Professional Documents
Culture Documents
Optimised Design of Energy Efficient Building Façades Via Evolutionary Neural Networks
Optimised Design of Energy Efficient Building Façades Via Evolutionary Neural Networks
Review
a r t i c l e i n f o a b s t r a c t
Article history: Buildings are required to be more and more energy efficient, in order to comply with restrictive require-
Received 19 April 2011 ments of building regulations and energy certifications. Optimisation algorithms have shown to be
Received in revised form 5 August 2011 effective in identifying good solutions for the design of efficient building services. In this article Evo-
Accepted 4 October 2011
lutionary Neural Network Design (ENN-Design) has been adopted to drive the design of a typical façade
module for an office building. This application is significant, since façades play a major role in the def-
Keywords:
inition of the energy performance of buildings. Both single-objective and multi-objective optimisations
Energy optimisation
have been carried out. The aim of the article is to introduce an innovative approach for improving the
Multi-objective optimisation
Neural networks
performance of building envelopes by means of a reasonable amount of calculation time.
CO2 emission reduction © 2011 Published by Elsevier B.V.
Energy efficient buildings
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3297
2. Optimised design of energy efficient building façades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3298
2.1. Single-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3298
2.2. Multi-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3298
3. Problem description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3299
4. Evolutionary Neural Network Design optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3299
4.1. Single-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3299
4.2. Multi-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3300
5. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3300
5.1. Single-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3300
5.2. Multi-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3301
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3301
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3301
1. Introduction
for a considerable percentage of the overall energy consumption
It has been shown that a new global economic strategy is and, consequently, carbon emissions [2]. Regulations are becom-
necessary in order to limit the effects of climate change: the ing more and more restrictive in limiting the energy requirements
business-as-usual path will lead to unbearable risks for climate and of buildings and energy certifications are linking the commercial
society [1]. A field that is strongly involved in this necessary change value of buildings with their level of sustainability.
of perspective is the building industry, as buildings are responsible This forces engineers and architects to adopt new tools, since
the traditional ways of designing buildings are no more effective
in meeting the sustainability requirements and, at the same time,
∗ Corresponding author. Tel.: +44 2077555042; fax: +44 2075803924. providing adequate levels of comfort for the occupants. The main
E-mail address: giovanni.zemella@arup.com (G. Zemella). issue is due to the difficulty in considering the overall behaviour
of the building interacting with different climates and different priori strategy assigns different weights to various criteria leading
seasons [3,4]. to the identification of a single-objective function to be opti-
Façades influence the energy performance of buildings since mised. The a posteriori optimisation addresses the multi-objective
they are responsible for heat losses and solar gains and allow for problem by providing a set of optimal solutions representing the
daylighting. Therefore the energy requirements for heating, cool- trade-off between the different considered criteria.
ing and artificial lighting are strongly driven by the performance of
the façade. Designing a façade for an energy efficient building is a
very challenging task, because it is necessary to consider how the 2.1. Single-objective optimisation
different aspects (heating, cooling and lighting) interact in the over-
all energy requirements of the building [5]. The main challenge is In this study the target is to identify the façade configuration
to find the best balance between contradictory requirements, such leading to the minimum amount of carbon emissions required in
as allowing for high levels of natural lighting without solar gains order to guarantee the target level of comfort for the occupants of
being too high during summer. At the same time, during winter an the building. The objective function which has been minimised is
adequate amount of solar gains can be beneficial for the reduction the annual amount of carbon dioxide emitted per floor area due to
of heating loads. An efficient design should not focus on one single the energy consumption for heating, cooling and artificial lighting,
aspect (e.g. minimising the cooling energy requirements) without and has been evaluated by means of the following equation [5]:
considering the interactions with the other sources of energy con-
fgas
QC
kg
sumption. On the contrary, it is extremely important to take into CO2
ECO2 = QH + fel QL + (1)
account all the elements at the same time. This cannot be done H COP m2
properly by means of the traditional procedures, which are mainly
based on experience and specific design conditions. A new, holistic where
approach is necessary.
Evolutionary optimisation approaches have been proven effec-
- QH , QL and QC [kWh/m2 ] are the annual energy consumptions due
tive in driving the design of energy efficient building services
to heating, artificial lighting and cooling respectively, calculated
because of their capability in handling extremely large numbers
by means of EnergyPlus simulations;
of variables and potential solutions [3,6,7]. However, evolution-
- fgas and fel [kgCO2 /kWh] are carbon intensity factors for gas and
ary algorithms have not yet been widely applied for the design of
electricity respectively; these parameters relate the amount of
the façades of energy efficient buildings, even if they represent an
carbon emissions to the energy consumption of gas (for heating)
important tool in developing these optimisations [2].
and electricity (for cooling and artificial electricity). These values
The study of evolutionary algorithms has shown the great help
have been assumed to be [5]: fgas = 0.194 kgCO2 /kWh and fel =
of statistical models in driving the evolution of the best solution
0.422 kgCO2 /kWh;
in large and complex search spaces [8,9]: Artificial Neural Net-
- H [–] is the overall efficiency of the heating system, which has
works (ANN) have been adopted to predict the whole solution
been assumed to be 0.89 [5];
space, calculated by means of time-consuming simulations. In this
- COP [–] is the coefficient of performance of the cooling system:
approach the optimisation algorithm can evolve towards the opti-
for London an average value of 3.4 has been assumed [5];
mal solutions by using the Response Surface Approximation (RSA)
- For the lighting system, a luminous efficacy of 75 lm/W has been
developed by the ANN.
assumed [12].
This paper introduces a novel approach for the design of façades
of energy efficient buildings and describes the development of an
evolutionary approach based on statistical models with the form of The optimisation of the single-objective function depends on the
predictive neural networks. This approach has been very successful weights that are assigned, a priori, to the different criteria, like the
in other fields in reaching the optimal solutions while using very constant COP of the cooling system or the carbon intensity factors.
limited computational resources [10].
The aim of this paper is to apply the optimisation procedures for
the design of a typical curtain wall façade. The evolutionary model 2.2. Multi-objective optimisation
based design has been developed according to two approaches,
which will be shown and compared: The multi-objective algorithm aims to optimise each criterion
separately, so there is no need to assume any weights. In this paper
the Pareto approach has been adopted [13]; it consists in deriving a
- a single-objective optimisation, which consists in designing a
curve (the Pareto front) made up of all the non-dominated solutions
building façade leading to the minimisation of the operational
of the search space, where a solution X1 is said to be non-dominated
carbon emissions;
by the solution X2 if both the following conditions are satisfied [13]:
- a multi-objective optimisation, which identifies the trade-off
curve in optimising energy consumption for cooling and artificial
lighting. 1. the solution X1 is no worse than X2 in all objectives;
2. the solution X1 is strictly better than X2 in at least one objective.
The approaches have been carried out by coupling the sim-
ulation tool EnergyPlus to a specifically developed optimisation Potentially, it is possible to consider several objectives simulta-
algorithm: the Evolutionary Neural Network Design (ENN-Design). neously; in this paper two objectives are considered since for an
office building there are two main sources of energy consumption:
2. Optimised design of energy efficient building façades the cooling system and the artificial lighting. The most difficult
challenge for a design team is often, in fact, a proper compro-
Optimisation procedures can be effectively adopted in order to mise between these two contradictory criteria. Therefore the Pareto
identify optimal – or very good – solutions for single and multi- front for this case will be the trade-off curve between the annual
objective problems. It is possible to choose between two strategies energy consumption for cooling and for lighting (i.e. the parameters
for the optimisation process [6,11]: a priori or a posteriori. The a QC and QL previously presented).
G. Zemella et al. / Energy and Buildings 43 (2011) 3297–3302 3299
Table 1 artificial lights is adjusted in order to reach the target level for
Solar performance of the considered glass types.
illuminance. It should be noted that internal blinds may be nec-
Glass type Variable g-Value LT essary in order to avoid glare issues, and this may have an impact
Planibel TopN Type 0 0.58 76% on the actual levels of daylighting. Since the focus of the paper
Iplus Sun Type 1 0.43 69% is on the optimisation methodology rather than on the detailed
CoolLite SKN 165B Type 2 0.34 58% energy simulation, the presence of the internal blinds has not
Ipasol 50/25 Type 3 0.28 48% been considered.
Ipasol 40/21 Type 4 0.22 40%
- Façade thermal transmittance: the opaque spandrel elements
have been assigned a centre pane U-value of 0.35 W/m2 K, and
for the double glazed unit a centre pane U-value of 1.1 W/m2 K
3. Problem description
has been assumed. Therefore, the overall thermal transmittance
of the façade varies between 0.9 W/m2 K and 1.3 W/m2 K for a
This study considers an office building located in London and,
vision area of 20% and 80% respectively, taking into account the
more specifically, a typical façade module for the west elevation,
thermal bridges due to the framing system.
made out of a visual area inserted between two opaque panels;
- Internal gains: the assumptions for the internal gains have been
in order to limit the solar gains, the presence of external shading
based on the indications reported in [15]; for people doing office
devices (i.e. a horizontal overhang and vertical fins) has been con-
activity, the internal gains are considered as 10 W/m2 as sensible
sidered. The façade module has been assumed to be 3 m wide and
gains and 8 W/m2 as latent gains. For the electrical equipment
3.8 m high; the depth of the module has been chosen in accordance
sensible heat gains of 15 W/m2 are considered. The maximum
with the prescriptions of [14], which identifies the first 4.5 m as the
amount of heat gains for artificial lighting is 12 W/m2 : this value is
perimeter area of the building, which is influenced by the façade
then modulated considering the amount of available daylighting.
performance.
The following variables have been considered for the optimisa-
tion process: 4. Evolutionary Neural Network Design optimisation
- visual area: the percentage varying between 20% and 80% of the The optimisation of complex combinatorial systems may be
façade surface in 43 steps; very difficult to be addressed under standard optimisation tech-
- horizontal overhang: the depth varying between 0.1 m and 1.0 m niques. An evolutionary technique is based on the Darwinian
in steps of 0.1 m; theory [17], which evolves populations of candidate solutions using
- vertical fins: the depth varying between 0.1 m and 1.0 m in steps genetic operators such as selection, recombination and mutation.
of 0.1 m; Genetic Algorithms [18], Evolutionary Algorithms [19] and other
- vertical fins: the inclination varying between 10◦ and 90◦ in steps approaches implement these mechanisms, inspired by the biolog-
of 10◦ ; ical evolution. Accordingly to this approach, a first population of
- the glass installed: five different coatings available in the mar- solutions is randomly chosen in the experimental space and eval-
ket have been considered to be applied to the double glazed unit. uated by means of a cost function or fitness function. The first
While the thermal transmittance of the glazed panel is constant population of solutions is then evolved accordingly to evolutionary
(i.e. 1.1 W/m2 K for an argon filled cavity), the solar performance genetic operators, through a sequence of generations.
varies in terms of total solar energy transmittance (g-value) and In general, recombination and mutation ensure diversity and
visual light transmittance (LT). The performance of the consid- variability inside the population (exploration of the space), while
ered glass types are summarised in Table 1. Please note that the selection boosts the quality of the response (exploitation of the best
table reports the performance for double glazed units, where the solution).
high performance coating is applied on the inner surface of the For the study addressed in this paper, a new evolutionary pro-
external glass pane. For the inner pane a clear laminated glass has cedure has been created, the Evolutionary Neural Network Design.
been considered. The values have been calculated in accordance This procedure is based on Artificial Neural Networks models [20]
with the European Standard EN 410. and the evolutionary approach. ENN-Design has been used both in
single-objective and in multi-objective optimisation showing good
In order to find the best combination of the variable levels, performance in discovering optimum solutions in a fast and reliable
193,500 potential solutions should be evaluated: all the possible way.
solutions have been tested in order to have a clear indication about
the actual performance of the optimisation process. 4.1. Single-objective optimisation
The thermal model has been built with the software tool Energy-
Plus calculating the annual energy consumption for heating, cooling The ENN-Design is based on a non-linear statistical model and
and artificial lighting. The thermal model refers to an office build- the search strategy evolves according to the information collected
ing space and the following assumptions have been considered, in throughout the different steps of the evolutionary procedure.
accordance with [15]: A three-layered neural network topology has been adopted,
where a sigmoidal activation function maps the input layer to the
- Internal environment conditions: the internal air temperature is hidden layer and a linear activation function maps the hidden layer
22 ◦ C and 24 ◦ C for winter and summer respectively; during the to the output layer [20].
hours when the space is not occupied, the temperature can drop The Darwinian paradigm, in this case, has been adopted to
in winter down to 12 ◦ C and in summer it can go up to 28 ◦ C. evolve the structure of the neural network by mutating the struc-
The design value that has been set for the illuminance is 500 lux tural parameters, generation by generation.
[16]: the illuminance level is evaluated at a reference point in Initially, a very small population is randomly selected: 50 combi-
the centre of the room and at a height of 0.8 m (i.e. the typical nations of variables are detected from 193,500 potential solutions.
desk level). This is to simulate the presence of a sensor connected The response is then calculated by means of EnergyPlus analysis
to the building management systems: considering the amount according to Eq. (1). For this population it is possible to build a set
of daylighting reaching the sensor, the power provided by the of neural network models with different elements (neurons) in the
3300 G. Zemella et al. / Energy and Buildings 43 (2011) 3297–3302
5. Results
Fig. 3. Increase of the number of solutions within the optimality regions. Fig. 5. Impact of different glazing types on the façade performance.
References
[1] N. Stern, The Economics of Climate Change—The Stern Review, Cambridge Uni-
versity Press, Cambridge, 2007.
[2] V. Sambou, B. Lartigue, F. Monchoux, M. Adj, Thermal optimization of mul-
tilayered walls using genetic algorithms, Energy and Buildings 41 (2009)
Fig. 4. Multi objective: evaluation of the performance of the algorithm. 1031–1036.
3302 G. Zemella et al. / Energy and Buildings 43 (2011) 3297–3302
[3] E. Znouda, N. Ghrab-Morcos, A. Hadj-Alouane, Optimization of Mediterranean [11] D.A. Van Veldhuizen, G.B. Lamont, Multi-objective evolutionary algorithms:
building design using genetic algorithms, Energy and Buildings 39 (2007) analyzing the state-of-the-art, Evolutionary Computation 8 (2) (2000)
148–153. 125–147.
[4] D.A. Coley, S. Schukat, Low-energy design: combining computer-based optimi- [12] CIBSE, Guide F. Energy Efficiency in Buildings, The Chartered Institution of
sation and human judgement, Building and Environment 37 (2002) 1241–1247. Building Services Engineers, 2004.
[5] M. Kragh, A. Simonella, The missing correlation between thermal insulation and [13] K. Deb, Multi-objective Optimization Using Evolutionary Algorithms, John
energy performance of office buildings, in: Proceedings of the International Wiley & Sons, New York, 2001.
Conference on Building Envelope Systems & Technology (ICBEST), Bath, UK, [14] British Council for Offices, Guide 2005 Best Practice in the Specification for
2007. Offices, BCO, 2009.
[6] J.A. Wright, H.A. Loosemore, R. Farmani, Optimization of building thermal [15] CIBSE, Guide A. Environmental Design, The Chartered Institution of Building
design and control by multi-criterion genetic algorithm, Energy and Buildings Services Engineers, 2006.
34 (2002) 959–972. [16] The Society of Light and Lighting, Lighting Guide 7: Office Lighting, The Char-
[7] A.J. Ardakani, F.F. Ardakani, S.H. Hosseinian, A novel approach for optimal chiller tered Institution of Building Services Engineers, 2005.
loading using particle swarm optimization, Energy and Buildings 40 (2008) [17] C. Darwin, On the Origin of Species, John Murray, London, 1859.
2177–2187. [18] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learn-
[8] L. Magnier, F. Haghighat, Multiobjective optimization of building design using ing, Addison–Wesley, New York, 1989.
TRNSYS simulations, genetic algorithm, and Artificial Neural Network, Building [19] T. Bäck, Evolutionary Algorithms in Theory and Practice: Evolution Strate-
and Environment 45 (2010) 739–746. gies, Evolutionary Programming Genetic Algorithms, University Press, Oxford,
[9] L. Zhou, F. Haghighat, Optimization of ventilation system design and opera- 1996.
tion in office environment. Part I. Methodology, Building and Environment 44 [20] B.D. Ripley, Pattern Recognition and Neural Networks, University Press, Cam-
(2009) 651–656. bridge, 1996.
[10] D. De March, M. Forlin, D. Slanzi, I. Poli, An evolutionary predictive approach [21] D.E. Rumelhart, G.E. Hinton, R.J. Williams, Learning internal representations by
to design high dimensional experiments, in: R. Serra, I. Poli, M. Villani (Eds.), error propagation, in: D.E. Rumelhart, J.L. McClelland (Eds.), Parallel Distributed
Artificial Life and Evolutionary Computation: Proceedings of WIVACE, World Processing, MIT Press, Cambridge, MA, 1986, pp. 318–362.
Scientific Publishing Company, Singapore, 2008, pp. 81–88.