additional investment cost (ICa), k€
Bubble size = Spread of overheating hours Bubble size = Deviation of overheating hours
(110-822 h/a) (112-822 h/a)
60 60

50 50
40 40
30 30
20 20
10 10
0 0
-5 5 15 25 35 45 55 -5 5 15 25 35 45 55
Bubble size = Spread of global cost Bubble size = Deviation of global cost
(22.5-50.5 k€) (38-80.34 k€)
100 100
80 80
60 60
40 40
20 20
0 0
-5 5 15 25 35 45 55 -5 5 15 25 35 45 55
Figure D.11 Pareto fronts for the policymaker and the homeowner calculated using the max-
min method and the best-case and worst-case methods. The figures row wise from top to
bottom represent CO2 emissions, overheating hours and global costs, respectively. The
figures column wise from left to right represent the max-min method and the best-case and
worst-case method. In all figures, the robustness is included as bubble size. The green bubble
represents the reference building design.
A few designs selected in different ranges of ICa from the Pareto front (see Figure
D.11) were compared and are presented here to enhance design decision-making
process (see Figure D.12 and Figure D.13). Details of these selected designs are
tabulated in Table D.3 and Table D.4.
254
The max-min method: In the first range of ICa, the existing building had less
overheating hours and the lowest spread of overheating, but resulted in the highest
spread of GC and CO2 emissions (see Figure D.11). In contrast, the design with an
ICa of 15.8 k€ had the lowest spread of GC and CO2 emissions, but resulted in the
highest spread of overheating hours. In such cases, both decision makers could
prioritize at least one of their preferred performance indicators and trade this off with
other performance indicators to find their preferred robust design. For instance, the
design with an ICa of 5.6 k€ was the most preferred robust design if overheating and
CO2 emissions were prioritized, because this design has the same spread of CO 2
emissions as that of the design with an ICa of 15.8 k€ and the same spread of
overheating hours as that of the design with an ICa of 0 k€. Conversely, if GC and
CO2 emissions were prioritized, then the design with an ICa of 15.8 k€ was the most
preferred robust design for both decision makers.
In the second range of ICa, the design with an ICa of 16.3 k€ was the most robust to
overheating, however, this design resulted in the highest spread for GC and CO 2
emissions. In contrast, the design with an ICa of 30.4 k€ had the lowest spread for
GC and CO2 emissions, but resulted in a higher spread of overheating compared to
the overheating spread for the design with an ICa of 16.3 k€. If both decision makers
accepted this overheating spread as a trade-off with improved robustness of GC and
CO2 emissions, then the design with an ICa of 30.4 k€ would be the most preferred
robust design. However, this design incurred very high ICa. As a compromise among
all these performance indicators and their corresponding robustness, the design with
an ICa of 20.6 k€ was the most preferred robust design, because it had lower ICa,
and smaller spreads of GC and CO2 emissions compared to the respective robust
designs for each of these performance indicators. Furthermore, this preferred robust
design had better predicted performance.
Similar to observations made in the first two ICa ranges, the design with an ICa of 40
k€ had the least spread of overheating, but had the highest spread for GC and CO 2
emissions in the last range of ICa. This lowest spread of overheating was attributed
to low insulation levels in combination with high infiltration rates. In contrast, the
design with an ICa of 46.7 k€ had the lowest spread of GC and CO2 emissions, but
the highest spread of overheating hours. This lowest spread of GC and CO2
emissions was attributed to the inclusion of large RES systems in combination with
very high insulation levels. However, these designs with very high insulation levels
resulted in the highest spread of overheating hours. This spread of overheating hours
255
Appendices
could be reduced to a large extent by including higher WWRs, as observed in the

design with an ICa of 50.8 k€. This design was the most preferred robust design as
it had a smaller spread of GC, CO2 emissions and overheating hours compared to
the respective robust designs, but incurred higher GC and ICa.

6000
5000
Spread of
4000
3000
2000
1000
0
0.0 5.6 10.1 15.8 16.3 20.6 25.3 30.4 35.8 40.0 46.7 50.8

Spread of 800
600
400
200
0
0.0 5.6 10.1 15.8 16.3 20.6 25.3 30.4 35.8 40.0 46.7 50.8
120
Spread of global cost (GC),
100
k€/30 years
80
60
40
20
0
0.0 5.6 10.1 15.8 16.3 20.6 25.3 30.4 35.8 40.0 46.7 50.8
Figure D.12 Variation of CO2 emissions, overheating hours, global cost (box plots-left
figures) and their corresponding spreads (bar plots-right figures) of the selected Pareto
solutions (see Table D.3).
256
Table D.3 Details of selected Pareto designs, for further analysis, in different additional investment cost ranges from the Pareto front (see
Figure D.11) of both decision makers calculated using the max-min method.
ICa, k€ 0.0 5.6 10.1 15.8 16.3 20.6 25.3 30.4 35.8 40.0 46.7 50.8
Renovation package (RP) 0 0 1 3 3 4 7 5 7 1 7 7
Infiltration, dm3/ds2 1 1 0.625 0.1 0.1 0.1 0.4 0.1 0.1 0.625 0.5 0.5
WWR, % 40 40 20 20 20 20 20 20 20 80 40 80
HVAC system HR107 HR107 HR107 GSHP HR107 GSHP GSHP HR107 GSHP GSHP GSHP GSHP
PV system, m2 0 28.8 0 19.2 22.4 28.8 32 28.8 32 32 32 32
SDHW system, m2 0 0 0 0 0 0 0 5 5 10 10 10
Orientation S S E S S S
Tilt angle, ° 90 0 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 0 0 0 0 0 12 12 20 20 20
Table D.4 Details of selected Pareto designs, for further analysis, in different additional investment cost ranges from the Pareto front (see
Figure D.11) of both decision makers calculated using the best-case and worst-case method.
ICa, k€ 0.0 5.2 10.6 14.8 15.9 21.0 25.9 30.2 35.8 40.6 44.4 49.9
Renovation package (RP) 0 0 1 1 0 3 5 7 1 4 7 7
0.62
Infiltration, dm3/ds2 1 1 0.1 0.625 1 0.1 0.1 0.1 1 0.625 0.625
5
WWR, % 40 40 20 20 40 20 20 20 80 20 20 80
HVAC system HR107
PV system, m2 0 25.6 9.6 32 32 32 28.8 32 32 32 32 32
SDHW system, m2 0 0 0 0 0 0 2.5 0 10 10 10 10
Orientation E S S E S S
Tilt angle, ° 45 45 45 45 45 45 45 45 45 45 45 45
Battery, kWh 0 0 0 0 20 4 4 8 16 20 20 20
257
Appendices
The best-case and worst-case method: In the first range of ICa, the design with an ICa of
5.2 k€ had the lowest deviation of CO2 emissions and overheating hours and was
accordingly the most robust to CO2 emissions and overheating hours. However, this
design had the highest deviation of GC (see Figure D.13). In contrast, the design with
an ICa of 10.6 k€ was the most robust to GC, but the least robust to overheating hours
and CO2 emissions. Among the four selected designs in this ICa range, the design
with an ICa of 14.8 k€ had better predicted performance and robustness of CO2
emissions, overheating and GC; hence, it was the most preferred design.
6000

5000
Deviation of 4000
3000
2000
1000
0
0.0 5.2 10.6 14.8 15.9 21.0 25.9 30.2 35.8 40.6 44.4 49.9
800
Deviation of
600
400
200
0
0.0 5.2 10.6 14.8 15.9 21.0 25.9 30.2 35.8 40.6 44.4 49.9
120
Deviation of global cost (GC),
100
80
k€/30 years
60
40
20
0
0.0 5.2 10.6 14.8 15.9 21.0 25.9 30.2 35.8 40.6 44.4 49.9
Figure D.13 Variation of CO2 emissions, overheating hours, global cost (box plots – left
figures) and their corresponding deviations (bar plots – right figures) of the selected Pareto
solutions (see Table D.4).
Similarly, comparing the four selected designs in the second ICa range, the design
with an ICa of 21 k€ had better predicted performance of GC and the lowest deviation
of GC, but this design resulted in the largest deviation for overheating and CO 2
emissions. In contrast, the design with an ICa of 15.9 k€ had better predicted
performance of CO2 emissions and overheating hours, and correspondingly lower
deviation. This lowest deviation in overheating was due to low insulation levels (see
Table D.4) and high infiltration rates, but resulted in very high operating costs and
consequently the largest deviation of GC.
258
In the last ICa range, the last three designs resulted in similar predicted performance
and corresponding deviation for CO2 emissions. The first design (ICa of 35.8 k€) had
the lowest deviation of overheating hours, but had the highest deviation of CO 2
emissions and GC, and hence may not be the preferred robust design. Therefore, the
selection of robust designs in this ICa range depends on overheating hours, GC and
their corresponding deviation. It can be inferred from Figure D.13 that the design
with an ICa of 49.9 k€ had lower deviation of overheating hours, but had higher
deviation of GC compared to other designs in this ICa range. Conversely, the designs
with an ICa of 40.6 and 44.4 k€ had similar deviations of GC, but the design with an
ICa of 40.6 k€ had the lowest deviation of overheating. Furthermore, it had better
predicted performance of overheating and GC and incurred a lower ICa among four
designs. Accordingly, it was the most preferred robust design.
D.2.2 Robust design options

The design options of the Pareto solutions calculated using the max-min method and
the best-case and worst-case method were compared with that of the minimax regret
method. This comparison is presented for CO2 emissions, overheating hours and
global costs in Figure D.14, Figure D.15 and Figure D.16, respectively.
The max-min method: Designs with renovation packages RP1 and RP3 resulted in the
smallest spread of CO2 emissions (see Figure D.14), whereas designs with RP1 had
the smallest spread of overheating hours (see Figure D.15), and designs with RP3-
RP7 had the smallest spread of global cost (see Figure D.16). However, the spread of
overheating hours for designs with RP3 is close to that of designs with RP1.
Therefore, RP3 would be the most preferred renovation package for both decision
makers if they chose a conservative approach in the decision-making process. It is
worth noting that renovation packages RP3-RP7 resulted in similar spreads of global
cost and CO2 emissions. Therefore, improving the insulation levels of existing
building beyond RP3 does not yield significant benefits compared to the required
ICa. Furthermore, the renovation packages with very high insulation levels (RP4-
RP7), especially deep renovation package (Rp7; Rc = 10/10/10 m2K/W; U =0.52
W/m2K), resulted in the largest spread of overheating hours and were therefore least
robust to overheating.
Similarly, designs with low infiltration rates resulted in the least spread of GC and
CO2 emissions. Contrariwise, designs with low infiltration resulted in the largest
spread of overheating and this spread of overheating reduced gradually with high
259
Appendices
infiltration rates. An infiltration rate of 0.625 dm3/ds2 represents a fair trade-off

among these performance indicators and was therefore the most preferred robust
design option. However, if designs had low insulation levels, such as RP1, designs
with an infiltration rate of 0.1 dm3/ds2 also resulted in a similar spread of overheating
hours. Designs with low WWRs resulted in the smallest spread of GC due to their
low ICa. Additionally, designs with high WWRs resulted in the smallest spread of
overheating due to the increased area of window opening allowing more ventilation
for natural cooling. CO2 emissions spread was least effected by the WWR. A WWR
of 40% represents a fair trade-off among all these preferred performance indicators
for both decision makers.
As noted earlier, the heating and RES systems had no influence on overheating
hours. As such, to find the most robust heating and RES systems, only GC and CO2
emissions were compared. Designs with the HR107 gas boiler resulted in the
smallest spread of CO2 emissions and designs with the GSHP resulted in the
smallest spread of GC, which was also observed in the robust design options for the
policymaker and homeowner when calculated separately. It can be noted that the
spreads of CO2 emissions for the designs with GSHP were also close to those of
designs with HR107 gas boiler (see Figure 6.22). Therefore, GSHP was the most
preferred robust heating system for both decision makers in the case of the max-min
method. As noted earlier for the policymaker and the homeowner separately, designs
with large RES systems were the most robust using the max-min method. It can be
inferred that when using the max-min method, the most robust design options for
both decision makers combined are renovation package RP3, an infiltration rate of
0.625 dm3/ds2, a WWR of 40%, a GSHP heating system and large RES system.
The best-case and worst-case method: Designs with RP6 had the lowest deviation of CO2
emissions (see Figure D.14), but resulted in high deviation of overheating (see Figure
D.15). Designs with renovation packages, RP1, and RP3-RP7 in combination with
different RES systems resulted in similar deviation of CO2 emissions. The same
observations can be made for GC (see Figure D.16). Among these renovation
packages, RP1 had the lowest deviation of overheating hours and was accordingly the
most preferred robust design. Furthermore, RP1 had the lowest ICa among these
renovation packages. Designs with high infiltration rates resulted in the highest
deviation of CO2 emissions and GC, but in the lowest deviation of overheating hours,
and the inverse is also true. Therefore, infiltration rates of 0.4-0.625 dm3/ds2 is a
reasonable trade-off among these performance indicators.
260
6000 6000 6000

5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
6000 6000 6000

1 = HR107

0 = N-S facing S

5000 2 = ASHP 5000 5000

90 = E-W facing E
4000 3 = GSHP 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
1 2 3 0 90 180 0 45 90
6000
6000 6000 6000


5000
5000 5000 5000
4000 4000 4000 4000
3000 3000 3000

3000
2000 2000 2000
2000 1000
1000 1000
0 1000 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0PV system size, m2 Solar DHW system size, m2 Battery capacity, kWh
20 40 60 80
Figure D.14 Variation of robustness of CO2 emissions for different design options of all
Pareto solutions for the policymaker and the homeowner calculated using three robustness
assessment methods (Triangles represent spread, rectangles represent deviation and circles
represent maximum regret).
6000
900 900 900

800 800 800
5000
700 700 700

600 4000 600 600
500 500 500
400 3000 400 400
300 300 300
200 2000 200 200
100 100 100
0 1000 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
0
20 40 60 80
Figure D.15 Variation of robustness of overheating hours for different design options of all
represent maximum regret). Only design options that influence overheating are shown here.
261
Appendices
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

0 = N-S facing S

1 = HR107
80 2 = ASHP 80 90 = E-W facing E 80
3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 0 45 90
100 6000 100 100


80 5000 80 80
60 4000 60 60
40 3000 40 40
20 20 20
2000
0 0 0
0
1000
6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0
20 40 60 80
Figure D.16 Variation of robustness of global cost for different design options of all Pareto
solutions for the policymaker and the homeowner calculated using three robustness
Designs with a low WWR (20%) reduced the deviation of GC, and designs with a
high WWR (80%) resulted in low deviation of overheating. Similar deviation of CO2
emissions can be observed for all considered WWRs. However, designs with a WWR
of 40% resulted in lower deviation of overheating hours than those of WWR 80 and
also had lower GC deviation, and were accordingly the most preferred. As expected,
designs with a large PV system (32 m2), large SDHW systems (7.5 m2) and bigger
battery capacities (16 kWh) were the most robust options for both decision makers.
In summary, the robust design options are the RP1 renovation package, a moderately
air tight building envelope, the HR107 gas boiler and south oriented large RES
systems at a tilt angle of 45°. These robust design options are similar to the robust
design options with the minimax regret method. These design options for three
robustness assessment methods are compared with the most robust designs selected
based on the MCDM approach (see Table 6.8).
262
D.2.3 The most robust design using the MCDM approach

The design scores of Pareto solutions calculated using the Hurwicz criterion
considering all preferred performance and robustness indicators for both decision
makers were compared for the max-min method and the best-case and worst-case
method. This comparison is presented in Figure D.17. As noted earlier with the
design scores for the minimax regret method (see Figure 6.25), the design score
gradually decreased beyond a certain ICa, indicating that these designs were not cost-
optimal robust solutions. This trend in design scores is in contrast to the design
scores for the policymaker but is similar to the design scores for the homeowner.
The most robust designs based on the highest design scores were the designs with
an ICa of 22.02 k€, and 22.05 k€ using the max-min method and the best-case and
worst-case method, respectively. Alternatively, the decision makers can choose
robust design options based on the highest design score (see Figure D.18) and can
trade these off with the required ICa. In Figure D.18, the design scores for different
design options are presented for the max-min method, and the best-case and worst-
case method. In addition, these designs scores were compared with those of the
design options for the minimax regret method.
The max-min method: It can be inferred from Figure D.18 that the designs with
renovation package RP1, an infiltration rate of 0.1 dm3/ds2 and a WWR of 20% had
the highest design score for the max-min method and were therefore the most
robust. In addition, the designs with the GSHP heating system were the most robust,
which is similar to the most robust design option using the trade-off approach.
Designs with a PV system of 28.8 m2 were the most robust, which contrasts with the
robust PV system for the minimax regret method. This relatively small PV system
was compensated by a small SDHW system of 2.5-5 m2 in the max-min method. It
was found that south facing PV and SDHW systems at a tilt angle of 45° were the
most robust design options. Designs with a battery capacity of 16 kWh were the most
robust.
The best-case and worst-case method: Designs with renovation package RP3, an
infiltration rate of 0.1 dm3/ds2 and a WWR of 20% were the most robust. These most
robust design options were similar to the most robust design options with the
minimax regret method. The HR 107 gas boiler was the most robust heating option
and south facing PV and SDHW systems at a tilt angle of 45° were the most robust
design options. As expected, designs including large PV system of 32 m 2 and a
battery capacity of 16 kWh were the most robust.
263
Appendices
Spread Deviation
1 1
0.9 0.9
0.8 0.8
0.7 0.7
Design score
Design score
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
-5 5 15 25 35 45 55 -5 5 15 25 35 45 55
Additional investment cost (IC a), k€ Additional investment cost (ICa), k€
Figure D.17 The design scores of Pareto solutions for both policymaker and homeowner
calculated using the Hurwicz criterion for the max-min and the best-case and worst-case
methods considering predicted performance and corresponding robustness. The reference
building design is indicated in green (ICa = 0k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 6 7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9
1 = HR107 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 0 45 90
1 6000 1 1
0.9 0.9 0.9

0.8 5000 0.8 0.8
0.7 0.7 0.7

0.6 4000 0.6 0.6
0.5 0.5 0.5
0.4 3000 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
2000
0.1 0.1 0.1
0 0 0
0 1000
6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
0
20 40 60 80
Figure D.18 The design scores of different design options of Pareto solutions for the
policymaker and the homeowner calculated using the Hurwicz criterion for three robustness
264
It is intriguing that robust design options differed greatly for the three methods. This
difference was attributed to the method of optimizing the robustness in these
methods. In the min-max method, spread minimizes variation across extreme
scenarios, leading to conservative design options. In contrast, using the minimax
regret method, robustness (maximum regret) is optimized with respect to optimal
performance. For instance, large battery capacities could enhance self-consumption
of electricity and consequently reduce variations (spread) in GC and CO 2 emission,
whereas relatively small battery capacities resulted in cost-optimal design options
and were thus robust to GC using the maximum regret method.
D.2.4 Scenario analysis
The sensitivity index is shown in Figure D.19 for all predicted performance
indicators and their corresponding robustness calculated using three robustness
assessment methods. It can be inferred from Figure D.19 that the number of
occupants (OS) and their corresponding behaviour, especially Ause, Luse and IHG, had
a significant influence on all performance indicators and their corresponding
robustness. Similarly, climate scenarios influenced all performance indicators and
their corresponding robustness.
However, as noted earlier using the minimax regret (see Figure 6.27), there were a
few scenarios that influenced predicted performance but not robustness. The
influence of net-metering was limited especially with the max-min method, because
designs with large battery systems overcome the disadvantage of the termination of
net-metering. Therefore, a battery could be a feasible option to tackle net-metering
termination in the future and to reduce grid stress by increasing self-consumption
of electricity. It was also intriguing that DHW scenarios had little influence on
robustness calculated using the max-min method. This little influence was attributed
to designs with large SDHW systems in the Pareto front calculated using the max-
min method.
In summary, occupant behaviour related scenarios and climate scenarios were the
most influential scenarios on all considered performance indicators, particularly
number of occupants and their corresponding Ause and Luse, and climate change.
265
Appendices
a) CO2 emissions
1.0
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
CO2 emissions Spread of CO2 emissions
Deviation of CO2 emissions Maximum regret of CO2 emissions
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
c) Global cost
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Figure D.19 Sensitivity of various scenarios to the predicted performance and corresponding
robustness of preferred performance indicators of both policymaker and homeowner
calculated using the Mann-Whitney test. Scenarios where p<0.05 (dotted line) are assumed
to be sensitive. These values are shown for three robustness assessment methods (Stars
represent predicted performance, triangles represent spread, and rectangles represent
deviation and circles represent maximum regret).
266
Appendix E. Demonstration of the CPRA approach using the new
house case study
This appendix briefly discusses the demonstration of the CPRA approach using the new
house case study for the policymaker, the homeowner and both decision makers. This
demonstration was summarized in Chapter 6.
E.1 Demonstration of the CPRA approach for the policymaker
E.1.1 Trade-off solutions using the Pareto front

The Pareto front for the policymaker calculated using three robustness assessment
methods is shown in Figure E.1. In this figure, graphs from left to right represent
Pareto fronts calculated using the max-min, the best-case and worst-case and the
minimax regret methods. Again, each bubble represents the median value of CO2
emissions and the bubble size depicts the corresponding robustness. It can be noted
from Figure E.1 that the CO2 emissions and their corresponding robustness
gradually decreased in proportion to increases in ICa up to 15 k€ with all three
methods. Beyond the 15 k€ ICa, CO2 emissions of the designs further decreased for
all three methods, but the improvement in robustness did not outweigh the required
ICa for the max-min method and the best-case and worst-case method. In contrast,
the maximum regret of CO2 emissions gradually decreased in line with an increase
in ICa and it was even close to zero for a few designs beyond the 15 k€ ICa.
a) Spread b) Deviation c) Maximum Regret

Bubble size = Spread of CO2 emissions Bubble size = Deviation of CO2 emissions Bubble size = Max.Regret of CO2 emissions
(2324-3884 kgCO2/a) (2247-5518 kgCO2/a) (5.34-3966 kgCO2/a)
4000 4000 4000
3500 3500 3500

3000 3000 3000
2500 2500 2500
2000 2000 2000
1500 1500 1500
1000 1000 1000
500 500 500
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Additional investment cost (ICa), k€ Additional investment cost (IC a), k€ Additional investment cost (IC a), k€
Figure E.1 Pareto fronts for the policymaker calculated using three robustness assessment
methods. The figures from left to right represent the max-min method, the best-case and
worst-case method and the minimax regret method. In all figures, the robustness is included
as bubble size. The green bubble represents the reference building design.
For all three methods, in the ICa range of 0-15 k€, the policymaker could prioritize
CO2 emissions and their corresponding robustness and then trade them off with the
267
Appendices
required ICa. In the next ICa range of 15-30 k€, since robustness is almost identical
for all designs, the policymaker may wish to prioritize predicted performance and
then trade it off with the required ICa. This observation was valid for the remaining
designs in the last ICa range. In the case of the minimax regret method, the
policymaker could prioritize CO2 emissions and corresponding maximum regrets
and trade them off with the required ICa.
E.1.2 Robust design options

The design options of the entire Pareto front calculated using three robustness
assessment methods were compared and this comparison is presented in Figure E.2.
For all three methods, designs with low infiltration rates of 0.1 and 0.15 dm3/ds2 were
found to be the most robust to CO2 emissions. Similarly, designs with WWRs of 20-
40% in buildings oriented towards the south were the most robust options. In
addition, south facing PV and SDHW systems at a tilt angle of 45° were the most
robust options. Also, the HR 107 gas boiler was the most robust heating system.
Designs with large RES systems (PV=32 m2, SDHW= 7.5-10 m2 and battery =20
kWh) were the most robust options for all three methods. The notable difference in
the results of the three robustness assessment methods are presented below.
The max-min method: Designs with building envelope package P3 had the lowest
spread of CO2 emissions, and accordingly this was the most robust building
envelope package (see Figure E.2). However, designs with P1 and large RES system
also resulted in a similar spread as that of designs with P3. The required ICa for P1
was lower compared to P3 and was thus the more preferred robust building envelope
package. Light-weight buildings oriented towards the south were the most robust.
The best-case and worst-case method: Designs with the building envelope packages P4
and P5 resulted in the lowest deviation (see Figure E.2). However, P4 was the most
preferred robust building envelope package due to its lower IC a compared to P5.
Heavy-weight buildings oriented towards the south resulted in the lowest deviation
of CO2 emissions and were hence the most robust.
The minimax regret method: It can be inferred from Figure E.2 that designs with
building envelope package P5 had zero maximum regret for CO 2 emissions.
Therefore, P5 was the most robust renovation package. However, designs with
building envelope packages P3 and P0 also resulted in close to zero maximum
regrets, which was attributed to the inclusion of large RES systems.
268
Demonstration of the CPRA approach using the new house case study
6000 6000 6000

5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
Building envelope package (P) Infiltration rate, dm3/ds2 WWR, (%)
6000 6000 6000

1 = HR107

0 = N-S facing S 1 = Heavy-weight

5000 2 = ASHP 5000 5000

90 = E-W facing E 2 = Light-weight
4000 3 = GSHP 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
1 2 3 0 90 180 1 2 3
HVAC system type Building orientation (°) Building thermal mass
6000 6000 6000


5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000
6000 6000
0 = N-S facing S
5000 5000 5000

90 = E-W facing E
4000 4000 4000
3000 3000
3000
2000 2000
2000
1000 1000
1000 0 0
0 90 180 0 45 90
0 Orientation of RES, (°) Tilt angle of RES, (°)
20 40 60 80
Figure E.2 Variation of robustness of CO2 emissions of different design options of all Pareto
solutions for the policymaker calculated using three robustness assessment methods
regret).
Similar to observations made about the robust designs for the renovation house case
study, designs with large RES systems were the most robust with all three methods.
The notable difference among the three robustness assessment methods was that
light-weight building designs with low insulation levels and large RES systems were
the most robust with the max-min method. In contrast, heavy-weight building
designs with high insulation levels in combination with large RES systems were the
most robust with the other two methods.
269
Appendices
E.1.3 The most robust design using the MCDM approach

It can be observed from Figure E.3 that the design score gradually increased in
proprtion to increases in ICa for all robustness assessmnet methods. The most robust
design in each of the three methods was the design with the highest ICa. However,
beyond a certain ICa level, the improvement in the design score did not outweigh the
required ICa. For instance, in the case of the max-min method, it can be noted that
the design with an ICa of 23 k€ resulted in a similar design score as that of the design
with an ICa of 28 k€. Similarly, in the case of the best-case and worst-case method, it
was found that the design score did not improve significantly beyond 23.6 k€ of IC a.
Therefore, the policymaker could prefer the most robust design based on the highest
design score or trade off the design score with the required ICa.
Figure E.4 provides an overview of the design scores of different design options of
all Pareto solutions calculated using three robustness assessment methods. The
most robust design options based on the highest design score were similar to the
robust design options using the trade-off approach. For instance, designs with large
RES systems were the most robust for all three methods. It is worth noting that
different building envelope packages yielded the same design score and this same
design score was attributed to the combination of other design options such as RES
systems. For instance, P1 and P3 envelope packages with SDHW systems of 10 m2
and 7.5 m2 have same design score. The main difference among robust designs based
on the highest design score for the three robustness assessment methods are
summarized below.
The max-min method: It can be noted from Figure E.4 that the designs with building
envelope packages P1 and P3 had the highest design score and were hence the most
robust. However, the policymaker would prefer P1 due to its low ICa. In addition,
light-weight buildings were the most robust.
The best-case and worst-case method: Designs with building envelope packages P4 and
P5 were the most robust, but the policymaker would prefer P4. In contrast to the
max-min method, heavy-weight buildings were the most robust with the best-case
and worst-case method.
The minimax regret method: The design with building envelope package P3 had the
highest design score and was accordingly the most robust. Similar to the best-case
and worst-case method, here, heavy-weight buildings were the most robust.
270

1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Design score
Design score
Design score
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Additional investment cost (IC a), k€ Additional investment cost (ICa), k€ Additional investment cost (ICa), k€
Figure E.3 The design scores of Pareto solutions for the policymaker calculated using the
performance and corresponding robustness. The reference building is indicated in green
(ICa = 0k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
1 = HR107 0 = N-S facing S 1 = Heavy-weight
0.9 0.9 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
2 = Light-weight
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 1 2 3
HVAC system type Building orientation, (°) Building thermal mass
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000 PV system size, m2 Solar DHW system size, m2 Battery capacity, kWh
5000 1 1
0.9
90 = E-W facing E
4000 0.8 0.8
0.7 0.7
0.6 0.6
3000
0.5 0.5
0.4 0.4
2000 0.3 0.3
0.2 0.2
1000 0.1 0.1
0 0
0 0 90 180 0 45 90
Orientation of RES, (°) Tilt angle of RES, (°)
20 40 60 80

Figure E.4 The design scores of different design options of Pareto solutions for the
policymaker calculated using the Hurwicz criterion for three robustness assessment methods
regret).
271
Appendices
E.1.4 Scenario analysis

Figure E.5 shows the sensitivity index (p) of various scenarios to the predicted
performance and robustness of Pareto solutions calculated using three robustness
assessment methods. It can be observed that the number of occupants and their
corresponding behavior, mostly regarding appliance use, lighting use, IHG and
DHW use were the most influential scenarios on CO2 emissions and their
corresponding robustness for all three robustness assessment methods. As expected,
ventilation and window opening had no influence on the robustness of CO 2
emissions for all three methods as these two parameters had little influence on
energy consumption. Similarly, the influence of net-metering was reduced to a great
extent by the inclusion of large battery systems.
It is worth noting that all scenarios, except heating setpoint, influenced predicted
performance. This observation contrasts with the renovation house case study
results, as the heating setpoints had little influence on CO 2 emissions for new
houses. This little influence was attributed to the Pareto solutions that were
dominated by building envelope packages with very high insulation levels. These
very high insulation levels resulted in very low heating demands and variations in
these low heating demands were less influenced by heating setpoints. Furthermore,
the variation in this low heating demand has little influence on CO 2 emissions
compared to the plug loads in these buildings.
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
Figure E.5 Sensitivity of various scenarios to the predicted performance and corresponding
robustness of the preferred performance indicators of the policymaker calculated using the
maximum regret).
272
E.2 Demonstration of the CPRA approach for the homeowner

The Pareto front for the homeowner calculated using three robustness assessment
methods is shown in Figure E.6. The graphs from left to right show the Pareto front
for different robustness assessment methods and top to bottom show the Pareto
front for different preferred performance indicators of the homeowner. It is worth
noting that different x-axis scales were used in these graphs to enhance visualization.
The max-min method yielded Pareto solutions across a wide range of IC a compared
to the other two methods. This difference among Pareto solutions was attributed to
the GC, which decreases gradually in proportion to increases in IC a with the max-
min method, as observed with renovation buildings (Figure D.2). In contrast, with
the best-case and worst-case method and the maximum regret method, designs
beyond 15 k€ ICa were not cost-optimal and hence do not feature on the
corresponding Pareto fronts.
a) Spread b) Deviation c) Maximum regret

Bubble size = Spread of overheating hours Bubble size = Deviation of overheating hours Bubble size = Max. Regret of overheating hours
(160-1251 h/a) (168-478 h/a) (11-294 h/a)
120 120 120
100 100 100
80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-5 5 15 25 35 -5 0 5 10 -5 0 5 10
Bubble size = Spread of global cost Bubble size = Deviation of global cost Bubble size =Max. Regret of global cost
(22.8-42.3 k€) (35.7-46.4 k€) (1.1-11 k€)
100 100 100
80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-5 5 15 25 35 -5 0 5 10 -5 0 5 10
Figure E.6 Pareto fronts for the homeowner calculated using three robustness assessment
methods. The figures from left to right represent the max-min method, the best-case and
worst-case method and the minimax regret method. The figures from top to bottom
represent overheating hours and global costs, respectively. In all figures, robustness is
included as bubble size. The green bubble represents the reference building design. Different
x-axis scales are used for these graphs to enhance visualization.
273
Appendices
It can be inferred from Figure E.6 that with all three robustness assessment
methods, the building designs with similar or better predicted performance and
corresponding robustness of overheating hours as those of the reference building
design (P0) can be found across the entire ICa range. As such, the homeowner could
prioritize a design that has better predicted performance and corresponding
robustness of GC and then trade them off with the required ICa. For instance, with
the max-min method, the designs within the ICa range of 25-35 k€ were the most
robust to GC, but these designs resulted in high GC. For the best-case and worst-
case method, deviation of GC did not reduce significantly with increases in ICa.
Therefore, the preferred robust designs could be found in the IC a range of 0-5 k€.
Similarly, using the minimax regret method, the designs in the ICa range of 5-10 k€
were the most robust. It is worth noting that compared to the renovation house case
study results, the designs with much lower ICa were more robust for new buildings
because the designs options of the new house had lower ICa (see Table C.3 and Table
C.4) compared to the design options of the reference building design (P0).

Figure E.7 and Figure E.8 provide an overview of different design options to aid the
homeowner to determine which design options lead to optimal performance and
which is the most robust design. For instance, it can be observed from Figure E.7
that the designs with building envelope packages with low insulation levels, P0 and
P1, were the most robust to overheating hours with all three methods. Similarly,
designs with high infiltration rates (0.625 dm3/ds2) were the most robust to
overheating hours, and designs with low infiltration rates were the most robust to
GC. Designs with WWRs of 20-40% were the most robust to both overheating and
GC. Interestingly, both building orientations (south and east) resulted in similar
robustness to GC and overheating hours. Similarly, a south facing PV system was
the most robust to GC with all three methods. The heavy-weight buildings were the
most robust to overheating hours with all three methods and the effect of the
building’s thermal mass on GC was limited.
In contrast to observations made with the renovation house case study results, the
HR 107 boiler was the most robust heating option for the new building with all three
methods. In the case of the renovation house case study results, GSHP was the most
robust heating option with the max-min method. The notable differences among
robust designs with all three methods are summarized below.
274
The max-min method: Designs with high insulation levels (P2-P4) were the most
robust. Similar to the renovation house case study results, designs with large RES
systems (PV systems of 32 m2, battery capacity of 16 kWh) were the most robust.
Interestingly, robustness of GC was least influenced by SDHW system, because all
sizes of SDHW system resulted in a similar spread of GC. Therefore, designs with
no SDHW system were the most preferred as they required no ICa. As expected, a
tilt angle of 45° was the most robust option for RES systems.
The best-case and worst-case method: Designs with building envelope package P1 were
the most robust to GC (see Figure E.8). In contrast to the max-min method, designs
with a small PV system of 12.8m2 were found to be the most robust. There were no
designs with a SDHW system and battery among the Pareto design options for the
homeowner. Surprisingly, a PV system at a tilt angle of 0° was the most robust
option. This difference was attributed to the increase in self-consumption of
electricity with 0° tilt for all the Pareto designs with a small PV system and no battery.
The minimax regret method: All the robust design options with the minimax regret
method were quite similar to the robust design options found with the best-case and
worst-case method. These similarities in the robust designs were attributed to the
optimization of robustness with respect to the best performance/optimal
performance in these methods.
1400 1400 1400
1200 1200 1200

1000 1000 1000

800 800 800
600 600 600
400 400 400
200 200 200
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
Building envelope package (P)
6000 Infiltration rate, dm3/ds2 WWR, (%)
1400 1400
5000 1 = Heavy-weight
0 = N-S facing S
1200 1200
4000 1000 1000
800 800
3000
600 600
2000 400 400
200 200
1000
0 0
0 0 90 180 1 2 3
20 Building orientation
40(°) Building
60 thermal mass 80

Figure E.7 Variation of robustness of overheating hours for different design options of all
Pareto solutions for the homeowner calculated using three robustness assessment methods
regret). Only design options that influence overheating are shown here.
275
Appendices
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

1 = Heavy-weight
0 = N-S facing S

1 = HR107
80 2 = ASHP 80 90 = E-W facing E 80 2 = Light-weight
3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 1 2 3
100 100 100


80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000 100 100

0 = N-S facing S
5000 80 90 = E-W facing E 80
4000 60 60
3000 40 40
20 20
2000
0 0
1000 0 90 180 0 45 90
0
20 40 60 80

Figure E.8 Variation of robustness of global cost for different design options of all Pareto
solutions for the homeowner calculated using three robustness assessment methods
regret).

The design score of Pareto solutions for the homeowner calculated using the three
robustness assessment methods is shown in Figure E.9. Similar to observations
made for the renovation house case study, the design score gradually decreased
beyond a certain level of ICa. The most robust design for the homeowner was the
design with an ICa of 15.2 k€ with the max-min method, and it was the design with
an ICa of 5.9k€ for the other two methods.
276

1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Design score
Design score
Design score
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-5 5 15 25 35 -5 5 15 25 35 -5 5 15 25 35
Figure E.9 The design scores of Pareto solutions for the homeowner calculated using the
performance and corresponding robustness. In all graphs, the reference building design is
indicated in green (ICa = 0k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9 0.9 0.9 2 = Light-weight
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 1 2 3
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000 PV
system size, m2 Solar DHW system size, m2 Battery capacity, kWh
5000 1 1
0.9
4000 0.8 90 = E-W facing E 0.8
0.7 0.7
3000 0.6 0.6
0.5 0.5
0.4 0.4
2000 0.3 0.3
0.2 0.2
1000 0.1 0.1
0 0
0 0 90 180 0 45 90
20 Orientation of RES,
40 (°) 60 of RES, (°)
Tilt angle 80

homeowner calculated using the Hurwicz criterion for three robustness assessment methods
regret).
277
Appendices
The robust design options based on the MCDM method (see Figure E.10) were quite
similar to the robust design options based on the trade-off solutions using the Pareto
front (Figure E.7 and Figure E.8). Furthermore, the robust design options of the new
house case study were in line with those of the renovation house case study.

Figure E.11 shows the sensitivity index for predicted performance and corresponding
robustness, calculated using the three robustness assessment methods.
a) Overheating hours
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
b) Global cost
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
robustness of the preferred performance indicators of the homeowner calculated using the
maximum regret).
278
As expected, the number of occupants and behavior related scenarios were found to
be the most influential on overheating and GC (see Figure E.11) in the case of the
max-min method and the best-case and worst-case method. This observation was
similar to that of the renovation building. As discussed in the case of the renovation
house case study, low-high scenarios for appliance and lighting use and their
corresponding IHG influenced GC, but did not influence the maximum regrets of
GC. OS, IHG and CS were the most influential scenarios for overheating.
Furthermore, NM had no influence on GC due designs among Pareto solutions
having very small RES systems.
E.3 Demonstration of the CPRA approach for both decision makers

The Pareto front for both decision makers calculated using three robustness
assessment methods was compared and is presented in Figure E.12. In this figure,
graphs from left to right show different robustness indicators and the graphs from
top to bottom represent different preferred performance indicators of both decision
makers. It can be observed that CO2 emissions gradually decreased in proportion to
increases in ICa, whereas GC gradually increased in proportion to increases in ICa.
In contrast, the designs with similar overheating hours can be found across the entire
ICa range of 0-30 k€. These observations were valid for three robustness assessment
methods.
In addition, designs in the high ICa range (35-45 k€) had very low CO2 emissions and
corresponding high robustness with all three methods, even though improvement
in robustness did not outweigh the required ICa with the max-min and the best-case
and worst-case methods. In contrast, in this ICa range, designs had close to zero
maximum regrets for CO2 emissions with the minimax regret method, which were
accordingly the most preferred robust designs for the policymaker. However, these
designs resulted in high overheating hours and high GC, and were also the least
robust to GC and overheating hours. As such, they may not be preferred by the
homeowner. The designs within the ICa range of 15-35 k€ represent compromise
among all performance indicators and thus, were the preferred robust designs for
both decision makers using the minimax regret method. Similar observations can be
made for the best-case and worst-case method. Contrariwise, in the case of the max-
min method, the spread of GC gradually decreased in proportion to increases in ICa.
Furthermore, CO2 emissions gradually decreased, but the reduction in spread of CO2
279
Appendices
emissions was not significant. These designs incurred very high GC. Therefore,
using the max-min method, both decision makers might prefer the designs in the
ICa range of 15-25 k€.
a) Spread b) Deviation c) Maximum Regret

Bubble size = Spread of CO2 emissions Bubble size = Deviation of CO2 emissions Bubble size =Max. Regret of CO2 emissions
(2239-4231 kgCO2/a) (2242-5525 kgCO2/a) (0-3984 kgCO2/a)
5000 5000 5000


4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Bubble size = Spread of overheating hours Bubble size = Deviation of overheating hours Bubble size = Max. Regret of overheating hours
(161-1271 h/a) (168-1272 h/a) (11-1110 h/a)
120 120 120

100 100 100
80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Bubble size = Spread of global cost Bubble size = Deviation of global cost Bubble size =Max. Regret of global cost
(21.9-42.4 k€) (35.4-64.7 k€) (5-48 k€)
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Figure E.12 The Pareto fronts for the policymaker and the homeowner calculated using three
robustness assessment methods. The figures row wise from top to bottom represent CO2
emissions, overheating hours and global costs, respectively. The figures column wise from
left to right represent max-min method, the best-case and worst-case method and the
minimax regret method. In all figures, the robustness is included as bubble size. The green

The robust design options for both decision makers using the three robustness
assessment methods are presented in Figure E.13, Figure E.14 and Figure E.15. These
figures present the variation of robustness of preferred performance indicators of
both decision makers calculated using three robustness assessment methods. It can
be noted that the design options that were the most robust to a performance indicator
could be the least robust to other performance indicators. For instance, with the
minimax regret method, designs with building envelope package P5 were the most
280
robust to CO2 emissions, whereas these designs were least robust to overheating
hours. Therefore, as discussed earlier, a compromise between these preferred
performance and robustness indicators should be made to find the preferred robust
design options for both decision makers. The preferred robust designs were
compared and are tabulated in Table 6.12.
6000 6000 6000

5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
6000 6000 6000

1 = HR107

0 = N-S facing S 1 = Heavy-weight
5000 2 = ASHP 5000 5000

4000 3 = GSHP 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
1 2 3 0 90 180 1 2 3
HVAC system type Building orientation (°) Building thermal mass
6000 6000 6000


5000 5000 5000
4000 4000 4000
3000 3000 3000
2000 2000 2000
1000 1000 1000
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000 6000 6000

0 = N-S facing S
5000 5000 5000

90 = E-W facing E
4000 4000
4000
3000 3000
3000 2000 2000
2000 1000 1000
0 0
1000 0 90 180 0 45 90
0
20 40 60 80
Figure E.13 Variation of robustness of CO2 emissions of different design options from all
281
Appendices
1400 1400 1400

1200 1200 1200
1000 1000 1000

800 800 800
600 600 600
400 400 400
200 200 200
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
6000
1400 1400
1 = Heavy-weight

0 = N-S facing S
5000 1200 1200

1000 1000
4000
800 800
3000 600 600
400 400
2000
200 200
1000 0 0
0 90 180 1 2 3
0 Building orientation (°) Building thermal mass
20 40 60 80
Figure E.14 Variation of robustness of overheating hours for different design options of all
282
100 100 100

80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
100 100 100

1 = Heavy-weight
0 = N-S facing S

1 = HR107
80 2 = ASHP 80 90 = E-W facing E 80 2 = Light-weight
3 = GSHP
60 60 60
40 40 40
20 20 20
0 0 0
1 2 3 0 90 180 1 2 3
100 100 100


80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000
100 100
5000 0 = N-S facing S
80 90 = E-W facing E 80
4000
60 60
3000
40 40
2000
20 20
1000
0 0
0 0 90 180 0 45 90
20 40 60 80
Figure E.15 Variation of robustness of global cost for different design options of all Pareto
solutions for the policymaker and the homeowner calculated using three robustness

considering all preferred performance and robustness indicators of both decision
makers were compared and are presented in Figure E.16.
283
Appendices
Spread Deviation Max.Regret

1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Design score
Design score
Design score
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-5 5 15 25 35 45 -5 5 15 25 35 45 -5 5 15 25 35 45
Figure E.16 The design scores of Pareto solutions for both policymaker and homeowner
calculated using the Hurwicz criterion for three robustness assessment methods considering
predicted performance and corresponding robustness. In all graphs, the reference building
design is indicated in green (ICa = 0k€).
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 20 40 60 80
1 1 1
0.9 0.9 0.9
0.8 2 = ASHP 0.8 90 = E-W facing E 0.8
2 = Light-weight
0.7 3 = GSHP 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
1 2 3 0 90 180 1 2 3
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7

0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 6.4 12.8 19.2 25.6 32 0 2.5 5 7.5 10 0 4 8 12 16 20
6000
5000 1 1
0.9
4000 0.8 90 = E-W facing E 0.8
0.7 0.7
3000 0.6 0.6
0.5 0.5
0.4 0.4
2000
0.3 0.3
0.2 0.2
1000 0.1 0.1
0 0
0 0 90 180 0 45 90
20 Orientation of RES,
40(°) 60 of RES, (°)
Tilt angle 80

policymaker and the homeowner calculated using the Hurwicz criterion for three robustness
284
Using the max-min method, the design scores beyond an ICa of 17.2 k€ gradually
decreased in proportion to increases in ICa. Similar observations can also be made
for the best-case and worst-case method (ICa of 21.5 k€) and the minimax regret
method (ICa of 19.1 k€). The most robust designs based on the highest design score
were the designs with an ICa of 17.2 k€, 21.5 k€ and 19.1 k€ with the max-min method,
the best-case and worst-case method and the minimax regret method respectively.
The robust designs were compared and are presented in Table 6.12. The robust
design options for the new house case study selected based on the MCDM approach
(see Figure E.17) were in line with those of the renovation house case study. In
addition, the building oriented towards the south was the most robust, except for the
minimax regret method, where an east orientation was the most robust option. The
influence of the thermal mass of the building was similar here to observations made
regarding thermal mass when the decision makers were considered separately.
The influence of scenarios on different performance indicators and their

corresponding robustness for both decision makers (see Figure E.18) was similar to
the influence observed for the homeowner (see Figure E.11) and the policymaker (see
Figure E.5) separately. The only exception was with the minimax regret method,
where the Ause, Luse and corresponding IHG had no influence on maximum regret of
GC for the homeowner (see Figure E.18), which is in contrast to the trend with these
scenarios for both decision makers combined. The notable difference between the
designs for the policymaker and designs for the homeowner was that Pareto
solutions had different battery capacities. In the case of the homeowner, there were
no designs with a battery system in the Pareto front. In such cases, the net imported
and exported energy of all designs were influenced by shell load dominated scenarios
such as Thsp and CS. In contrast, the plug load dominated scenarios such as low-high
Luse and Ause resulted in the same net imported and exported energy because these
profiles were the same for all designs and therefore had little influence.
285
Appendices
a) CO2 emissions
1.0
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
c) Global cost
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Scenarios
robustness of the preferred performance indicators of the homeowner calculated using the
maximum regret).
286
Appendix F. User group meetings: response to the main feedback
In this appendix, the details of user group meetings and the responses to the main feedback
are summarized.
F.1 First user group meeting

Meeting date 02-07-2015
Meeting location Vabi Elements, Delft, The Netherlands
Demonstration of initial version of the CPRA approach
Presented topic (CPRA0) with the main focus on the max-min robustness
assessment method.
The response to the main feedback from the first user group meeting is summarized
below:
1. Is the CPRA approach applicable for new buildings or renovated buildings?

The proposed CPRA approach is applicable to all types of residential buildings. The
CPRA approach illustrates the optimal balance between building envelope
performance improvements and corresponding onsite energy generation systems to
achieve more robust designs. The CPRA approach is generic and can be applied to
renovations as well as to new buildings.
2. Why is the chosen case study very old?
The case study chosen for the demonstration of the CPRA approach is based on
[Agentschap NL, 2013c]. Only the building layout is taken from this case study. The
building envelope properties are varied to obtain designs ranging from traditional
buildings to passive house standards. Moreover, the age of case study is not crucial
since, as stated earlier, the current CPRA approach can be applied to all types of
residential buildings.
3. In occupant scenarios, should an elderly family not be included?
The schedule of all-day presence and high setpoint temperatures represents retired
(elderly) people.
287
Appendices
4. What is the practical use of this CPRA approach?
Using the current CPRA approach, the decision maker can select robust designs that
ensure desired performance in operation. The proposed CPRA approach also
provides a decision maker with information to trade-off investment in improving
building envelope with that of energy generation systems (energy balance) and
robustness of design.
5. How can a decision maker/ designer choose a robust design using the presented CPRA
approach?
The CPRA approach illustrates the selection of a robust designs using multiple
performance indicators and performance spread across all scenarios. A robust design
is chosen based on the lowest median value of a performance indicator that has the
minimum spread. The robustness of the design can be traded off with required
additional investment cost.
F.2 Second user group meeting

Meeting location Smits van Burgst, Zoetermeer
Demonstration of the updated version of the CPRA
Presented topic approach (CPRA1) using case studies with the main focus on
the minimax regret method.
The response to the main feedback from the second user group meeting is
summarized below:
1. There is only one comfort-related parameter (overheating hours); why? The first
priority of homeowners is to have a comfortable environment that includes different
parameters (T, CO2, RH…)
The choice of performance indicator depends on the end user. In this case study,
overheating hours are used for demonstration purposes. In addition, adequate
ventilation is provided to avoid any CO2 related discomfort issues. Ideal heating
systems are used in this case study to avoid any underheating hours, However, the
developed CPRA approach remains the same for other comfort related performance
indicators.
2. How do you calculate overheating?
Thermal comfort is evaluated based on maximum and minimum acceptable indoor

temperatures with respect to reference outdoor temperature, as proposed by [Peeters
288
User group meetings: response to the main feedback
et al., 2009]. Overheating hours are the total number of hours exceeding the
allowable indoor temperatures during occupancy in a year. A weighting factor is
assigned for every excess degree above allowable indoor temperature limits.
3. Why take electricity consumption into account if your case-study buildings are always
NZEB?
The electricity consumption of a design was used to demonstrate the building-
dependency on the grid; the higher the electricity consumption of a building, the
higher its dependency on the grid. However, as rightly pointed out in the meeting,
this performance indicator doesn’t quantify the net-imports and exports from and to
the grid. Based on suggestions during the meeting, other performance indicators
that can account for energy balance such as net balance, onsite energy-matching and
off-site energy matching index will be assessed.
4. Robustness could also mean the risk of setting the system incorrectly (higher risk for
more complicated systems, risk of failure of complex systems)?
Yes. In this research, the building’s performance robustness is being assessed rather
than the system’s robustness. It would be interesting to assess the system’s
robustness with respect to incorrect settings; however, this area of investigation falls
outside the scope of this research.
5. Can we compare more performance indicators at once?
Yes, the multi-criteria assessment implemented in this CPRA approach allows

decision makers to compare multiple performance indicators at once. Based on the
decision maker’s preference, performance indicators can be prioritized and trade-off
solutions can be found from the Pareto front calculated using multiple performance
indicators and their corresponding performance robustness.
289
Appendices
F.3 Third user group meeting

Meeting location Halmos, Den Haag
Presented topic Demonstration of the CPRA approach for practical use.
The response to the main feedback from the third user group meeting is
summarized below:
1. The developed CPRA approach is demonstrated only for buildings. Can this approach
be used for designing robust energy systems such as HVAC or renewable energy
systems?
Yes, as demonstrated in the presentation, the developed CPRA approach is generic
and can be used for performance robustness assessment in a holistic approach i.e.
entire building design, as well as for performance robustness assessment of
individual systems such as HVAC, PV and SDHW system.
Using this approach, a designer can identify robust designs of energy systems with
respect to type of systems, size of systems etc. based on their performance
robustness.
2. What is robustness in this work? Does the robustness assessment also consider the risk
that a building will not function properly due to failure of systems such as a heat
pump?
In this research, robustness is defined as the ability of a building to maintain the
desired performance under uncertainties in building operation and external
conditions such as occupant behavior, climate change and policy change. The focus
is on performance robustness assessment rather than system robustness. Risk
assessment considering failure/malfunctioning of systems is not considered in the
present study, because it is not within the focus of this work.
3. Can this CPRA approach be used to improve energy performance predictions with EPC
assessment to get closer to actual performance i.e. to reduce the performance gap?
Currently, in EPC assessment, which is used to determine which energy label to
assign to a building, fixed assumptions regarding building operation, such as set
point temperatures and internal heat gains are used in the calculation of energy
performance predictions. However, uncertainties in operation lead to different
energy performances in different buildings with the same energy label [Majcen et al.,
2013]. By including different (low-high) scenarios considering uncertainties in
290
User group meetings: response to the main feedback
operation, performance range can be predicted as explained in the CPRA approach.

In future work, this performance range could be used to asses building labels to
reduce the performance gap and thus improve the robustness of buildings.
4. How can this approach be demonstrated to clients in simple way (e.g. one sheet/slide)
to enhance the decision-making process?
In this work, different visualization methods are considered to present the results to
decision-makers to enhance the decision-making process. These methods are not
presented in this meeting. However, based on suggestions in the feedback meeting,
future presentations will include them. For instance, providing an overview of design
options of the Pareto solutions of the entire design space in a single sheet would
enable decision makers to select robust design options that lead to optimal and
robust performance.
5. Did you consider the risk of not meeting a required design parameter, for instance not
meeting the required airtightness of a building, which is very high in practice?
Uncertainties in thermo-physical parameters are not considered in this study. As
pointed out in the discussion, the risk of not meeting the required design parameter
in the construction phase is very high, especially for design parameters such as
infiltration rates. Even though this CPRA approach is aimed at providing design
decision support at the design phase, the use of sensitivity analysis in this CPRA
approach enables the easy identification of the most influential design parameters
and the corresponding risks associated with them. This information can be used in
the decision-making process to stress the importance of different design parameters.
F.4 Fourth user group meeting

Meeting location TUe, Eindhoven
Demonstration of the updated version of the CPRA
Presented topic
approach using the renovation case study.
The response to the main feedback from the fourth user group meeting is
summarized below:
1. Why not consider underheating hours as a performance indicator? Buildings with

low insulation levels are prone to underheating discomfort and might not be robust?
Yes, the existing buildings and the buildings with low insulation levels are prone to
underheating discomfort. However, to reduce the complexity of considering too
291
Appendices
many indicators in the study, ideal heating systems are considered, and therefore,
underheating hours are transformed into equivalent operating costs.
It is noteworthy that the developed CPRA approach would not change by including
an additional performance indicator.
2. The outcome of the presented case study indicates that low insulation levels and large
energy systems are found to be robust? Does it mean that leaving existing buildings as
is a good option?
From the homeowner’s perspective, the most robust renovation measures for an
existing building built between 1975-91 is to double the insulation levels and install
large RES systems (e.g. PV system of 32 m2).
3. How can outcomes of this research be transferred to the end user/customers? For
instance, how can this approach be used by a real estate company in selling a house?
Using this approach, performance range can be given as a contracting option rather
than nominal/deterministic performance. In other words, regret/spread/deviation
can be added in contracting options in addition to predicted performance. The
developed CPRA approach can be used to find risk averse solutions. The developed
CPRA approach was tested on similar applications by a PDEng graduate for the
Homij MorgenWonen case study.
In addition, as one of the user group members pointed out, it is becoming

increasingly common to guarantee the required performance in operation. In
response to this observation, it was explained that the CPRA approach is intended to
find robust design options that deliver the required performance in operation.
292
References
Agentschap NL. 2011. “Ministerie van Binnenlandse Zaken En Koninkrijksrelaties

Voorbeeldwoningen : Bestaande Bouw.”
Agentschap NL. 2013a. “Ministerie van Binnenlandse Zaken En Koninkrijksrelaties:
Nieuwbouw.”
Agentschap NL. 2013b. “Ministerie van Economische Zaken: Referentiewoningen
Nieuwbouw,” 37.
Agentschap NL. 2013c. “Referentiewoningen Nieuwbouw (BENG).”
http://www.rvo.nl/sites/default/files/2013/09/BENG Tussenwoning.pdf .
Aissi, Hassene, Cristina Bazgan, and Daniel Vanderpooten. 2009. “Min-Max and Min-Max
Regret Versions of Combinatorial Optimization Problems: A Survey.” European
Journal of Operational Research 197 (2): 427–38. doi:10.1016/j.ejor.2008.09.012.
Anderies, John M. 2014. “Embedding Built Environments in Social–ecological Systems:
Resilience-Based Design Principles.” Building Research & Information 42 (2). Taylor &
Francis: 130–42. doi:10.1080/09613218.2013.857455.
Anderies, John M., Carl Folke, Brian Walker, and Elinor Ostrom. 2013. “Aligning Key
Concepts for Global Change Policy: Robustness, Resilience,and Sustainability.”
Ecology and Society 18 (2): 8–24. doi:10.5751/es-05178-180208.
Andersson, P. 1997. “Robustness of Technical Systems in Relation to Quality, Reliability and
Associated Concepts.” Journal of Engineering Design 8 (3): 277–88.
doi:10.1080/09544829708907966.
Ascione, Fabrizio, Nicola Bianco, Rosa Francesca De Masi, Gerardo Maria Mauro, and
Giuseppe Peter Vanoli. 2017. “Resilience of Robust Cost-Optimal Energy Retrofit of
Buildings to Global Warming: A Multi-Stage, Multi-Objective Approach.” Energy and
Buildings 153: 150–67. doi:10.1016/j.enbuild.2017.08.004.
ASHRAE. 2010. “Thermal Environmental Conditions for Human Occupancy,
ANSI/ASHRAE Standard 55-2010.”
Attia, Shady, Elisabeth Gratia, André De Herde, and Jan L M Hensen. 2012. “Simulation-
Based Decision Support Tool for Early Stages of Zero-Energy Building Design.” Energy
and Buildings 49. Elsevier B.V.: 2–15. doi:10.1016/j.enbuild.2012.01.028.
Averbakh, Igor. 2000. “Minmax Regret Solutions for Minimax Optimization Problems with
Uncertainty.” Operations Research Letters 27 (2): 57–65. doi:10.1016/S0167-
6377(00)00025-0.
Baker, Jack W., Matthias Schubert, and Michael H. Faber. 2008. “On the Assessment of
Robustness.” Structural Safety 30 (3): 253–67. doi:10.1016/j.strusafe.2006.11.004.
Bankes, Steven. 2010. “Robustness, Adaptivity, and Resiliency Analysis.” Association for the
Advancement of Artificial Intelligence, 2–7. (www.aaai.org).
Becchio, Cristina, Paolo Dabbene, Enrico Fabrizio, Valentina Monetti, and Marco Filippi.
2015. “Cost Optimality Assessment of a Single Family House: Building and Technical
Systems Solutions for the nZEB Target.” Energy and Buildings 90: 173–87.
doi:10.1016/j.enbuild.2014.12.050.
Berger, Tania, Christoph Amann, Herbert Formayer, Azra Korjenic, Bernhard Pospischal,
293
References
Christoph Neururer, and Roman Smutny. 2014. “Impacts of Climate Change upon
Cooling and Heating Energy Demand of Office Buildings in Vienna, Austria.” Energy
and Buildings 80: 517–30. doi:10.1016/j.enbuild.2014.03.084.
Betlem, Nienke, Hans Van Eck, and Raymond Beuken. 2010. “Energy Performance of
Buildings: Implementation of the EPBD in The Netherlands Status in November
2010.”
Bettis, Richard A., and Michael A. Hitt. 1995. “The New Competitive Landscape.” Strategic
Management Journal 16 (S1): 7–19. doi:10.1002/smj.4250160915.
Beyer, Hans Georg, and Bernhard Sendhoff. 2007. “Robust Optimization - A
Comprehensive Survey.” Computer Methods in Applied Mechanics and Engineering 196
(33–34): 3190–3218. doi:10.1016/j.cma.2007.03.003.
Bickel, Peter J. 1976. “Another Look at Robustness: A Review of Reviews and Some New
Developments.” Scandinavian Journal of Statistics 3 (4): 145–58.
Blasco, X, J M Herrero, J Sanchis, and M Martínez. 2008. “A New Graphical Visualization of
N -Dimensional Pareto Front for Decision-Making in Multiobjective Optimization”
178: 3908–24. doi:10.1016/j.ins.2008.06.010.
Blom, Inge, Laure Itard, and Arjen Meijer. 2011. “Environmental Impact of Building-Related
and User-Related Energy Consumption in Dwellings.” Building and Environment 46
(8). Elsevier Ltd: 1657–69. doi:10.1016/j.buildenv.2011.02.002.
BPIE. 2010. “Cost Optimality - Discussing Methodology and Challenges within the Recast
EPBD,” 40.
Burhenne, Sebastian, Dirk Jacob, and Gregor P Henze. 2011. “Sampling Based on SOBOL
Sequences for Monet Carlo Techniques Applied to Building Simulations.” In
Proceedings of Building Simulation 2011: 12th Conference of International Building
Performance Simulation Association, Sydney, 14-16 November., 14–16.
Burhenne, Sebastian, Olga Tsvetkova, Dirk Jacob, Gregor P. Henze, and Andreas Wagner.
2013. “Uncertainty Quantification for Combined Building Performance and Cost-
Benefit Analyses.” Building and Environment 62: 143–54.
doi:10.1016/j.buildenv.2013.01.013.
Buso, Tiziana, Valentina Fabi, Rune K. Andersen, and Stefano P. Corgnati. 2015. “Occupant
Behaviour and Robustness of Building Design.” Building and Environment 94. Elsevier
Ltd: 694–703. doi:10.1016/j.buildenv.2015.11.003.
Butera, Federico M. 2013. “Zero-Energy Buildings: The Challenges.” Advances in Building
Energy Research 7 (1): 51–65. doi:10.1080/17512549.2012.756430.
CBS. 2015. “Aardgas En Elektriciteit, Gemiddelde Prijzen van Eindverbruikers.”
CBS. 2016a. “CBS StatLine - Aardgas En Elektriciteit, Gemiddelde Prijzen van
Eindverbruikers.”
CBS. 2016b. “Central Bearue of Statistics.”
http://statline.cbs.nl/StatWeb/publication/?DM=SLNL&PA=7409WBO&D1=40-
59&D2=0&D3=17-.
CBS. 2016c. “Central Bureau of Statistics Netherlands - Households.”
http://statline.cbs.nl/StatWeb/publication/?VW=T&DM=SLEN&PA=82905ENG&LA=
EN.
Chalupnik, Marek J., David C. Wynn, and P. John Clarkson. 2013. “Comparison of Ilities for
294
Protection Against Uncertainty in System Design.” Journal of Engineering Design 24
(12): 814–29. doi:10.1080/09544828.2013.851783.
Chien, Chen Fu, and Jia Nian Zheng. 2012. “Mini-Max Regret Strategy for Robust Capacity
Expansion Decisions in Semiconductor Manufacturing.” Journal of Intelligent
Manufacturing 23 (6): 2151–59. doi:10.1007/s10845-011-0561-1.
Chinazzo, Giorgia, Parag Rastogi, and Marilyne Andersen. 2015a. “Assessing Robustness
Regarding Weather Uncertainties for Energy- Efficiency-Driven Building
Refurbishments.” Energy Procedia 78: 931–36. doi:10.1016/j.egypro.2015.11.021.
Chinazzo, Giorgia, Parag Rastogi, and Marilyne Andersen. 2015b. “Robustness Assessment
Methodology for the Evaluation of Building Performance with a View to Climate
Uncertainties.” In Proceedings of BS2015: 14th Conference of International Building
Performance Simulation Association, Hyderabad, India, Dec. 7-9, 2015., 947–54.
CIBSE. 2006. CIBSE Guide A: Environmental Design. 2nded. London; UK: The Chartered
Institution of Building Services Engineers London.
Clarke, J.A., and J.L.M. Hensen. 2015. “Integrated Building Performance Simulation:
Progress, Prospects and Requirements.” Building and Environment 91. Elsevier: 294–
306. doi:10.1016/j.buildenv.2015.04.002.
Clevenger, Caroline M, and John Haymaker. 2006. “The Impact Of The Building Occupant
On Energy Modeling Simulations.” In Joint International Conference on Computing and
Decision Making in Civil and Building Engineering, 1–10.
CLG. 2010. “Code for Sustainable Homes: Technical Guide.” doi:ISBN 978 1 85946 331 4.
Coakley, Daniel, Paul Raftery, and Marcus Keane. 2014. “A Review of Methods to Match
Building Energy Simulation Models to Measured Data.” Renewable and Sustainable
Energy Reviews 37: 123–41. doi:10.1016/j.rser.2014.05.007.
Cole, Raymond J., and Laura Fedoruk. 2015. “Shifting from Net-Zero to Net-Positive Energy
Buildings.” Building Research & Information 43 (1): 111–20.
doi:10.1080/09613218.2014.950452.
Coley, David, Tristan Kershaw, and Matt Eames. 2012. “A Comparison of Structural and
Behavioural Adaptations to Future Proofing Buildings against Higher Temperatures.”
Building and Environment 55. Elsevier Ltd: 159–66. doi:10.1016/j.buildenv.2011.12.011.
Compendium voor de Leefomgeving. 2017. “Energielabels van Woningen, 2007 - 2016.”
http://www.clo.nl/indicatoren/nl0556-energielabels-woningen.
Costa, Sergio. 2017. “Planning for the Future: Nearly Zero Energy Building Retrofits for.”
MSc Thesis. Eindhoven University of Technology, The Netherlands.
http://www.janhensen.nl/team/past/master/Costa_2017.pdf.
Cowling, Peter, Graham Kendall, and Limin Han. 2002. “An Investigation of a
Hyperheuristic Genetic Algorithm Applied to a Trainer Scheduling Problem.”
Proceedings of the 2002 Congress on Evolutionary Computation, CEC 2002 2: 1185–90.
doi:10.1109/CEC.2002.1004411.
Crawley, Drury B, Jon W Hand, Michael Kummert, and Brent T Griffith. 2006. “Contrasting
the Capabilities of Building Energy Performance Simulation Programs.” Building and
Environment 43 (4): 661–73. doi:10.1016/j.buildenv.2006.10.027.
de Wilde, P. 2014. “The Gap between Predicted and Measured Energy Performance of
Buildings: A Framework for Investigation.” Automation in Construction 41: 40–49.
295
References
doi:10.1016/j.autcon.2014.02.009.
de Wilde, P, and David Coley. 2012. “The Implications of a Changing Climate for Buildings.”
Building and Environment 55: 1–7. doi:10.1016/j.buildenv.2012.03.014.
de Wilde, P, Y Rafiq, and M Beck. 2008. “Uncertainties in Predicting the Impact of Climate
Change on Thermal Performance of Domestic Buildings in the UK.” Building Services
Engineering Research & Technology 29 (1): 7–26. doi:Doi 10.1177/0143624407087261.
de Wilde, P, and Wei Tian. 2009. “Identification of Key Factors for Uncertainty in the
Prediction of the Thermal Performance of an Office Building under Climate Change.”
Building Simulation 2 (3): 157–74. doi:10.1007/s12273-009-9116-1.
De Wit, Sten, and Godfried Augenbroe. 2002. “Analysis of Uncertainty in Building Design
Evaluations and Its Implications.” Energy and Buildings 34 (9): 951–58.
doi:10.1016/S0378-7788(02)00070-1.
Deb, Kalyanmoy, Sunith Bandaru, David Greiner, Antonio Gaspar-Cunha, and Cem Celal
Tutum. 2014. “An Integrated Approach to Automated Innovization for Discovering
Useful Design Principles: Case Studies from Engineering.” Applied Soft Computing
Journal 15. Elsevier B.V.: 42–56. doi:10.1016/j.asoc.2013.10.011.
Deb, Kalyanmoy, and Himanshu Jain. 2013. “An Evolutionary Many-Objective Optimization
Algorithm Using Reference-Point Based Non-Dominated Sorting Approach, Part II:
Handling Constraints and Extending to an Adaptive Approach.” Ieeexplore.Ieee.Org 18
(c): 1–1. doi:10.1109/TEVC.2013.2281534.
Deb, Kalyanmoy, Amrit Pratap, Sameer Agarwal, and T. Meyarivan. 2002. “A Fast and Elitist
Multiobjective Genetic Algorithm: NSGA-II.” IEEE Transactions on Evolutionary
Computation 6 (2): 182–97. doi:10.1109/4235.996017.
Dey, Balaram, Bipradas Bairagi, Bijan Sarkar, and Subir Kumar Sanyal. 2016. “Multi
Objective Performance Analysis: A Novel Multi-Criteria Decision Making Approach
for a Supply Chain.” Computers & Industrial Engineering 94: 105–24.
doi:10.1016/j.cie.2016.01.019.
DHPA. 2015. “Heat Pumps in Domestic Housing and Demand Management Summary
Housing and Demand.”
dubbelglasweetjes. 2016. “Window Prices.” dubbelglasweetjes.nl.
E.ON. 2016. “EON Premium Solar System.”
http://www.eon.nl/thuis/nl/zonnepanelen/onzezonneproducten/premium.html.
E.ON. 2017a. “Costs and Savings of Solar Panels.”
https://www.eon.nl/energieproducten/zonnepanelen/kosten-en-besparingen/.
E.ON. 2017b. “Net-Metering Scheme.”
https://www.eon.nl/energieproducten/zonnepanelen/kosten-en-
besparingen/terugleveren-en-salderen/.
Eck, Randy van. 2016. “Investigating the Influence of Occupants on the Indoor Climate and
Heating Energy for the ‘ Morgen Woningen .’” Eindhoven University of Technology,
The Netherlands.
EFFRA. 2013. Factories of the Future - Multi-Annual Roadmap for the Contractual PPP under
Horizon 2020. European Factories of the Future Research Association (EFFRA).
doi:10.2777/29815.
Ehrgott, Matthias, Jonas Ide, and Anita Schöbel. 2014. “Minmax Robustness for Multi-
296
Objective Optimization Problems.” European Journal of Operational Research 239 (1):
17–31. doi:10.1016/j.ejor.2014.03.013.
Eisenhower, B., Z. O. Neill, S. Narayanan, V. a. Fonoberov, and I. Mezic. 2011. “A
Comparative Study on Uncertainty Propagation in High Performance Building
Design.” Building Simulation 2011, 14–16.
Eisenhower, B, Zheng O Neill, Vladimir a Fonoberov, and Igor Mezi. 2011. “Uncertainty and
Sensitivity Decomposition of Building Energy Models.” Journal of Building Performance
Simulation 5 (3): 171–84. doi:10.1080/19401493.2010.549964.
EPBD. 2010. “Directive 2010/31/EU of the European Parliament and of the Council of 19
May 2010 on the Energy Performance of Buildings (Recast).” Official Journal of the
European Union, 13–35. doi:doi:10.3000/17252555.L_2010.153.eng.
Erhorn, Hans, and Heike Erhorn-Kluttig. 2011. “Terms and Definitions for High
Performance Buildings.” Concerted Action: Energy Performance of Buildings. www.epbd-
ca.org.
ESTECO. 2015. “modeFRONTIERTM 4 User Manual.” http://www.esteco.com/modefrontier.
European Commision. 2002. “Directive 2002/91/EC of the European Parliament and of the
Council of 16 December 2002 on the Energy Performance of Buildings.” Official
Journal Of The European Union, 65–71. doi:10.1039/ap9842100196.
European Commission. 2009. Energy-Efficient Buildings PPP: Multi-Annual Road Map and
Lonegr Term Strategy.
European Commission. 2011. “A Roadmap for Moving to a Competitive Low Carbon
Economy in 2050.” http://eur-lex.europa.eu/legal-
content/EN/TXT/PDF/?uri=CELEX:52011DC0112&from=EN.
Evins, Ralph. 2013. “A Review of Computational Optimisation Methods Applied to
Sustainable Building Design.” Renewable and Sustainable Energy Reviews 22: 230–45.
doi:10.1016/j.rser.2013.02.004.
Evins, Ralph. 2015. “Multi-Level Optimization of Building Design, Energy System Sizing and
Operation.” Energy 90. Elsevier Ltd: 1775–89. doi:10.1016/j.energy.2015.07.007.
Fadeyi, Moshood Olawale. 2014. “Initial Study on the Impact of Thermal History on
Building Occupants’ Thermal Assessments in Actual Air-Conditioned Office
Buildings.” Building and Environment 80: 36–47. doi:10.1016/j.buildenv.2014.05.018.
Fawcett, William, Martin Hughes, Hannes Krieg, Stefan Albrecht, and Anders Vennström.
2012. “Flexible Strategies for Long-Term Sustainability under Uncertainty.” Building
Research & Information 40 (5): 545–57. doi:10.1080/09613218.2012.702565.
Filippidou, Faidra, Nico Nieboer, and Henk Visscher. 2016. “Energy Efficiency Measures
Implemented in the Dutch Non-Profit Housing Sector.” Energy & Buildings 132: 107–
16. doi:10.1016/j.enbuild.2016.05.095.
Fleiter, Tobias, Jan Steinbach, Mario Ragwitz, Marlene Arens, Ali Aydemir, Rainer Elsland,
Clemens Frassine, et al. 2016. “Mapping and Analyses of the Current and Future
(2020-2030) Heating/cooling Fuel Deployment (Fossil/renewables). Work Package 2:
Assessment of the Technologies for the Year 2012.”
Francis, Royce, and Behailu Bekera. 2014. “A Metric and Frameworks for Resilience
Analysis of Engineered and Infrastructure Systems.” Reliability Engineering and System
Safety 121: 90–103. doi:10.1016/j.ress.2013.07.004.
297
References
Gaetani, Isabella, P Hoes, and Jan L.M. Hensen. 2016a. “Occupant Behavior in Building
Energy Simulation: Towards a Fit-for-Purpose Modeling Strategy.” Energy and
Buildings 121. Elsevier B.V.: 188–204. doi:10.1016/j.enbuild.2016.03.038.
Gaetani, Isabella, P Hoes, and Jan L.M. Hensen. 2016b. “On the Sensitivity to Different
Aspects of Occupant Behaviour for Selecting the Appropriate Modelling Complexity in
Building Performance Predictions Modelling Complexity in Building Performance
Predictions.” Journal of Building Performance Simulation, 1–11.
doi:10.1080/19401493.2016.1260159.
Gang, Wenjie, Shengwei Wang, Godfried Augenbroe, and Fu Xiao. 2016. “Robust Optimal
Design of District Cooling Systems and the Impacts of Uncertainty and Reliability.”
Energy and Buildings 122: 11–22. doi:10.1016/j.enbuild.2016.04.012.
Gang, Wenjie, Shengwei Wang, Fu Xiao, and Dian-ce Gao. 2015. “Robust Optimal Design of
Building Cooling Systems Considering Cooling Load Uncertainty and Equipment
Reliability.” Applied Energy 159. Elsevier Ltd: 265–75.
doi:10.1016/j.apenergy.2015.08.070.
Gang, Wenjie, Shengwei Wang, Chengchu Yan, and Fu Xiao. 2015. “Robust Optimal Design
of Building Cooling Systems Concerning Uncertainties Using Mini-Max Regret
Theory.” Science and Technology for the Built Environment 21 (6): 789–99.
doi:10.1080/23744731.2015.1056657.
Gaspars-Wieloch, Helena. 2014. “Modifications of the Hurwiczs Decision Rule.” Central
European Journal of Operations Research 22 (4): 779–94. doi:10.1007/s10100-013-0302-
y.
Geraedts, Rob P. 2008. “Design for Change: Flexibility Key Performance Indicators.” In
I3CON Conference, Loughborough, May 2008, 11–21.
Gosling, Jonathan, Mohamed Naim, Paola Sassi, Laura Iosif, and Robert Lark. 2007.
“Flexible Buildings for an Adaptable and Sustainable Future.” Arcom 2011, 115–24.
Gosling, Jonathan, Paola Sassi, Mohamed Naim, and Robert Lark. 2013. “Adaptable
Buildings: A Systems Approach.” Sustainable Cities and Society 7: 44–51.
doi:10.1016/j.scs.2012.11.002.
Gram-Hanssen, Kirsten. 2013. “Efficient Technologies or User Behaviour, Which Is the
More Important When Reducing Households’ Energy Consumption?” Energy
Efficiency 6 (3): 447–57. doi:10.1007/s12053-012-9184-4.
Gram-hanssen, Kirsten, and Susse Georg. 2017. “Energy Performance Gaps: Promises ,
People , Practices.” Building Research & Information, 1–9.
doi:10.1080/09613218.2017.1356127.
Guerra-Santin, O, and Laure Itard. 2010. “Occupants’ Behaviour: Determinants and Effects
on Residential Heating Consumption.” Building Research and Information 38 (3): 318–
38. doi:10.1080/09613211003661074.
Guerra-Santin, O, Laure C.M. Itard, Henk Visscher, Olivia Guerra-Santin, Laure C.M. Itard,
D. Majcen, Laure C.M. Itard, et al. 2010. “Occupants’ Behaviour: Determinants and
Effects on Residential Heating Consumption.” Building Research and Information 41
(11). Elsevier: 318–38. doi:10.1080/09613211003661074.
Guerra-Santin, O, and S. Silvester. 2016. “Development of Dutch Occupancy and Heating
Profiles for Building Simulation.” Building Research & Information, 1–18.
doi:10.1080/09613218.2016.1160563.
298
Gunawan, S., and S. Azarm. 2005. “Multi-Objective Robust Optimization Using a Sensitivity
Region Concept.” Structural and Multidisciplinary Optimization 29 (1): 50–60.
doi:10.1007/s00158-004-0450-8.
Hamdy, M, Salvatore Carlucci, Pieter Jan Hoes, and Jan L.M. Hensen. 2017. “The Impact of
Climate Change on the Overheating Risk in dwellings—A Dutch Case Study.”
Building and Environment 122: 307–23. doi:10.1016/j.buildenv.2017.06.031.
Hamdy, M, Ala Hasan, and Kai Siren. 2013. “A Multi-Stage Optimization Method for Cost-
Optimal and Nearly-Zero-Energy Building Solutions in Line with the EPBD-Recast
2010.” Energy and Buildings 56: 189–203. doi:10.1016/j.enbuild.2012.08.023.
Hamdy, M, and J.L.M; Hensen. 2015. “Ranking of Dwelling Types in Terms of Overheating
Risk and Sensitivity to Climate Change.” Building Simulation’15 15: 8.
Hamdy, M, Anh-Tuan Nguyen, and Jan L.M. Hensen. 2016. “A Performance Comparison of
Multi-Objective Optimization Algorithms for Solving Nearly-Zero-Energy-Building
Design Problems.” Energy and Buildings 121: 57–71. doi:10.1016/j.enbuild.2016.03.035.
Hassler, Uta, and Niklaus Kohler. 2014. “Resilience in the Built Environment.” Building
Helton, J. C., J. D. Johnson, C. J. Sallaberry, and C. B. Storlie. 2006. “Survey of Sampling-
Based Methods for Uncertainty and Sensitivity Analysis.” Reliability Engineering and
System Safety 91 (10–11): 1175–1209. doi:10.1016/j.ress.2005.11.017.
Heo, Y., R. Choudhary, and G. A. Augenbroe. 2012. “Calibration of Building Energy Models
for Retrofit Analysis under Uncertainty.” Energy and Buildings 47: 550–60.
doi:10.1016/j.enbuild.2011.12.029.
Hermelink, A, S Schimschar, T Boermans, L Pagliano, P Zangheri, R Armani, K Voss, and E
Musall. 2013. “Towards Nearly Zero- Energy Buildings Definition of Common
Principles under the EPBD.” Ecofys by Order of European Commission.
https://ec.europa.eu/energy/sites/ener/files/documents/nzeb_full_report.pdf.
Heyden, Y Vander, and D L Massart. 1996. “Review of the Use of Robustness and
Ruggedness in Analytical Chemistry.” Data Handling in Science and Technology 19: 79–
147.
Hoes, P. 2014. “Computational Performance Prediction of the Potential of Hybrid Adaptable
Thermal Storage Concepts for Lightweight Low-Energy Houses.” Eindhoven
Univesrity of Technology. doi:10.13140/2.1.2329.8562.
Hoes, P., J. L M Hensen, M. G L C Loomans, B. de Vries, and D. Bourgeois. 2009. “User
Behavior in Whole Building Simulation.” Energy and Buildings 41 (3): 295–302.
doi:10.1016/j.enbuild.2008.09.008.
Hoes, P, M Trcka, J L M Hensen, and B Hoekstra Bonnema. 2011. “Optimizing Building
Designs Using a Robustness Indicator with Respect to User Behavior.” Proceedings of
the 12th Conference of the International Building Performance Simulation Association, 14–
16.
Hollnagel, Erik. 2014. “Resilience Engineering and the Built Environment.” Building
Holmes, Michael J., and Jacob N. Hacker. 2007. “Climate Change, Thermal Comfort and
Energy: Meeting the Design Challenges of the 21st Century.” Energy and Buildings 39
(7): 802–14. doi:10.1016/j.enbuild.2007.02.009.
299
References
Hong, Tianzhen, Sarah C. Taylor-Lange, Simona D’Oca, Da Yan, and Stefano P. Corgnati.
2016. “Advances in Research and Applications of Energy-Related Occupant Behavior
in Buildings.” Energy and Buildings 116: 694–702. doi:10.1016/j.enbuild.2015.11.052.
Hopfe, C.J. 2009. “Uncertainty and Sensitivity Analysis in Building Performance
Simulation for Decision Support and Design Optimization.” PhD Thesis. Eindhoven
University of Technology. http://www.bwk.tue.nl/bps/hensen/team/past/Hopfe.pdf.
Hopfe, C.J., and Jan L M Hensen. 2011. “Uncertainty Analysis in Building Performance
Simulation for Design Support.” Energy and Buildings 43 (10): 2798–2805.
doi:10.1016/j.enbuild.2011.06.034.
Hopfe, C J., Godfried L M Augenbroe, and Jan L M Hensen. 2013. “Multi-Criteria Decision
Making under Uncertainty in Building Performance Assessment.” Building and
Environment 69: 81–90. doi:10.1016/j.buildenv.2013.07.019.
Hu, Huafen, and Godfried Augenbroe. 2012. “A Stochastic Model Based Energy
Management System for off-Grid Solar Houses.” Building and Environment 50.
Elsevier Ltd: 90–103. doi:10.1016/j.buildenv.2011.10.011.
Huang, Beiqing, and Xiaoping Du. 2007. “Analytical Robustness Assessment for Robust
Design.” Structural and Multidisciplinary Optimization 34 (2): 123–37.
doi:10.1007/s00158-006-0068-0.
Hurwicz, L. 1952. “A Criterion for Decision Making under Uncertainty.” Technical Report
355, Cowles Commission.
IEA. 2013. “Annex 55 Reliability of Energy Efficient Building Retrofitting - Probability
Assessment of Performance & Cost (RAP-RETRO) - Probabilistic Tools.” Energy
Conservation in Buildings and Community Systems.
IEEE. 1994. IEEE Standard Glossary of Computer Hardware Terminology. IEEE Standards
Association. doi:10.1109/IEEESTD.1995.79522.
ISSO. 2011a. “HANDLEIDING ENERGIEPRESTATIE ADVIES WONINGEN.”
http://www.isso.nl/kennis-voor-u/docent/winkel/detail/Shop/isso-publicatie-823-
handleiding-energieprestatie-advies-woningen-%0A2011.
ISSO. 2011b. ISSO Publicatie 32, Herziene Versie.
Jafari, Amirhosein, and Vanessa Valentin. 2017. “An Optimization Framework for Building
Energy Retrofits Decision-Making.” Building and Environment 115: 118–29.
doi:10.1016/j.buildenv.2017.01.020.
Janssen, Hans. 2013. “Monte-Carlo Based Uncertainty Analysis: Sampling Efficiency and
Sampling Convergence.” Reliability Engineering and System Safety 109: 123–32.
doi:10.1016/j.ress.2012.08.003.
Jenkins, David P., Sandhya Patidar, Phil Banfill, and Gavin Gibson. 2014. “Developing a
Probabilistic Tool for Assessing the Risk of Overheating in Buildings for Future
Climates.” Renewable Energy 61: 7–11. doi:10.1016/j.renene.2012.04.035.
Karjalainen, Sami. 2016. “Should We Design Buildings That Are Less Sensitive to Occupant
Behaviour? A Simulation Study of Effects of Behaviour and Design on Office Energy
Consumption.” Energy Efficiency, no. 1992. Energy Efficiency: 1–14.
doi:10.1007/s12053-015-9422-7.
KEMA. 2016. “Nationaal Actieplan Zonnestroom 2016.”
Kibert, Charles J., and Maryam Mirhadi Fard. 2012. “Differentiating among Low-Energy,
300
Low-Carbon and Net-Zero-Energy Building Strategies for Policy Formulation.”
Building Research and Information 40 (5): 625–37. doi:10.1080/09613218.2012.703489.
Kim, Sean Hay. 2013. “An Evaluation of Robust Controls for Passive Building Thermal Mass
and Mechanical Thermal Energy Storage under Uncertainty.” Applied Energy 111: 602–
23. doi:10.1016/j.apenergy.2013.05.030.
Kingspan insulation. 2016. “Prijs- En Assortimentslijst Kool Therm ® April 2016
Inhoudsopgave.”
Kitano, Hiroaki. 2004. “Biological Robustness.” Nature Review Genetics 5: 826–37.
doi:doi:10.1038/nrg1471.
Kiureghian, Armen Der, and Ove Ditlevsen. 2009. “Aleatory or Epistemic? Does It Matter?”
Structural Safety 31 (2): 105–12. doi:10.1016/j.strusafe.2008.06.020.
KNMI. 2014. “KNMI’14 Climate Scenarios for the Netherlands; A Guide for Professionals in
Climate Adaptation,” 34. doi:http://www.climatescenarios.nl/.
Kolokotsa, D., D. Rovas, E. Kosmatopoulos, and K. Kalaitzakis. 2011. “A Roadmap towards
Intelligent Net Zero- and Positive-Energy Buildings.” Solar Energy 85 (12). Elsevier Ltd:
3067–84. doi:10.1016/j.solener.2010.09.001.
Koninkrijksrelaties Ministerie van Binnenlandse Zaken. 2016. “Cijfers over Wonen En
Bouwen 2016.”
Kotireddy, Rajesh, P Hoes, and Jan L. M. Hensen. 2018. “A Methodology for Performance
Robustness Assessment of Low-Energy Buildings Using Scenario Analysis.” Applied
Energy 212: 428–42. doi:https://doi.org/10.1016/j.apenergy.2017.12.066.
Kotireddy, Rajesh, P Hoes, and Jan L.M. Hensen. 2015. “Optimal Balance between Energy
Demand and Onsite Energy Generation for Robust Net Zero Energy Buildings
Considering Future Scenarios.” In In the Proceedings of 14th IBPSA Conference, 1970–
77. doi:http://www.ibpsa.org/proceedings/BS2015/p2376.pdf.
Kotireddy, Rajesh, P Hoes, and Jan L.M. Hensen. 2017a. “Robust Net-Zero Energy Buildings
– A Methodology for Designers to Evaluate Robustness.” REHVA Journal 54 (4): 9–19.
http://www.rehva.eu/fileadmin/REHVA_Journal/REHVA_Journal_2017/RJ4/p.9/09-
19_RJ1704_WEB.pdf.
Kotireddy, Rajesh, P Hoes, and Jan L.M. Hensen. 2017b. “Simulation-Based Comparison of
Robustness Assessment Methods to Identify Robust Low Energy Building Designs.”
In Proceedings of 15th IBPSA Conference, SanFrancisco, CA, USA, 892–901.
doi:10.26868/25222708.2017.240.
Kumar, Abhishek, Bikash Sah, Arvind R. Singh, Yan Deng, Xiangning He, Praveen Kumar,
and R. C. Bansal. 2017. “A Review of Multi Criteria Decision Making (MCDM)
towards Sustainable Renewable Energy Development.” Renewable and Sustainable
Kumari, A. Charan, K. Srinivas, and M. P. Gupta. 2013. “Software Module Clustering Using
a Hyper-Heuristic Based Multi-Objective Genetic Algorithm.” Proceedings of the 3rd
IEEE International Advance Computing Conference, IACC 2013, 813–18.
doi:10.1109/IAdCC.2013.6514331.
Kurnitski, Jarek, Francis Allard, Derrik Braham, Guillaume Goeders, Per Heiselberg,
Lennart Jagemar, Risto Kosonen, et al. 2012. “How to Define Nearly Net Zero Energy
Buildings nZEB.” REHVA Journal, 6–12.
301
References
Larhlimi, Abdelhalim, Sylvain Blachon, Joachim Selbig, and Zoran Nikoloski. 2011.
“Robustness of Metabolic Networks: A Review of Existing Definitions.” BioSystems 106
(1): 1–8. doi:10.1016/j.biosystems.2011.06.002.
Lee, Bruno, and Jan L.M. Hensen. 2015. “Developing a Risk Indicator to Quantify Robust
Building Design.” Energy Procedia 78: 1895–1900. doi:10.1016/j.egypro.2015.11.357.
Leichenko, Robin. 2011. “Climate Change and Urban Resilience.” Current Opinion in
Environmental Sustainability 3 (3): 164–68. doi:10.1016/j.cosust.2010.12.014.
Lesne, Annick. 2008. “Robustness: Confronting Lessons from Physics and Biology.”
Biological Reviews 83 (4): 509–32. doi:10.1111/j.1469-185X.2008.00052.x.
Leyten, Joe L., and Stanley R. Kurvers. 2006. “Robustness of Buildings and HVAC Systems
as a Hypothetical Construct Explaining Differences in Building Related Health and
Comfort Symptoms and Complaint Rates.” Energy and Buildings 38 (6): 701–7.
doi:10.1016/j.enbuild.2005.11.001.
Li, Guijie, Zhenzhou Lu, Luyi Li, and Bo Ren. 2012. “Aleatory and Epistemic Uncertainties
Analysis Based on Non-Probabilistic Reliability and Its Kriging Solution.” Applied
Mathematical Modelling 40: 5703–16. doi:10.1016/j.apm.2016.01.017.
Loonen, R.C.G.M., Fabio Favoino, Jan L.M. Hensen, and Mauro Overend. 2017. “Review of
Current Status, Requirements and Opportunities for Building Performance
Simulation of Adaptive Facades.” Journal of Building Performance Simulation 10 (2):
205–23. doi:10.1080/19401493.2016.1152303.
Loonen, R.C.G.M., M. Trcka, D Cóstola, and J L M Hensen. 2013. “Climate Adaptive
Building Shells : State-of-the-Art and Future Challenges.” Renewable and Sustainable
Lu, Yuehong, Shengwei Wang, Chengchu Yan, and Zhijia Huang. 2017. “Robust Optimal
Design of Renewable Energy System in Nearly/net Zero Energy Buildings under
Uncertainties.” Applied Energy 187. Elsevier Ltd: 62–71.
doi:10.1016/j.apenergy.2016.11.042.
Lu, Yuehong, Shengwei Wang, Chengchu Yan, and Kui Shan. 2015. “Impacts of Renewable
Energy System Design Inputs on the Performance Robustness of Net Zero Energy
Buildings.” Energy 93. Elsevier Ltd: 1595–1606. doi:10.1016/j.energy.2015.10.034.
Lusby, Richard M., Jesper Larsen, and Simon Bull. 2017. “A Survey on Robustness in
Railway Planning.” European Journal of Operational Research 0: 1–15.
doi:10.1016/j.ejor.2017.07.044.
Lysen, E. 1996. “The Trias Energetica: Solar Energy Strategies for Developing Countries.” In
In the Proceedings of Eurosun Conference 1996, 16–19.
Maas, W., Zijlema, P. 2017. “Greenhouse Gas Emissions in the Netherlands.” National
Inventory Report,. doi:10.1016/B978-0-12-409548-9.05178-2.
Macdonald, Iain A. 2009. “Comparison of Sampling Techniques on the Performance of
Monte-Carlo Based Sensitivity Analysis.” BS2009: 11th Conference of International
Building Performance Simulation Association, Glasgow, Scotland, July 27-30, no. 3: 992–
99.
Macdonald, and Paul Strachan. 2001. “Practical Application of Uncertainty Analysis.” Energy
and Buildings 33 (3): 219–27. doi:10.1016/S0378-7788(00)00085-2.
Machairas, Vasileios, Aris Tsangrassoulis, and Kleo Axarli. 2014. “Algorithms for
302
Optimization of Building Design: A Review.” Renewable and Sustainable Energy Reviews
31 (1364): 101–12. doi:10.1016/j.rser.2013.11.036.
Maier, T., M. Krzaczek, and J. Tejchman. 2009. “Comparison of Physical Performances of
the Ventilation Systems in Low-Energy Residential Houses.” Energy and Buildings 41
(3): 337–53. doi:10.1016/j.enbuild.2008.10.007.
Majcen, D., L. C.M. Itard, and H. Visscher. 2013. “Theoretical vs. Actual Energy
Consumption of Labelled Dwellings in the Netherlands: Discrepancies and Policy
Implications.” Energy Policy 54. Elsevier: 125–36. doi:10.1016/j.enpol.2012.11.008.
Manrique, Benjamin, Rajesh Kotireddy, Sunliang Cao, Ala Hasan, P Hoes, Jan L.M.
Hensen, and Sire. 2018. “Lifecycle Cost and CO2 Emissions of Residential Heat and
Electricity : A Study of Prosumers in Finland and the Netherlands.” Energy Conversion
and Management 160: 495–508. doi:https://doi.org/10.1016/j.enconman.2018.01.069.
Marszal, A. J., P. Heiselberg, J. S. Bourrelle, E. Musall, K. Voss, I. Sartori, and A. Napolitano.
2011. “Zero Energy Building - A Review of Definitions and Calculation
Methodologies.” Energy and Buildings 43 (4). Elsevier B.V.: 971–79.
doi:10.1016/j.enbuild.2010.12.022.
Martinaitis, Vytautas, Edmundas Kazimieras Zavadskas, Violeta Motuzienė, and Tatjana
Vilutienė. 2015. “Importance of Occupancy Information When Simulating Energy
Demand of Energy Efficient House: A Case Study.” Energy and Buildings 101: 64–75.
doi:10.1016/j.enbuild.2015.04.031.
MathWorks. 2016. “MATLAB (Matrix Laboratory).” www.mathworks.nl.
Mavrogianni, A., M. Davies, J. Taylor, Z. Chalabi, P. Biddulph, E. Oikonomou, P. Das, and B.
Jones. 2014. “The Impact of Occupancy Patterns, Occupant-Controlled Ventilation and
Shading on Indoor Overheating Risk in Domestic Environments.” Building and
Mavrotas, George, José Rui Figueira, and Eleftherios Siskos. 2015. “Robustness Analysis
Methodology for Multi-Objective Combinatorial Optimization Problems and
Application to Project Selection.” Omega 52: 142–55. doi:10.1016/j.omega.2014.11.005.
McLeod, Robert S., C.J. Hopfe, and Alan Kwan. 2013. “An Investigation into Future
Performance and Overheating Risks in Passivhaus Dwellings.” Building and
McManus, Hugh, and Daniel Hastings. 2006. “A Framework for Understanding
Uncertainty and Its Mitigation and Exploitation in Complex Systems.” IEEE
Engineering Management Review 34 (3): 81–94. doi:10.1109/EMR.2006.261384.
Medineckiene, M., E. K. Zavadskas, F. Björk, and Z. Turskis. 2015. “Multi-Criteria Decision-
Making System for Sustainable Building Assessment/certification.” Archives of Civil
and Mechanical Engineering 15 (1). Politechnika Wrocławska: 11–18.
doi:10.1016/j.acme.2014.09.001.
Meier, Alan; Olofsson, Thomas; Lamberts, Roberto. 2002. “What Is an Energy-Efficient
Building ?” In IX Encontro Naciona de Tecnologia Do Ambiente Construído, 3–12.
Meijer, Frits, Laure Itard, and Minna Sunikka-Blank. 2009. “Comparing European
Residential Building Stocks: Performance, Renovation and Policy Opportunities.”
Building Research and Information 37 (5–6): 533–51. doi:10.1080/09613210903189376.
Mekdeci, Brian, Adam M. Ross, Donna H. Rhodes, and Daniel E. Hastings. 2015. “Pliability
and Viable Systems: Maintaining Value under Changing Conditions.” IEEE Systems
303
References
Journal 9 (4): 1173–84. doi:10.1109/JSYST.2014.2314316.

Mela, Kristo, Teemu Tiainen, and Markku Heinisuo. 2012. “Comparative Study of Multiple
Criteria Decision Making Methods for Building Design.” Advanced Engineering
Informatics 26 (4): 716–26. doi:10.1016/j.aei.2012.03.001.
Middendorp, Gerton van. 2018. “Future Proofing Residential Apartment Buildings:
Consequences of a Guaranteed Net-Zero Energy Performance for a Mid-Rise
Residential Apartment Building.” Eindhoven University of Technology, The
Netherlands. http://www.janhensen.nl/team/master/Middendorp.pdf.
Ministerie van VROM. 2009. “Energiegedrag in De Woning.”
Mlecnik, Erwin. 2012. “Defining Nearly Zero-Energy Housing in Belgium and the
Netherlands.” Energy Efficiency 5 (3): 411–31. doi:10.1007/s12053-011-9138-2.
Mlecnik, T. Schütze, S. J T Jansen, G. De Vries, H. J. Visscher, and A. Van Hal. 2012. “End-
User Experiences in Nearly Zero-Energy Houses.” Energy and Buildings 49: 471–78.
doi:10.1016/j.enbuild.2012.02.045.
Mondal, S.C., P.K. Ray, and J. Maiti. 2014. “Modelling Robustness for Manufacturing
Processes: A Critical Review.” International Journal of Production Research 52 (2): 521–
38. doi:10.1080/00207543.2013.837588.
Morgan, Craig, and Richard de Dear. 2003. “Weather, Clothing and Thermal Adaptation to
Indoor Climate.” Climate Research 24 (3): 267–84. doi:10.3354/cr024267.
“MorgenWonen.” 2016. doi:http://www.morgenwonen.nl/.
Moss, Richard H, Jae a Edmonds, Kathy a Hibbard, Martin R Manning, Steven K Rose,
Detlef P van Vuuren, Timothy R Carter, et al. 2010. “The next Generation of Scenarios
for Climate Change Research and Assessment.” Nature 463 (7282). Nature Publishing
Group: 747–56. doi:10.1038/nature08823.
Mulliner, Emma, Naglis Malys, and Vida Maliene. 2016. “Comparative Analysis of MCDM
Methods for the Assessment of Sustainable Housing Affordability.” Omega 59: 146–
56. doi:10.1016/j.omega.2015.05.013.
Mullins, Joshua, You Ling, Sankaran Mahadevan, Lin Sun, and Alejandro Strachan. 2016.
“Separation of Aleatory and Epistemic Uncertainty in Probabilistic Model Validation.”
Reliability Engineering and System Safety 147: 49–59. doi:10.1016/j.ress.2015.10.003.
Mulville, Mark, and Spyridon Stravoravdis. 2016. “The Impact of Regulations on
Overheating Risk in Dwellings.” Building Research & Information 3218: 1–15.
doi:10.1080/09613218.2016.1153355.
Musall, E, and K Voss. 2013. “Understanding Net ZEB - Overview of Existing Definitions in
EU / Europe.”
NEN. 2008. “Netherlands Norm, NEN 5060- Hygrothermal Performance of Buildings -
Climatic Reference Data.”
NEN7120. 2011. “Energy Performance of Buildings- Determination Method.” Nederlandse
Norm NEN 7120: Energy Performance of Buildings- Determination Method. Vol. 7120.
NEN7120+C2. 2012. “Energy Performance of Buildings-Determination Method.” Vol. 6.
Nguyen, Anh-Tuan, Sigrid Reiter, and Philippe Rigo. 2014. “A Review on Simulation-Based
Optimization Methods Applied to Building Performance Analysis.” Applied Energy 113:
1043–58. doi:10.1016/j.apenergy.2013.08.061.
304
Nicol, Lee Ann, and Peter Knoepfel. 2014. “Resilient Housing: A New Resource-Oriented
Approach.” Building Research & Information 42 (2). Taylor & Francis: 229–39.
doi:10.1080/09613218.2014.862162.
“Nieuw Utrechts Peil.” 2017. http://nieuwutrechtspeil.nl/.
Nik, Vahid M., and Angela Sasic Kalagasidis. 2013. “Impact Study of the Climate Change on
the Energy Performance of the Building Stock in Stockholm Considering Four
Climate Uncertainties.” Building and Environment 60. Elsevier Ltd: 291–304.
doi:10.1016/j.buildenv.2012.11.005.
Nik, Vahid M., Erika Mata, and Angela Sasic Kalagasidis. 2015. “A Statistical Method for
Assessing Retrofitting Measures of Buildings and Ranking Their Robustness against
Climate Change.” Energy and Buildings 88. Elsevier B.V.: 262–75.
doi:10.1016/j.enbuild.2014.11.015.
Nuon. 2017. “Net-Metering Scheme.”
https://www.nuon.nl/producten/zonnepanelen/salderen/.
O’Neill, Zheng, and B Eisenhower. 2013a. “Leveraging the Analysis of Parametric
Uncertainty for Building Energy Model Calibration.” Building Simulation, 1–13.
doi:10.1007/s12273-013-0125-8.
O’Neill, Zheng, and Bryan Eisenhower. 2013b. “Leveraging the Analysis of Parametric
Uncertainty for Building Energy Model Calibration.” Building Simulation, 1–13.
doi:10.1007/s12273-013-0125-8.
Olewnik, Andrew, Trevor Brauen, Scott Ferguson, and Kemper Lewis. 2004. “A Framework
for Flexible Systems and Its Implementation in Multiattribute Decision Making.”
Journal of Mechanical Design 126 (3): 412–19. doi:10.1115/1.1701874.
Østergård, Torben, Rasmus L. Jensen, and Steffen E. Maagaard. 2017. “Early Building
Design: Informed Decision-Making by Exploring Multidimensional Design Space
Using Sensitivity Analysis.” Energy and Buildings 142. Elsevier B.V.: 8–22.
doi:10.1016/j.enbuild.2017.02.059.
Oudenand, E.d., and R. Gal. 2014. Vision and Roadmap Eindhoven Energy-Neutral 2045.
Papachristos, George. 2015. “Household Electricity Consumption and CO2 Emissions in the
Netherlands: A Model-Based Analysis.” Energy and Buildings 86: 403–14.
doi:10.1016/j.enbuild.2014.09.077.
Parys, Wout, Hilde Breesch, Hugo Hens, and Dirk Saelens. 2012. “Feasibility Assessment of
Passive Cooling for Office Buildings in a Temperate Climate through Uncertainty
Analysis.” Building and Environment 56: 95–107. doi:10.1016/j.buildenv.2012.02.018.
Passipedia. 2017. “Passivehouse.” https://passipedia.org/basics/the_passive_house_-
_definition.
Pavzek, Karmen, and Crtomir Rozman. 2009. “Decision Making Under Conditions of
Uncertainty in Agriculture: A Case Study of Oil Crops.” Poljoprivreda 15 (1): 45–50.
http://hrcak.srce.hr/index.php?show=clanak&id_clanak_jezik=61873.
Peeters, Leen, Richard de Dear, Jan Hensen, and William D’haeseleer. 2009. “Thermal
Comfort in Residential Buildings: Comfort Values and Scales for Building Energy
Simulation.” Applied Energy 86 (5): 772–80. doi:10.1016/j.apenergy.2008.07.011.
PEP. 2008. “Promotion of European Passive Houses.”
Perera, A. T D, R. A. Attalage, K. K C K Perera, and V. P C Dassanayake. 2013. “A Hybrid
305
References
Tool to Combine Multi-Objective Optimization and Multi-Criterion Decision Making

in Designing Standalone Hybrid Energy Systems.” Applied Energy 107. Elsevier Ltd:
412–25. doi:10.1016/j.apenergy.2013.02.049.
PHI. 1990. “Passivehaus Requirements.”
http://passivehouse.com/02_informations/02_passive-house-
requirements/02_passive-house-requirements.htm.
PHPP. 1998. “Passivehaus Planning Package.”
http://www.passiv.de/en/04_phpp/04_phpp.htm.
Plas, J Jesse. 2017. “A Comparative Assessment of Ventilative and Mechanical Cooling for
Residential Zero Energy Buildings Considering the Future Climate.” Eindhoven
University of Technology.
Polasky, Stephen, Stephen R. Carpenter, Carl Folke, and Bonnie Keeler. 2011. “Decision-
Making under Great Uncertainty: Environmental Management in an Era of Global
Change.” Trends in Ecology and Evolution 26 (8): 398–404.
doi:10.1016/j.tree.2011.04.007.
Polatidis, Heracles, Dias A. Haralambopoulos, Giussepe Munda, and Ron Vreeker. 2006.
“Selecting an Appropriate Multi-Criteria Decision Analysis Technique for Renewable
Energy Planning.” Energy Sources, Part B: Economics, Planning, and Policy 1 (2): 181–93.
doi:10.1080/009083190881607.
Preston, Alice. 2015. “Types of Usability Methods - STC Usability SIG.” 06-01-2015; 22-08-
2017.
Puranik, Sanket. 2017. “‘Risk Averse’ Design Solutions for HOMIJ Using Building
Performance Simulations.” Eindhoven University of Technology, The Netherlands.
http://www.janhensen.nl/team/past/Puranik.pdf.
Ramallo-González, A. P., T. S. Blight, and D. A. Coley. 2015. “New Optimisation
Methodology to Uncover Robust Low Energy Designs That Accounts for Occupant
Behaviour or Other Unknowns.” Journal of Building Engineering 2: 59–68.
doi:10.1016/j.jobe.2015.05.001.
Rasouli, Mohammad, Gaoming Ge, Carey J. Simonson, and Robert W. Besant. 2013.
“Uncertainties in Energy and Economic Performance of HVAC Systems and Energy
Recovery Ventilators due to Uncertainties in Building and HVAC Parameters.” Applied
Thermal Engineering 50 (1): 732–42. doi:10.1016/j.applthermaleng.2012.08.021.
Rezaee, Roya, Jason Brown, Godfried Augenbroe, and Jinsol Kim. 2015. “Assessment of
Uncertainty and Confidence in Building Design Exploration.” Artificial Intelligence for
Engineering Design, Analysis and Manufacturing 29 (4): 429–41.
doi:10.1017/S0890060415000426.
Roberz, F., R. C.G.M. Loonen, P. Hoes, and J. L.M. Hensen. 2017. “Ultra-Lightweight
Concrete: Energy and Comfort Performance Evaluation in Relation to Buildings with
Low and High Thermal Mass.” Energy and Buildings 138: 432–42.
doi:10.1016/j.enbuild.2016.12.049.
Rodrigues, Lucelia Taranto, Mark Gillott, and David Tetlow. 2013. “Summer Overheating
Potential in a Low-Energy Steel Frame House in Future Climate Scenarios.”
Sustainable Cities and Society 7: 1–15. doi:10.1016/j.scs.2012.03.004.
Ross, Adam M., Donna H. Rhodes, and Daniel E. Hastings. 2008. “Defining Changeability:
Reconciling Flexibility, Adaptability, Scalability, Modifiability, and Robustness for
Maintaining System Lifecycle Value.” Systems Engineering 11 (3): 246–61.
306
RVO. 2013. “Infoblad Trias Energetica,” 4–5. http://www.rvo.nl/sites/default/files/Infoblad
Trias Energetica en energieneutraal bouwen-juni 2013.pdf.
RVO. 2014. “Rapport - Actualisatie Investeringskosten Maatregelen EPA-Maatwerkadvies
Bestaande Woningbouw 2014.”
RVO. 2015a. “Handreiking BENG.”
https://www.rvo.nl/sites/default/files/2015/10/Handreiking BENG.pdf.
RVO. 2015b. “Hernieuwbare Energie in Bijna Energieneutrale Gebouwen (BENG).”
RVO. 2015c. “Nul Op de Meter,” 24. doi:RVO-059-1501/BR-DUZA.
RVO. 2016a. “Current Dutch Building Standards.”
https://www.rvo.nl/onderwerpen/duurzaam-ondernemen/gebouwen/wetten-en-
regels-gebouwen/nieuwbouw/energieprestatie-epc/referentiewoningen-
epc/tussenwoning.
RVO. 2016b. “RVO BENG Referentiegebouwen.”
doi:http://www.rvo.nl/initiatieven/energiezuiniggebouwd/hoekwoning-m.
RVO. 2017a. “Energiebesparingsverkenner.”
https://energiebesparingsverkenner.rvo.nl/Verken/Verkenning?mode=2.
RVO. 2017b. “Energy Prices.” https://energiecijfers.databank.nl/dashboard/Energieprijzen--
cgdgchagcrhiijugj/.
Rysanek, A. M., and R. Choudhary. 2013. “Optimum Building Energy Retrofits under
Technical and Economic Uncertainty.” Energy and Buildings 57: 324–37.
doi:10.1016/j.enbuild.2012.10.027.
Saaty, R. W. 1987. “The Analytic Hierarchy Process-What It Is and How It Is Used.”
Mathematical Modelling 9 (3–5): 161–76. doi:10.1016/0270-0255(87)90473-8.
Saint-Gobian Isover. 2016. “Saint-Gobian Isover.” Catalogus Bouw 2016. Etten-Leur.
Sameni, Seyed Masoud Tabatabaei, Mark Gaterell, Azadeh Montazami, and Abdullahi
Ahmed. 2015. “Overheating Investigation in UK Social Housing Flats Built to the
Passivhaus Standard.” Building and Environment 92: 222–35.
doi:10.1016/j.buildenv.2015.03.030.
Sartori, I., and A. G. Hestnes. 2007. “Energy Use in the Life Cycle of Conventional and Low-
Energy Buildings: A Review Article.” Energy and Buildings 39 (3): 249–57.
doi:10.1016/j.enbuild.2006.07.001.
Sautua, Santiago I. 2017. “Does Uncertainty Cause Inertia in Decision Making? An
Experimental Study of the Role of Regret Aversion and Indecisiveness.” Journal of
Economic Behavior & Organization 136: 1–14. doi:10.1016/j.jebo.2017.02.003.
Savage, L. 1951. “The Theory of Statistical Decision.” Journal of the American Statistical
Association 46: 55–67.
Shahrokni, Ali, and Robert Feldt. 2013. “A Systematic Review of Software Robustness.”
Information and Software Technology 55 (1): 1–17. doi:10.1016/j.infsof.2012.06.002.
Silva, Arthur Santos, and Enedir Ghisi. 2014. “Uncertainty Analysis of User Behaviour and
Physical Parameters in Residential Building Performance Simulation.” Energy and
Solar Energy Laboratory University of Wisconsin-Madison, GmbH - TRANSSOLAR
Energietechnik, CSTB - Centre Scientifique et Technique du Bâtiment, and TESS –
307
References
Thermal Energy Systems Specialists. 2009. “TRNSYS 17 – a TRaNsientSYstem.

Simulation Program.” Simulation 1: 1–79.
Solé, Ricard V., Martí Rosas-Casals, Bernat Corominas-Murtra, and Sergi Valverde. 2008.
“Robustness of the European Power Grids under Intentional Attack.” Physical Review
E - Statistical, Nonlinear, and Soft Matter Physics 77 (2): 1–7.
doi:10.1103/PhysRevE.77.026102.
Sousa, Joana. 2012. “Energy Simulation Software for Buildings: Review and Comparison.”
Information Technology for Energy Applications 2012, 6–7.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.416.7812&rep=rep1&type=p
df.
Struck, C., J.L.M; Hensen, and P. Kotek. 2009a. “On the Application of Uncertainty and
Sensitivity Analysis with Abstract Building Performance Simulation Tools.” Journal of
Building Physics 33 (1): 5–27. doi:10.1177/1744259109103345.
Struck, C, J Hensen, and P Kotek. 2009b. “On the Application of Uncertainty and Sensitivity
Analysis with Abstract Building Performance Simulation Tools.” Journal of Building
Physics 33 (1): 5–27. doi:10.1177/1744259109103345.
Struck, C, and J L M Hensen. 2013. “Scenario Analysis for the Robustness Assessment of
Building Design Alternatives – a Dutch Case Study.” In Proceedings of CISBAT 2013,
939–44.
Sun, Kaiyu, and Tianzhen Hong. 2017. “A Framework for Quantifying the Impact of
Occupant Behavior on Energy Savings of Energy Conservation Measures.” Energy and
Sun, Y, Li Gu, C. F Jeff Wu, and Godfried Augenbroe. 2014. “Exploring HVAC System
Sizing under Uncertainty.” Energy and Buildings 81: 243–52.
doi:10.1016/j.enbuild.2014.06.026.
Sun, Y, Pei Huang, and Gongsheng Huang. 2015. “A Multi-Criteria System Design
Optimization for Net Zero Energy Buildings under Uncertainties.” Energy and
Sun, Y, Heng Su, C. F Jeff Wu, and Godfried Augenbroe. 2015. “Quantification of Model
Form Uncertainty in the Calculation of Solar Diffuse Irradiation on Inclined Surfaces
for Building Energy Simulation.” Journal of Building Performance Simulation 8 (4):
253–65. doi:10.1080/19401493.2014.914247.
Taguchi, G. 1987. System of Experimental Design. Vol 1. Kraus International, New York.
Taguchi, G., S. Chowdhury, and S. Taguchi. 2000. Robust Engineering. New York: McGraw-
Hill.
Tesla. 2017. “Powerwall: Batteries for Home.”
https://www.tesla.com/nl_NL/powerwall?redirect=no#design.
Tian, Wei. 2013. “A Review of Sensitivity Analysis Methods in Building Energy Analysis.”
Renewable and Sustainable Energy Reviews 20: 411–19. doi:10.1016/j.rser.2012.12.014.
Torcellini, P, S Pless, and M Deru. 2006. “Zero Energy Buildings: A Critical Look at the
Definition Preprint.” ACEE Summer Studay 2: 15. doi:10.1016/S1471-0846(02)80045-
2.
Trcka, Marija, and Jan L M Hensen. 2010. “Overview of HVAC System Simulation.”
Automation in Construction 19 (2): 93–99. doi:10.1016/j.autcon.2009.11.019.
308
Tuohy, Paul. 2009. “Regulations and Robust Low-Carbon Buildings.” Building Research &
Information 37 (4): 433–45. doi:10.1080/09613210902904254.
USGBC. 2017. “Green Building.” https://www.usgbc.org/articles/what-green-building.
van den Hurk, Bart, Albert Klein Tank, Geert Lenderink, Aad van Ulden, Geert Jan van
Oldenborgh, Caroline Katsman, Henk van den Brink, et al. 2006. “KNMI Climate
Change Scenarios 2006 for the Netherlands.”
Van Gelder, L. 2014. “A Probabilistic Design Methodology for Building Performance
Optimisation.” KU Leuven.
Van Gelder, L, Hans Janssen, and Staf Roels. 2014. “Probabilistic Design and Analysis of
Building Performances: Methodology and Application Example.” Energy and Buildings
79: 202–11. doi:10.1016/j.enbuild.2014.04.042.
Voss, Karsten, Sebastian Herkel, Jens Pfafferott, Gunter Lohnert, and Andreas Wagner.
2007. “Energy Efficient Office Buildings with Passive Cooling - Results and
Experiences from a Research and Demonstration Programme.” Solar Energy 81 (3):
424–34. doi:10.1016/j.solener.2006.04.008.
Voss, Karsten, Eike Musall, and Markus Lichtmeß. 2011. “From Low-Energy to Net Zero-
Energy Buildings: Status and Perspectives.” Journal of Green Building 6 (1): 46–57.
doi:10.3992/jgb.6.1.46.
Voss, Karsten, Igor Sartori, and Roberto Lollini. 2012a. “Nearly-Zero , Net Zero and Plus
Energy Buildings.” REHVA Journal 49 (6): 23–28.
http://www.rehva.eu/en/648.nearly-zero-net-zero-and-plus-energy-buildings-how-
definitions-regulations-affect-the-solutions.
Voss, Karsten, Igor Sartori, and Roberto Lollini. 2012b. “Nearly-Zero , Net Zero and Plus
Energy Buildings.” REHVA Journal 49 (6): 23–28.
VROM. 2010. “Duurzaam Bouwen En Verbouwen: Trias Energetica.”
Wald, A. 1945. “Statistical Decision Functions Which Minimize the Maximum Risk.” The
Annals of Mathematics 46 (2): 265–80.
Walsh, Matthew M., Evan H. Einstein, and Kevin A. Gluck. 2013. “A Quantification of
Robustness.” Journal of Applied Research in Memory and Cognition 2 (3). The Society for
Applied Research in Memory and Cognition: 137–48.
doi:10.1016/j.jarmac.2013.07.002.
Wan, Kevin K W, Danny H W Li, and Joseph C. Lam. 2011. “Assessment of Climate Change
Impact on Building Energy Use and Mitigation Measures in Subtropical Climates.”
Energy 36 (3): 1404–14. doi:10.1016/j.energy.2011.01.033.
Wang, Jiang Jiang, You Yin Jing, Chun Fa Zhang, and Jun Hong Zhao. 2009. “Review on
Multi-Criteria Decision Analysis Aid in Sustainable Energy Decision-Making.”
Renewable and Sustainable Energy Reviews 13 (9): 2263–78.
doi:10.1016/j.rser.2009.06.021.
Wang, Liping, Paul Mathew, and Xiufeng Pang. 2012. “Uncertainties in Energy
Consumption Introduced by Building Operations and Weather for a Medium-Size
Office Building.” Energy and Buildings 53: 152–58. doi:10.1016/j.enbuild.2012.06.017.
Wiberg, H.A., Laurent Georges, Tor Helge Dokka, Matthias Haase, Berit Time, Anne G.
Lien, Sofie Mellegard, and Mette Maltha. 2014. “A Net Zero Emission Concept
Analysis of a Single-Family House.” Energy and Buildings 74: 101–10.
309
References
doi:10.1016/j.enbuild.2014.01.037.
Wilcoxon, Frank. 1945. “Individual Comparisons by Ranking Methods.” Biometrics Bulletin 1
(6): 80–83.
Woloszyn, Monika, and Ian Beausoleil-Morrison. 2017. “Treating Uncertainty in Building
Performance Simulation.” Journal of Building Performance Simulation 10 (1): 1–2.
doi:10.1080/19401493.2017.1261641.
Xidonas, Panos, George Mavrotas, Christis Hassapis, and Constantin Zopounidis. 2017.
“Robust Multiobjective Portfolio Optimization: A Minimax Regret Approach.”
European Journal of Operational Research 262 (1): 299–305.
doi:10.1016/j.ejor.2017.03.041.
Yan, Da, William O’Brien, Tianzhen Hong, Xiaohang Feng, H. Burak Gunay, Farhang
Tahmasebi, and Ardeshir Mahdavi. 2015. “Occupant Behavior Modeling for Building
Performance Simulation: Current State and Future Challenges.” Energy and Buildings
107: 264–78. doi:10.1016/j.enbuild.2015.08.032.
Yun, Geun Young, Paul Tuohy, and Koen Steemers. 2009. “Thermal Performance of a
Naturally Ventilated Building Using a Combined Algorithm of Probabilistic Occupant
Behaviour and Deterministic Heat and Mass Balance Models.” Energy and Buildings 41
(5): 489–99. doi:10.1016/j.enbuild.2008.11.013.
Zero Carbon Hub. 2015. “Overheating in Homes: The Big Picture.”
doi:10.1017/CBO9781107415324.004.
Zero Carbon Hub. 2016. “Solutions to Overheating in Homes, Evidence Review.”
http://www.zerocarbonhub.org/sites/default/files/resources/reports/ZCH-
OverheatingEvidenceReview.pdf.
Zhang, Sheng, Pei Huang, and Yongjun Sun. 2016. “A Multi-Criterion Renewable Energy
System Design Optimization for Net Zero Energy Buildings under Uncertainties.”
Energy 94: 654–65. doi:10.1016/j.energy.2015.11.044.
310
Publications list
Journal papers
Published
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen (2018). “A methodology for

performance robustness assessment of low-energy buildings using scenario
analysis”. Applied Energy 212: 428-442.
• Benjamin Manrique Delgado, Rajesh Kotireddy, Sunliang Cao, Ala Hasan,

Pieter-Jan Hoes, Jan L.M. Hensen, Kai Sirén (2018). “Lifecycle cost and CO2
emissions of residential heat and electricity prosumers in Finland and the
Netherlands”. Energy Conversion and Management 160: 495-508.
Under review
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen. “Integration of robustness

indicators in optimization framework to find robust optimal low-energy
building designs”. Under revision in Journal of Building Performance
Simulation.
• Rajesh Kotireddy, Roel Loonen, Pieter-Jan Hoes, Jan Hensen. “Prediction of

building performance robustness under uncertainties: Why? What? How?”.
Under review in Building and Environment.
Under preparation
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen. “Multi-criteria assessment

and multi-criteria decision making considering robustness to identify robust
low-energy building designs”.
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen. “Robust renovation

measures for Dutch dwellings: demonstration using a case study”.
Professional journal
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen (2017). Robust net zero
energy buildings: A methodology for designers to evaluate robustness.
REHVA Journal 54 (4): 9-19.
311
Publications list
Conference papers (peer reviewed)

• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen (2017). “Simulation-based
comparison of robustness assessment methods to identify robust low energy
buildings”. In the proceedings of 15th International Building Simulation
Conference, San Francisco, USA. 892-901.
• Rajesh Kotireddy, Pieter-Jan Hoes, Jan Hensen (2015). “Optimal balance

between energy demand and onsite energy generation for robust net zero
energy buildings considering future scenarios”. In the proceedings of 14th
International Building Simulation Conference, Hyderabad, India. 1970-1977.
312
Curriculum Vitae
Rajesh Kotireddy was born in Kotireddygaripalli, India. He completed his MS

(Master of Science) in Mechanical Engineering at the Indian Institute of Technology
(IIT), Madras, India. He worked on performance and reliability assessment of a novel
mixed refrigerant cascade refrigerator for low temperature applications for his
graduation thesis. Up on his graduation, he was recruited by the Swiss Center for
Electronics and Micro-Technology (CSEM). Kotireddy worked as a R&D Engineer in
the field of sustainable energy for 3.5 years with CSEM in the United Arab Emirates.
During his tenure at CSEM, he worked on techno-economic feasibility assessment
of energy efficiency measures and renewable energy integration in the built
environment for local conditions.
Rajesh started his PhD at the end of 2013 in the computational building performance
group under the supervision of Professor Jan Hensen and dr.ir. Pieter-Jan Hoes. He
worked on robust low-energy houses for his PhD dissertation and this project is part
of the green tech initiative by Eurotech energy efficient buildings and communities.
He has published several conference and journal papers.
313
Bouwstenen is een publicatiereeks
van de Faculteit Bouwkunde,
Technische Universiteit Eindhoven.
Zij presenteert resultaten van
onderzoek en andere activiteiten op
het vakgebied der Bouwkunde,
uitgevoerd in het kader van deze
Faculteit.
Bouwstenen zijn telefonisch te

bestellen op nummer
040 - 2472383
Kernredactie
MTOZ
Reeds verschenen in de serie
Bouwstenen
nr 1 nr 9
Elan: A Computer Model for Building Strukturering en Verwerking van
Energy Design: Theory and Validation Tijdgegevens voor de Uitvoering
Martin H. de Wit van Bouwwerken
H.H. Driessen ir. W.F. Schaefer
R.M.M. van der Velden P.A. Erkelens
nr 2 nr 10
Kwaliteit, Keuzevrijheid en Kosten: Stedebouw en de Vorming van
Evaluatie van Experiment Klarendal, een Speciale Wetenschap
Arnhem K. Doevendans
J. Smeets
C. le Nobel nr 11
M. Broos Informatica en Ondersteuning
J. Frenken van Ruimtelijke Besluitvorming
A. v.d. Sanden G.G. van der Meulen
nr 3 nr 12
Crooswijk: Staal in de Woningbouw,
Van ‘Bijzonder’ naar ‘Gewoon’ Korrosie-Bescherming van
Vincent Smit de Begane Grondvloer
Kees Noort Edwin J.F. Delsing
nr 4 nr 13
Staal in de Woningbouw Een Thermisch Model voor de
Edwin J.F. Delsing Berekening van Staalplaatbetonvloeren
onder Brandomstandigheden
nr 5 A.F. Hamerlinck
Mathematical Theory of Stressed
Skin Action in Profiled Sheeting with nr 14
Various Edge Conditions De Wijkgedachte in Nederland:
Andre W.A.M.J. van den Bogaard Gemeenschapsstreven in een
Stedebouwkundige Context
nr 6 K. Doevendans
Hoe Berekenbaar en Betrouwbaar is R. Stolzenburg
de Coëfficiënt k in x-ksigma en x-ks?
K.B. Lub nr 15
A.J. Bosch Diaphragm Effect of Trapezoidally
Profiled Steel Sheets:
nr 7 Experimental Research into the
Het Typologisch Gereedschap: Influence of Force Application
Een Verkennende Studie Omtrent Andre W.A.M.J. van den Bogaard
Typologie en Omtrent de Aanpak
van Typologisch Onderzoek nr 16
J.H. Luiten Versterken met Spuit-Ferrocement:
Het Mechanische Gedrag van met
nr 8 Spuit-Ferrocement Versterkte
Informatievoorziening en Beheerprocessen Gewapend Betonbalken
A. Nauta K.B. Lubir
Jos Smeets (red.) M.C.G. van Wanroy
Helga Fassbinder (projectleider)
Adrie Proveniers
J. v.d. Moosdijk
nr 17 nr 27
De Tractaten van Het Woonmilieu op Begrip Gebracht:
Jean Nicolas Louis Durand Een Speurtocht naar de Betekenis van het
G. van Zeyl Begrip 'Woonmilieu'
Jaap Ketelaar
nr 18
Wonen onder een Plat Dak: nr 28
Drie Opstellen over Enkele Urban Environment in Developing Countries
Vooronderstellingen van de editors: Peter A. Erkelens
Stedebouw George G. van der Meulen (red.)
K. Doevendans
nr 29
nr 19 Stategische Plannen voor de Stad:
Supporting Decision Making Processes: Onderzoek en Planning in Drie Steden
A Graphical and Interactive Analysis of prof.dr. H. Fassbinder (red.)
Multivariate Data H. Rikhof (red.)
W. Adams
nr 30
nr 20 Stedebouwkunde en Stadsbestuur
Self-Help Building Productivity: Piet Beekman
A Method for Improving House Building
by Low-Income Groups Applied to Kenya nr 31
1990-2000 De Architectuur van Djenné:
P. A. Erkelens Een Onderzoek naar de Historische Stad
P.C.M. Maas
nr 21
De Verdeling van Woningen: nr 32
Een Kwestie van Onderhandelen Conjoint Experiments and Retail Planning
Vincent Smit Harmen Oppewal
nr 22 nr 33
Flexibiliteit en Kosten in het Ontwerpproces: Strukturformen Indonesischer Bautechnik:
Een Besluitvormingondersteunend Model Entwicklung Methodischer Grundlagen
M. Prins für eine ‘Konstruktive Pattern Language’
in Indonesien
nr 23 Heinz Frick arch. SIA
Spontane Nederzettingen Begeleid:
Voorwaarden en Criteria in Sri Lanka nr 34
Po Hin Thung Styles of Architectural Designing:
Empirical Research on Working Styles
nr 24 and Personality Dispositions
Fundamentals of the Design of Anton P.M. van Bakel
Bamboo Structures
Oscar Arce-Villalobos nr 35
Conjoint Choice Models for Urban
nr 25 Tourism Planning and Marketing
Concepten van de Bouwkunde Benedict Dellaert
M.F.Th. Bax (red.)
H.M.G.J. Trum (red.) nr 36
Stedelijke Planvorming als Co-Produktie
nr 26 Helga Fassbinder (red.)
Meaning of the Site
Xiaodong Li
nr 37 nr 48
Design Research in the Netherlands Concrete Behaviour in Multiaxial
editors: R.M. Oxman Compression
M.F.Th. Bax Erik van Geel
H.H. Achten
nr 49
nr 38 Modelling Site Selection
Communication in the Building Industry Frank Witlox
Bauke de Vries
nr 50
nr 39 Ecolemma Model
Optimaal Dimensioneren van Ferdinand Beetstra
Gelaste Plaatliggers
J.B.W. Stark nr 51
F. van Pelt Conjoint Approaches to Developing
L.F.M. van Gorp Activity-Based Models
B.W.E.M. van Hove Donggen Wang
nr 40 nr 52
Huisvesting en Overwinning van Armoede On the Effectiveness of Ventilation
P.H. Thung Ad Roos
P. Beekman (red.)
nr 53
nr 41 Conjoint Modeling Approaches for
Urban Habitat: Residential Group preferences
The Environment of Tomorrow Eric Molin
George G. van der Meulen
Peter A. Erkelens nr 54
Modelling Architectural Design
nr 42 Information by Features
A Typology of Joints Jos van Leeuwen
John C.M. Olie
nr 55
nr 43 A Spatial Decision Support System for
Modeling Constraints-Based Choices the Planning of Retail and Service Facilities
for Leisure Mobility Planning Theo Arentze
Marcus P. Stemerding
nr 56
nr 44 Integrated Lighting System Assistant
Activity-Based Travel Demand Modeling Ellie de Groot
Dick Ettema
nr 57
nr 45 Ontwerpend Leren, Leren Ontwerpen
Wind-Induced Pressure Fluctuations J.T. Boekholt
on Building Facades
Chris Geurts nr 58
Temporal Aspects of Theme Park Choice
nr 46 Behavior
Generic Representations Astrid Kemperman
Henri Achten
nr 59
nr 47 Ontwerp van een Geïndustrialiseerde
Johann Santini Aichel: Funderingswijze
Architectuur en Ambiguiteit Faas Moonen
Dirk De Meyer
nr 60 nr 72
Merlin: A Decision Support System Moisture Transfer Properties of
for Outdoor Leisure Planning Coated Gypsum
Manon van Middelkoop Emile Goossens
nr 61 nr 73
The Aura of Modernity Plybamboo Wall-Panels for Housing
Jos Bosman Guillermo E. González-Beltrán
nr 62 nr 74
Urban Form and Activity-Travel Patterns The Future Site-Proceedings
Daniëlle Snellen Ger Maas
Frans van Gassel
nr 63
Design Research in the Netherlands 2000 nr 75
Henri Achten Radon transport in
Autoclaved Aerated Concrete
nr 64 Michel van der Pal
Computer Aided Dimensional Control in
Building Construction nr 76
Rui Wu The Reliability and Validity of Interactive
Virtual Reality Computer Experiments
nr 65 Amy Tan
Beyond Sustainable Building
editors: Peter A. Erkelens nr 77
Sander de Jonge Measuring Housing Preferences Using
August A.M. van Vliet Virtual Reality and Belief Networks
co-editor: Ruth J.G. Verhagen Maciej A. Orzechowski
nr 66 nr 78
Das Globalrecyclingfähige Haus Computational Representations of Words
Hans Löfflad and Associations in Architectural Design
Nicole Segers
nr 67
Cool Schools for Hot Suburbs nr 79
René J. Dierkx Measuring and Predicting Adaptation in
Multidimensional Activity-Travel Patterns
nr 68 Chang-Hyeon Joh
A Bamboo Building Design Decision
Support Tool nr 80
Fitri Mardjono Strategic Briefing
Fayez Al Hassan
nr 69
Driving Rain on Building Envelopes nr 81
Fabien van Mook Well Being in Hospitals
Simona Di Cicco
nr 70
Heating Monumental Churches nr 82
Henk Schellen Solares Bauen:
Implementierungs- und Umsetzungs-
nr 71 Aspekte in der Hochschulausbildung
Van Woningverhuurder naar in Österreich
Aanbieder van Woongenot Gerhard Schuster
Patrick Dogge
nr 83 nr 94
Supporting Strategic Design of Human Lighting Demands:
Workplace Environments with Healthy Lighting in an Office Environment
Case-Based Reasoning Myriam Aries
Shauna Mallory-Hill
nr 95
nr 84 A Spatial Decision Support System for
ACCEL: A Tool for Supporting Concept the Provision and Monitoring of Urban
Generation in the Early Design Phase Greenspace
Maxim Ivashkov Claudia Pelizaro
nr 85 nr 96
Brick-Mortar Interaction in Masonry Leren Creëren
under Compression Adri Proveniers
Ad Vermeltfoort
nr 97
nr 86 Simlandscape
Zelfredzaam Wonen Rob de Waard
Guus van Vliet
nr 98
nr 87 Design Team Communication
Een Ensemble met Grootstedelijke Allure Ad den Otter
Jos Bosman
Hans Schippers nr 99
Humaan-Ecologisch
nr 88 Georiënteerde Woningbouw
On the Computation of Well-Structured Juri Czabanowski
Graphic Representations in Architectural
Design nr 100
Henri Achten Hambase
Martin de Wit
nr 89
De Evolutie van een West-Afrikaanse nr 101
Vernaculaire Architectuur Sound Transmission through Pipe
Wolf Schijns Systems and into Building Structures
Susanne Bron-van der Jagt
nr 90
ROMBO Tactiek nr 102
Christoph Maria Ravesloot Het Bouwkundig Contrapunt
Jan Francis Boelen
nr 91
External Coupling between Building nr 103
Energy Simulation and Computational A Framework for a Multi-Agent
Fluid Dynamics Planning Support System
Ery Djunaedy Dick Saarloos
nr 92 nr 104
Design Research in the Netherlands 2005 Bracing Steel Frames with Calcium
editors: Henri Achten Silicate Element Walls
Kees Dorst Bright Mweene Ng’andu
Pieter Jan Stappers
Bauke de Vries nr 105
Naar een Nieuwe Houtskeletbouw
nr 93 F.N.G. De Medts
Ein Modell zur Baulichen Transformation
Jalil H. Saber Zaimian
nr 106 and 107 nr 120
Niet gepubliceerd A Multi-Agent Planning Support
System for Assessing Externalities
nr 108 of Urban Form Scenarios
Geborgenheid Rachel Katoshevski-Cavari
T.E.L. van Pinxteren
nr 121
nr 109 Den Schulbau Neu Denken,
Modelling Strategic Behaviour in Fühlen und Wollen
Anticipation of Congestion Urs Christian Maurer-Dietrich
Qi Han
nr 122
nr 110 Peter Eisenman Theories and
Reflecties op het Woondomein Practices
Fred Sanders Bernhard Kormoss
nr 111 nr 123
On Assessment of Wind Comfort User Simulation of Space Utilisation
by Sand Erosion Vincent Tabak
Gábor Dezsö
nr 125
nr 112 In Search of a Complex System Model
Bench Heating in Monumental Churches Oswald Devisch
Dionne Limpens-Neilen
nr 126
nr 113 Lighting at Work:
RE. Architecture Environmental Study of Direct Effects
Ana Pereira Roders of Lighting Level and Spectrum on
Psycho-Physiological Variables
nr 114 Grazyna Górnicka
Toward Applicable Green Architecture
Usama El Fiky nr 127
Flanking Sound Transmission through
nr 115 Lightweight Framed Double Leaf Walls
Knowledge Representation under Stefan Schoenwald
Inherent Uncertainty in a Multi-Agent
System for Land Use Planning nr 128
˙
Liying Ma Bounded Rationality and Spatio-Temporal
Pedestrian Shopping Behavior
nr 116 Wei Zhu
Integrated Heat Air and Moisture
Modeling and Simulation nr 129
Jos van Schijndel Travel Information:
Impact on Activity Travel Pattern
nr 117 Zhongwei Sun
Concrete Behaviour in Multiaxial
Compression nr 130
J.P.W. Bongers Co-Simulation for Performance
Prediction of Innovative Integrated
nr 118 Mechanical Energy Systems in Buildings
The Image of the Urban Landscape �
Marija Trcka
Ana Moya Pellitero
nr 131
nr 119 Niet gepubliceerd
The Self-Organizing City in Vietnam
Stephanie Geertman
nr 132 nr 143
Architectural Cue Model in Evacuation Modelling Life Trajectories and Transport
Simulation for Underground Space Design Mode Choice Using Bayesian Belief Networks
Chengyu Sun Marloes Verhoeven
nr 133 nr 144
Uncertainty and Sensitivity Analysis in Assessing Construction Project
Building Performance Simulation for Performance in Ghana
Decision Support and Design Optimization William Gyadu-Asiedu
Christina Hopfe
nr 145
nr 134 Empowering Seniors through
Facilitating Distributed Collaboration Domotic Homes
in the AEC/FM Sector Using Semantic Masi Mohammadi
Web Technologies
Jacob Beetz nr 146
An Integral Design Concept for
nr 135 Ecological Self-Compacting Concrete
Circumferentially Adhesive Bonded Glass Martin Hunger
Panes for Bracing Steel Frame in Façades
Edwin Huveners nr 147
Governing Multi-Actor Decision Processes
nr 136 in Dutch Industrial Area Redevelopment
Influence of Temperature on Concrete Erik Blokhuis
Beams Strengthened in Flexure
with CFRP nr 148
Ernst-Lucas Klamer A Multifunctional Design Approach
for Sustainable Concrete
nr 137 Götz Hüsken
Sturen op Klantwaarde
Jos Smeets nr 149
Quality Monitoring in Infrastructural
nr 139 Design-Build Projects
Lateral Behavior of Steel Frames Ruben Favié
with Discretely Connected Precast Concrete
Infill Panels nr 150
Paul Teewen Assessment Matrix for Conservation of
Valuable Timber Structures
nr 140 Michael Abels
Integral Design Method in the Context
of Sustainable Building Design nr 151
Perica Savanovic´ Co-simulation of Building Energy Simulation
and Computational Fluid Dynamics for
nr 141 Whole-Building Heat, Air and Moisture
Household Activity-Travel Behavior: Engineering
Implementation of Within-Household Mohammad Mirsadeghi
Interactions
Renni Anggraini nr 152
External Coupling of Building Energy
nr 142 Simulation and Building Element Heat,
Design Research in the Netherlands 2010 Air and Moisture Simulation
Henri Achten Daniel Cóstola
nr 153 nr 165
Adaptive Decision Making In Beyond Uniform Thermal Comfort
Multi-Stakeholder Retail Planning on the Effects of Non-Uniformity and
Ingrid Janssen Individual Physiology
Lisje Schellen
nr 154
Landscape Generator nr 166
Kymo Slager Sustainable Residential Districts
Gaby Abdalla
nr 155
Constraint Specification in Architecture nr 167
Remco Niemeijer Towards a Performance Assessment
Methodology using Computational
nr 156 Simulation for Air Distribution System
A Need-Based Approach to Designs in Operating Rooms
Dynamic Activity Generation Mônica do Amaral Melhado
Linda Nijland
nr 168
nr 157 Strategic Decision Modeling in
Modeling Office Firm Dynamics in an Brownfield Redevelopment
Agent-Based Micro Simulation Framework Brano Glumac
Gustavo Garcia Manzato
nr 169
nr 158 Pamela: A Parking Analysis Model
Lightweight Floor System for for Predicting Effects in Local Areas
Vibration Comfort Peter van der Waerden
Sander Zegers
nr 170
nr 159 A Vision Driven Wayfinding Simulation-System
Aanpasbaarheid van de Draagstructuur Based on the Architectural Features Perceived
Roel Gijsbers in the Office Environment
Qunli Chen
nr 160
'Village in the City' in Guangzhou, China nr 171
Yanliu Lin Measuring Mental Representations
Underlying Activity-Travel Choices
nr 161 Oliver Horeni
Climate Risk Assessment in Museums
Marco Martens nr 172
Modelling the Effects of Social Networks
nr 162 on Activity and Travel Behaviour
Social Activity-Travel Patterns Nicole Ronald
Pauline van den Berg
nr 173
nr 163 Uncertainty Propagation and Sensitivity
Sound Concentration Caused by Analysis Techniques in Building Performance
Curved Surfaces Simulation to Support Conceptual Building
Martijn Vercammen and System Design
Christian Struck
nr 164
Design of Environmentally Friendly nr 174
Calcium Sulfate-Based Building Materials: Numerical Modeling of Micro-Scale
Towards an Improved Indoor Air Quality Wind-Induced Pollutant Dispersion
Qingliang Yu in the Built Environment
Pierre Gousseau
nr 175 nr 185
Modeling Recreation Choices A Distributed Dynamic Simulation
over the Family Lifecycle Mechanism for Buildings Automation
Anna Beatriz Grigolon and Control Systems
Azzedine Yahiaoui
nr 176
Experimental and Numerical Analysis of nr 186
Mixing Ventilation at Laminar, Transitional Modeling Cognitive Learning of Urban
and Turbulent Slot Reynolds Numbers Networks in Daily Activity-Travel Behavior
Twan van Hooff ¸
Sehnaz Cenani Durmazoglu �
nr 177 nr 187
Collaborative Design Support: Functionality and Adaptability of Design
Workshops to Stimulate Interaction and Solutions for Public Apartment Buildings
Knowledge Exchange Between Practitioners in Ghana
Emile M.C.J. Quanjel Stephen Agyefi-Mensah
nr 178 nr 188
Future-Proof Platforms for Aging-in-Place A Construction Waste Generation Model
Michiel Brink for Developing Countries
Lilliana Abarca-Guerrero
nr 179
Motivate: nr 189
A Context-Aware Mobile Application for Synchronizing Networks:
Physical Activity Promotion The Modeling of Supernetworks for
Yuzhong Lin Activity-Travel Behavior
Feixiong Liao
nr 180
Experience the City: nr 190
Analysis of Space-Time Behaviour and Time and Money Allocation Decisions
Spatial Learning in Out-of-Home Leisure Activity Choices
Anastasia Moiseeva Gamze Zeynep Dane
nr 181 nr 191
Unbonded Post-Tensioned Shear Walls of How to Measure Added Value of CRE and
Calcium Silicate Element Masonry Building Design
Lex van der Meer Rianne Appel-Meulenbroek
nr 182 nr 192
Construction and Demolition Waste Secondary Materials in Cement-Based
Recycling into Innovative Building Materials Products:
for Sustainable Construction in Tanzania Treatment, Modeling and Environmental
Mwita M. Sabai Interaction
Miruna Florea
nr 183
Durability of Concrete nr 193
with Emphasis on Chloride Migration Concepts for the Robustness Improvement
Przemys�aw Spiesz of Self-Compacting Concrete:
Effects of Admixtures and Mixture
nr 184 Components on the Rheology and Early
Computational Modeling of Urban Hydration at Varying Temperatures
Wind Flow and Natural Ventilation Potential Wolfram Schmidt
of Buildings
Rubina Ramponi
nr 194 nr 204
Modelling and Simulation of Virtual Natural Geometry and Ventilation:
Lighting Solutions in Buildings Evaluation of the Leeward Sawtooth Roof
Rizki A. Mangkuto Potential in the Natural Ventilation of
Buildings
nr 195 Jorge Isaac Perén Montero
Nano-Silica Production at Low Temperatures
from the Dissolution of Olivine - Synthesis, nr 205
Tailoring and Modelling Computational Modelling of Evaporative
Alberto Lazaro Garcia Cooling as a Climate Change Adaptation
Measure at the Spatial Scale of Buildings
nr 196 and Streets
Building Energy Simulation Based Hamid Montazeri
Assessment of Industrial Halls for
Design Support nr 206
Bruno Lee Local Buckling of Aluminium Beams in Fire
Conditions
nr 197 Ronald van der Meulen
Computational Performance Prediction
of the Potential of Hybrid Adaptable nr 207
Thermal Storage Concepts for Lightweight Historic Urban Landscapes:
Low-Energy Houses Framing the Integration of Urban and
Pieter-Jan Hoes Heritage Planning in Multilevel Governance
Loes Veldpaus
nr 198
Application of Nano-Silica in Concrete nr 208
George Quercia Bianchi Sustainable Transformation of the Cities:
Urban Design Pragmatics to Achieve a
nr 199 Sustainable City
Dynamics of Social Networks and Activity Ernesto Antonio Zumelzu Scheel
Travel Behaviour
Fariya Sharmeen nr 209
Development of Sustainable Protective
nr 200 Ultra-High Performance Fibre Reinforced
Building Structural Design Generation and Concrete (UHPFRC):
Optimisation including Spatial Modification Design, Assessment and Modeling
Juan Manuel Davila Delgado Rui Yu
nr 201 nr 210
Hydration and Thermal Decomposition of Uncertainty in Modeling Activity-Travel
Cement/Calcium-Sulphate Based Materials Demand in Complex Uban Systems
Ariën de Korte Soora Rasouli
nr 202 nr 211
Republiek van Beelden: Simulation-based Performance Assessment
De Politieke Werkingen van het Ontwerp in of Climate Adaptive Greenhouse Shells
Regionale Planvorming Chul-sung Lee
Bart de Zwart
nr 212
nr 203 Green Cities:
Effects of Energy Price Increases on Modelling the Spatial Transformation of
Individual Activity-Travel Repertoires and the Urban Environment using Renewable
Energy Consumption Energy Technologies
Dujuan Yang Saleh Mohammadi
nr 213 nr 223
A Bounded Rationality Model of Short and Personalized Route Finding in Multimodal
Long-Term Dynamics of Activity-Travel Transportation Networks
Behavior Jianwe Zhang
Ifigeneia Psarra
nr 224
nr 214 The Design of an Adaptive Healing Room
Effects of Pricing Strategies on Dynamic for Stroke Patients
Repertoires of Activity-Travel Behaviour Elke Daemen
Elaheh Khademi
nr 225
nr 215 Experimental and Numerical Analysis of
Handstorm Principles for Creative and Climate Change Induced Risks to Historic
Collaborative Working Buildings and Collections
Frans van Gassel Zara Huijbregts
nr 216 nr 226
Light Conditions in Nursing Homes: Wind Flow Modeling in Urban Areas Through
Visual Comfort and Visual Functioning of Experimental and Numerical Techniques
Residents Alessio Ricci
Marianne M. Sinoo
nr 227
nr 217 Clever Climate Control for Culture:
Woonsporen: Energy Efficient Indoor Climate Control
De Sociale en Ruimtelijke Biografie van Strategies for Museums Respecting
een Stedelijk Bouwblok in de Amsterdamse Collection Preservation and Thermal
Transvaalbuurt Comfort of Visitors
Hüseyin Hüsnü Yegenoglu Rick Kramer
nr 218 nr 228
Studies on User Control in Ambient Fatigue Life Estimation of Metal Structures
Intelligent Systems Based on Damage Modeling
Berent Willem Meerbeek Sarmediran Silitonga
nr 219 nr 229
Daily Livings in a Smart Home: A multi-agents and occupancy based
Users’ Living Preference Modeling of Smart strategy for energy management and
Homes process control on the room-level
Erfaneh Allameh Timilehin Moses Labeodan
nr 220 nr 230
Smart Home Design: Environmental assessment of Building
Spatial Preference Modeling of Smart Integrated Photovoltaics:
Homes Numerical and Experimental Carrying
Mohammadali Heidari Jozam Capacity Based Approach
Michiel Ritzen
nr 221
Wonen: nr 231
Discoursen, Praktijken, Perspectieven Performance of Admixture and Secondary
Jos Smeets Minerals in Alkali Activated Concrete:
Sustaining a Concrete Future
nr 222 Arno Keulen
Personal Control over Indoor Climate in
Offices:
Impact on Comfort, Health and Productivity
Atze Christiaan Boerstra
nr 232 nr 241
World Heritage Cities and Sustainable Gap-Theoretical Analyses of Residential
Urban Development: Satisfaction and Intention to Move
Bridging Global and Local Levels in Monitor- Wen Jiang
ing the Sustainable Urban Development of
World Heritage Cities nr 242
Paloma C. Guzman Molina Travel Satisfaction and Subjective Well-Being:
A Behavioral Modeling Perspective
nr 233 Yanan Gao
Stage Acoustics and Sound Exposure in
Performance and Rehearsal Spaces for nr 243
Orchestras: Building Energy Modelling to Support
Methods for Physical Measurements the Commissioning of Holistic Data Centre
Remy Wenmaekers Operation
Vojtech Zavrel
nr 234
Municipal Solid Waste Incineration (MSWI) nr 244
Bottom Ash: Regret-Based Travel Behavior Modeling:
From Waste to Value Characterization, An Extended Framework
Treatments and Application Sunghoon Jang
Pei Tang
nr 245
nr 235 Towards Robust Low-Energy Houses:
Large Eddy Simulations Applied to Wind A Computational Approach for Performance
Loading and Pollutant Dispersion Robustness Assessment using Scenario
Mattia Ricci Analysis
Rajesh Reddy Kotireddy
nr 236
Alkali Activated Slag-Fly Ash Binders:
Design, Modeling and Application
Xu Gao
nr 237
Sodium Carbonate Activated Slag:
Reaction Analysis, Microstructural
Modification & Engineering Application
Bo Yuan
nr 238
Shopping Behavior in Malls
Widiyani
nr 239
Smart Grid-Building Energy Interactions:
Demand Side Power Flexibility in Office
Buildings
Kennedy Otieno Aduda
nr 240
Modeling Taxis Dynamic Behavior in
Uncertain Urban Environments
Zheng Zhong
The built environment is moving towards energy efficient buildings and
communities to address the increasing concerns of climate change. Con-
sidering the substantial cost associated with the development of these
buildings, future operational performance is a significant criterion in the
design process. The operational performance of low-energy houses is influ-
enced by a multitude of dynamic factors, including occupant behaviour, fu-
ture climate conditions and economic factors. At the time of designing, it is
largely unknown to designers how these influences will unfold over a build-
ing’s life-span. Therefore, performance assessment of low-energy houses
taking into account these uncertainties should be assessed in the design
phase to identify designs capable of delivering the desired performance.
The aim of this research is to develop a computational performance ro-

bustness assessment (CPRA) approach that integrates uncertainties in
multi-criteria assessment using scenario analysis to quantify robustness
and facilitate the selection of robust designs for various decision makers.
The developed approach is demonstrated using new and renovation Dutch
houses for the policymaker and the homeowner.
The proposed CPRA approach deals with one of the major issues that the
building industry is facing in the development of low-energy houses; per-
formance deviation during operation compared to predicted performance.
This CPRA approach can be used by designers and consultants among
other actors involved in the decision-making process to identify robust de-
signs that have the potential to deliver the desired performance in opera-
tion and can thus improve customers satisfaction.
/ Department of the Built Environment

Kotireddy

Uploaded by

Copyright:

Available Formats

You might also like

Kotireddy

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Kotireddy

Uploaded by

Copyright:

Available Formats

Towards Robust Low-Energy Houses

A Computational Approach for Performance Robustness

Rajesh Reddy Kotireddy

geboren te Kotireddygaripalli, Andhra Pradesh, India

voorzitter : prof.ir. E.S.M. Nelissen

leden : prof.dr.ir.arch. S. Roels (KU Leuven)

Het onderzoek of ontwerp dat in dit proefschrift wordt beschreven is uitgevoerd in

© Rajesh Kotireddy, 2018

All rights reserved. No part of this publication may be reproduced, distributed, or

Cover image: Licensed from IStock (Getty Images)

A catalogue record is available from the Eindhoven University of Technology Library

Published under Bouwstenen series 245 of the Department of the Built

Printed by TU/e Print service (Dereumaux), The Netherlands.

In my view, a PhD is a “process of learning”.

I am very thankful to Dr. Pieter-Jan, who was my daily supervisor. Pieter-Jan is a

My family’s endless support ensured the successful completion of this process of

The built environment is under an obligation to move towards low-energy buildings

In low-energy houses, these deviations are a serious concern, as their performance

However, in many cases, the probability of occurrence of these uncertainties is

The computational performance robustness assessment (CPRA) approach

The developed CPRA approach is implemented using a simulation framework, and

Nomenclature .......................................................................................................... XVII

2. State of the art: performance robustness assessment .......................................17

3. Methodology to develop and test the CPRA approach ...................................... 45

4. Implementation of the CPRA approach in the simulation framework ........... 67

5. Description of case studies and decision makers’ preferences ...................... 107

5.5 Performance indicators ..................................................................................... 125

7. Suitability and usability assessment of the CPRA approach.......................... 205

8. Conclusions, original contributions, limitations and future work ................. 215

Appendices ................................................................................................................ 225

References ................................................................................................................. 293

Publications list ...........................................................................................................311

Curriculum Vitae ....................................................................................................... 313

RES : Renewable Energy and Storage System

1.1 Need for low-energy houses

many sustainability frameworks, standards, codes and regulations such as Trias

A low-energy house is defined as a building “built according to special design criteria

1.2 Problem definition

1.2.1 Issues in current design practice

Overheating issues in low-energy houses are likely due to a number of reasons,

However, in the conventional design approach, building performance is predicted

To ensure intended performance, buildings can also be designed to have a relatively

Therefore, performance assessment of low-energy houses taking into account these

1.2.2 Quantification of impact of uncertainties

measures and renewable energy technologies considered for low-energy house

1.2.3 Performance robustness assessment based on scenario analysis

The non-probabilistic decision rules have been implemented to identify robust

1.2.4 Multi-criteria assessment and multi-criteria decision making

1.3 Research aim and objectives

The objectives of this research are:

1. Develop a computational performance robustness assessment (CPRA)

2. Develop a simulation framework to implement the CPRA approach in order

The specific objectives of this research are:

4. Identify uncertainties in building operation and external conditions such as

5. Identify appropriate robustness assessment methods for scenario analysis

6. Carry out multi-criteria performance assessment of various low-energy

7. Implement multi-criteria decision making using the trade-off approach on

8. Carry out sensitivity analysis to identify the most influential scenarios on