304 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO.

1, JANUARY 2017

New Multistage and Stochastic Mathematical Model

for Maximizing RES Hosting Capacity—Part I:
Problem Formulation
Sérgio F. Santos, Desta Z. Fitiwi, Student Member, IEEE, Miadreza Shafie-Khah, Member, IEEE,
Abebe W. Bizuayehu, Member, IEEE, Carlos M. P. Cabrita, and João P. S. Catalão, Senior Member, IEEE

Abstract—This two-part work presents a new multistage and i/Ωi Index/set of buses.
stochastic mathematical model, developed to support the decision- g/Ωg / ΩDG Index/set of DGs.
making process of planning distribution network systems (DNS)
for integrating large-scale “clean” energy sources. Part I is de- k/Ωk Index/set of branches.
voted to the theoretical aspects and mathematical formulations in s/Ωs Index/set of scenarios.
a comprehensive manner. The proposed model, formulated from t/Ωt Index/set of time stages.
the system operator’s viewpoint, determines the optimal sizing, w/Ωw Index/set of snapshots.
timing, and placement of distributed energy technologies (partic- ς/Ως Index/set of substations.
ularly, renewables) in coordination with energy storage systems
and reactive power sources. The ultimate goal of this optimization b) Parameters
work is to maximize the size of renewable power absorbed by the Cij Cost of branch i-j (€).
m in m ax
system, while maintaining the required/standard levels of power Ees,i , Ees,i Energy storage limits (MWh).
quality and system stability at a minimum possible cost. From the ERg , ERgE ,
N
Emission rates of new and
methodological perspective, the entire problem is formulated as a
mixed integer linear programming optimization, allowing one to ERςSS existing DGs, and energy purchased at
obtain an exact solution within a finite simulation time. Moreover, sub-stations, respectively (tons of CO2
it employs a linearized ac network model which captures the inher- equivalent—tCO2 e/MWh).
ent characteristics of electric networks and balances well accuracy gk , bk , Skm ax Conductance, susceptance and flow
with computational burden. The IEEE 41-bus radial DNS is used limit of branch k (Ω, Ω, MVA).
to test validity and efficiency of the proposed model, and carry out
the required analysis from the standpoint of the objectives set. Nu- ICg ,i , ICk , Investment cost of DG, line, energy
merical results are presented and discussed in Part II of this paper ICes,i , ICc,i storage system and capacitor banks,
to unequivocally demonstrate the merits of the model. respectively.
Index Terms—Distributed generation, distribution network L Total number of linear segments (€).
systems, energy storage systems, integrated planning, stochastic LTg , LTk , LTes, LTc Lifetimes of DG, line, energy storage
programming, variability and uncertainty. system and capacitor banks, respec-
tively (years).
NOMENCLATURE M Cc , M Ces Maintenance cost of capacitor bank and
a) Sets/Indices energy storage system per year (€).
c/Ωc Index/set of capacitor banks. M CgN , M CgE Maintenance costs of new and existing
es/Ωes Index/set of energy storages. DGs per year (€).
M CkN , M CkE Maintenance costs of new and existing
Manuscript received January 17, 2016; revised May 16, 2016 and June 29, branch k per year (€).
2016; accepted August 4, 2016. Date of publication August 5, 2016; date of M Pk , M Qk Big-M parameters associated to ac-
current version December 14, 2016. This work was supported by FEDER funds tive and reactive power flows through
through COMPETE and by Portuguese funds through FCT, under Projects
FCOMP-01-0124-FEDER-020282 (Ref. PTDC/EEA-EEL/118519/2010) and branch k, respectively.
UID/CEC/50021/2013, and in part by the EU Seventh Framework Programme Ni , Nς Number of buses and substations,
FP7/2007-2013 under Grant Agreement No. 309048. The work of S. F. San- respectively.
tos was supported by the UBI/Santander Totta doctoral incentive grant in the
Engineering Faculty. Paper no. TSTE-0053-2016. OCgN,i,s,w ,t , Operation cost of unit energy pro-
S. F. Santos, D. Z. Fitiwi, M. Shafie-Khah, A. W. Bizuayehu, and J. P.S. OCgE,i,s,w ,t duction by new and existing DGs
Catalão are with the INESC TEC and Faculty of Engineering, University of
Porto, Porto 4200-465, Portugal, with the C-MAST, University of Beira Interior,
(€/MWh).
ch,m ax dch,m ax
Covilhã 6201-001, Portugal, and also with the INESC-ID, Instituto Superior Pes,i , Pes,i Charging and discharging power limits
Técnico, University of Lisbon, Lisbon 1049-001, Portugal (e-mail: sdfsantos@ of a storage system (MW).
gmail.com; dzf@ubi.pt; miadreza@gmail.com; buzeabebe@gmail.com;
catalao@ubi.pt).
Psol,h Hourly solar PV output (MW).
C. M. P. Cabrita is with the CISE—Electromechatronic Systems Research Pr Rated power of a DG unit (MW).
Centre, University of Beira Interior, Covilha 6201-001, Portugal (e-mail: Pwnd,h Hourly wind power output (MW).
cabrita@ubi.pt).
Color versions of one or more of the figures in this paper are available online
Q0c Rating of minimum capacitor bank.
at http://ieeexplore.ieee.org. Rc Certain radiation point (often taken to
Digital Object Identifier 10.1109/TSTE.2016.2598400 be 150 W/m2 ).
1949-3029 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY 305

Rh Hourly solar radiation (W/m2 ). θi , θj Voltage angles at nodes i and j

rk , xk Resistance and reactance of branch k, (radians).
respectively. d) Functions
Rstd Solar radiation in standard condition ECtSS Expected cost of energy purchased
(usually set to 1000 W/m2 ). from upstream (€).
vci Cut-in wind speed (m/s). EN SCt Expected cost of unserved power (€).
vco Cut-out wind speed (m/s). EmiCtDG Expected emission cost of power
vh Observed/sampled hourly wind speed production using DG (€).
(m/s). EmiCtN , EmiCtE Expected emission cost of power pro-
Vnom Nominal voltage (kV). duction using new and existing DGs,
vr Rated wind speed (m/s). respectively (€).
Zij Impedance of branch i-j (Ω). EmiCtSS Expected emission cost of purchased
αl , βl Slopes of linear segments. power (€).
ch
ηes dch
, ηes Charging and discharging efficiencies InvCtCAP , M ntCtCap NPV investment/maintenance cost of
of a storage system (%). capacitor banks (€).
λC O2 e
s,w ,t Price of emissions (€/tCO2 e). InvCtDG , M ntCtDG , NPV investment/maintenance/
λdch
es,i,s,w ,t Cost of energy discharged from storage ECtDG energy cost of DGs, respectively (€).
system (€/MWh). InvCtES , M ntCtES , NPV investment/maintenance/
λςs,w ,t Price of electricity purchased from expected energy
upstream (€/MWh). ECtES cost of an energy storage system,
ρs , πw Probability of scenario s and weight (in respectively (€).
hours) of snapshot group w. InvCtLN , M ntCtLN NPV investment/maintenance cost of a
υs,w ,t Penalty for unserved power (€/MWh). distribution line (€).
c) Variables I. INTRODUCTION
i i
Ds,w ,t , Qs,w ,t Active and reactive power demand at
node i (MW, MVAr). A. Motivation and Aims
Ees,i,s,w ,t Stored energy (MWh). OWADAYS, the issue of integrating distributed genera-
ch
Ies,i,s,w
N E
dch
,t , Ies,i,s,w ,t
Pg ,i,s,w ,t , Pg ,i,s,w ,t
Charge-discharge indicator variables.
Active power produced by new and
N tions (DGs) (renewable DGs, in particular) is globally
gaining momentum because of several techno-economic and
existing DGs (MW). environmental factors. Since recent years, the size of DGs inte-
PςSS SS
,s,w ,t , Qς ,s,w ,t Active and reactive power imported grated into distribution systems has been increasing. This trend
from grid (MW). is more likely to continue in the years to come because it is
Pk , Qk , θk Active and reactive power flows, and now widely accepted that DGs bring wide-range benefits to the
voltage angle difference of link k, system. However, given the current set-up of distribution net-
respectively (MW, MVAr, radians). works (which are generally passive), large-scale DG integration
pk ,s,w ,t,l , qk ,s,w ,t,l Step variables used in linearization of is not technically possible because this brings about tremendous
quadratic flows (MW, MVAr). challenges to the system operation, especially in undermining
P Lk , QLk Active and reactive power losses, the power system quality and stability.
respectively (MW, MVAr). Such challenges/limitations are expected to be alleviated
P Lς ,s,w ,t , QLς ,s,w ,t Active and reactive losses at substation when distribution networks undergo the anticipated evolution-
ς (MW, MVAr). ary process from passive to active networks or smart grids. This
ch dch
Pes,i,s,w ,t , Pes,i,s,w ,t Power charged to and discharged from transition is expected to result in a system that is adequately
storage system (MW). equipped with appropriate technologies, state-of-the-art solu-
Qci,s,w ,t Reactive power injected by capacitor tions and a new operational philosophy that is totally different
bank at node i (MVAr). from the current “fit and forget” approach. This is expected to
QN E
g ,i,s,w ,t , Qg ,i,s,w ,t Reactive power consumed/produced offer sufficient flexibility and control mechanism in the system.
by new and existing DGs (MVAr). Nevertheless, the process is not straightforward as it demands
ug ,i,t , uk ,t Utilization (status) variables of DG exceptionally huge investments in smart-grid technologies and
and line (1 if it is connected; 0, concepts to fully automate the system, and this should be accom-
otherwise). panied by a new operational philosophy. Therefore, the whole
V i , Vj Voltage magnitudes at nodes i and j transformation process (i.e., the transformation of current dis-
(kV). tribution systems to full-scale smart-grids) might be very slow,
xg ,i,t , xes,i,t , Investment variables for DG, energy and its realization might take several decades.
xc,i,t , xk ,t storage system, capacitor banks and However, given the techno-economic factors and global con-
distribution lines. cerns about environmental issues, the integration of renewable
δi,s,w ,t Unserved power at node i (MW). energy sources (RES) cannot be postponed. It is likely that the
306
integration of DGs in distribution systems will go ahead along
with smart-grid enabling technologies that have the capability
to alleviate the negative consequences of large-scale integration
of DGs. In other words, in order to facilitate (speed up) the
much-needed transformation of conventional (passive) DNSs
and support large-scale RES integration, different smart-grid
enabling technologies such as reactive power sources, advanced
switching and storage devices are expected to be massively de-
ployed in the near term.
To this end, developing strategies, methods and tools to max-
imize the penetration level of DGs (particularly, RES) has be-
come very crucial to guide such a complex decision-making
process.
In this respect, this work focuses on the development of
multi-stage mathematical models to determine the optimal siz-
ing, timing and placement of energy storage systems (ESS) and
reactive power sources as well as that of RES in distribution
networks. The ultimate goal of this optimization work is to
maximize the RES power absorbed by the system at a minimum
cost while maintaining the power quality and stability at the
required/standard levels.
The problem is formulated from the system perspective
(i.e., in centralized planning framework). In a deregulated
environment and from the smart-grid context where the current
regulatory and technical challenges are expected to be fully
resolved, planning will most likely involve distributed decision-
making processes. One may rightfully argue that the investment
decisions obtained from a coordinated planning model may not
be implemented in reality. This is because distributed decisions
often lead to sub-optimal solutions, which may be different from
the ones obtained by the centralized planning model. However,
this does not mean that the outcomes of the coordinated plan-
ning model cannot be used. For instance, these outcomes can be
regarded as the best investment targets. Given these targets, dis-
tributed decisions (solutions) can be systematically made to ap-
proach one or more of these targets (for example, via incentive or
market-based mechanisms). From another perspective, in the ab-
sence of “attractive” market environment (seen from the private
investors), distribution network systems may not see significant
breakthrough when it comes to investments in DGs and ESS. In
this case, DSOs may, instead, be given additional roles and re-
sponsibilities that include investing in DGs and ESSs in coordi-
nation with network investments. DSOs may also oversee invest-
ments in DGs and ESSs. In addition, DSOs may also be required
to manage these assets in a coordinated manner to keep the sys-
tem integrity, stability and power quality at the required/standard
levels. Another issue which explains the versatility of the co-
ordinated approach is related to the realization of smart-grids.
Even if there is a general consensus on the smartification of
power systems (distribution networks, in particular), and there
are signs that some systems are evolving into smart-grids, the
whole process is going to probably take very long time. Based
on the aforementioned reasons, a centralized (coordinated)
approach of the planning problem can provide vital solutions.
B. State-of-the-Art Literature Review and Background
Reducing fossil fuel dependence and mitigating climate
change have led to an increased pressure to change the cur-
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017
rent generation paradigm. The compounded effect of increasing
demand for electricity, environmental and climate change con-
cerns is triggering a policy shift all over the world, especially
when it comes to energy production. Integration of DG, partic-
ularly, RES, in electric distribution network systems is gaining
momentum. In particular, the recent developments in a climate
change conference held in Paris (COP21) are expected to accel-
erate renewable integration. It is highly expected that large-scale
DG integration will be one of the solutions capable of mitigating
the aforementioned problems and overcoming the challenges.
Because of this, Governments of various nations have introduced
targets to achieve large-scale integration of DGs. In particular,
in the European Union, which strongly advocates the impor-
tance of integrating renewables, RES are expected to cover 20%
and 50% of the overall energy consumption by 2020 and 2050,
respectively.
DGs offer a more environmentally friendly option through
great opportunities with renewable-enabling technologies such
as wind, photovoltaic, biomass, etc. RES are abundant in nature,
which, under a favorable RES integration policy and incentive
mechanism, makes it attractive for the large-scale power gen-
eration sector. Nevertheless, there is no rule or partial rule on
the DG unit’s connection; typically, these are traditionally con-
nected at the end of radial feeder systems or nodes with greater
load on the distribution system. This is realized often with sev-
eral impeding restrictions put in place to alleviate the negative
consequences of integrating DGs in the system.
The optimal and dynamic planning of the DG placement and
sizing is becoming extremely important for energy producers,
consumers and network operators in technical and economic
terms. There are many studies in the literature on this topic, yet
most of them only consider the optimal location of a single DG
unit or do not consider simultaneously positioning and sizing
RES units, mainly due to their high dispatch unpredictability.
The increase in DG penetration increases the uncertainty and the
fluctuations of power production. If the placement and proper
sizing is not taken into account, the benefits of DG integration
can be lost, leading to inefficient operation and increasing the
electricity cost and energy losses.
Another major concern with the wide DG penetration is sys-
tem reliability. The penetration of distributed systems can result
in the degradation of power quality, particularly in cases of
slightly meshed networks [1] or microgrids. In this paradigm,
the use of ESS has been seen as one of the viable options to
mitigate the aforementioned concerns.
The DG allocation and sizing subject have attracted special
interest from researchers in recent years. An excellent review
of previous works related to this subject area, published prior to
the year 2013, is presented in [2].
In [2] and [3], an analysis of several techniques used on the
DG impact assessment in the electrical system is presented.
Most of these techniques analyze the distribution system to
determine rules that can be used for DG integration [4]–[8].
Important issues related to the connection of DG units are
the network topology, DG capacity and suitable location; be-
cause, each bus in the system has an optimal level of DG in-
tegration. If the value surpasses this level, system losses can
increase [9], [10].
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY
Recently, several methods have been proposed for planning
and operation or in some cases for both location and sizing of
DGs in the distribution system. In general, these methods can be
classified as heuristic [11]–[26], numerical [27]–[34] and ana-
lytical [35], [36] based methods. Heuristic based methods apply
advanced artificial intelligence algorithms, such as genetic algo-
rithms (GA) [11]–[14], particle swarm optimization [15]–[19],
harmony search [20], [21] and big bang crunch [22]–[24].
Numerical methods are algorithms that seek numerical results
for different problems in particular to the problem in question.
Some of the most recently works use nonlinear mixed inte-
ger programming [27]–[29], mixed integer linear programming
(MILP) [30], [31], quadratic programming [32] and ac optimal
power flow-based [33].
The exhaustive search methods seek the optimal DG location
for a given DG size under different load models. Therefore,
these methods fail to represent accurately the behavior of the
DG optimization problem involving two discrete variables, both
for optimum DG size and optimal DG location. In [35], Atwa
and El-Saadany present a technique with a probabilistic basis
for determining the capacity and optimal placement of wind
DG units to minimize energy loss in the distribution system. A
sensitivity analysis is presented in [36] for DG placement and
sizing in the network.
Despite many studies in the literature on areas related to DG
placement and sizing problem, most of them only consider the
optimal location of a single DG unit, mostly of conventional
DGs. The simultaneous consideration of placement, timing and
sizing of DG units (especially RES), along with the placement,
timing and sizing of smart-grid enabling technologies, seems
to be far from being addressed in the literature. The increase
in RES-based DG penetration increases the uncertainty and
the fluctuations of the system production. If the placement and
proper sizing is not taken into account, the benefits of DG in-
tegration may not be exploited; instead, this may result in the
degradation of system efficiency, increased cost of electricity
and energy losses. Another major concern with the wide-range
DG penetration is system reliability. However, the simultaneous
investment planning of DGs, ESSs and reactive power sources is
expected to significantly alleviate these challenges and increase
the penetration level of RES-based DGs.
C. Contributions
The novel contributions of this work are threefold:
1) A new multi-stage and stochastic optimization model is
proposed, which simultaneously considers the integration
of smart-grid enabling technologies such as ESSs, reac-
tive power sources, and network reinforcements to support
large-scale RES-based DG integration. To the best of the
authors’ knowledge, this is the first time such a joint op-
timization framework is proposed to such a problem.
2) A MILP programming is proposed based on a linearized
ac network model which captures well the inherent char-
acteristics of electric networks. To the best of the authors’
knowledge, this network model has not been used in the
literature related to the distribution network systems.
307
Fig. 1. Illustration of variability and uncertainty in wind power output.
3) The solution framework simultaneously finds the optimal
sizing, time and placement of DGs, ESSs and reactive
power sources in distribution networks, which is another
innovative step forward.
D. Paper Organization
The rest of this paper is organized as follows. Section III
presents a brief description of how the uncertainty and the
variability pertaining to variable energy sources and electricity
demand are handled in the planning process. The mathematical
formulation and detailed description of the proposed model
are presented in Section IV. The last section draws some
conclusions.
II. UNCERTAINTY AND VARIABILITY MANAGEMENT
A. Description of Terminologies
The terminologies uncertainty and variability are sometimes
incorrectly used interchangeably in the literature despite the
fact that they are different. Variability, as defined in [37], refers
to the natural variation in time of a specific uncertain parame-
ter, whereas uncertainty refers to “the degree of precision with
which the parameter is measured” or predicted. We follow these
terminologies in our paper when referring to operational vari-
ability and uncertainty, which are introduced by model param-
eters. For example, wind power output is characterized by both
phenomena; its hourly variation corresponds to the variability
while its partial unpredictability (i.e., the error introduced in
predicting the wind power output) is related to uncertainty. The
schematic illustration in Fig. 1 clearly distinguishes both termi-
nologies. As illustrated in this figure, the hourly differences in
wind power outputs are due to the natural variability of primary
energy source (wind speed); whereas, the likelihood of having
different power outputs at a given hour is a result of uncertainty
(partial unpredictability) in wind speed.
Other terminologies used in this paper are snapshot, sce-
nario and time stage. A snapshot refers to an hourly operational
situation. Alternatively, it can be understood as a demand—
generation pattern at a given hour. A scenario, on the other
hand, denotes the evolution of an uncertain parameter over a
given time horizon (often yearly). For example, the hourly vari-
ations of wind power production and electricity consumption
collectively form a group of snapshots; whereas, the annual de-
mand growth (which is subject to uncertainty) and RES power
output uncertainty are represented by a number of possible
308
storylines (scenarios) [38]. Time stage (also referred to as deci-
sion stage) stands for the yearly decision stages throughout the
planning horizon. The length of planning horizon in the present
work three years, which is divided into yearly decision stages.
B. Uncertainty and Variability
There are various sources of uncertainty and variability in a
distribution systems planning problem, particularly with inter-
mittent renewable sources. These are related to the variability in
time and the randomness of operational situations [39]. In addi-
tion, there are other uncertainties mostly related to the long-term
electricity, carbon and fuel prices, rules, regulations and poli-
cies, etc. To account for uncertainty in demand, two demand
scenarios are formed by assuming a ±5% prediction error mar-
gin from a long real-life demand profile (which is 8760 long).
This gives a total of three demand profile scenarios, which are
kept the same throughout the planning horizon. Due to the lack
of sufficient historical data, first, synthetic hourly wind speed
and solar radiation series (20 for each) are generated using the
methods in [1] and [2], respectively, considering the autocorrela-
tions and diurnal patterns of each parameter. Then, the average
wind speed and solar radiation profiles are determined from
the generated series. The power outputs corresponding to each
wind speed and solar radiation are then determined by plugging
in these values in the respective power curves given by equations
(1) and (2). Note that these power output samples (snapshots)
cannot be used directly in the planning process because the re-
sulting wind power output profiles may not respect the natural
correlations that exist between them and the average electricity
demand profile. These samples should be readjusted to reflect
the temporal correlations that naturally exist among demand,
solar radiation and wind speed series. To this end, the correla-
tion between wind and solar sources is considered to be −0.3
while that of wind and demand is 0.28, which is in line with the
results in [40]. A correlation of 0.5 is assumed between solar
and demand, according to [41].
Given this desired correlation matrix, the wind and the solar
power output profiles can be transformed into new ones, respect-
ing the correlation among them. Such adjustments in the corre-
lation of data series are performed using Cholesky factorization,
a method used for generating correlated random variables. The
method works as follows. Given a desired correlation matrix R,
and uncorrelated data D, a new data Z whose correlation ma-
trix is R can be generated by simply multiplying the Cholesky
decomposition of R by D.
Then, the hourly wind and solar power output are determined
by plugging in these readjusted series into their corresponding
power curves given by
0, 0 ≤ vh ≤ vci
 
Pr A + Bvh , vci ≤ vh ≤ vr
3
Pwnd,h = . (1)
Pr , vr ≤ vh ≤ vco
0, vh ≤ vco
In the above equation, A and B are parameters represented
by the expressions in [42] and [43]. Similarly, the hourly solar
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017
Fig. 2 Wind and solar PV power output uncertainty characterization example.
power output Psol,h is determined by plugging in the hourly
solar radiation levels in the solar power output expression given
in (2) [44]:
⎧ P R2
⎪
⎪
r h
, 0 ≤ Rh ≤ Rc
⎨ R s t d ∗R c
Psol,h = P R
Rc ≤ Rh ≤ Rstd . (2)
Rstd ,
r h
⎪
⎪
⎩
Pr , Rh ≥ Rstd
In the present work, uncertainty in wind speed and solar radi-
ation is assumed to lead to a ±15% deviation on average from
the corresponding average power output profiles. This approx-
imately translates to ±5% forecasting error in wind speed or
solar radiation. Note that such error or higher is induced even
by the most advanced forecasting tools available today. Based
on the assumptions, two hourly profiles of wind power output
are generated by considering the ±15% margin (one above the
average and one below the average profile). In total, this leads to
three wind power output profiles (including the average profile).
These are defined as wind scenarios. Similarly, to account for
the uncertainty in solar power outputs, two profiles are gener-
ated for each by considering a ±15% uncertainty margin. The
two generated solar power output profiles along with the av-
erage profile form the set of solar scenarios. An illustration of
uncertainty characterization of wind and solar power outputs is
shown in Fig. 2.
Note that each of these scenarios has 8760 snapshots of de-
mand, wind and solar power outputs. These individual scenarios
are combined to form a set of 27 scenarios (3 x 3 x 3), which are
used in the analysis. These scenarios are assumed to be equally
probable; hence, the probability of realization of each scenario
ρs is given by 1/27. To ensure problem tractability, the multi-
dimensional input data (27 x 8760) is clustered into 27X200
groups via a standard clustering technique (k-means algorithm
[45], which has been applied in investment planning problems
as in [46]). Here, each cluster represents a group of similar oper-
ational situations. A representative snapshot, the medoid in this
case, is then selected from each cluster. A weight is assigned
to each representative snapshot, which is proportional to the
number of operational situations in its group. Note that while
clustering such a large number of operational situations in this
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY 309

manner is critically important to guarantee problem tractability,

the chronological orders (and by implication the autocorrela-
tions) of the considered data (time series) are not unfortunately
preserved in the reduced number of operational situations. In
other words, such information is lost during the clustering pro-
cess. In the context of medium- to long-term planning problems,
the impact of such information loss on the planning outcome
may not be significant. However, if this is a concern, the chrono-
logical information can be somehow recovered by methods as
in [47]. A comprehensive analysis on issues related to this will
be addressed in future works by the authors.

III. MATHEMATICAL FORMULATION OF THE PROBLEM

A. Brief Description of the Problem
As mentioned earlier, this work develops an integrated opti-
mization model that simultaneously finds the optimal locations
and sizes of installed DG power (particularly, focusing on wind
and solar), ESS and reactive power sources such as capacitor
banks. The optimal deployment of the aforementioned enabling
technologies should inherently meet the goal of maximizing
the renewable power integrated/absorbed into the system. The Fig. 3. A schematic representation of (a) possible future scenarios’ trajectories
with multiple scenario spots along the planning horizon and (b) a decision
entire model is formulated as a stochastic mixed integer linear structure at each stage.
programming optimization. In addition, instead of the custom-
ary direct current (dc) network models, a linearized ac model
planning horizon, we obtain a single investment solution which
is used here to better capture the inherent characteristics of the
is good enough for all scenarios [38], [48]. Note that while op-
network system. An ideal representation of the network system
erational variables depend on each scenario and snapshot, the
would be to use the full ac network model. However, embedding
investment decision variables only depend on the time stage in-
this model in planning problems is computationally unafford-
dex. This means that the investment solution obtained should
able. Because of their appealing computational performances,
satisfy all conditions in every scenario, making the solution ro-
dc based optimization models are commonly used in distri-
bust against any realization of the considered scenarios. It should
bution systems planning problems. However, the dc network
be noted here that the robustness of the solution is directly related
model does not consider reactive power flows, which can have
with the level of details of uncertainty and variability charac-
significant impact on the planning solutions, such as invest-
terization. Generally, the higher the numbers of snapshots and
ment decisions related to capacitor banks. In addition, voltage
scenarios considered are, the more robust the solution is. How-
magnitudes are often considered to be the same throughout the
ever, there is always a threshold beyond which adding more
system. Voltage angles are also sometimes neglected in dc mod-
snapshots and scenarios does not significantly change the so-
els. All these simplifying assumptions may result in sub-optimal
lution but increases unnecessary computational burden. If the
solutions (often underinvestment) when embedded in planning
scenarios considered in the planning are carefully selected to be
models. The linearized ac model on the contrary acknowledges
representative enough of all possible uncertainty realizations,
the presence of both active and reactive power flows, voltage
then, the robustness and reliability of the solution can be more
magnitude and angle differences among nodes in the system.
guaranteed.
The model captures the physical characteristics of the network
The length of the planning horizon in the present work is
system in a better way when compared to dc network models,
assumed to be three years, which is then divided into yearly
and very close to the ideal ac network model. Computationally
decision stages. In each stage, investment decisions related to
speaking, the linearized ac model is more expensive than the
DGs, ESSs, capacitor banks and lines are made. These deci-
dc network model because it involves a more detailed network
sions can be regarded as here-and-now because such decisions
representation than the dc model. However, embedding the lin-
are independent of any scenario or snapshot. However, opera-
earized ac model in planning problems yields far better solutions
tion variables (such as actual power productions, storage level,
than when using the dc one. Hence, the linearized ac model can
power flows, etc.) depend on scenarios and snapshots, as well
be generally regarded as a bridge between the dc and the ac
as decision stages.
network models. Planning models based on this linearized ac
model are tractable enough, and the results are accurate enough.
The schematic representation in Fig. 3 illustrates the multi- B. Objective Function
stage and multi-scenario modeling framework and the expansion As mentioned earlier, the objective of this work is to max-
solution structure (i.e., Xt ’s, where Xt represents the solution imize RES integration in DNS from the system perspective
vectors of several investment variables). At each stage of the (or, from the distribution system operators’ point of view) by
310
Fig. 4. Illustration of cost components within and outside the planning
horizon.
optimally deploying different smart-grid enabling technologies
at a minimum cost. Here, it is assumed that the DSO owns some
generation sources and ESSs.
The resulting problem is formulated as a multi-stage stochas-
tic MILP an overall cost minimization as an objective (3). The
objective function in (3) is composed of net present value (NPV)
of five cost terms each multiplied by a certain relevance factor
αj ∀j ∈ {1, 2, . . . , 5}. Note that, in this work, all cost terms
are assumed to be equally important; hence, these factors are
set equal. However, depending on the relative importance of the
considered costs, different coefficients (relevance factors) can
be adopted in the objective function. Note that all cost terms
have the same units (Euros). In reality, the objective function
is one: the total cost in the system which is the sum of vari-
ous cost components (operation, maintenance, emission, invest-
ment, etc.). However, a decision-maker may not be interested in
some of these costs, for instance, because their values (and/their
expected influences on the solution) are negligible compared
with others. Such cost terms would then have their relevance
factors set to zero. It is for this sole purpose that the relevance
factors (which should not be confused with weights like in the
Pareto-type optimization) are included in the formulation.
In the present work, a perpetual planning horizon [49] is as-
sumed when formulating the integrated planning problem, as
in [50]. This is purposely done to balance different cost terms
within and outside the actual planning horizon. To further clarify
this, consider the illustrative example in Fig. 4. It is understood
that investments are made in a specific year within the planning
horizon (the second year in this case) and the investment costs
are prorated throughout its lifetime, i.e., distributed into equal
payments among the years within the life span of the asset.
However, the maintenance and operation (O&M) related costs
are incurred every year within and after the planning horizon.
 (1 + r)−t
T InvC = (InvCtDG + InvCtLN + InvCtES + InvCtCAP )
t
r
t∈Ω
N P V of in v estm en t cos t
 −t
TMC = (1 + r) (M ntCtDG + M ntCtLN + M ntCtES + M ntCtCap )
t∈Ω 
N P V of m ain ten an ce costs
(1 + r)−T
+ (M ntCTDG + M ntCTLN + M ntCTES + M ntCTCap )
r
N P V of m ain ten an ce costs in cu r r ed af ter stag e T
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017
To balance these cost terms, a perpetual planning horizon, i.e.,
an endless payment of fixed payments is assumed. Based on
the finance theory [49], the present value of perpetuity, which
is the sum of the net worth of infinite annual fixed payments,
is determined by dividing the fixed payment at a given period
by the interest rate r. Based on this, the O&M costs include the
associated annual costs within (part I) and outside the planning
horizon (part II). The latter (part II) are determined by the per-
petuity of the costs in the last planning stage updated by NPV
factor in this case (1 + r)−3 . Note that after the lifetime of a
given asset elapses, investments will be made in the same asset
with the same cost, leading to a seemingly perpetual planning
horizon.
The first term in (3), T InvC, represents the total investment
costs under the assumption of perpetual planning horizon [49].
In other words, “the investment cost is amortized in annual
installments throughout the lifetime of the installed component”,
as is done in [50]. Here, the total investment cost is the sum of
investment costs of new and existing DGs, feeders, ESS and
capacitor banks, as in (4).
The second term, T M C, in (3) denotes the total maintenance
costs, which is given by the sum of individual maintenance costs
of new and existing DGs as well as that of feeders, ESS and ca-
pacitor banks in the system at each stage and the corresponding
costs incurred after the last planning stage, as in (5). Note that
the latter costs depend on the maintenance costs of the last plan-
ning stage. Here, a perpetual planning horizon is assumed. The
third term T EC in (3) refers to the total cost of energy in the
system, which is the sum of the cost of power produced by new
and existing DGs, purchased from upstream and supplied by
ESS at each stage as in (6). Equation (6) also includes the total
energy costs incurred after the last planning stage under a per-
petual planning horizon. These depend on the energy costs of
the last planning stage. The fourth term T EN SC represents the
total cost of unserved power in the system and is calculated as
in (7). The last term T EmiC. gathers the total emission costs in
the system, given by the sum of emission costs for the existing
and new DGs as well that of power purchased from the grid
at the substations. equation (4), (5) as shown at the bottom of
the page.
M inimize T C = α1 ∗ T InvC + α2 ∗ T M C + α3 ∗ T EC
+ α4 ∗ T EN SC + α5 ∗ T ImiC (3)
(4)
(5)
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY 311

 (xg ,i,t − xg ,i,t−1 ) in Eq. (9). All the differences for this particu-
T EC = (1 + r)−t (ECtDG + ECtES + ECtSS ) lar example are (xg ,i,1 − xg ,i,0 ) = 0, (xg ,i,2 − xg ,i,1 ) = 1, and
t∈Ω t
(xg ,i,3 − xg ,i,2 ) = 0, which indicates that the investment cost is
N P V of oper ation costs considered only once at the second stage. Instead of defining the
−T variable xg ,i,t as a binary variable, one may allow it to have any
(1 + r)
+ (ECTDG + ECTES + ECTSS ) (6) integer value as far as it is deemed optimal. In this case, for the
r above example, suppose the optimal solution is to install one DG
N P V oper ation costs in cu r r ed af ter stag e T in the second stage and one more DG in the third stage which
 means {xg ,i,1 ; xg ,i,2 ; xg ,i,3 } is equal to {0; 1; 2}. Note that xg ,i,3
EN SC = (1 + r)−t EN SCt +
should be equal to 2 because the investment decision made in the
t∈Ω t
preceding stages should be also available in the third stage. For
N P V of r eliability costs
this example, (xg ,i,1 − xg ,i,0 ) = 0, (xg ,i,2 − xg ,i,1 ) = 1, and
(1 + r)−T (xg ,i,3 − xg ,i,2 ) = 1, showing that the investment cost each DG
EN SCT (7) is considered only once in the summation.
r
In general, the formulation remains valid regardless of how
N P V r eliability costs in cu r r ed af ter stag e T
the investment variables are defined. Note that investment

T EmiC = (1 + r)−t (EmiCtDG + EmiCtSS ) variables refer to the decision variables corresponding to
t∈Ω t investments in DGs, ESS, capacitor banks and distribution lines
in each of the decision stages along the three-year planning
N P V em ission costs
horizon.
−T
(1 + r) Equation (13) stands for the maintenance costs of new and
+ (EmiCTDG + EmiCTSS ) . (8)
r existing DGs at each time stage. The maintenance cost of a
N P V em ission costs in cu r ed af ter stag e T new/existing feeder is included only when its corresponding
investment/utilization variable is different from zero. Similarly,
The individual cost components in (4)–(8) are computed by
the maintenance costs of new and existing feeders at each stage
the following expressions. Equations (9)–(12) represent the in-
are given by Eq. (14). Equations (15) and (16) are related to the
vestment costs of DGs, feeders, ESS and capacitor banks, re-
maintenance costs at each stage of energy storage and capacitor
spectively. Notice that all investment costs are weighted by the
r (1+r ) L T
banks, respectively
capital recovery factor, (1+r ) L T −1
. The formulations in (9)–(12)
ensure that the investment cost of each asset added to the system   r(1 + r)L T g
is considered only once in the summation. In this regard, there InvCtDG = ICg ,i (xg ,i,t − xg ,i,t−1 ),
g ∈Ω g i∈Ω i (1 + r)L T g − 1
are two issues that need to be taken care of in the formulation.
On one hand, it is required that investment decisions already where xg ,i,0 = 0 (9)
made at a given stage cannot be reversed back (divested) in the  r(1 + r)L T k
subsequent stages. This condition is met by the set of logical InvCtLN = ICk (xk ,t − xk ,t−1 ) ,
constraints described in the following subsection in the model k ∈Ω k
(1 + r)L T k − 1
formulation, for example, xg ,i,t ≥ xg ,i,t−1 . Such a logical con-
where xk ,0 = 0 (10)
straint states that the investment decision at a planning stage
t should be at least equal to the investment decision in the pre-   r(1 + r) LTe s

ceding stage t − 1. In other words, xg ,i,t should be equal to the InvCtES = ICes (xes,i,t −xes,i,t−1 ),
es∈Ω e s i∈Ω i (1 + r)L T e s − 1
investments made in the preceding stages plus the additional
investment in stage t. On the other hand, only the investment where xes,i,0 = 0 (11)
costs for the marginal (additional) investment made at each   r(1 + r)L T c
stage should be considered in the investment cost summations in InvCtCAP = ICc (xc,i,t − xc,i,t−1 ) ,
(9)–(12). In the example, the additional (marginal) investments c∈Ω c i∈Ω i (1 + r)L T c − 1
made at each stage are given by: (xg ,i,t − xg ,i,t−1 ). This is why
where xc,i,0 = 0 (12)
the investment cost function in (9) contains this expression.    
Now, suppose the decision variable on DG investments xg ,i,t M ntCtDG = M CgN xg ,i,t + M CgE ug ,i,t (13)
is defined as a binary one. This means that only one DG of type g ∈Ω g i∈Ω i g ∈Ω g i∈Ω i
g can be installed at node i in either of the planning stages.  
Suppose it becomes most economical to install it in the sec- M ntCtLN = M CkE uk ,t + M CkN xk ,t (14)
ond year, i.e., xg ,i,2 = 0. The logical constraint in (70) leads k ∈Ω k k ∈Ω k
to xg ,i,3 ≥ xg ,i,2 , i.e., xg ,i,3 ≥ 1. For this particular example,  
M ntCtES = M Ces xes,i,t (15)
the binary variable for each stage, i.e., {xg ,i,1 ; xg ,i,2 ; xg ,i,3 } is es∈Ω e s i∈Ω i
equal to {0; 1; 1}, respectively. Recall that the investment cost  
of this DG should be considered only once (in the second year) M ntCtCap = M Cc xc,i,t . (16)
in the summation, and this is taken care of by the expression c∈Ω c i∈Ω i
312 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017

The total cost of power produced by new and existing DGs is assumption is valid in distribution systems, where the active
given by Eq. (17). Note that these costs depend on the amount of power flow dominates the total apparent power in lines. Fur-
power generated at each scenario, snapshot and stage. Therefore, thermore, the voltage magnitude at bus i can be expressed as
these costs represent the expected costs of operation. Similarly, the sum of the nominal voltage and a small deviation ΔVi ,
Eqs. (18) and (19), respectively, account for the expected costs as in (27):
of energy supplied by the ESS, and that purchased from up-
stream (i.e., transmission grid). The penalty for the unserved Pk = Vi2 gk − Vi Vj (gk cos θk + bk sin θk ) (25)
power, given by (20), is also dependent on the scenarios, and
time stages. Equation (20) therefore gives the expected cost of Qk = −Vi2 bk + Vi Vj (bk cos θk − gk sin θk ) (26)
unserved energy in the system. The expected emission costs Vi = Vnom + ΔVi ,
of power generated by new and existing DGs are given by
(21)–(23), and that of energy purchased from the grid is cal- where ΔV m in ≤ ΔVi ≤ ΔV m ax . (27)
culated using (24). Note that, for the sake of simplicity, a linear
emission cost function is assumed here. In reality, the emission Note that the voltage deviations at each node ΔVi are ex-
cost function is highly nonlinear and nonconvex, as in [44], [51] pected to be very small. Substituting (27) in (25) and (26) and
    neglecting higher order terms, we get:
ECtDG = ρs πw (OCgN,i,s,w ,t PgN,i,s,w ,t  2 
s∈Ω s w ∈Ω w g ∈Ω g i∈Ω i Pk ≈ Vnom + 2Vnom ΔVi gk
+ OCgE,i,s,w ,t PgE,i,s,w ,t ) (17) − (Vnom
2
+ Vnom ΔVi + Vnom ΔVj )(gk + bk θk ) (28)
     2 
ECtES = ρs πw Qk ≈ − Vnom + 2Vnom ΔVi bk
s∈Ω s w ∈Ω w c∈Ω c i∈Ω i 2
+ (Vnom + Vnom ΔVi + Vnom ΔVj )(bk − gk θk ). (29)
× λdch dch
es,i,s,w ,t Pes,i,s,w ,t (18)
   Note that Eqs. (28) and (29) still contain nonlinearities be-
ECtSS = ρs πw λςs,w ,t PςSS
,s,w ,t (19) cause of the products of two continuous variables—voltage
s∈Ω s w ∈Ω w ς ,Ω ς deviations and angle differences. However, since these vari-
   ables (ΔVi , ΔVj and θk ) are very small, their products can be
EN SCt = ρs πw υs,w ,t δi,s,w ,t (20)
neglected. Hence, the above flow equations become
s∈Ω s w ∈Ω w i∈Ω i

EmiCtDG = EmiCtN + EmiCtE (21) Pk ≈ Vnom (ΔVi − ΔVj ) gk − Vnom

2
bk θ k (30)
   
EmiCtN = ρs πw Qk ≈ − Vnom (ΔVi − ΔVj ) bk − Vnom
2
gk θk . (31)
s∈Ω s w ∈Ω w g ∈Ω g i∈Ω i
The linear planning model proposed here is based on the
× λC O2 e N N
s,w ,t ERg Pg ,i,s,w ,t (22)
above linearized flow equations. This linearization approach
   
EmiCtE = ρs πw was first introduced in [52] in the context of transmission expan-
s∈Ω s w ∈Ω w g ∈Ω g i∈Ω i sion planning problem. When the investment planning problem
includes network switching, reinforcement, replacement and ex-
× λC O2 e E E
s,w ,t ERg Pg ,i,s,w ,t (23) pansion of feeders, Eqs. (30) and (31) must be multiplied by the
    corresponding binary variables as in (32)–(35). This is to make
EmiCtE = ρs πw
sure the flow through an existing/a new feeder is zero when its
s∈Ω s w ∈Ω w g ∈Ω g i∈Ω i
switching/investment variable is zero; otherwise, the flow in that
× λC O2 e E E
s,w ,t ERg Pg ,i,s,w ,t . (24) feeder should obey the Kirchhoff’s law

C. Constraints Pk ≈ uk ,t {Vnom (ΔVi − ΔVj ) gk − Vnom

2
bk θ k } (32)
1) Kirchhoff’s Voltage Law: The customary ac power flow Qk ≈ uk ,t {−Vnom (ΔVi − ΔVj ) bk − Vnom
2
gk θk } (33)
equations, given by (25) and (26), are highly non-linear and non- Pk ≈ xk ,t {Vnom (ΔVi − ΔVj ) gk − Vnom
2
bk θ k } (34)
convex. Understandably, using these flow expressions in power
system planning applications is increasingly difficult. Because Qk ≈ xk ,t {−Vnom (ΔVi − ΔVj ) bk − Vnom
2
gk θk }. (35)
of this, Eqs. (25) and (26) are often linearized by considering
two practical assumptions. The first assumption is concerning The bilinear products, involving binary with voltage deviation
the bus voltage magnitudes, which in distribution systems are and angle difference variables, introduces undesirable nonlin-
expected to be close to the nominal value Vnom . The second as- earity to the problem. This nonlinearity can be avoided using the
sumption is in relation to the voltage angle difference θk across big-M formulation, i.e., by reformulating the above equations
a line which is practically small, leading to the trigonomet- into their respective disjunctive equivalents as in (36)–(39). As a
ric approximations sin θk ≈ θk and cos θk ≈ 1. Note that this rule-of-thumb, the big-M parameter is often set to the maximum
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY 313

transfer capacity in the system 3) Line Losses: The active and reactive power losses in line
k can be approximated as follows:
|Pk ,s,w ,t − {Vnom (ΔVi,s,w ,t − ΔVj,s,w ,t ) gk
− Vnom
2
bk θk ,s,w ,t )} ≤ M Pk (1 − uk ,t ) (36) P Lk = Pk,ij + Pk,ji ≈ 2Vnom
2
gk (1 − cos θk )

|Qk ,s,w ,t − {−Vnom (ΔVi,s,w ,t − ΔVj,s,w ,t ) bk ≈ Vnom

2
gk θk2 (47)

−Vnom
2
gk θk ,s,w ,t } ≤ M Qk (1 − uk ,t ) (37) QLk = Qk,ij + Qk,ji ≈ −2Vnom
2
bk (1 − cos θk )

|Pk ,s,w ,t − {Vnom (ΔVi,s,w ,t − ΔVj,s,w ,t ) gk ≈ −bk Vnom

2
θk2 . (48)

− Vnom
2
bk θk ,s,w ,t } ≤ M Pk (1 − xk ,t ) (38) Clearly, Eqs. (47) and (48) are nonlinear and nonconvex func-
|Qk ,s,w ,t − {−Vnom (ΔVi,s,w ,t − ΔVj,s,w ,t ) bk tions, making the problem more complex to solve. This can be
overcome by having the quadratic angle differences piecewise-
− Vnom
2
gk θk ,s,w ,t } ≤ M Qk (1 − xk ,t ) . (39) linearized, as it is done for the quadratic flows in the above.
2) Flow Limits: The apparent power flow through a line Sk However, instead of doing this, the expressions in (47) and
 (48) can be expressed in terms of the active and the reactive
is given by Pk2 + Q2k and this has to be less than or equal to
the rated value which is denoted as: power flows as in (49) and (50). Note that Eq. (49) can be
easily obtained by multiplying the squared expressions of both
Pk2 + Q2k ≤ (Skm ax )2 . (40) sides of the equations in (30) and (31) by the resistance of the
branch, combining the resulting equations, neglecting higher
Considering line switching/investment, Eq. (38) can be order terms and reordering both sides of the resulting equation.
rewritten as: Equation (50) is also obtained in a similar fashion but by mul-
tiplying the squared expressions by reactance. For the sake of
Pk2,s,w ,t + Q2k ,s,w ,t ≤ uk ,t (Skm ax )2 (41)
completeness, details concerning the derivations (49) and (50)
Pk2,s,w ,t + Q2k ,s,w ,t ≤ xk ,t (Skm ax )2 . (42) are presented in Appendix A
  2
The quadratic expressions of active and reactive power flows P Lk ,s,w ,t = rk Pk2,s,w ,t + Q2k ,s,w ,t /Vnom (49)
in (41) through (42) can be easily linearized using piecewise   2
linearization, considering a sufficiently large number of lin- QLk ,s,w ,t = xk Pk2,s,w ,t + Q2k ,s,w ,t /Vnom . (50)
ear segments, L. There are a number of ways of linearizing
such functions such as incremental, multiple choice, convex Note that expressing the losses as a function of flows has
combination and other approaches in the literature [53]. Here, two advantages. First, doing so reduces the number of nonlinear
the first approach (which is based on first-order approximation terms that has to be linearized, which in turn results in a model
of the nonlinear curve) is used because of its relatively sim- with a reduced number of equations and variables. For example,
ple formulation. To this end, two non-negative auxiliary vari- if Eqs. (47) and (48) are used instead, in addition to the quadratic
ables are introduced for each of the flows Pk and Qk such that power flow terms Pk2 and Q2k , the quadratic angle differences θk2
Pk = Pk+ − Pk− and Qk = Q+ − should also be linearized to make the problem linear and convex.
k − Qk .
Note that these auxiliary variables (i.e., Pk+ , Pk− , Q+ − On the contrary, when Eqs. (49) and (50) are used, one is only
k and Qk )
represent the positive and negative flows of Pk and Qk , respec- required to linearize Pk2 and Q2k . Second, it avoids unnecessary
tively. This helps one to consider only the positive quadrant of constraints on the angle differences when a line between two
the nonlinear curve, resulting in a significant reduction in the nodes is not connected or remains not selected for investment.
mathematical complexity, and by implication the computational This is often avoided by introducing binary variables and using
burden. In this case, the associated linear constraints are: a so-called big-M formulation [52]. However, this adds extra
complexity to the problem.

L 4) Kirchhoff’s Current Law (Active and Reactive Load Bal-
Pk2,s,w ,t ≈ αk ,l pk ,s,w ,t,l (43) ances): All the time, load balance should be respected at each
l=1 node, i.e., the sum of all injections should be equal to the sum

L of all withdrawals at each node. This is enforced by adding the
Q2k ,s,w ,t ≈ βk ,l qk ,s,w ,t,l (44) following two constraints:
l=1
     dch 

L PgE,i,s,w ,t +PgN,i,s,w ,t + Pes,i,s,w ,t − Pes,i,s,w
ch
,t
Pk+,s,w ,t + Pk−,s,w ,t = pk ,s,w ,t,l (45) g ∈Ω D G es∈Ω e s
l=1  
+ PςSS
,s,w ,t + Pk ,s,w ,t − Pk ,s,w ,t + δi,s,w ,t

L
in,k ∈i
− out,k∈i
Q+
k ,s,w ,t + Qk ,s,w ,t = qk ,s,w ,t,l (46)
l=1
1
i
= Ds,w ,t + P Lς ,s,w ,t + P Lk ,s,w ,t ∀ς ∀ς ∈ i (51)
2
where pk ,s,w ,t,l ≤ Pkm ax /L and qk ,s,w ,t,l ≤ Qm
k
ax
/L. k∈i
314 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017

  Notice that inequalities (53) and (54) involve products of

(QE N
g ,i,s,w ,t + Qg ,i,s,w ,t )+ Qci,s,w ,t + QSS
ς ,s,w ,t charging/discharging binary variables and investment variable.
g ∈Ω D G c∈Ω c
In order to linearize these, new continuous positive variables
  ch dch
+ Qk ,s,w ,t − Qk ,s,w ,t = Qis,w ,t + QLς ,s,w ,t zes,i,s,w ,t , and zes,i,s,w ,t , which replaces the bilinear products
in,k∈i out, k ∈i
in each constraint, is introduced such that the set of linear
constraints in (60) and (61) hold. For instance, the product
 1  1 dch dch
+ QLk ,s,w ,t+ QLk ,s,w ,t ∀ς, ∀ς ∈ i. (52) Ies,i,s,w ,t xes,i,t is replaced by the positive variable zes,i,s,w ,t .
2 2 Then, the bilinear product is decoupled by introducing the set
in , k ∈i out,k ∈i
of constraints in (60) [56].
Equations (51) and (52) stand for the active and the reactive
,t ≤ xes Ies,i,s,w ,t ; zes,i,s,w ,t ≤ xes,i,t ; zes,i,s,w ,t
dch m ax dch dch dch
power balances at each node, respectively. zes,i,s,w
5) Bulk Energy Storage Model Constraints: The generic   m ax
≥ xes,i,t − 1 − Ies,i,s,w
dch
,t xes . (60)
bulk ESS is modeled by
ch
0 ≤ Pes,i,s,w ch,m ax Similarly, the product Ies,i,s,w ,t xes,i,t is decoupled by includ-
,t ≤ Ies,i,s,w ,t xes,i,t Pes,i
ch ch
(53)
ing the following set of constraints:
ch,m ax
0 ≤ Pes,i,s,w
dch
,t ≤ Ies,i,s,w ,t xes,i,t Pes,i
dch
(54)
,t ≤ xes Ies,i,s,w ,t ; zes,i,s,w ,t ≤ xes,i,t ; zes,i,s,w ,t
ch m ax ch ch ch
zes,i,s,w
ch
Ies,i,s,w + dch
Ies,i,s,w ≤1 (55)   m ax
,t ,t ≥ xes,i,t − 1 − Ies,i,s,w
ch
,t xes . (61)
ch ch
Ees,i,s,w ,t = Ees,i,s,w −1,t + ηes Pes,i,s,w ,t The large number of discrete variables in the storage model
− dch dch
βes Pes,i,s,w ,t , where dch
βes presented above can render significant computational burden.
To overcome this, a relaxed ESS model can be formed without
dch
= 1/ηes (56) the charging and discharging indicator variables as:
≤ Ees,i,s,w ,t ≤ (53 )
m in m ax ch,m ax
Ees,i xes,i,t xes,i,t Ees,i (57) 0 ≤ Pes,i,s,w
ch
,t ≤ xes,i,t Pes,i
m ax
Ees,i,s,w 0 ,T 1 = μes xes,i,T 1 Ees,i (58) 0 ≤ Pes,i,s,w ch,m ax
(54 )
,t ≤ xes,i,t Pes,i
dch

Ees,i,s,w 1 ,t+1 = Ees,i,s,W,t ; Ees,i,s,W,T

Ees,i,s,w ,t = Ees,i,s,w −1,t + ηes Pes,i,s,w ,t − βes
ch ch dch dch
Pes,i,s,w ,t ,
= Ees,i,s,w 0 ,T 1 . (59) dch
where βes dch
= 1/ηes . (56 )
The limits on the capacity of ESS while being charged and
discharged are considered in Eqs. (53) and (54), respectively. Under normal conditions, the ESS model in (53 ), (54 ) and
Inequality (55) prevents simultaneous charging and discharging (56 ) is exact because by the principle of optimality, at most
ch dch
operation of ESS at the same operational time w. The amount of one of the variables Pes,i,s,w ,t and Pes,i,s,w ,t can be greater than
stored energy within the reservoir of bulk ESS at the operational zero. In other words, it does not economic sense to have both
time w as a function of energy stored until w − 1 is given by (56). variables to be greater than zero at the same time.
The maximum and minimum levels of storages in operational 6) Active and Reactive Power Limits of DGs: The active and
time w are also considered through inequality (57). Equation reactive power limits of existing generators are given by (62) and
(58) shows the initial level of stored energy in the bulk ESS (63), respectively. In the case of new (candidate) generators, the
as a function of its maximum reservoir capacity. In a multi- corresponding constraints are (64) and (65). Note that the binary
stage planning approach, Eq. (59) ensures that the initial level variables multiply the minimum and the maximum generation
of energy in the bulk ESS at a given year is equal to the final level limits to make sure that the power generation variable is zero
of energy in the ESS in the preceding year and the reservoir level when the generator remains either unutilized or unselected for
at the end of the planning horizon should be equal to the initial investment
level. The latter constraint guarantees that the optimal solution PgE,m in E,m ax
,i,s,w ,t ug ,i,t ≤ Pg ,i,s,w ,t ≤ Pg ,i,s,w ,t ug ,i,t
E
(62)
returned by the solution algorithm is not because of the initial
reservoir level. For the sake of simplicity , ηes dch
is assumed QgE,m in E E,m ax
,i,s,w ,t ug ,i,t ≤ Qg ,i,s,w ,t ≤ Qg ,i,s,w ,t ug ,i,t (63)
ch ch
to be equal to ηes , as in [54] and [55]. Both the charging ηes
dch
and discharging ηes efficiencies are expressed in terms of the PgN,m in N N,m ax
,i,s,w ,t xg ,i,t ≤ Pg ,i,s,w ,t ≤ Pg ,i,s,w ,t xg ,i,t (64)
energy at the nodes where the storage system is connected to.
QN,m in N,m ax
g ,i,s,w ,t xg ,i,t ≤ Qg ,i,s,w ,t ≤ Qg ,i,s,w ,t xg ,i,t .
N
(65)
Because of this, a certain percentage of the energy fed to the
storage system will be stored while the remaining will be lost It should be noted that these constraints are applicable only
in the form of losses (electrical, chemical, heat, etc.). This is for conventional DGs which have reactive power support ca-
related to the charging efficiency, which should then be less pabilities. In the case of variable generation sources, slight
than 1. On the other hand, in order to withdraw a given amount modifications are required. For instance, for wind and solar
of energy from the storage system, more energy is needed to PV generators, the upper bound Pgm,i,s,w
ax
,t should be equal to the
dch
cover the discharging losses. This is why we have 1/ηes in Eq. actual production level at a specific hour, which in turn depends
(56) associated with the energy at the output side of the ESS. on the level of primary energy source (wind speed and solar
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY 315

radiation). The lower bound Pgm,i,s,w

in
,t in this case is simply set
investment already made cannot be divested.
to zero. In addition, conventional wind and solar PV sources xk ,t ≥ xk ,t−1 (69)
do not often have the capability to provide reactive power sup-
port; hence, they are operated at a constant and lagging or unity xg ,i,t ≥ xg ,i,t−1 (70)
power factor. Under such an operation, the following constraints xes,i,t ≥ xes,i,t−1 (71)
should be used:
−1 xc,i,t ≥ xc,i,t−1 . (72)
g ,i,s,w ,t = tan(cos (pfg )) ∗ Pg ,i,s,w ,t
QE E
(63 )
10) Radiality Constraints: Distribution networks are struc-
QN −1
g ,i,s,w ,t = tan(cos (pfg )) ∗ Pg ,i,s,w ,t
N
(65 ) turally meshed but predominantly operated in a radial manner
where pfg is the power factor of the wind or solar type because of technical reasons (particularly related to the protec-
generator g. Under normal cases, PgE,i,s,w ,t = PgE,i,s,w
, m ax tion). The presence of DGs and reactive power sources in the
,t and
system nodes other than substation nodes (which is the subject
PgN,i,s,w ,t = PgN,i,s,w
, m ax
,t . of the present work) may lead to islanding, i.e., some loads may
On the other hand, for wind and solar PV type DGs with be isolated from the mains (substations) and/or loops, breaking
reactive power support capabilities such as doubly fed induction the radiality of the network. This is not desired mainly because
generator based wind turbine and voltage source inverter based of the aforementioned technical reasons. To ensure radiality,
PV, the following constraints are used: two conditions must be fulfilled. First, the solution must have
− tan(cos−1 (pfg )) ∗ PgE,i,s,w ,t ≤ QE g ,i,s,w ,t
Ni − Nς circuits. Second, the final topology should be con-
nected. Equation (73) represents the first necessary condition
≤ tan(cos−1 (pfg )) ∗ PgE,i,s,w ,t (63 ) for maintaining the radial topology of DNS

− tan(cos−1 (pfg )) ∗ PgN,i,s,w ,t ≤ QN g ,i,s,w ,t OR(xk ,t , uk ,t ) = Ni − Nς ∀t. (73)
≤ tan(cos−1 (pfg )) ∗ PgN,i,s,w ,t . (65 ) k ∈Ω i j

Note that the above equation assumes line investment is pos-

The above two inequalities show that the wind and solar type
DGs are capable of operating between pfg leading power factor
(capacitive) and pfg lagging power factor (reactive). This means
such DGs are capable of “producing” and “consuming” reactive
power depending on operational situations in the system.
7) Reactive Power Limit of Capacitor Banks: Inequality
(66) ensures that the reactive power produced by the reactive
power sources (capacitor banks) is bounded between zero and
the maximum possible capacity
0 ≤ Qci,s,w ,t ≤ xc,i,t Q0c . (66)
8) Active and Reactive Power Limits of Power Purchased:
For technical reasons, the power that can be purchased from the
transmission grid could have minimum and maximum limits,
which is enforced by (67) and (68). However, it is understood
that, in reality, setting such limits is difficult. They are included
here only for the sake of completeness
PςS,s,w
S, m in
,t ≤ PςSS S S, m ax
,s,w ,t ≤ Pς ,s,w ,t (67)
QSς ,s,w
S, m in S S, m ax
,t ≤ Qς ,s,w ,t ≤ Qς ,s,w ,t .
SS
(68)
For the analysis, the active power production limits are simply
set to 1.5 times the minimum and the maximum levels of total
load in the system.
Note that the multiplier is higher than one in the system be-
cause of the losses, which needs to be covered by generating ex-
tra power. The reactive power limits are determined by the power
−1
factor of the substation as: QSς ,s,w
S, m in
,t = −tan(cos (pfss )) ∗
S S, m ax −1
Pς ,s,w ,t and Qς ,s,w ,t = tan(cos (pfss )) ∗ Pς ,s,w ,t , where
SS SS
pfss is a given power factor at the substation, which is assumed
to be 0.9 throughout the analysis in this paper.
9) Logical Constraints: The following logical constraints
ensure that an investment decision cannot be reversed, i.e., an
sible in all corridors. Hence, in a given corridor, we can have
either an existing branch, a new one, or both connected in par-
allel, depending on the economic benefits of the final setup
(solution) brings about to the system. The radiality constraint in
(73) then has to accommodate this condition. One way to do this
is using the Boolean logic operation, as in (73). Unfortunately,
this introduces nonlinearity.
We show how this logic can be linearized using an additional
auxiliary variable zk ,t and the binary variables associated to
existing and new branches (i.e., uk ,t and xk ,t ), respectively.
Given zk ,t := OR(xk ,t , uk ,t ), this Boolean operation can be
expressed using the following set of linear constraints:
zk ,t ≤ xk ,t + uk ,t ; zk ,t ≥ xk ,t ; zk ,t ≥ uk ,t ; 0 ≤ zk ,t ≤ 1; ∀t.
(74)
Note that the auxiliary variable zk ,t is automatically con-
strained to be binary. Hence, it is not necessary to explicitly
define zk ,t as a binary variable; instead, defining it as a con-
tinuous positive variable is sufficient. Alternatively, if zk ,t is
defined to be binary variable from the outset, then, Eq. (73) can
be converted into a single range constraint as:
0 ≤ 2zk ,t − xk ,t − uk ,t ≤ 1; ∀t. (75)
Then, the radiality constraints in (73) can be reformulated using
the zk ,t variables as:

zk ,t = Ni − Nς ; ∀t. (76)
k ∈Ω i j
When all loads in the DNS are only fed by power from substa-
tions, the final solution obtained automatically satisfies the two
aforementioned conditions; hence, no additional constraints are
required, i.e., (74) or (75) along with (76) are sufficient to guar-
antee radiality. However, it should be noted that in the presence
316 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017

of DGs and reactive power sources, these constraints alone may

not ensure the radiality of the distribution network, as pointed (Qk )2
out in [57] and further discussed in [58]. This is however out of 2
≈ [(ΔVi − ΔVj ) bk ]2
Vnom
the scope of this work. If this is a critical issue, additional con- I
straints need to be added to guarantee that all buses are linked, + 2 ∗ bk Vnom gk θk ∗ (ΔVi − ΔVj )
as proposed in [50], [58]–[60].
The optimization problem, developed here, encompasses II

Eqs. (3)–(24) and constraints (36)–(39), (41)–(46), (49)–(72), 2

+ (Vnom gk θk ) . (A.2)
(74) or (75) and (76) when all considered DGs are conven-
tional generators with reactive power support capabilities. Oth- Since the variables θk , ΔVi and ΔVj are very small, the
erwise, constraints (63 ) and (65 ) need to be added in case second order terms (i.e., products of these variables) are close
RES-based DGs which do not have such capabilities are present to zero. Hence, the first and the second terms in (A.1) and
in the system or included in the planning. In addition, if there (A.2) can be neglected, leading to the following expressions,
are RES-based DGs which are instead capable of operating respectively:
as “generators” or “consumers” of reactive power depending (Pk )2
on system operational situations, constraints (63 ) and (65 ) 2
≈ (Vnom bk θk )2 (A.3)
Vnom
should be included and applied. In addition, if a relaxed ESS
model is sought, constraints (53)–(61) can be replaced with the (Qk )2
≈ (Vnom gk θk )2 . (A.4)
constraints in (53 ), (54 ) and (56 ). 2
Vnom
IV. CONCLUSION Multiplying both sides of (A.3) and (A.4) by rk and summing
This work has developed a new multi-stage and stochastic gives:
 2  2
model for jointly optimizing the integration of smart-grid en- Pk Qk
abling technologies such as ESS, reactive power sources, and rk + rk ≈ rk (Vnom bk θk )2
Vnom Vnom
network switching, reinforcement and/or expansion to support
large-scale renewable integration. The novel integrated planning + rk (Vnom gk θk )2 . (A.5)
model, proposed here, simultaneously determines the optimal After rearranging Eq. (A.5), we get:
size, time and placement of ESSs and reactive power sources  
as well as that of conventional DGs and RES in distribution   2 (bk )2
rk Pk2 + Q2k /Vnom ≈ gk (Vnom θk )2 rk + gk .
networks, which can be considered as a major leap forward. gk
The ultimate goal of this optimization work is to maximize (A.6)
the RES power absorbed by the system while maintaining the 2
One can easily verify that rk ( (bgkk) + gk ) = 1, reducing
power quality and stability within the required/standard levels
Eq. (A.5) to:
at a minimum possible cost. From a methodological perspec-   2
tive, the model has been formulated as a MILP optimization, rk Pk2 + Q2k /Vnom ≈ gk (Vnom θk )2 . (A.7)
employing a linearized ac network model which captures well
Recall that the right hand side of (A.7) corresponds to the
the inherent characteristics of power network systems. The lin-
active power losses expression in (49), which proves the deriva-
earized ac model also strikes the right balance between accuracy
tion. The flow-based reactive power losses in (50) are derived in
and computational burden. The standard IEEE 41-bus distribu-
a similar way. Multiplying both sides of (A.3) and (A.4) by xk
tion system is used to test the developed model and carry out the
instead of rk , adding both and rearranging the resulting equation
required thorough analysis from the standpoint of the objectives
leads to:
set in this work. Test results and discussions are presented in   2
Part II of this paper. rk Pk2 + Q2k /Vnom ≈ −bk Vnom
2
θk2 xk
 
APPENDIX × −bk + (gk )2 /(−bk ) . (A.8)
FLOW-BASED LOSSES
Note that, in Eq. (A.8), xk [−bk + (gk )2 /(−bk )] = 1. Hence,
The derivations related to the losses equations in (49) and (50) the equation reduces to:
are provided here. Squaring both sides of the flow equations in   2
(49) and (50) and dividing each by Vnom 2
, we get: rk Pk2 + Q2k /Vnom ≈ −bk Vnom2
θk2 . (A.9)
Notice that the right hand side of Eq. (A.8) is equal to the
(Pk )2
2
≈ [(ΔVi − ΔVj ) gk ]2 reactive losses expression in (50).
Vnom
I
REFERENCES
− 2 ∗ gk Vnom bk θk ∗ (ΔVi − ΔVj )
[1] P. K. Ray, S. R. Mohanty, and N. Kishor, “Classification of power quality
II disturbances due to environmental characteristics in distributed genera-
tion system,” IEEE Trans. Sustain. Energy, vol. 4, no. 2, pp. 302–313,
+ (Vnom bk θk )2 (A.1) Apr. 2013.
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY
[2] P. S. Georgilakis and N. D. Hatziargyriou, “Optimal distributed gener-
ation placement in power distribution networks: Models, methods, and
future research,” IEEE Trans. Power Syst., vol. 28, no. 3, pp. 3420–3428,
Aug. 2013.
[3] P. Paliwal, N. P. Patidar, and R. K. Nema, “Planning of grid integrated
distributed generators: A review of technology, objectives and techniques,”
Renew. Sustain. Energy Rev., vol. 40, pp. 557–570, Dec. 2014.
[4] D. Q. Hung, N. Mithulananthan, and R. C. Bansal, “An optimal investment
planning framework for multiple distributed generation units in industrial
distribution systems,” Appl. Energy, vol. 124, pp. 62–72, Jul. 2014.
[5] F. S. Abu-Mouti and M. E. El-Hawary, “Heuristic curve-fitted technique
for distributed generation optimisation in radial distribution feeder sys-
tems,” IET Gener. Transm. Distrib., vol. 5, no. 2, pp. 172–180, 2011.
[6] B. C. Pal and R. A. Jabr, “Ordinal optimisation approach for locating
and sizing of distributed generation,” IET Gener. Transm. Distrib., vol. 3,
no. 8, pp. 713–723, Aug. 2009.
[7] R. S. Maciel, M. Rosa, V. Miranda, and A. Padilha-Feltrin, “Multi-
objective evolutionary particle swarm optimization in the assessment of
the impact of distributed generation,” Electr. Power Syst. Res., vol. 89,
pp. 100–108, Aug. 2012.
[8] K. Vinothkumar and M. P. Selvan, “Fuzzy embedded genetic algorithm
method for distributed generation planning,” Electr. Power Compon. Syst.,
vol. 39, no. 4, pp. 346–366, Feb. 2011.
[9] F. Ugranlı and E. Karatepe, “Convergence of rule-of-thumb sizing and
allocating rules of distributed generation in meshed power networks,”
Renew. Sustain. Energy Rev., vol. 16, no. 1, pp. 582–590, Jan. 2012.
[10] V. V. S. N. Murty and A. Kumar, “Optimal placement of DG in radial dis-
tribution systems based on new voltage stability index under load growth,”
Int. J. Electr. Power Energy Syst., vol. 69, pp. 246–256, Jul. 2015.
[11] A. Bagheri, H. Monsef, and H. Lesani, “Integrated distribution network
expansion planning incorporating distributed generation considering un-
certainties, reliability, and operational conditions,” Int. J. Electr. Power
Energy Syst., vol. 73, pp. 56–70, Dec. 2015.
[12] W. Sheng, K. Liu, Y. Liu, X. Meng, and Y. Li, “Optimal placement
and sizing of distributed generation via an improved nondominated
sorting genetic algorithm II,” IEEE Trans. Power Del., vol. 30, no. 2,
pp. 569–578, Apr. 2015.
[13] R. Mena, M. Hennebel, Y.-F. Li, C. Ruiz, and E. Zio, “A risk-based
simulation and multi-objective optimization framework for the integration
of distributed renewable generation and storage,” Renew. Sustain. Energy
Rev., vol. 37, pp. 778–793, Sep. 2014.
[14] J. D. Foster, A. M. Berry, N. Boland, and H. Waterer, “Comparison of
mixed-integer programming and genetic algorithm methods for distributed
generation planning,” IEEE Trans. Power Syst., vol. 29, no. 2, pp. 833–843,
Mar. 2014.
[15] A. Ameli, S. Bahrami, F. Khazaeli, and M.-R. Haghifam, “A multiobjec-
tive particle swarm optimization for sizing and placement of DGs from
DG Owner’s and distribution company’s viewpoints,” IEEE Trans. Power
Del., vol. 29, no. 4, pp. 1831–1840, Aug. 2014.
[16] S. Wen, H. Lan, Q. Fu, D. C. Yu, and L. Zhang, “Economic allocation for
energy storage system considering wind power distribution,” IEEE Trans.
Power Syst., vol. 30, no. 2, pp. 644–652, Mar. 2015.
[17] M. Sedghi, M. Aliakbar-Golkar, and M.-R. Haghifam, “Distribution net-
work expansion considering distributed generation and storage units us-
ing modified PSO algorithm,” Int. J. Electr. Power Energy Syst., vol. 52,
pp. 221–230, Nov. 2013.
[18] H. Saboori, R. Hemmati, and V. Abbasi, “Multistage distribution network
expansion planning considering the emerging energy storage systems,”
Energy Convers. Manage., vol. 105, pp. 938–945, Nov. 2015.
[19] M. M. Aman, G. B. Jasmon, A. H. A. Bakar, and H. Mokhlis, “A new ap-
proach for optimum simultaneous multi-DG distributed generation Units
placement and sizing based on maximization of system loadability using
HPSO (hybrid particle swarm optimization) algorithm,” Energy, vol. 66,
pp. 202–215, Mar. 2014.
[20] R. S. Rao, K. Ravindra, K. Satish, and S. V. L. Narasimham, “Power
loss minimization in distribution system using network reconfiguration in
the presence of distributed generation,” IEEE Trans. Power Syst., vol. 28,
no. 1, pp. 317–325, Feb. 2013.
[21] K. Nekooei, M. M. Farsangi, H. Nezamabadi-Pour, and K. Y. Lee, “An
improved multi-objective harmony search for optimal placement of DGs in
distribution systems,” IEEE Trans. Smart Grid, vol. 4, no. 1, pp. 557–567,
Mar. 2013.
[22] M. Sedighizadeh, M. Esmaili, and M. Esmaeili, “Application of the hybrid
Big Bang-Big Crunch algorithm to optimal reconfiguration and distributed
generation power allocation in distribution systems,” Energy, vol. 76,
pp. 920–930, Nov. 2014.
317
[23] A. Y. Abdelaziz, Y. G. Hegazy, W. El-Khattam, and M. M. Othman,
“Optimal allocation of stochastically dependent renewable energy based
distributed generators in unbalanced distribution networks,” Electr. Power
Syst. Res., vol. 119, pp. 34–44, Feb. 2015.
[24] M. M. Othman, W. El-Khattam, Y. G. Hegazy, and A. Y. Abdelaziz,
“Optimal placement and sizing of distributed generators in unbalanced
distribution systems using supervised big bang-big crunch method,” IEEE
Trans. Power Syst., vol. 30, no. 2, pp. 911–919, Mar. 2015.
[25] D. Rama Prabha, T. Jayabarathi, R. Umamageswari, and S. Saranya,
“Optimal location and sizing of distributed generation unit using intelli-
gent water drop algorithm,” Sustain. Energy Technol. Assessments, vol. 11,
pp. 106–113, Sep. 2015.
[26] A. Mohamed Imran, M. Kowsalya, and D. P. Kothari, “A novel integration
technique for optimal network reconfiguration and distributed generation
placement in power distribution networks,” Int. J. Electr. Power Energy
Syst., vol. 63, pp. 461–472, Dec. 2014.
[27] A. K. Singh and S. K. Parida, “Allocation of distributed generation using
proposed DMSP approach based on utility and customers aspects under
deregulated environment,” Int. J. Electr. Power Energy Syst., vol. 68,
pp. 159–169, Jun. 2015.
[28] R. S. Al Abri, E. F. El-Saadany, and Y. M. Atwa, “Optimal placement
and sizing method to improve the voltage stability margin in a distribution
system using distributed generation,” IEEE Trans. Power Syst., vol. 28,
no. 1, pp. 326–334, Feb. 2013.
[29] Y. M. Atwa, E. F. El-Saadany, M. M. A. Salama, and R. Seethapathy,
“Optimal renewable resources mix for distribution system energy loss
minimization,” IEEE Trans. Power Syst., vol. 25, no. 1, pp. 360–370,
Feb. 2010.
[30] Y. Yang, S. Zhang, and Y. Xiao, “An MILP (mixed integer linear pro-
gramming) model for optimal design of district-scale distributed energy
resource systems,” Energy, vol. 90, pp. 1901—1915, Jul. 2015.
[31] S. Montoya-Bueno, J. I. Munoz, and J. Contreras, “A stochastic investment
model for renewable generation in distribution systems,” IEEE Trans.
Sustain. Energy, vol. 6, no. 4, pp. 1466–1474, Oct. 2015.
[32] E. E. Sfikas, Y. A. Katsigiannis, and P. S. Georgilakis, “Simulta-
neous capacity optimization of distributed generation and storage in
medium voltage microgrids,” Int. J. Electr. Power Energy Syst., vol. 67,
pp. 101–113, May 2015.
[33] K. Mahmoud, N. Yorino, and A. Ahmed, “Optimal distributed generation
allocation in distribution systems for loss minimization,” IEEE Trans.
Power Syst., vol. 31, no. 2, pp. 960–969, Mar. 2016.
[34] K. Vinothkumar and M. P. Selvan, “Hierarchical agglomerative clustering
algorithm method for distributed generation planning,” Int. J. Electr. Power
Energy Syst., vol. 56, pp. 259–269, Mar. 2014.
[35] Y. M. Atwa and E. F. El-Saadany, “Probabilistic approach for optimal
allocation of wind-based distributed generation in distribution systems,”
IET Renew. Power Gener., vol. 5, no. 1, pp. 79–88, 2011.
[36] V. V. S. N. Murty and A. Kumar, “Mesh distribution system analysis in
presence of distributed generation with time varying load model,” Int.
J. Electr. Power Energy Syst., vol. 62, pp. 836–854, Nov. 2014.
[37] G. van Belle, Statistical Rules of Thumb, 2nd ed. New York, NY, USA:
Wiley, 2008.
[38] F. Desta Zahlay, “Strategy, methods and tools for solving long-term trans-
mission expansion planning in large-scale power systems,” Ph.D. disser-
tation, Dept. Elect. Eng., Comillas Pontifical Univ., Madrid, 2016.
[39] P. S. Georgilakis and N. D. Hatziargyriou, “A review of power distri-
bution planning in the modern power systems era: Models, methods
and future research,” Electr. Power Syst. Res., vol. 121, pp. 89–100,
Apr. 2015.
[40] G. Sinden, “Characteristics of the UK wind resource: Long-term patterns
and relationship to electricity demand,” Energy Policy, vol. 35, no. 1,
pp. 112–127, Jan. 2007.
[41] S. Subhadarshi, “Minding the P’s and Q’s: Real and reactive power assess-
ment of hybrid energy conversion systems with wind and solar resources,”
Ph.D. dissertation, Dept. Electr. Comput. Eng., Iowa State Univ., Ames,
IA, USA, 2013.
[42] Y. Z. Li, Q. H. Wu, M. S. Li, and J. P. Zhan, “Mean-variance model for
power system economic dispatch with wind power integrated,” Energy,
vol. 72, pp. 510–520, Aug. 2014.
[43] M. Zhao, Z. Chen, and F. Blaabjerg, “Probabilistic capacity of a grid
connected wind farm based on optimization method,” Renew. Energy,
vol. 31, no. 13, pp. 2171–2187, Oct. 2006.
[44] M. Aien, M. Rashidinejad, and M. Fotuhi-Firuzabad, “On possibilistic
and probabilistic uncertainty assessment of power flow problem: A review
and a new approach,” Renew. Sustain. Energy Rev., vol. 37, pp. 883–895,
Sep. 2014.
318
[45] J. Hartigan and M. Wong, “Algorithm AS 136: A k-means clustering
algorithm,” Appl. Stat., vol. 28, no. 1, pp. 100–108, 1979.
[46] L. Baringo and A. J. Conejo, “Correlated wind-power production and
electric load scenarios for investment decisions,” Appl. Energy, vol. 101,
pp. 475–482, Jan. 2013.
[47] S. Wogrin, P. Duenas, A. Delgadillo, and J. Reneses, “A new approach to
model load levels in electric power systems with high renewable penetra-
tion,” IEEE Trans. Power Syst., vol. 29, no. 5, pp. 2210–2218, Sep. 2014.
[48] F. Desta Zahlay, F. de Cuadra, L. Olmos, M. Rivier, and I. J. Perez-Arriaga,
“A formulation for large-scale transmission expansion planning problem
and a solution strategy,” in Proc. 10th Int. Conf. Eur. Energy Market, 2013,
pp. 1–8.
[49] L. Blank and A. Tarquin, Engineering Economy, 7th ed. New York, NY,
USA: McGraw-Hill, 2011.
[50] G. Munoz-Delgado, J. Contreras, and J. M. Arroyo, “Joint expansion
planning of distributed generation and distribution networks,” IEEE Trans.
Power Syst., vol. 30, no. 5, pp. 2579–2590, Sep. 2015.
[51] P. Phonrattanasak, “Optimal placement of DG using multiobjective parti-
cle swarm optimization,” in Proc. 2nd Int. Conf. Mech. Electr. Technol.,
2010, pp. 342–346.
[52] H. Zhang, G. T. Heydt, V. Vittal, and J. Quintero, “An improved net-
work model for transmission expansion planning considering reactive
power and network losses,” IEEE Trans. Power Syst., vol. 28, no. 3,
pp. 3471–3479, Aug. 2013.
[53] J. P. Vielma, S. Ahmed, and G. Nemhauser, “Mixed-integer models for
nonseparable piecewise-linear optimization: Unifying framework and ex-
tensions,” Oper. Res., vol. 58, no. 2, pp. 303–315, Oct. 2009.
[54] C. Abbey and G. Joos, “A stochastic optimization approach to rating of
energy storage systems in wind-diesel isolated grids,” IEEE Trans. Power
Syst., vol. 24, no. 1, pp. 418–426, Feb. 2009.
[55] D. Pozo, J. Contreras, and E. E. Sauma, “Unit commitment with ideal and
generic energy storage units,” IEEE Trans. Power Syst., vol. 29, no. 6,
pp. 2974–2984, Nov. 2014.
[56] H. P. Williams, Model Building in Mathematical Programming, 4th ed.
New York, NY, USA: Wiley, 1999.
[57] E. Romero-Ramos, J. Riquelme-Santos, and J. Reyes, “A simpler and exact
mathematical model for the computation of the minimal power losses tree,”
Electr. Power Syst. Res., vol. 80, no. 5, pp. 562–571, May 2010.
[58] M. Lavorato, J. F. Franco, M. J. Rider, and R. Romero, “Imposing radiality
constraints in distribution system optimization problems,” IEEE Trans.
Power Syst., vol. 27, no. 1, pp. 172–180, Feb. 2012.
[59] J. Zhang, X. Yuan, and Y. Yuan, “A novel genetic algorithm based on
all spanning trees of undirected graph for distribution network reconfig-
uration,” J. Mod Power Syst. Clean Energy, vol. 2, no. 2, pp. 143–149,
Jun. 2014.
[60] H. Ahmadi and J. R. Martı́, “Mathematical representation of radiality
constraint in distribution system reconfiguration problem,” Int. J. Electr.
Power Energy Syst., vol. 64, pp. 293–299, Jan. 2015.
Sérgio F. Santos received the B.Sc. and M.Sc. de-
grees from the University of Beira Interior (UBI),
Covilhã, Portugal, in 2012 and 2014, respectively,
where he is currently working toward the Ph.D. de-
gree under UBI/Santander Totta doctoral incentive
grant in the Engineering Faculty. He was a Researcher
between 2013 and 2015 in the EU FP7 Project SiN-
GULAR with the Sustainable Energy Systems Lab,
UBI. His research interests include planning and op-
eration optimization of distribution system and smart
grid technologies integration.
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 8, NO. 1, JANUARY 2017
Desta Z. Fitiwi received the B.Sc. degree in elec-
trical and computer engineering from Addis Ababa
University, Addis Ababa, Ethiopia, in 2005, the
M.Sc. degree in electrical and electronics engineer-
ing from Universiti Teknologi PETRONAS, Tronoh,
Malaysia, in 2009. From 2005 to 2007, he was a Su-
pervisory Control and Data Acquisition and Adap-
tation Engineer with the Ethiopian Load Dispatch
Center Project. He is currently an Erasmus Mundus
Joint Doctorate in Sustainable Energy Technologies
and Strategies, jointly offered by Comillas Pontifical
University, KTH Royal Institute of Technology of Stockholm, and Delft Univer-
sity of Technology. He actively participated in the realization of four EU-funded
projects. He is currently a Researcher at the University of Beira Interior, Cov-
ilhã, Portugal. His research interests include regulation and economics of the
power industry, transmission expansion planning, sustainable energy model-
ing and strategic planning, artificial-intelligence applications, and mathematical
optimizations for power systems.
Miadreza Shafie-Khah (S’08–M’13) received the
M.Sc. and Ph.D. degrees in electrical engineering
from Tarbiat Modares University, Tehran, Iran, in
2008 and 2012, respectively. He received his first
postdoctoral from the University of Beira Interior,
Covilha, Portugal, in 2015, while working on the EU-
funded FP7 Project SiNGULAR. He is currently pur-
suing his second postdoctoral studies with the Uni-
versity of Salerno, Salerno, Italy. His research inter-
ests include power market simulation, market power
monitoring, power system optimization, operation of
electricity markets, price forecasting, and smart grids.
Abebe W. Bizuayehu (M’15) received the M.Sc. de-
gree in mechanical engineering, in the specialty of
industrial mechanic, from José Antonio Echevarrı́a
Higher Polytechnic Institute, Havana, Cuba, in 1989,
and the Ph.D. degree in renewable energy and en-
ergetic efficiency from the University of Zaragoza,
Zaragoza, Spain, in 2012. He received his postdoc-
toral studies from the University of Beira Interior,
Covilha, Portugal, in 2015, while working on the
EU-funded FP7 Project SiNGULAR.
SANTOS et al.: NEW MULTISTAGE AND STOCHASTIC MATHEMATICAL MODEL FOR MAXIMIZING RES HOSTING CAPACITY
Carlos M. P. Cabrita was born in Lisbon, Portugal,
on July 2, 1951. He received the E.E. Diploma and
the Ph.D. on E.E. from the Technical University of
Lisbon, Lisbon, in 1976 and 1988, respectively. From
1978 to 1996, he was with the Technical University
of Lisbon. Since 1997, he has been with the Uni-
versity of Beira Interior, Covilha, Portugal, where
he is currently a Full Professor in the Department
of Electromechanical Engineering. He has published
about 250 papers in technical journals and conference
proceedings. His research interests include electrical
machines design and maintenance processes.
319
João P. S. Catalão (M’04–SM’12) received the
M.Sc. degree from the Instituto Superior Técnico
(IST), Lisbon, Portugal, in 2003, and the Ph.D. degree
and Habilitation for Full Professor ("Agregação")
from the University of Beira Interior (UBI), Cov-
ilha, Portugal, in 2007 and 2013, respectively.
He is currently a Professor with the Faculty of En-
gineering, University of Porto, Porto, Portugal, and a
Researcher at INESC TEC, INESC-ID/IST-UL, and
C-MAST/UBI. He was the Primary Coordinator of
the EU-funded FP7 Project SiNGULAR ("Smart and
Sustainable Insular Electricity Grids Under Large-Scale Renewable Integra-
tion"), a 5.2-million-euro project involving 11 industry partners. He has au-
thored or coauthored more than 470 publications, including 154 journal papers,
281 conference proceedings papers, 23 book chapters, and 11 technical re-
ports, with an h-index of 27 (according to Google Scholar), having supervised
more than 45 postdoctoral, Ph.D., and M.Sc. students. He is the Editor of the
books entitled Electric Power Systems: Advanced Forecasting Techniques and
Optimal Generation Scheduling and Smart and Sustainable Power Systems: Op-
erations, Planning and Economics of Insular Electricity Grids (Boca Raton, FL,
USA: CRC Press, 2012 and 2015, respectively). His research interests include
power system operations and planning, hydro and thermal scheduling, wind
and price forecasting, distributed renewable generation, demand response, and
smart grids.
Prof. Catalão is an Editor of the IEEE TRANSACTIONS ON SMART GRID,
an Editor of the IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, and an As-
sociate Editor of the IET Renewable Power Generation. He was the Guest
Editor-in-Chief for the Special Section on "Real-Time Demand Response" of
the IEEE TRANSACTIONS ON SMART GRID, published in December 2012, and
the Guest Editor-in-Chief for the Special Section on "Reserve and Flexibil-
ity for Handling Variability and Uncertainty of Renewable Generation" of the
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, published in April 2016. He
received the 2011 Scientific Merit Award UBI-FE/Santander Universities and
the 2012 Scientific Award UTL/Santander Totta. Also, he has won four Best
Paper Awards at IEEE Conferences.