Electrical Power and Energy Systems 75 (2016) 194–204

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

A two-point estimate method for uncertainty modeling

in multi-objective optimal reactive power dispatch problem
Seyed Masoud Mohseni-Bonab a, Abbas Rabiee a,⇑, Behnam Mohammadi-Ivatloo b, Saeid Jalilzadeh a,
Sayyad Nojavan b
a
Department of Electrical Engineering, Faculty of Engineering, University of Zanjan, Zanjan, Iran
b
Faculty of Electrical and Computer Engineering, University of Tabriz, Tabriz, Iran

a r t i c l e i n f o a b s t r a c t

Article history: Due to nonlinear and discrete variables and constraints, optimal reactive power dispatch (ORPD) is a
Received 20 October 2014 complex optimization problem in power systems. In this paper, the purpose is to solve multi objective
Received in revised form 8 August 2015 ORPD (MO-ORPD) problem considering bus voltage limits, the limits of branches power flow, generators
Accepted 21 August 2015
voltages, transformers tap changers and the amount of compensation on weak buses. The objectives of
Available online 25 September 2015
this paper are real power losses and voltage deviations from their corresponding nominal values, which
are conflicting objectives. Because of the stochastic behavior of loads, the MO-ORPD problem requires a
Keywords:
probabilistic approach. Hence, in this paper, a two-point estimate method (TPEM) is proposed to model
Monte Carlo simulation (MCS)
Multi objective optimal reactive power
the load uncertainty in MO-ORPD problem. Moreover, the proposed method is compared with some other
dispatch (MO-ORPD) methods such as deterministic approaches and Monte Carlo simulations (MCS). The obtained results
Real power loss (PL) approve the efficiency of the proposed methodology. The proposed models are implemented and solved
Two-point estimate method (TPEM) using GAMS optimization package and verified using IEEE 14-bus and IEEE 30-bus standard test systems.
Uncertainty Ó 2015 Elsevier Ltd. All rights reserved.
Voltage deviation (VD)

Introduction to the reactive power dispatch problem with a particle swarm opti-
mization approach based on multi-agent systems is presented in
Optimal power flow (OPF) is one of the main problems in power [4]. In [5], a model for ORPD is presented for minimization of the
system operation, which was introduced by Carpentier for first total costs. The total cost is defined as cost of energy loss of trans-
time about 50 years ago [1]. Generally this problem categorized mission network and the costs of adjusting the control devices. In
into two sub-problems, namely optimal reactive power dispatch [6], a harmony search algorithm is implemented for solution of
(ORPD) and optimal real power dispatch [2]. ORPD is important ORPD problem. In this paper, different objective functions
for security and economy of power systems. The ORPD determines including power transmission loss, voltage stability and voltage
the optimal amount of reactive power generation at different profile are optimized separately. Hybrid methods are also used
places, which is used for minimization of real power transmission for solution of ORPD problem to provide the advantage of different
losses and total voltage deviation with considering different methods simultaneously. Hybridization of modified teaching
equality and inequality constraints. Nonlinear objective function learning algorithm and double differential evolution algorithm
and different type of constraints makes the ORPD problem a has been used in [7] for effective solution of ORPD problem. In
large-scale nonlinear optimization problem. [8], hybrid standard real-coded genetic algorithm and simulated
The ORPD problem is modeled for different objective functions annealing method is used to solve ORPD problem. In [9], applica-
and various methods are used for its solution. As presented in [3], tion of chance-constrained programming method to handle the
the reactive power generation management can be employed to uncertainties in ORPD problem is studied. Uncertain nodal power
improve the voltage stability margin of power systems. A solution injections and random branch outages are considered as uncer-
tainty sources. The problem is solved by combining probabilistic
power flow and genetic algorithm. The differential evolution
⇑ Corresponding author at: Department of Electrical Engineering, Faculty of
Engineering, University of Zanjan, P.O. Box: 45371-38111, Zanjan, Iran. Tel.: +98
algorithm for optimal settings of reactive power dispatch control
2433054208; fax: +98 2433052762. variables is employed in [10].
E-mail addresses: [email protected] (S.M. Mohseni-Bonab), rabiee@ ORPD problem is modeled as multi-objective optimization
znu.ac.ir (A. Rabiee), [email protected] (B. Mohammadi-Ivatloo), jalilzadeh@ problem and solved using different methods in literature. A
znu.ac.ir (S. Jalilzadeh), [email protected] (S. Nojavan).

http://dx.doi.org/10.1016/j.ijepes.2015.08.009
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204 195

Nomenclature

Sets Q min max

Gi =Q Gi minimum/maximum value for reactive power of the
NB =N j set of buses i-th bus
NL set of branches (transmission lines) V min
i =V max
i minimum/maximum value for voltage magnitude of
NG set of generating units the i-th bus
ND set of load buses Smax
‘ maximum value of power flow of ‘-th transmission line
w‘ set of buses adjacent to ‘-th branch Q Ci VAR compensation capacity in each step at bus i
NT set of tap changing transformers Amin =Amax minimum/maximum reactive power compensation
i i
Nsh set of VAR compensators step at bus i
NO set of objective functions
NP set of Pareto optimal solutions
Variables
x vector of dependent variables
Indices u vector of control variables
k index of Pareto optimal solutions Tt value of t-th tap changer setting
i=j index of bus number where i ¼ 1; 2; . . . ; NB V i =V j voltage magnitude of bus i/j
‘ index of transmission lines hi =hj voltage angle at bus i/j
sl index of slack bus S‘ power flow of ‘-th transmission line
r index of objective functions Q Gi reactive power generation in bus i
t index of on-load tap changing transformers Ai reactive power compensation step at bus i
Q shi reactive power compensation at bus i
Parameters
w1 weight of objective 1 (real power loss) Functions
w2 weight of objective 2 (voltage deviation) J total objective function
y‘ =g ‘ =b‘ Admittance/conductance/susceptance of ‘-th line J1 first objective function (PL = real power loss)
Y ij ¼ Gij þ jBij ij-th element of system YBUS matrix J2 second objective function (VD = voltage deviation)
PGi active power production at bus i J pu normalized objective function (PLpu and VDpu )
Pmin
Gi =P Gi
max
minimum/maximum value for active power
J max
r =J min
r maximum/minimum value for r-th objective function
T min =T max minimum/maximum value for t-th tap changer
t t
PLmin =PLmax minimum/maximum value for PL
settings
P Di real power of the i-th bus VDmin =VDmax minimum/maximum value for VD
Q Di reactive power of the i-th bus

strength Pareto evolutionary algorithm is proposed in [11] to han- solve multi objective reactive power dispatch considering load
dle the ORPD problem considering the real power loss and the bus uncertainty. Load forecast can be obtained using historical load
voltage deviations as objective functions. In [12], real power loss, data and whether forecast data using different methods. But,
voltage deviation and voltage stability index are considered as always the forecast is not perfect and there is an inaccuracy in
objective functions and the obtained multi-objective problem is the forecasted data. Therefore it is necessary to consider the effect
solved using teaching learning based optimization algorithm. of uncertain loads in the problem.
Improving voltage stability margin of power system [13] by con- Uncertain parameters in power systems can be divided into two
trolling VAR sources is studied in [14,15]. In [15], L-index is used categories: The first one is technical parameters like outages,
as voltage stability index and is incorporated in multi-objective demand and generation and second one is economical parameters
ORPD problem considering active power losses as another objec- like as inflation rate or price levels. There are different methodolo-
tive. The problem is solved using chaotic PSO based multi- gies for handling uncertainties in power systems that is based on
objective optimization method. In [16], ORPD problem is modeled aforementioned parameters. Stochastic programming is widely
as fuzzy goal programming problem and solved using genetic algo- used in power system planning and operation for uncertainty mod-
rithm. ORPD problem considering static voltage stability and volt- eling [23–25]. In stochastic programming based methods, the
age deviation is solved using a seeker optimization algorithm uncertain parameters are modeled using discrete scenarios with
(SOA) in [17]. The multi-objective ORPD problem considering their occurrence probability. Information gap decision theory
active power losses and voltage stability index as objective func- (IGDT) is a non-possibilistic uncertainty modeling method, which
tions is solved using modified NSGA-II in [18]. In [19], a hybrid does not require probability distribution of the uncertain parame-
fuzzy multi-objective evolutionary algorithm based approach is ters. IGDT method is used for modeling wind power generation
proposed for solution of multi-objective ORPD problems. Hybrid uncertainty in OPF problem in presence of HVDC lines [26]. This
modified imperialist competitive algorithm and invasive weed method is also used for modeling price uncertainty in operation
optimization is implemented in [20] for multi-objective ORPD of generation companies [27] and distribution companies [28].
(MO-ORPD) problem solution. In [21], different constraint handling Robust optimization is another decision making tool in uncertain
methods in ORPD problem including feasible solutions, self- environments. This method is utilized in [29] for market price
adaptive penalty, e-constraint, stochastic ranking, and the uncertainty modeling in optimal self-scheduling of a hydro-
ensemble of constraint handling techniques is evaluated. A multi thermal generating company. In [30], robust optimization method
objective chaotic parallel vector evaluated interactive honey bee is used for decision making of a retailer in energy market. An
mating optimization algorithm is presented in [22] to solve the updated review of the uncertainty modeling methods in energy
MO-ORPD problem with considering operational constraints of systems are provided in [31].
the generators. The aim of this paper is determining optimal values of control
Therefore, it is observed that the MO-ORPD problem has been variables in order to achieve the objectives such as reducing real
solved so far with many intelligent algorithms but none of them power losses and minimizing voltage deviation considering the
196 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204

technical constraints as well as some existing uncertainties. In X

ND

order to model the load uncertainty, the Hong’s two-point estimate J 2 ¼ VDðx; uÞ ¼ jV i  V spc
i j ð2Þ
i¼1
method (TPEM) is used. The main feature of the proposed TPEM is
that it only requires resolving 2  m deterministic MO-ORPD to
obtain the behavior of m random variable. Since this paper focuses Constraints
on the uncertainties involved by the load, it is assumed that their
statistical features are estimated or measured. The main contribu- Equality constraints (AC power flow equations)
tions of this work are summarized as follows: The AC power flows equations are expressed as follows.
X
NB
(1) The stochastic behavior of load is modeled using TPEM. P Gi  P D i ¼ V i V j ½Gij cosðhi  hj Þ þ Bij sinðhi  hj Þ
(2) Similar to the Monte Carlo simulation (MCS) approaches, j¼1
TPEM use the deterministic routines for solving the proba- ð3Þ
X
NB
bilistic MO-ORPD. However, they require a much lower com- Q Gi  Q Di þ Q shi ¼ V i V j ½Gij sinðhi  hj Þ  Bij cosðhi  hj Þ
putational burden. j¼1
(3) Furthermore, PEM overcome the difficulties associated with
the lack of perfect knowledge of the probability functions of
Operational limits
stochastic variables, since these functions are approximated
The generators active and reactive power outputs along with
using only their first few statistical moments (i.e., mean,
bus voltages should be hold in a pre-specified interval, as follows:
variance, skewness, and kurtosis). Therefore, a smaller level
of data information is needed. Pmin max
Gi 6 P Gi 6 P Gi ; 8i ¼ fslg ð4Þ

The rest of the paper is organized as follows: Section ‘ORPD

Q min max
Gi 6 Q Gi 6 Q Gi ; 8i 2 N G ð5Þ
problem formulation’ and ‘Multi objective optimal reactive power
dispatch (MO-ORPD)’ describe the reactive power dispatch
problem formulation and MO-ORPD problem, respectively. V min
i 6 V i 6 V max
i ; 8i 2 N B ð6Þ
Implementation of TPEM on the problem is presented in Section Also, the line flow limits are as follows.
‘Implementation of TPEM’. Section ‘Case study’ describes the
numerical example used in this paper. A brief summary of the jS‘ j 6 Smax
‘ ; 8‘ 2 N L ð7Þ
simulation results, the obtained numerical results, and some Also, on-load tap changing transformers (OLTC) settings are mod-
other observations and discussions, are also included in this eled in this paper. As it is depicted in Fig. 1, the OLTC connected
section. Finally, the contributions and conclusions of this paper between buses i and j will change three elements of system YBUS
are summarized in last section. corresponding to ii-th, ij-th and ji-th elements.
2 3
 
ORPD problem formulation 6 7
6 i# j# 7
6 7
A system operator usually has various objectives such as mini- 6 i ! Ty2‘  Ty‘t 7
Y BUS ¼6
6
7 ð8Þ
7
t
mization of sum of system transmission loss and voltage deviation 6 7
6 7
of load buses from their desired values. These objective functions 4 j !  Ty‘t y‘ 5
may conflict with each other. Hence, at the first, the confliction
 ...
among them is investigated. The conventional ORPD model can
be described mathematically as follows. The OLTC settings should be restricted by their lower and upper
limits as follows:
ORPD objective functions T min 6 T t 6 T max ; 8t 2 N T ð9Þ
t t

In this paper the objective functions are minimization of real It is worth to note that the reactive power output of VAR
power losses and voltage deviation in load buses. compensation devices are modeled as a multi-step compensation,
i.e. a discrete variable is defined for each VAR compensation node
Minimization of total real power losses as follows, which determine how many steps of VAR injections are
With the increasing rate of energy consumption, the amount of necessary.
power losses increased too, making the reduction of power losses Q shi ¼ Q C i  Ai ; 8i 2 Nsh ð10Þ
as an important aim for system operators [32,33]. The active power
losses can be expressed as follows. The reactive power compensation steps are limited as follows.

X
NL Amin
i 6 Ai 6 Amax
i ; 8i 2 N sh ð11Þ
J 1 ¼ PLðx; uÞ ¼ g ‘ ½V 2i þ V 2j  2V i V j cosðhi  hj Þ ð1Þ
‘¼1
i;j2W‘

Minimization of voltage deviation at load bus

The second objective of ORPD problem is to maintain a proper
voltage level at load buses. Electrical equipment is designed for
optimum operation of nominal voltage. The deviation from the
nominal voltage will decrease the efficiency and life of the electri-
cal devices. Thus, the voltage profile of the system could be opti-
mized by minimization of the sum voltage deviations from the
corresponding rated values at load buses. This objective function
is defined as follows: Fig. 1. OLTC model used in this paper.
 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204 197

Multi objective optimal reactive power dispatch (MO-ORPD) Implementation of TPEM

Various methods are available to solve multi-objective Monte-Carlo simulation is an iterative approach which utilizes
optimization problems such as weighted sum approach [34], cumulative density function (CDF) of random variables to deter-
e-constraint method [24], and evolutionary algorithms [35]. In this mine the final result. This method is widely used for uncertainty
paper, the proposed multi-objective model of the MO-ORPD is modeling. Requiring great number of iterations to reach the
solved using the weighted sum method. desired convergence is the main drawback of the MCS. The TPEM
In the weighted sum method, different weights are used for the is an approximate method. The information provided by central
conflicting objective functions to generate different Pareto optimal moments is used to find some representative points (s points for
solutions and then the different weights selects the most satisfac- each variable) named concentrations. These representative points
tory solution from the optimal Pareto set. In the weighted sum are used for solution of the model and the statistical information
method, the problem is solved as follows: of the random output variable is calculated using the solutions
obtained for representative points [40].
min½Jðx; uÞ ¼ w1 J 1;pu ðx; uÞ þ w2 J 2;pu ðx; uÞ ð12Þ Assume that Xfx1 ; x2 ; . . . ; xl ; . . . ; xm g is a random variable with a
mean value lxl and standard deviation rxl : Moreover, Z is a random
where
function of X (i.e. Z = F(X)). Each of the s concentrations of variables
w1 þ w2 ¼ 1 ð13Þ xl can be defined as a pair composed of a location xl;s and a weight
The aforementioned MO-ORPD problem can be formulated wl;s : The proposed method uses a particular case of point estimate
mathematically as a nonlinear constrained optimization problem, method, known as Hong’s two-point estimate method (HTPEM).
which can be expressed as: Using HTPEM, function F has to be evaluated only s times for each
h i random input variable xl at the points made up of the sth location
xT ¼ ½V L T ; ½Q G T ; ½SL T of random input variable xl and the mean value (lxl ) of remaining
h i ð14Þ input variables. Therefore, the total number of evaluations is 2  m.
uT ¼ ½V G T ; ½Q C T ; ½TT The location xl;s is determined as follows [41]:
xl;s ¼ lxl þ nl;s  rxl ð19Þ
Fuzzy modeling for normalizing objective functions In which, nl;s is the standard location of random variable xl . The
standard locations and weights of random variable of xl are
Since the objective functions (1) and (2) are not in the same computed by:
range and dimension, a fuzzy satisfying method is proposed to sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2
calculate the normalized form of the objective functions in (12). kl;3 kl;3 kl;3 kl;3
The fuzzy membership of each objective function maps it to the nl;1 ¼ þ mþ ; nl;2 ¼  mþ ð20Þ
2 2 2 2
interval [0, 1]. More generally, the i-th objective function of J i is
normalized as follows. and
8 nl;2 nl;1
>
>
>1 J ðkÞ min
r 6 Jr wl;1 ¼  ; wl;2 ¼ ð21Þ
< ðkÞ
mðnl;1  nl;2 Þ mðnl;1  nl;2 Þ
ðkÞ J r J max
Jr;pu ¼ r
J min
6 6 J rðkÞ J max ; 8r ¼ 1; . . . ; NO ; 8k ¼ 1; . . . ; NP
>
> J min J max r r
where kl;3 denotes the skewness of the random variable xl :
>
r r
: ðkÞ max h i
0 Jr P Jr
E ðxl  lxl Þ3
ð15Þ kl;3 ¼ ð22Þ
ðrxl Þ3
For the entire Pareto optimal set, the number of best compro-
mise solution (BCS) is obtained by using min–max criterion [36] The algorithm of solving the MO-ORPD problem by means of
as follows. HTPEM is shown in Fig. 2.
  In the MO-ORPD problem, the load is modeled as random
max minðJ r;pu Þ ð16Þ variable with known probability distribution. The locations and
k r
weights have to be computed as described previously. A determin-
This means that the solution which has the largest value of istic MO-ORPD must be run for each concentration. The solution of
minr ðJr;pu Þ, is the BCS. In this paper for objective functions (1) and an MO-ORPD is:
(2), the normalized values are expressed as: Z l;s ¼ Ffxl;1 ; xl;2 ; . . . ; xl;s ; . . . ; xm;s g ð23Þ
PL  PLmax where Z l;s , is the vector of random output variables associated with
PLpu ¼ J 1;pu ¼ ð17Þ
PLmin  PLmax the sth concentration of random input variable and represents the
nonlinear relation between the input and output variables in
VD  VDmax the MO-ORPD. The raw moments of output random variables are
VDpu ¼ J 2;pu ¼ ð18Þ determined as:
VDmin  VDmax
X
EðZÞ ffi EðZÞ þ wl;s  Z l;s ð24Þ
s
Solution tool
To clarify the flowchart of algorithm, the solution steps are
In this paper the stochastic multi-objective reactive power dis- summarized as follows:
patch problem is formulated as a mixed integer nonlinear pro-
gramming (MINLP) problem, and it is solved using Generalized Step 1: Set the first and second moments of sth output random
Algebraic Modeling Systems (GAMS) software [37]. Also, CONOPT variables to zero: EðZÞ ¼ 0.
[38] and SBB [39] solvers are utilized for dealing with nonlinear Step 2: Select the input random variable xl :
programming (NLP) and MINLP problems, respectively. Step 3: Compute kl;3 ; nl;s ; wl;s based on (19)–(22).
198 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204

Fig. 2. Flowchart of the proposed algorithm.

Step 4: Determine the two estimated locations of xl;s . Step 7: Repeat the steps 2–6 until all concentrations of all input
Step 5: Solve the deterministic MO-ORPD for each concentration. random variables are taken into account. Finally, compute the
Step 6: Update the raw moments of output variables. statistical information of output random variables.
 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204 199

Case study Table 1 shows the values of both objective functions for all 11
Pareto optimal solutions. Among these optimal solutions, Solu-
Test systems tion#1 is the minimum power loss, with the equal to
13.08294 MW and the VD of 2.9% (minimum VD).
Simulation is carried out on IEEE 14-bus and IEEE 30-bus stan- As explained in Section ‘Fuzzy modeling for normalizing objec-
dard test systems. tive functions’, in order to select the best solution among the
Case I: IEEE 14-bus system [10,42] consists of 14 buses, which 5 obtained Pareto optimal set, a fuzzy satisfying method is utilized
of them are generator buses. Bus 1 is the slack bus, 2, 3, 6 and 8 are here. It is evident from the last column of Table 1 that the BCS is
taken as PV buses and the remaining 9 buses are PQ buses. The net- Solution#7, with the maximum weakest membership function of
work has 20 branches, 3 transformers and 4 capacitor banks. The 0.82695. The corresponding MO-ORPD problem real power loss
three branches 4–7, 4–9, 5–6 are equipped with under load tap and VD are equal to 13.29397 MW and 0.00628 pu, respectively.
changing transformers which their taps is within the interval Correspondingly, the Pareto optimal front of the two objective
[0.9, 1.1]. Two capacitor banks are available at weak buses 9 and functions is derived, which is depicted in Fig. 3. This Pareto front
14. Each capacitor bank consists of three 6 MVAr steps. The dimen- consists of 11 Pareto optimal solutions.
sion of control variables is 17, which consist of five PV generator
voltages within the range of [0.9, 1.1] and power output of slack Uncertainty characterization using TPEM
bus generator in the interval [0, 323.4] MW, three tap changing Due to the stochastic behavior of the load, the MO-ORPD prob-
transformers within the range of [0.9, 1.1] and two shunt compen- lem analysis requires a probabilistic approach. In this section, a
sation capacitor banks. The topology and initial operating condi- TPEM is employed to model the uncertainty in load with consider-
tions of this system are given in [42]. ing the mean and standard deviation values of load. In this case,
Case II: IEEE 30-bus system [42] consists of 30 buses, out of the expected objective functions are higher than the deterministic
which 6 are generator buses. Bus 1 is the slack bus. The network case. The mean values of real power loss and voltage deviation of
has 41 branches, 4 transformers and 9 capacitor banks. The dimen- MO-ORPD for this case are also summarized in Table 2. According
sion of control variable is 25. The initial operating conditions of the to Table 2 Solution#7 is the BCS. The Pareto optimal front of the
system are given by [42]. According to [12], shunt capacitors are two objective functions is derived, which is depicted in Fig. 4. This
installed at buses 10, 12, 15, 17, 20, 21, 23 and 29. Also, each Pareto front consists of 11 Pareto optimal solutions.
VAR compensation device is assumed to be 6 steps of 6 MVAR.
In order to clearly illustrate the effectiveness of proposed Uncertainty characterization using MCS
method, a comparison among the results of three different cases: In this section, MCS is used to deal with the aforementioned
uncertainties. The MCS is a numerical simulation procedure
(I) Deterministic optimization (ignoring the uncertainty in applied to the problems involving random variables with known
input parameter).
(II) Uncertainty characterization using TPEM.
(III) Uncertainty characterization using MCS.
14.5
The simulation results are described as follows.

Case I: IEEE-14 bus system 14

PL (MW)

Deterministic optimization
In Case I, only the mean value of load is considered in the MO-
ORPD problem. Real power loss and voltage deviation are consid- 13.5
ered as objective functions simultaneously.
In order to solve the multi-objective RPD problem by weighted
sum method, maximum and minimum values of the expected real
power loss (i.e. f1) and voltage deviation (i.e. f2) are calculated, 13
which are equal to 14.33382 MW, 13.08294 MW, 0.02925 pu and 0 0.005 0.01 0.015 0.02 0.025 0.03

0.00147 pu, respectively. These border values are achieved by VD (pu)

maximizing and minimizing the objective functions of MO-ORPD
Fig. 3. Pareto front of deterministic case for IEEE 14-bus system.
individually.

Table 1
Pareto optimal solution of deterministic case for IEEE 14-bus system.

# w1 w2 PL (MW) VD (pu) J 1;pu ¼ PLPLmax PL

J 2;pu ¼ VDVDmax VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 13.08294 0.02925 1 0 0

2 0.9 0.1 13.10148 0.01631 0.98518 0.46584 0.46584
3 0.8 0.2 13.13466 0.01196 0.95865 0.62236 0.62236
4 0.7 0.3 13.1677 0.00972 0.93224 0.70321 0.70321
5 0.6 0.4 13.19957 0.00839 0.90676 0.75105 0.75105
6 0.5 0.5 13.24408 0.00719 0.87118 0.79415 0.79415
7 0.4 0.6 13.29397 0.00628 0.83130 0.82695 0.82695
8 0.3 0.7 13.45718 0.00435 0.70082 0.89629 0.70082
9 0.2 0.8 13.64897 0.00293 0.54750 0.94755 0.54750
10 0.1 0.9 13.92241 0.00190 0.32890 0.98465 0.32890
11 0.0 1.0 14.33382 0.00147 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.
200 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204

Table 2
Pareto optimal solution of TPEM for IEEE 14-bus system.

# w1 w2 PL (MW) VD (pu) J1;pu ¼ PLPLmax PL

J 2;pu ¼ VDVDmax VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 13.09301 0.02856 1 0 0

2 0.9 0.1 13.11238 0.0162 0.98454 0.45635 0.45635
3 0.8 0.2 13.14655 0.01185 0.95725 0.61699 0.61699
4 0.7 0.3 13.18318 0.00947 0.928 0.70488 0.70488
5 0.6 0.4 13.21372 0.00823 0.90363 0.7506 0.7506
6 0.5 0.5 13.2563 0.00712 0.86962 0.79179 0.79179
7 0.4 0.6 13.30514 0.00625 0.83064 0.82377 0.82377
8 0.3 0.7 13.47842 0.00427 0.69229 0.89704 0.69229
9 0.2 0.8 13.67139 0.00288 0.53822 0.94833 0.53822
10 0.1 0.9 13.94164 0.00189 0.32245 0.98492 0.32245
11 0.0 1.0 14.3455 0.00148 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.

or assumed probability distributions. It consists of repeating a 14.5

deterministic simulation process, where in each simulation, a par- BCS characteristics:
ticular set of values for the random variables are generated accord- PL mean 13.30991
PL SD 0.71726
ing to the corresponding probability distributions. The result of PL Var 0.51446
MCS is similar to a sample of an experimental observation. By col- 14 VD mean 0.006128
VD SD 0.00033

PL (MW)
lecting the results of many such simulations, it is possible to apply VD Var 0
the methods of statistical estimation and inference to the data set
obtained. BCS
For load buses, the random variable to be treated by the 13.5

simulation procedure is load considering its stochastic behavior.

It is assumed that this variable is normally distributed with a
known mean value (corresponding to the forecasted value) and a
13
5% standard deviation. 0 0.005 0.01 0.015 0.02 0.025 0.03
The appropriate values of random variables are generally VD (pu)
achieved by inverting the cumulative distribution function. In
Fig. 5. Pareto front of MCS for IEEE 14-bus system.

14.5 Table 4
Control variables for the best solution in IEEE 14-bus system.

Control variables Deterministic approach MCS (mean) TPEM (mean)

Generator variables
14
Vg1 (pu) 1.06 1.05921 1.059895
PL (MW)

Vg2 (pu) 1.03528 1.03455 1.035015

Vg3 (pu) 0.99919 0.99862 0.99898
Vg6 (pu) 1.02437 1.02407 1.02404
Vg8 (pu) 1.00945 1.0096 1.008995
13.5
X: 0.006251 Pg1 (MW) 227.294 227.158 227.3203
Y: 13.31
Shunt compensation (A  QC)
Qc9 (MVar) 0 0 0
Qc14 (MVar) 16 0.9462  6 16
13
0 0.005 0.01 0.015 0.02 0.025 0.03 Transformer tap changer
T4–7 (pu) 0.96816 0.96732 0.96883
VD (pu)
T4–9 (pu) 1.08422 1.08153 1.082185
T5–6 (pu) 0.98957 0.98926 0.989645
Fig. 4. Pareto front of TPEM for IEEE 14-bus system.

Table 3
Pareto optimal solution of MCS for IEEE 14-bus system.

# w1 w2 PL (MW) VD (pu) J 1;pu ¼ PLPLmax PL

J2;pu ¼ VDVD max VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 13.08053 0.028124 1 0 0

2 0.9 0.1 13.0999 0.016391 0.98430 0.44102 0.44102
3 0.8 0.2 13.13422 0.012028 0.95648 0.60500 0.60500
4 0.7 0.3 13.16835 0.009778 0.92881 0.68957 0.68957
5 0.6 0.4 13.20198 0.008429 0.90154 0.74028 0.74028
6 0.5 0.5 13.2492 0.007181 0.86326 0.78720 0.78720
7 0.4 0.6 13.30991 0.006128 0.81405 0.82676 0.81405
8 0.3 0.7 13.47235 0.004255 0.68236 0.89716 0.68236
9 0.2 0.8 13.66353 0.002876 0.52738 0.94901 0.52738
10 0.1 0.9 13.91945 0.001927 0.31991 0.98468 0.31991
11 0.0 1.0 14.31407 0.001519 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.
 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204 201

Table 5
Pareto optimal solution of deterministic case for IEEE 30-bus system.

# w1 w2 PL (MW) VD (pu) J1;pu ¼ PLPL max PL

J2;pu ¼ VDVDmax VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 4.27644 0.02331 1 0 0

2 0.9 0.1 4.31683 0.01284 0.94982 0.47461 0.47461
3 0.8 0.2 4.40142 0.00569 0.84474 0.79893 0.79893
4 0.7 0.3 4.43982 0.00411 0.79904 0.87062 0.79904
5 0.6 0.4 4.49653 0.00269 0.72659 0.93511 0.72659
6 0.5 0.5 4.52067 0.00228 0.69660 0.95364 0.6966
7 0.4 0.6 4.54198 0.00205 0.67013 0.96407 0.67013
8 0.3 0.7 4.57093 0.00184 0.63417 0.97332 0.63417
9 0.2 0.8 4.61642 0.00165 0.57765 0.98226 0.57765
10 0.1 0.9 4.71905 0.00142 0.45016 0.99265 0.45016
11 0.0 1.0 5.08142 0.00126 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.

5.2 5.3

5.1 5.2

5 5.1

4.9 5

4.8 4.9
PL (MW)

PL (MW)
4.7 4.8

4.6 4.7

4.5 4.6
X: 0.004075
4.4 4.5 Y: 4.464

4.3 4.4

4.2 4.3
0 0.005 0.01 0.015 0.02 0.025 0 0.005 0.01 0.015 0.02 0.025
VD (pu) VD (pu)

Fig. 6. Pareto front of deterministic case for IEEE 30-bus system. Fig. 7. Pareto front of TPEM for IEEE 30-bus system.

Table 6
Pareto optimal solution of TPEM for IEEE 30-bus system.

# w1 w2 PL (MW) VD (pu) J1;pu ¼ PLPL max PL

J2;pu ¼ VDVDmax VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 4.30268 0.02379 1 0 0

2 0.9 0.1 4.33289 0.01425 0.96803 0.42378 0.42378
3 0.8 0.2 4.41618 0.00651 0.87991 0.76750 0.76750
4 0.7 0.3 4.46399 0.00408 0.82932 0.87547 0.82932
5 0.6 0.4 4.50329 0.00294 0.78774 0.92578 0.78774
6 0.5 0.5 4.5299 0.00244 0.75959 0.94828 0.75959
7 0.4 0.6 4.55421 0.00213 0.73386 0.96167 0.73386
8 0.3 0.7 4.58568 0.00188 0.70057 0.97295 0.70057
9 0.2 0.8 4.63158 0.00165 0.65201 0.98306 0.65201
10 0.1 0.9 4.73173 0.00141 0.54604 0.99381 0.54604
11 0.0 1.0 5.24781 0.00127 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.

Table 7
Pareto optimal solution of MCS for IEEE 30-bus system.

# w1 w2 PL (MW) VD (pu) J1;pu ¼ PLPL max PL

J2;pu ¼ VDVDmax VD minðJ1;pu ; J 2;pu Þ
max PLmin max VDmin

1 1.0 0.0 4.29858 0.02343 1 0 0

2 0.9 0.1 4.32877 0.01442 0.96781 0.40721 0.40721
3 0.8 0.2 4.41204 0.00661 0.87901 0.76024 0.76024
4 0.7 0.3 4.45897 0.00421 0.82896 0.86850 0.82896
5 0.6 0.4 4.49929 0.00301 0.78596 0.92281 0.78596
6 0.5 0.5 4.52747 0.00246 0.75591 0.94768 0.75591
7 0.4 0.6 4.55584 0.00209 0.72565 0.96452 0.72565
8 0.3 0.7 4.58361 0.00187 0.69604 0.97424 0.69604
9 0.2 0.8 4.62892 0.00165 0.64773 0.98444 0.64773
10 0.1 0.9 4.7242 0.00144 0.54612 0.99394 0.54612
11 0.0 1.0 5.23631 0.00130 0 1 0

The bold numbers correspond to the best compromise Pareto optimal solution.
202 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204

5.6 solutions, Solution#7 is the BCS. The Pareto optimal front of the
BCS characteristics
PL mean 4.45897
objective functions is derived, which is depicted in Fig. 5. The char-
5.4 PL SD 0.21608
PL Var 0.04669
acteristics of this solution, i.e. its mean and standard deviation for
VD mean 0.00421 both objectives are also given in this figure.
5.2 VD SD 0.00072
VD Var 0

Control variables in Case I

PL (MW)

5
Table 4 summarizes the obtained control variables for Case I. It
4.8 is observed that the obtained results by TPEM method is very close
BCS
to those obtained by MCS, which shows the efficiency of the TPEM
4.6
method.
4.4
Case II: IEEE-30 bus system
4.2
0 0.005 0.01 0.015 0.02 0.025
Deterministic optimization
VD (pu)
Similar to Case-I, the Pareto front is obtained for IEEE 30-bus
Fig. 8. Pareto front of MCS for IEEE 30-bus system. test system without considering load uncertainty. Table 5 summa-
rizes the information of the Pareto solutions for this case. Fig. 6
shows the Pareto front for Case II. It is evident from the last column
of Table 5 that the BCS is Solution#4, with the maximum weakest
Table 8
Control variables for the best solution in IEEE 30-bus system. membership function of 0.79904. The corresponding MO-ORPD
problem PL and VD are equal to 4.43982 MW and 0.00411 pu,
Control variables Deterministic approach MCS (mean) TPEM (mean)
respectively.
Generator variables
Vg1 (pu) 1.02287 1.01984 1.01962
Uncertainty characterization using TPEM
Vg2 (pu) 1.01621 1.01302 1.01296
Vg5 (pu) 1.00636 1.00347 1.00342 The load uncertainty is modeled using TPEM for Case II and
Vg8 (pu) 0.99956 0.099632 0.99638 results of Pareto solutions are presented in Table 6. Fig. 7 shows
Vg11 (pu) 1.00577 1.00389 1.00322 the Pareto front for Case II with considering load uncertainty using
Vg13 (pu) 1.01696 1.01397 1.01353
TPEM.
Pg1 (MW) 60.30136 60.8497 60.9537
Shunt compensation (QC  A)
Uncertainty characterization using MCS
Qc10 (MVar) 0 0 0
Qc12 (MVar) 0 0 0
The MCS method is used for load uncertainty modeling in this
Qc15 (MVar) 16 0.7481  6 16 case. Similar to Case I, 10,000 different samples are selected for
Qc17 (MVar) 16 0.9484  6 16 loads based on the normal distribution. Table 7 summarizes the
Qc20 (MVar) 0 0.0515  6 0 obtained results using MCS for Case II. Also, the optimal Pareto
Qc21 (MVar) 16 0.9785  6 16
front is depicted in Fig. 8. It is observed from Table 7 that the
Qc23 (MVar) 0 0 0
Qc24 (MVar) 16 0.8756  6 16 BCS is Solution#4 in this case. The characteristics of this solution,
Qc29 (MVar) 0 0.5748  6 0.5  6 i.e. its mean and standard deviation for both objectives are also
Transformer tap changer given in Fig. 8.
T6–9 0.97062 0.9685 0.96949
T6–10 1.1 1.0998 1.1 Control variables in Case II
T4–12 0.9809 0.9746 0.98136
Table 8 summarizes the obtained control variables for Case II.
T28–27 0.99198 0.9947 0.99654
Similar to Case I, the results obtained by TPEM approach is close
to those obtained by MCS, which means that the TPEM method
could be employed to deal with uncertainties in the case of uncer-
particular, the MATLAB function RANDN provides normally tain MO-ORPD problem.
distributed random numbers directly.
In this method, 10,000 random variables are selected for consid- Comparison and discussion
ering the stochastic behavior of the loads. Due to the large number
of MCS samples, just some statistical parameters such as mean, Table 9 compares the obtained results for Case I and Case II with
standard deviation (SD) and variance (Var) of the answers are pre- the previously published works. As can be observed from this table,
sented here. Table 3 shows the mean value of both objective func- the proposed model can obtain better results compared with the
tions for all 11 Pareto optimal solutions. Among these optimal heuristic methods.

Table 9
Comparison of obtained results for deterministic cases with previously published methods.

Case # Solution Values

PL (MW) VD (pu)
Case I Method Proposed DE [10] PSO [10] ACO [7] Proposed
PL minimization 13.08294 13.239 13.25 13.1226 0.02925
VD minimization 14.33382 0.00147
BCS 13.29397 0.00628
Case II Method Proposed DE [43] QOTLBO [12] HBMO [22] Proposed QOTLBO [12] HBMO [22] DE [43]
PL minimization 4.27644 4.555 4.5594 4.40867 0.02331 1.9057 0.87364 1.9589
VD minimization 5.08142 6.4755 6.4962 5.20924 0.00126 0.0856 0.2106 0.0911
BCS 4.43982 5.2594 5.53522 0.00411 0.1210 0.87664
 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204 203

Table 10 [6] Khazali A, Kalantar M. Optimal reactive power dispatch based on harmony
Comparison between the results of TPEM and other methods in Case I (IEEE 14-bus search algorithm. Int J Electr Power Energy Syst 2011;33:684–92.
system). [7] Ghasemi M, Ghanbarian MM, Ghavidel S, Rahmani S, Mahboubi Moghaddam E.
Modified teaching learning algorithm and double differential evolution
Case I Deterministic approach TPEM MCS algorithm for optimal reactive power dispatch problem: a comparative
study. Inform Sci 2014;278:231–49.
PL (MW) 13.29397 13.30514 13.30991
[8] Das DB, Patvardhan C. Reactive power dispatch with a hybrid stochastic search
VD (pu) 0.00628 0.00625 0.006128 technique. Int J Electr Power Energy Syst 2002;24:731–6.
Number of run 1 56 10,000 [9] Hu Z, Wang X, Taylor G. Stochastic optimal reactive power dispatch:
Computing time (min) 0.374 14.862 315.278 formulation and solution method. Int J Electr Power Energy Syst
2010;32:615–21.
[10] Varadarajan M, Swarup K. Differential evolutionary algorithm for optimal
Table 11 reactive power dispatch. Int J Electr Power Energy Syst 2008;30:
Comparison between the results of TPEM and other methods in Case II (IEEE 30-bus 435–41.
system). [11] Abido M, Bakhashwain J. Optimal VAR dispatch using a multiobjective
evolutionary algorithm. Int J Electr Power Energy Syst 2005;27:13–20.
Case II Deterministic approach TPEM MCS [12] Mandal B, Roy PK. Optimal reactive power dispatch using quasi-oppositional
teaching learning based optimization. Int J Electr Power Energy Syst
PL (MW) 4.43982 4.46399 4.45897 2013;53:123–34.
VD (pu) 0.00411 0.00408 0.00421 [13] Darabian M, Mohseni Bonab SM, Mohammadi-Ivatloo B. Improvement of
Number of run 1 120 10,000 power system stability by optimal SVC controller design using shuffled frog-
Computing time (min) 1.399 147.32 1515.02 leaping algorithm. IETE J Res 2015;61(2):160–9.
[14] Rabiee A, Vanouni M, Parniani M. Optimal reactive power dispatch for
improving voltage stability margin using a local voltage stability index. Energy
Convers Manage 2012;59:66–73.
Tables 10 and 11 compare the obtained objective functions (J1 [15] Chen G, Liu L, Song P, Du Y. Chaotic improved PSO-based multi-objective
and J2) value for the BCS in TPEM, MCS and deterministic approach optimization for minimization of power losses and L index in power systems.
for Cases I and II, respectively. On the other hand, because of consid- Energy Convers Manage 2014;86:548–60.
[16] Pal BB, Biswas P, Mukhopadhyay A. GA based FGP approach for optimal
ering the load uncertainty in the TPEM and MCS, the total expected
reactive power dispatch. Proc Technol 2013;10:464–73.
real power losses in both cases are higher than the deterministic [17] Dai C, Chen W, Zhu Y, Zhang X. Reactive power dispatch considering voltage
approach. Moreover, it is inferred from these tables that the results stability with seeker optimization algorithm. Electr Power Syst Res
2009;79:1462–71.
obtained by the TPEM and MCS are very close. This means the fact
[18] Jeyadevi S, Baskar S, Babulal C, Willjuice Iruthayarajan M. Solving
that the TPEM is an accurate method for dealing with such a prob- multiobjective optimal reactive power dispatch using modified NSGA-II. Int J
abilistic model. However, the number of runs and execution time of Electr Power Energy Syst 2011;33:219–28.
TPEM is much less than MCS. Therefore, with a reasonable approx- [19] Saraswat A, Saini A. Multi-objective optimal reactive power dispatch
considering voltage stability in power systems using HFMOEA. Eng Appl
imation for both objective functions values, the performance of Artif Intell 2013;26(1):390–404.
TPEM is desired in probabilistic MO-ORPD problem. [20] Ghasemi M, Ghavidel S, Ghanbarian MM, Habibi A. A new hybrid algorithm for
optimal reactive power dispatch problem with discrete and continuous control
variables. Appl Soft Comput 2014;22:126–40.
Conclusions
[21] Mallipeddi R, Jeyadevi S, Suganthan PN, Baskar S. Efficient constraint handling
for optimal reactive power dispatch problems. Swarm Evol Comput
Multi objective reactive power dispatch (MO-ORPD) problem is 2012;5:28–36.
studied in this paper considering the load uncertainty. The objec- [22] Ghasemi A, Valipour K, Tohidi A. Multi objective optimal reactive power
dispatch using a new multi objective strategy. Int J Electr Power Energy Syst
tive functions used in the proposed probabilistic MO-ORPD prob- 2014;57:318–34.
lem are real power losses and voltage deviations (from their [23] Soroudi A, Rabiee A. Optimal multi-area generation schedule considering
corresponding nominal values). The stochastic behavior of load is renewable resources mix: a real-time approach. IET Gener Transm Distrib
2013;7(9):1011–26.
simulated using two-point estimate method (TPEM) and Monte [24] Mohseni-Bonab SM, Rabiee A, Mohammadi-Ivatloo B. Voltage stability constrained
Carlo simulation (MCS). Mixed integer nonlinear programming multi-objective optimal reactive power dispatch under load and wind power
model is developed for the proposed MO-ORPD problem. The pro- uncertainties: a stochastic approach. Renew Energy 2016;85:598–609.
[25] Alipour M, Mohammadi-Ivatloo B, Zare K. Stochastic risk-constrained short-
posed method is implemented and analyzed on two standard test term scheduling of industrial cogeneration systems in the presence of demand
cases, and its effectiveness is verified using different simulations response programs. Appl Energy 2014;136:393–404.
and comparisons. The proposed model yields better results in com- [26] Rabiee A, Soroudi A, Keane A. Information gap decision theory based OPF with
HVDC connected wind farms. IEEE Trans Power Syst 2014.
parison with previously proposed heuristic algorithms for deter-
[27] Kazemi M, Mohammadi-Ivatloo B, Ehsan M. Risk-constrained strategic bidding
ministic cases. Also, the results of case studies show that the of GenCos considering demand response. IEEE Trans Power Syst 2014;30
obtained values for both objective functions in MO-ORPD using (1):376–84.
[28] Soroudi A, Ehsan M. IGDT based robust decision making tool for DNOs in load
TPEM is very close to the corresponding values obtained by MCS.
procurement under severe uncertainty. IEEE Trans Smart Grid 2013;4
However, the number of runs and the execution time of TPEM (2):886–95.
are much less than the MCS. Therefore, in order to save the compu- [29] Soroudi A. Robust optimization based self scheduling of hydro-thermal Genco
tation time while maintaining the reasonable approximation for in smart grids. Energy 2013;61:262–71.
[30] Nojavan S, Mohammadi-Ivatloo B, Zare K. Robust optimization based price-
the objective functions value, the TPEM is preferred to deal with taker retailer bidding strategy under pool market price uncertainty. Int J Electr
the probabilistic MO-ORPD problem. Power Energy Syst 2015;73:955–63.
[31] Soroudi A, Amraee T. Decision making under uncertainty in energy systems:
References state of the art. Renew Sustain Energy Rev 2013;28:376–84.
[32] Baran ME, Wu FF. Network reconfiguration in distribution systems for loss
reduction and load balancing. IEEE Trans Power Deliv 1989;4(2):1401–7.
[1] Zehar K, Sayah S. Optimal power flow with environmental constraint using a
[33] Ochoa LF, Harrison GP. Minimizing energy losses: optimal accommodation and
fast successive linear programming algorithm: application to the algerian
smart operation of renewable distributed generation. IEEE Trans Power Syst
power system. Energy Convers Manage 2008;49(11):3362–6.
2011;26(1):198–205.
[2] Shi L, Wang C, Yao L, Ni Y, Bazargan M. Optimal power flow solution
[34] Mohseni-Bonab SM, Rabiee A, Jalilzadeh S, Mohammadi-Ivatloo B, Nojavan S.
incorporating wind power. IEEE Syst J 2012;6:233–41.
Probabilistic multi objective optimal reactive power dispatch considering load
[3] Rabiee A, Parniani M. Optimal reactive power dispatch using the concept of
uncertainties using Monte Carlo simulations. J Oper Autom Power Eng 2015;3
dynamic VAR source value. In: Power & energy society general meeting, 2009
(1):83–93.
PES’09 IEEE. IEEE; 2009. p. 1–5.
[35] Deb K. Multi-objective optimization using evolutionary algorithms. John Wiley
[4] Zhao B, Guo C, Cao Y. A multiagent-based particle swarm optimization
& Sons; 2001.
approach for optimal reactive power dispatch. IEEE Trans Power Syst 2005;20
[36] Rabiee A, Soroudi A, Mohammadi-Ivatloo B, Parniani M. Corrective voltage
(2):1070–8.
control scheme considering demand response and stochastic wind power. IEEE
[5] Zhang Y-j, Ren Z. Optimal reactive power dispatch considering costs of
Trans Power Syst 2014;29:2965–73.
adjusting the control devices. IEEE Trans Power Syst 2005;20(3):1349–56.
204 S.M. Mohseni-Bonab et al. / Electrical Power and Energy Systems 75 (2016) 194–204

[37] Brooke A, Kendrick D, Meeraus A. GAMS release 2.25: a user’s [41] Su C-L. Probabilistic load-flow computation using point estimate method. IEEE
guide. Washington, DC: GAMS Development Corporation; 1996. Trans Power Syst 2005;20(4):1843–51.
[38] Drud AS. CONOPT—a large-scale GRG code. ORSA J Comput 1994;6:207–16. [42] Zimmerman RD, Murillo-Sánchez CE, Thomas RJ. MATPOWER: steady-state
[39] The GAMS Software Website; 2013. <http://www.gams.com/dd/docs/solvers/ operations, planning, and analysis tools for power systems research and
sbb.pdf1>. education. IEEE Trans Power Syst 2011;26:12–9.
[40] Morales JM, Perez-Ruiz J. Point estimate schemes to solve the probabilistic [43] El Ela AA, Abido M, Spea S. Differential evolution algorithm for optimal
power flow. IEEE Trans Power Syst 2007;22(4):1594–601. reactive power dispatch. Electr Power Syst Res 2011;81(2):458–64.