Electrical Power and Energy Systems 81 (2016) 64–77

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Optimal power flow with emission and non-smooth cost functions using
backtracking search optimization algorithm
A.E. Chaib a,⇑, H.R.E.H. Bouchekara a,b, R. Mehasni a, M.A. Abido c
a
Constantine Electrical Engineering Laboratory, LEC, Department of Electrical Engineering, University of Freres Mentouri Constantine, 25000 Constantine, Algeria
b
Laboratory of Electrical Engineering of Constantine, LGEC, Department of Electrical Engineering, University of Freres Mentouri Constantine, 25000 Constantine, Algeria
c
Electrical Engineering Department, King Fahd University of Petroleum and Minerals, Dhahran 31261, Saudi Arabia

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

Article history: Economic operation of electric energy generating systems is one of the prevailing problems in energy
Received 11 December 2014 systems. In this paper, a new method called the Backtracking Search Optimization Algorithm (BSA) is
Received in revised form 14 January 2016 proposed for solving the optimal power flow problem. This method is tested for 16 different cases on
Accepted 4 February 2016
the IEEE 30-bus, IEEE 57-bus, and IEEE 118-bus test systems. In addition to the traditional generating fuel
cost, multi-fuels options, valve-point effect and other complexities have been considered. Furthermore,
different objectives such as voltage profile improvement, voltage stability enhancement and emission
Keywords:
reduction are considered. The obtained results are compared with those obtained using some
Backtracking search optimization
Energy efficiency
well-known optimization algorithms. This comparison highlights the effectiveness of the BSA method
Optimal power flow for solving different OPF problems with complicated and non-smooth objective functions.
Power system optimization Ó 2016 Elsevier Ltd. All rights reserved.
Environmental pollution
Valve-point effect

Introduction Frank et al. in [4] presented a comprehensive survey of various tra-

There are three types of problems commonly referred to in
power system literature: power flow, economic dispatch, and opti-
mal power flow [1]. Economic dispatch (ED), applies several for-
mulations to determine the least-cost generation dispatch to
satisfy the total required demand, however these formulations
simplify or even ignore power flow constraints [1]. The optimal
power flow (OPF) has been initially proposed by French scholar
Carpentier in 1962 [1]. Since then, it becomes one of the most
important functions of operation, production, control and monitor-
ing of power in modern energy systems [1,2].
A basic OPF problem seeks optimal distribution of the real
power and/or the reactive power by adjusting the control variables,
so that a specific objective in the operation of electric power sys-
tem is optimized. During the optimization, the power flow balance,
generator capability, transmission capability, voltage profile con-
straints must be satisfied.
Various traditional optimization techniques have been used to
solve the OPF problem. These include linear programming
(1979), Newton methods (1992), interior point methods (1998)
and dynamic programming (2001). Pandya and Joshi in [3] and
⇑ Corresponding author. Tel.: +213 777036601; fax: +213 31908113.
E-mail address: chaiballaeddine1986@hotmail.fr (A.E. Chaib).
ditional optimization methods used to solve OPF problems. How-
ever, in practice, traditional methods suffer from some
inadequacy. Some of their shortcomings among others are: firstly,
they do not guarantee finding the global optimum, secondly, tradi-
tional methods involve complex calculations with long time, and
they are not adapted for discrete variables [4].
Over the past few decades, many powerful metaheuristics have
been developed. Some of them have been applied to the OPF prob-
lem with impressive success. Some of the recent applications of
metaheuristics to OPF problem are: Artificial Bee Colony (ABC)
[5,6], Black Hole (BH) [7], Teaching Learning Based Optimization
(TLBO) [8], League Championship Algorithm (LCA) [9], Differential
Search Algorithm (DSA) [10], Krill Herd Algorithm (KHA) [11],
Gravitational Search Algorithm (GSA) [12,13], Imperialist Compet-
itive Algorithm (ICA) [14] and Group Search Optimization (GSO)
[15]. A review of many metaheuristics applied to solve the OPF
problem is given in [16,17]. However, due to the variability of
the objectives while solving OPF problems, no algorithm is the best
in solving all OPF problems. Therefore, there is always a need for a
new algorithm, which can efficiently solve the majority of the OPF
problems.
The Backtracking Search Algorithm (BSA) which is developed by
Civicioglu [18] is a new evolutionary algorithm (EA) developed to
solve real-valued numerical optimization problems. As other EAs
like genetic algorithm (GA) and differential evolution (DE), BSA is

http://dx.doi.org/10.1016/j.ijepes.2016.02.004
0142-0615/Ó 2016 Elsevier Ltd. All rights reserved.
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 65

Nomenclature

ai ; bi ; ci cost coefficient of the it ith generator QC shunt VAR compensation

D dimensions (the number of parameter to be opti- Q G, Q D reactive power generation and load demand, respec-
mized) tively
di ; e i coefficients reflecting the valve-point effect Sli ; Smax
li apparent power flow of ith line and its maximum va-
F Scale factor lue
g ðx; uÞ set of equality constraints T max
i ; T min
i upper and lower limits of tap settings of regulating
Gij , Bij conductance and susceptance of the admittance ma- transformer i, respectively
trix entry ij, respectively u vector of independent variables or control variables
hðx; uÞ set of inequality constraints upj predefined upper limits of problem
J ðx; uÞ objective function V max
Gi ; V Gi
min
upper and lower limits of voltage magnitude upper
Lj voltage stability local indicator of bus j limits of bus i
N population size (the number of individuals) V Li ; V lim
Li voltage magnitude of load bus j and its limit
NB number of buses V max
Li ; V Li
min
upper and lower limits of voltage magnitude load bus
NC number of VAR compensators j, respectively
NG number of generators V Lj voltage magnitude at load bus j
NL number of load buses VD load bus voltage deviation
nl number of transmission lines VG voltage magnitude at PV buses
NT the number of regulating transformers x vector of dependent variables or state variables
ObjFun objective function used in the problem xmax ; xmin upper and lower limits of variable x
oldP historical population Y ij admittance matrix between bus i and bus j
P population ai ; bi ; ci ; xi and li coefficients of the ith generator emission char-
PG1 ; Plim
G1 active power generation of slack bus and its limit acteristics
Pmax min
Gi ; P Gi upper and lower limits of active power generation of hi phase angle of bus i
bus i, respectively hij phase angle difference between buses i and j
PG , PD active power generation and load demand, respec- kLmax weighting factor of the Lmax term compared with the
tively cost term
Ploss, Qloss active and reactive power transmission losses, respec- kEmission weighting factor of the emission term compared with
tively the cost term
Q max min
Ci ; Q Ci upper and lower limits of reactive power generation kVD weighting factor of the VD term compared with the
of the compensator capacitors j, respectively cost term.
Q Gi ; Q lim
Gi reactive power generation of bus j and its limit kP ; kQ ; kS ; kV penalty factors of P, Q, S, V respectively.
Q max
Gi ; Q Gi
min
upper and lower limits of reactive power generation
of unit i, respectively

based on three basic and well-known operators that are selection,
mutation and crossover. Furthermore, unlike many other meta-
heuristics, the BSA algorithm has only one control parameter and
it is not very sensitive to the initial value of this parameter as
reported in [18]. Since it was introduced, the BSA has attracted
many researches and it has been applied to various optimization
problems. The following are some successful examples. In [19] a
comparative analysis of BSA with other evolutionary algorithms
for global continuous optimization is given. In [20] the BSA has
been used for antenna array design. In [21] the BSA was employed
for the design of robust Power System Stabilizers (PSSs) in
multimachine power systems. In [22] it has been used for the
allocation of multi-type distributed generators along distribution
networks.
Therefore, this paper seeks to apply to the OPF problem a new
evolutionary algorithm that has not received yet much attention
in the power systems community that is the BSA. Furthermore,
in this paper, not only the basic OPF problem is investigated, but
also some complex formulations with non-smooth cost functions
and different objective functions are considered along with the
environmental concern imposing the reduction of emission.
The remainder of the paper is organized as follows. First, the
OPF formulation is presented in brief in section 2. Then, the main
features of the BSA algorithm are presented in section 3. Next,
the results after solving different cases of OPF problem using BSA
are discussed in section 4. Finally, conclusions are drawn in the last
section of this paper.
Problem formulation
The OPF is a power flow problem which gives the optimal set-
tings of the control variables for a given settings of load by mini-
mizing a predefined objective function such as the cost of active
power generation or transmission losses. OPF considers the operat-
ing limits of the system and it can be formulated as a nonlinear
constrained optimization problem as follows:
Minimize J ðx; uÞ ð1Þ
Subject to g ðx; uÞ ¼ 0 ð2Þ
and hðx; uÞ 6 0 ð3Þ
where u is the vector of independent variables or control variables,
x is the vector of dependent variables or state variables, Jðx; uÞ is the
objective function, g ðx; uÞ is the set of equality constraints and
hðx; uÞ is the set of inequality constraints.
The control variables u and the state variables x of the OPF
problem are defined as follows.
Control variables
These are the set of variables which can be adjusted to satisfy
the load flow equations [9]. The set of control variables in the
OPF problem formulation are:
66 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

V min max
Gi 6 V Gi 6 V Gi ; i ¼ 1; . . . ; NG ð8Þ
PG active power generation at the PV buses except at the
slack bus.
VG voltage magnitude at PV buses. P min max
Gi 6 P Gi 6 P Gi ; i ¼ 1; . . . ; NG ð9Þ
T tap settings of transformer.
QC shunt VAR compensation. Q min max
Gi 6 Q Gi 6 Q Gi ; i ¼ 1; . . . ; NG ð10Þ

(b) Transformer constraints

Hence, u can be expressed as:

 Transformer tap settings must be restricted within their
uT ¼ PG2    PGNG ; V G1    V GNG ; Q C1    Q C NC ; T 1    T NT ð4Þ specified lower and upper limits as follows:

where NG, NT and NC are the number of generators, the number of T min
i 6 T i 6 T max
i ; i ¼ 1; . . . ; NT ð11Þ
regulating transformers and the number of VAR compensators,
(c) Shunt VAR compensator constraints
respectively.
Shunt VAR compensators must be bounded by their lower
and upper limits as follows:
State variables
Q min max
Ci 6 Q Ci 6 Q Ci ; i ¼ 1; . . . ; NC ð12Þ
These are the set of variables which describe any unique state of
the system [9]. The set of state variables for the OPF problem for- (d) Security constraints
mulation are: These constraints can be mathematically formulated as
follows:
P G1 active power at slack bus. V min 6 V Li 6 V max
Li Li ; i ¼ 1; . . . ; NL ð13Þ
VL voltage magnitude at PQ buses (load buses).
QG reactive power generation of all generator units.
Sli 6 Smax
li ; i ¼ 1; . . . ; nl ð14Þ
Sl transmission line loading (or line flow).

It is worth mentioning that control variables are self-

Hence, x can be expressed as: constrained. The inequality constraints of dependent variables,

 which contain load bus voltage magnitude; real power generation
xT ¼ PG1 ; V L1    V LNL ; Q G1    Q GNG ; Sl1    Slnl ð5Þ
at slack bus, reactive power generation and line loading can be
where, NL, and nl are the number of load buses and the number of included into the objective function as quadratic penalty terms.
transmission lines, respectively. In these terms, a penalty factor multiplied with the square of the
disregard value of dependent variable is added to the objective
Constraints function and any unfeasible solution obtained is declined [12].
Details about how penalty factors can be selected are given in
OPF constraints can be classified into equality and inequality [23]. Mathematically, penalty function can be expressed as
constraints, as detailed in the following sections. follows:
 2 NL 
X 2
Equality constraints Penalty ¼ kP PG1  Plim þ kV V Li  V lim
G1 Li
Set of constraints that embody the typical nonlinear power flow i¼1
equations that control the power system, given as follows. NG 
X 2 nl 
X 2
þ kQ Q Gi  Q lim
Gi þ kS Sli  Smax
li ð15Þ
(a) Real power constraints: i¼1 i¼0

X
NB

    where kP ; kV ; kQ and kS are penalty factors and xlim is the violated
PGi  PDi  V i V j Gij cos hij þ Bij sin hij ¼ 0 ð6Þ limit value of the dependent variable x and can be defined as
j¼i
follows:
(b) Reactive power constraints: 
xmax ; x > xmax
xlim ¼ ð16Þ
X
NB

    xmin ; x < xmin
Q Gi  Q Di  V i V j Gij sin hij  Bij cos hij ¼ 0 ð7Þ
j¼i
Objective function
where hij ¼ hi  hj , NB is the number of buses, P D is the active load
demand, Q D is the reactive load demand, Gij and Bij are the ele- Several objective functions are considered in this paper which
 
ments of the admittance matrix Y ij ¼ Gij þ jBij representing the are: cost reduction (with and without multi-fuels options or
conductance and susceptance between bus i and bus j, valve-point effect), voltage profile improvement, voltage stability
respectively. enhancement and emission reduction. Moreover, some combina-
tions between these objective functions are considered also.

Inequality constraints Backtracking Search Algorithm (BSA)

Set of constraints which reflect the system operational and the
physical limits of the system given as follows.
(a) Generator constraints
For all generators including the slack: voltage, active and
reactive outputs ought to be restricted by their lower and
upper limits as follows:
As previously mentioned, BSA is a new EA developed by Civi-
cioglu for solving real-valued numerical optimization problems
[18]. It has only one tuning parameter and it uses the three basic
operators that are selection, mutation and crossover. The general
structure of the BSA is given in Algorithm 1 shown below. From
this algorithm, it can be noticed that the optimization process of
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 67

BSA is achieved following five steps namely: initialization, Algorithm 2. Pseudocode of BSA [18].
selection-I, mutation, crossover, and selection-II. The following
subsections describe in details these steps. 1 function bsa (ObjFun, N, D, maxcycle, low, up)
Input: ObjFun, N, D, maxcycle, mixrate, low1:D, up1: D
Algorithm 1. General structure of BSA [18]. Output: globalminimum, globalminimizer
// INITIALIZATION
1. Initialization 2 globalminimum = inf
repeat 3 for i from 1 to N do
2. Selection-I 4 for j from 1 to D do
Generation of Trial-Population 5 Pi;j ¼ rnd:ðupj — lowj Þ þ lowj // Initialization of
3. Mutation population, P.
4. Crossover 6 oldPi;j ¼ rnd:ðupj — lowj Þ þ lowj // Initialization of oldP.
end 7 end
5. Selection-II 8 fitness Pi = ObjFun ðPi Þ // Initial-fitness values of P.
until stopping conditions are met 9 end
10 for iteration from 1 to maxcycle do
// SELECTION-I
11 if (a < bja, b  U(0,1)) then oldP:= P end
Initialization
12 oldP: = permuting (oldP)
13 Generation of Trial-Population
As many other metaheuristics, in the initialization step the BSA
// MUTATION
generates a random initial population P of N individuals in the
14 mutant = P + F. (oldP — P)
search space that has D dimensions as shown in lines (3–9) of
// CROSSOVER
Algorithm 2.
15 Map1:N;l:D ¼ 1 // Initial-map is an N-by-D matrix of
ones.
Selection-I
16 if (c < djc, d  U(0,1)) then
17 for i from 1 to N do
In this second step, a historical population called oldP is deter-
18
mined for calculation of the search direction. oldP is initialized like
mapi;uð1:dmixrate:rnd:De ¼ 0ju ¼ permutingðh1; 2; 3; . . . . . . ; DiÞ
the population P as shown in line (6) of Algorithm 2. However, oldP
can be redefined in every iteration using an ‘if–then’ rule as shown 19 end
in line (11) of Algorithm 2. After oldP is determined, the order of 20 else
individuals inside it is randomly permutated (line (12) of 21 for i from 1 to N do, mapi, rndi(D) = 0, end
Algorithm 2). 22 end
// Generation of Trial Population, T
Mutation 23 T: = mutant
24 for i from 1 to N do
In the mutation step, a mutant is generated as the initial form of 25 for j from i to D do
trial population (T) using the old population (oldP) and the popula- 26 if mapi, j = 1 then Ti,j:= Pij,
tion (P) itself as follows [18]: 27 end
28 end
Mutant ¼ P þ F  ðoldP  P Þ ð17Þ // Boundary Control Mechanism
where F is the scale factor used to control the search direction 29 for i from 1 to N do
matrix (oldP-P). It is worth to mention here that using the old pop- 30 for j from 1to D do
ulation in mutation allows BSA to benefit from the experience of 31 if (Ti, j <lowj) or (Ti, j > upj) then
previous generations [18]. The mutation step is shown in line 14 32 Ti, j= rnd. (upj —lowj) + lowj
in Algorithm 2. 33 end
34 end
Crossover 35 end
36 end
The crossover operator used in BSA is different from the one // SELECTION-II
used in DE or its variants. The crossover process represents the last 37 fitness T = ObjFn (T)
step while generating the trial population. This process starts by 38 for i from 1 to N do
generating a binary integer-valued matrix of size (N:D) denoted 39 if fitness Ti < fitness Pi then
mapn;m using two strategies as shown in Algorithm 2 (lines from 40 fitness Pi:= fitness Ti
(15) through (22)). This matrix indicates the individuals of T to Pi = Ti
41 end
be manipulated using the relevant individuals of P as shown in line
(26) of Algorithm 2 [18]. 42 end
43 fitness Pbest = min (fitness P) n best 2 {1, 2, 3. . . N}
Selection II 44 if fitness Pbest < globalminimum then
45 globalminimum: =fitness Pbest
In this step, the trial population T is used to update the popula- globalminimizer: = Pbest
// Export globalminimum and globalminimizer
tion P in a greedy manner as shown in Algorithm 2 (lines from (37)
through (42)). If two individuals have the same order, one from the 46 end
47 end
trial population T i and one from the population Pi , and T i has a bet-
ter fitness than P i , therefore, P i is replaced by T i [18].
68 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

Application and results Table 2

The main characteristics of the IEEE 30-bus test system.

It is worth mentioning that, the developed software program is System characteristics Value Details
written in the commercial MATLAB computing environment using Buses 30 –
some MATPOWER programs. It is applied on a 2.20 GHz i7 personal Branches 41 –
computer with 8.00 GB-RAM and using parallel processing to run Generators 6 Buses: 1, 2, 5, 8, 11 and 13
Shunts 9 Buses: 10, 12, 15, 17, 20, 21, 23, 24 and 29
the different runs.
Transformers 4 Branches: 11, 12, 15 and 36
In this paper, the proposed BSA algorithm has been applied on Control variables 24 –
three test systems that are the IEEE 30-bus, the IEEE 57-bus, and
IEEE-118 bus test systems. Moreover, 16 different case studies
are investigated. These cases start from basic cases then different in Table 5 where the optimal generation fuel cost obtained is
complexities are included and other objectives are considered. (799.0760 $/h). Furthermore, the evolution of the objective func-
These cases are summarized in Table 1. In our implementation, tion over iterations is sketched in Fig. 1. This figure shows the evo-
the transformer tap settings have been considered as discrete vari- lution of cost (without penalty) and the penalty term over
able with a step of 0.125 p.u. iterations. It can be seen that the penalty term tends to zero after
34 iterations only. In other words, the BSA finds a feasible solution
IEEE 30-bus test system for this case after 34 iterations.

The IEEE 30-bus test system has a total generation capacity of
435.0 MW and its main characteristics are given in Table 2.
Detailed data can be derived from [24]. Cost and emission coeffi-
cients used in this paper are given in Table 3. Table 4 gives cost
coefficients when considering multi-fuels options for generators
1 and 2.
CASE 1: OPF by considering cost reduction
The first case investigated in this paper is the base case, which
consists of minimizing the generation fuel cost expressed by a
quadratic function. Hence, the objective function (or more pre-
cisely the augmented objective function) for this case is:
! i
X
NG
Jðx; uÞ ¼ ai þ bi P Gi þ ci P2Gi þ Penalty
i¼1
where ai ; bi and ci are the cost coefficients of the ith generator.
CASE 2: OPF by considering cost reduction and voltage profile
improvement
Bus voltage is one of the most important and significant indica-
tion of safety and service quality of power systems. The solution
found in CASE 1 is a purely cost based objective function. Such
solution may have an undesirable voltage profile [25]. Thus, this
example aims at minimizing fuel cost with a flatter voltage profile
by considering a twofold objective function [26]. The voltage pro-
file is optimized by minimizing the load bus voltage deviation (VD)
from 1.0 p.u, which is given by:
X
NL
VD ¼ V L  1 ð19Þ
j¼1
ð18Þ
Therefore, the objective function can be expressed as:
!
X
NG

The proposed BSA technique has been run for this case and the
Jðx; uÞ ¼ ai þ bi PGi þ ci P2Gi þ kVD ðVDÞ þ Penalty ð20Þ
i¼1
optimal control variables obtained after optimization are tabulated

Table 1
Different case studies investigated in this paper.

Name Cost Voltage profile Voltage stability Multi-fuels Valve-point effect Emission System
CASE 1 – – – – – IEEE 30-bus test system

CASE 2 – – – –

CASE 3 – – – –

CASE 4 – – – –

CASE 5 – – –

CASE 6 – – –

CASE 7 – – – –

CASE 8 – – –

CASE 9 – – –

CASE 10 – – – –

CASE 11 – – – – – IEEE 57-bus test system

CASE 12 – – – –

CASE 13 – – – –

CASE 14 – – – –

CASE 15 – – – –

CASE 16 IEEE 118-bus test system

 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 69

Table 3
Cost and emission coefficients for the IEEE 30-bus test system.

Generator Bus a b c d e a b c x l
G1 1 0 2 0.00375 18 0.037 4.091 5.554 6.49 2.00E04 2.857
G2 2 0 1.75 0.0175 16 0.038 2.543 6.047 5.638 5.00E04 3.333
G3 5 0 1 0.0625 14 0.04 4.258 5.094 4.586 1.00E06 8
G4 8 0 3.25 0.00834 12 0.045 5.326 3.55 3.38 2.00E03 2
G5 11 0 3 0.025 13 0.042 4.258 5.094 4.586 1.00E06 8
G6 13 0 3 0.025 13.5 0.041 6.131 5.555 5.151 1.00E05 6.667

Table 4
Cost coefficients for generator 1 and 2 of the IEEE 30-bus test system for multi-fuels sources.

Generator Bus a b c P min P max a b c P min P max

G1 1 55 0.7 0.005 50 140 82.5 1.05 0.0075 140 200
G2 2 40 0.3 0.01 20 55 80 0.6 0.02 55 80

Table 5
Optimal settings of control variables found for CASE 1 through CASE 10.

CASE 1 CASE 2 CASE 3 CASE 4 CASE 5 CASE 6 CASE 7 CASE 8 CASE 9 CASE 10
PG1 177.3838 173.1714 172.0277 139.9204 139.5920 139.7202 198.7223 197.0323 197.6731 112.9189
PG2 48.8335 48.3093 48.8610 54.9886 54.2062 54.7933 44.3031 41.8254 46.0134 59.3719
PG5 21.2907 23.4476 22.3206 23.2095 24.4755 27.7068 18.5637 22.0463 15.6953 27.6576
PG8 21.0186 22.1097 20.9493 35.0000 28.6711 31.0154 10.0000 10.7320 12.4700 34.9989
PG11 11.4675 13.3916 13.5766 18.5930 22.2083 16.7496 10.1017 10.8690 10.0000 27.0652
PG13 12.0602 12.3455 14.1551 18.3118 21.5854 19.8743 12.0000 12.0000 12.0000 26.4502
VG1 1.1000 1.0441 1.0880 1.0863 1.0356 1.0902 1.1000 1.0514 1.0954 1.1000
VG2 1.0806 1.0245 1.0678 1.0699 1.0177 1.0720 1.0778 1.0278 1.0754 1.0855
VG5 1.0545 1.0037 1.0488 1.0403 1.0037 1.0463 1.0520 1.0086 1.0455 1.0606
VG8 1.0633 1.0005 1.0510 1.0532 1.0012 1.0568 1.0574 1.0018 1.0382 1.0757
VG11 1.0946 1.0316 1.0979 1.0679 1.0240 1.1000 1.0802 1.0263 1.1000 1.1000
VG13 1.1000 1.0049 1.0972 1.0541 1.0111 1.0903 1.0803 1.0061 1.0985 1.1000
T11(69) 1.0250 1.0500 0.9750 1.0625 1.0375 0.9625 1.0000 1.0250 0.9750 1.0000
T12(610) 0.9000 0.9000 0.9000 0.9125 0.9000 0.9250 1.0125 0.9125 0.9000 0.9500
T15(412) 0.9625 0.9625 0.9625 1.0000 0.9875 0.9625 1.0250 0.9625 0.9750 1.0000
T36(2827) 0.9625 0.9625 0.9500 0.9875 0.9625 0.9500 1.0000 0.9750 0.9375 0.9625
QC10 4.2998 5.0000 4.5315 5.0000 4.3687 4.4714 4.3411 3.8622 4.5090 3.4844
QC12 4.6378 0.7241 4.5338 5.0000 5.0000 4.5103 4.9527 1.9742 4.9177 4.5129
QC15 4.9106 3.7630 4.6750 5.0000 3.3418 5.0000 4.2358 2.4068 4.1622 4.7990
QC17 5.0000 2.3539 4.2197 5.0000 0.0000 5.0000 4.7605 0.0000 5.0000 4.9965
QC20 4.0889 4.9912 5.0000 3.1123 5.0000 4.9360 4.0597 5.0000 5.0000 3.9809
QC21 5.0000 3.6589 5.0000 5.0000 3.9460 3.8837 4.5901 5.0000 4.6871 4.7684
QC23 3.1843 4.9775 4.7058 3.9314 4.9261 4.8542 4.1971 4.5359 4.5496 3.8535
QC24 4.8423 4.8500 3.6102 5.0000 5.0000 3.4071 5.0000 4.9781 2.9377 4.2332
QC29 2.5810 2.2713 4.8119 1.4393 3.7995 4.1926 4.1450 4.2463 3.9664 1.6339
Cost ($/h) 799.0760 803.4294 800.3340 646.1504 653.1019 648.1424 830.7779 836.8811 832.7029 835.0199
VD 1.9129 0.1147 1.9855 1.0273 0.1161 1.9825 1.2050 0.1194 1.8985 1.9214
Lmax 0.1273 0.1484 0.1259 0.1378 0.1478 0.1258 0.1363 0.1491 0.1262 0.1266
Ploss (MW) 8.6543 9.3751 8.4904 6.6233 7.3386 6.4595 10.2908 11.1050 10.4519 5.0626
Qloss (MVar) 36.5617 41.0206 34.0897 30.6217 32.6230 26.2444 40.4990 45.2534 41.0987 22.1368
Emission (ton/h) 0.3671 0.3546 0.3514 0.2833 0.2805 0.2821 0.4377 0.4300 0.4343 0.2425

where kVD is a weighting factor which has to be selected carefully in

order to give the desired amount of weight or importance of the VD
×108
900 2 term compared with the cost term. After many tests, kVD is selected
as 1000 in this study to balance between both objectives.
Cost
The proposed BSA technique has been run for this case and the
Penalty
optimal control variables obtained are given in Table 5. The first
comment that can be concluded from this table is that VD has been
optimized compared with CASE 1, it has been reduced from
Cost ($/h)

Penalty

1
1.9129 p.u to 0.1147 p.u. This reduction can be visually verified
in Fig. 2 where the optimized voltage profile of load buses is greatly
improved compared to that of CASE 1. Furthermore, the evolution
of both objective functions over iterations is sketched in Fig. 3.

CASE 3: OPF by considering cost reduction and voltage stability

800
enhancement
0
1 125 250 375 500 Prediction of voltage instability is an issue of paramount impor-
Iterations tance in power systems. In [27] Kessel and Glavitch have devel-
oped a voltage stability index called Lmax which is defined based
Fig. 1. Objective function variation for CASE 1. on local indicators Lj and it is given by:
70 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

1.1

1.05
Voltage Profile (p.u.)

0.95

0.9
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
PQ Bus Number
(a)
1.1

1.05
Voltage Profile (p.u.)

0.95

0.9
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
PQ Bus Number
(b)
Fig. 2. Comparison of voltage profiles between (a) CASE 1 and (b) CASE 2.

860 1.2 where H matrix is generated by the partial inversion of Y bus . More
Cost details can be found in [27].
850 VD 1 The indicator Lmax varies between 0 and 1 where the lower the
indicator, the more the system stable. Therefore, enhancing voltage
840 0.8 stability can be achieved through the minimization of Lmax of the
VD (p.u.)
Cost ($/h)

whole system [28]. Thus, the objective function in this case is:
830 0.6
!
X
NG
Jðx; uÞ ¼ ai þ bi PGi þ ci P2Gi þ kLmax ðLmax Þ þ Penalty ð23Þ
820 0.4 i¼1

where kLmax is a weighting factor selected as 6000 in this study.

810 0.2 The results of the optimization study are given in Table 5 while
the trend of convergence is shown in Fig. 4. It appears that the Lmax
800 0 has been reduced from 0.1273 to 0.1259 compared with CASE 1.
1 125 250 375 500
Iterations
CASE 4: OPF by considering multi-fuels
Fig. 3. Objective function variation for CASE 2. In some practical cases, thermal generating plants may have
multi-fuel sources like coal, natural gas and oil. Hence, the fuel cost
  curve can be expressed by a piecewise quadratic functions as fol-
Lmax ¼ max Lj j ¼ 1; 2; . . . ; NL ð21Þ
lows [28]:
where Lj is the local indicator of bus j and it is represented as
follows: f i ¼ aik þ bik PGi þ cik P2Gi if Pmin min
Gik 6 P Gi 6 P Gik ð24Þ

XNG
V i where k is the fuel option. In this study, generators 1 and 2 have two

Lj ¼ 1  HLGji j ¼ 1; 2; . . . ; NL ð22Þ fuel options (k = 2). The generator fuel cost coefficients are given in
i¼1
V j
Table 4. The objective function of CASE 4 is given by:
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 71

!
900 0.14 NG 
X    
2 min
Jðx; uÞ ¼ ai þ bi P Gi þ ci PGi þ di  sin ei  PGi  PGi
Cost i¼1
L max
þ Penalty
ð28Þ
Cost ($/h)

Lmax
where di and ei are the coefficients that represent the valve-point
850 0.13
effect.
The OPF problem of CASE 7 has been solved using the BSA algo-
rithm and the optimal control variables obtained are displayed in
Table 5. The optimal fuel cost obtained in this case has increased
from 799.0760 $/h to 830.7779 $/h compared to the basic case
i.e., CASE 1 due to the consideration of the valve-point effect.
800 0.12
1 125 250 375 500
CASE 8: OPF by considering valve-point effect and voltage profile
Iterations
improvement
Fig. 4. Objective function variation for CASE 3. This case is similar to CASE 2 where the aim is to optimize both
the cost and to improve the voltage profile. However, in this case
the valve point effect is taken into consideration. The optimal
Jðx; uÞ ¼ J CostG1 and G2
þ J CostRemaining Generators
þ Penalty ð25Þ results obtained using BSA for this case are given in Table 5.

where CASE 9: OPF by considering valve-point effect and voltage stability

X
2 enhancement
JCostG1 and G2
¼ aik þ bik P Gi þ cik P 2Gi This case is similar to CASE 3 where the aim is to optimize both
i¼1 ð26Þ the fuel cost considering the valve-point effect and the voltage sta-
if Pmin
Gik 6 PGi 6 P min
Gik for fuel option k; k ¼ 1 and 2 bility index. The optimal results obtained using BSA, for this case
are tabulated in Table 5.
and
X
NG CASE 10: OPF by considering emission
J CostRemaining Generators
¼ ai þ bi PGi þ ci P2Gi ð27Þ With growing political concerns about the environment, it is
i¼3 desirable to adjust the optimal power flow to account for emissions
The results obtained for this case are tabulated in Table 5. The [32]. The total ton/h emission of the atmospheric pollutants such
optimal cost obtained for this case is 646.1504 $/h. as sulfur oxides SOx and nitrogen oxides NOx caused by fossil-
fueled thermal units can be expressed as [32]:
CASE 5: OPF by considering multi-fuels and voltage profile
improvement
This case is similar to CASE 2 however, instead of the basic 900 0.3
quadratic cost curve, the cost curve is expressed by a piecewise Cost
quadratic functions due to considering multi-fuels option. The Emission
BSA has been run for this case and the optimal control variables

Emission (ton/h)
obtained are tabulated in Table 5. We can see from this table that
Cost ($/h)

the voltage profile has been improved compared with CASE 4

because VD has been reduced from (1.0273 p.u) to (0.1161 p.u). 850 0.25
However, the generation cost has slightly raised from 646.1504
$/h to 653.1019 $/h.

CASE 6: OPF by considering multi-fuels and voltage stability

enhancement
The aim of this case is to minimize the total generating cost 800 0.2
considering multi-fuels option and enhance the voltage stability 1 125 250 375 500
of the whole system. The results of this case are tabulated in Iterations
Table 5. It can be seen from this table that Lmax has been reduced
from 0.1378 to 0.1258 compared with CASE 4. However, the gener- Fig. 5. Objective function variation for CASE 10.
ation cost has been slightly raised from 646.1504 $/h to 648.1424
$/h.
Table 6
The main characteristics of the IEEE 57-bus test system.
CASE 7: OPF by considering valve-point effect
In order to have a realistic and more efficient modeling of gen- System Value Details
Characteristics
erator cost functions, the valve point–effect has to be taken into
account. The generating units with multi-valve steam turbines Buses 57 [33]
exhibit a greater variation in the fuel-cost functions and produce Branches 80 [33]
Generators 7 Buses: 1, 2, 3, 6, 8, 9 and 12
a ripple-like effect [29]. In order to take into account the valve- Shunts 3 Buses: 18, 25 and 53
point effect of generating units, a recurring rectifying sinusoidal Transformers 17 Branches: 19, 20, 31, 35, 36, 37, 41, 46, 54, 58, 59,
term is added to the cost function [30]. Hence, the objective func- 65, 66, 71, 73, 76 and 80
tion of CASE 5 is not continuous and non-linear and can be Control 33 –
variables
expressed as follows [31]:
72 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

Table 7
Cost and emission coefficients for the IEEE 57-bus test system.

Generator Bus a b c d e a b c x l
G1 1 0 2 0.00375 18 0.037 4.091 5.554 6.49 2.00E04 2.86E01
G2 2 0 1.75 0.0175 16 0.038 2.543 6.047 5.638 5.00E04 3.33E01
G3 3 0 3 0.025 13.5 0.041 6.131 5.555 5.151 1.00E05 6.67E01
G4 6 0 2 0.00375 18 0.037 3.491 5.754 6.39 3.00E04 2.66E01
G5 8 0 1 0.0625 14 0.04 4.258 5.094 4.586 1.00E06 8.00E01
G6 9 0 1.75 0.0195 15 0.039 2.754 5.847 5.238 4.00E04 2.88E01
G7 12 0 3.25 0.00834 12 0.045 5.326 3.555 3.38 2.00E03 2.00E01

Table 8
Optimal settings of control variables found for CASE 11 through CASE 15.

CASE 11 CASE 12 CASE 13 CASE 14 CASE 15

PG1 537.1555 532.0307 527.1008 509.4963 380.4090
PG2 100.0000 100.0000 100.0000 99.9995 100.0000
PG3 68.4433 57.4381 67.7407 64.7584 118.8113
PG6 100.0000 100.0000 100.0000 99.9946 99.9954
PG8 165.0000 165.0000 165.0000 165.0006 165.0018
PG9 100.0000 99.7228 99.9533 100.0000 100.0000
PG12 218.4831 236.3389 230.3504 246.1003 310.8383
VG1 1.0600 1.0307 1.0453 1.0600 1.0600
VG2 1.0531 1.0235 1.0276 1.0548 1.0560
VG3 1.0315 1.0111 1.0097 1.0382 1.0433
VG6 1.0152 1.0039 1.0161 1.0257 1.0237
VG8 1.0147 1.0229 1.0271 1.0278 1.0223
VG9 0.9962 1.0026 1.0031 1.0082 1.0058
VG12 1.0092 1.0200 1.0233 1.0204 1.0197
T19(418) 0.9250 1.0000 0.9250 0.9500 0.9750
T20(418) 0.9750 0.9625 0.9250 0.9625 0.9500
T31(2120) 1.0125 0.9750 1.1000 1.0250 1.0250
T35(2425) 0.9250 0.9625 0.9375 0.9500 0.9625
T36(2425) 0.9375 0.9625 0.9500 0.9125 0.9125
T37(2426) 0.9875 1.0375 1.0375 0.9875 1.0000
T41(729) 0.9500 0.9500 0.9375 0.9500 0.9500
T46(3432) 0.9125 0.9250 0.9000 0.9250 0.9250
T54(1141) 0.9125 0.9000 0.9000 0.9000 0.9250
T58(1545) 0.9500 0.9375 0.9500 0.9625 0.9625
T59(1446) 0.9375 0.9625 0.9250 0.9375 0.9375
T65(1051) 0.9500 1.0000 0.9375 0.9500 0.9500
T66(1349) 0.9125 0.9000 0.9125 0.9125 0.9125
T71(1143) 0.9375 0.9375 0.9250 0.9375 0.9375
T73(4056) 1.0000 1.0250 1.0875 1.0125 1.0125
T76(3957) 0.9625 0.9125 1.0500 0.9625 0.9750
T80(955) 0.9375 0.9750 0.9625 0.9500 0.9500
QC18 5.0000 3.7851 3.1020 4.9582 4.9368
QC25 4.9944 5.0000 1.7607 4.7275 4.9580
QC53 4.9773 5.0000 5.0000 5.0000 5.0000
Cost ($/h) 6411.0043 6436.7551 6428.9946 6462.4093 6652.9484
VD 1.1009 0.6888 1.4009 1.2425 1.2286
Lmax 0.2819 0.2941 0.2746 0.2777 0.2782
Ploss (MW) 38.2819 39.7304 39.3452 34.5497 24.2558
Qloss (MVar) 158.2351 165.0756 160.8380 143.0694 101.2434
Emission (ton/h) 1.9726 1.9600 1.9225 1.8333 1.2796

X
NG     The results obtained after optimization using the BSA algorithm
Emission ¼ 102 ai þ bi PGi þ ci P 2Gi þ xi eðli PGi Þ ð29Þ are displayed in Table 5 and the trend of optimization is repre-
i¼1
sented in Fig. 5. The results show that the emission has been
reduced from (0.3671 ton/h) to (0.2425 ton/h) i.e. by 33.94%, how-
where ai ; bi ; ci ; xi and li are coefficients of the ith generator emis-
ever, the total fuel cost has increased from (799.0760 $/h) to
sion characteristics.
(835.0199 $/h) i.e. by 4.50% compared with CASE 1.
Therefore, the objective function for this case can be expressed
by:
IEEE 57-bus test system
!
X
NG
Jðx; uÞ ¼ ai þ bi P Gi þ ci P2Gi þ kEmission ðEmissionÞ þ Penalty In order to test the scalability of the proposed BSA algorithm, a
i¼1 larger test system is considered in this paper, which is the IEEE 57-
ð30Þ bus test system. This system has a total generation capacity of
1975.9 MW and its main characteristics are given in Table 6. Cost
where kEmission is a weighting factor and it is equal to 1000 in this and emission coefficients for this system are given in Table 7.
case. Detailed data can be derived from [33].
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 73

1.1

1.05
Voltage Profile (p.u.)

0.95

0.9
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

PQ Bus Number

(a)
1.1

1.05
Voltage Profile (p.u.)

0.95

0.9
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

PQ Bus Number

(b)
Fig. 6. Comparison of voltage profile between (a) CASE 11 and (b) CASE 12.

Table 9
The main characteristics of the IEEE 118-bus test system. is given by (20) where kVD ¼ 10; 000 in this case to balance
between the objectives. The results of such optimization using
System Value Details
characteristics
the proposed BSA algorithm are tabulated in Table 8. It appears
from this table that the VD has been reduced from (1.1009 p.u.)
Buses 118 [33]
Branches 186 [33]
to (0.6888 p.u.) compared with CASE 11. However, the cost has
Generators 54 Buses: 1, 4, 6, 8, 10, 12, 15, 18, 19, 24, 25, 26, 27, slightly increased from (6411.0043 $/h) to (6436.7551 $/h) com-
31, 32, 34, 36, 40, 42, 46, 49, 54, 55, 56, 59, 61, pared with CASE 11. The improvement of the voltage profile is
62, 65, 66, 69, 70, 72, 73, 74, 76, 77, 80, 85, 87, illustrated in Fig. 6.
89, 90, 91, 92, 99, 100, 103, 104, 105, 107, 110,
111, 112, 113 and 116
Shunts 14 Buses: 5, 34, 37, 44, 45, 46, 48, 74, 79, 82, 83,
105, 107 and 110
CASE 13: OPF by considering cost reduction and voltage stability
Transformers 9 Branches: 8, 32, 36, 51, 93, 95, 102, 107 and 127 enhancement
Control variables 130 – The voltage stability enhancement has been also tested for the
IEEE 57-test system in this case where the aim is to minimize
the cost and at the same time to improve the vulnerability of the
system by enhancing its voltage stability. Thus, the objective func-
CASE 11: OPF by considering cost reduction tion of CASE 13 is given by (23) with kLmax ¼ 10; 000. The BSA is run
The aim here is to minimize the total generating fuel cost. for this case and the obtained control variables are tabulated in
Therefore, the objective function of this case is given by (18). The Table 8. These results show clearly that the voltage stability of
BSA is run in order to find the optimal settings for CASE 11 and the system is enhanced with a slight augmentation in cost.
the obtained results are given in Table 8. The cost obtained for this
case is (6411.0043 $/h) while VD and Lmax are (1.1009 p.u.) and
(0.2819), respectively. CASE 14: OPF by considering valve-point effect
The effect of valve-point loading is tested in this case. Therefore,
CASE 12: OPF by considering cost reduction and voltage profile the objective function of this case is given by (28). The obtained
improvement results using BSA are given in Table 8. In this case the cost has mar-
As in CASE 2, the goal here is to optimize both the total gener- ginally augmented from (6411.0043 $/h) to (6462.4093 $/h) com-
ating fuel cost and the voltage profile. Hence the objective function pared with CASE 11.
74 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

Table 10 Table 10 (continued)

Optimal settings of control variables found for CASE 16.
CASE 16
CASE 16
VG49 1.0087
PG1 378.1328 VG54 1.0038
PG4 30.0000 VG55 0.9989
PG6 30.4589 V56 0.9998
PG8 30.3166 V59 1.0064
PG10 30.7436 VG61 1.0037
PG12 314.6145 VG62 0.9993
PG15 68.4755 VG65 1.0114
PG18 30.2733 VG66 1.0187
PG19 30.5375 VG69 1.0518
PG24 30.0000 VG70 1.0131
PG25 30.2062 VG72 0.9993
PG26 151.3036 VG73 1.0029
PG27 220.6014 VG74 0.9945
PG31 30.2083 VG76 0.9808
PG32 32.1285 VG77 1.0045
PG34 30.1760 VG80 1.0133
PG36 30.4660 VG85 0.9932
PG40 30.2968 VG87 1.0099
PG42 30.2084 VG89 1.0025
PG46 30.8136 VG90 0.9940
PG49 35.8272 VG91 1.0070
PG54 166.5605 VG92 0.9974
PG55 47.4169 VG99 1.0108
PG56 30.6967 VG100 1.0099
PG59 30.8771 VG103 1.0107
PG61 119.9260 VG104 1.0033
PG62 124.4234 VG105 1.0008
PG65 30.5199 VG107 1.0059
PG66 285.5017 VG110 1.0030
PG69 289.5617 VG111 1.0126
PG70 30.2547 VG112 0.9983
PG72 30.5085 VG113 1.0009
PG73 30.5679 VG116 0.9988
PG74 30.0000 T8(85) 0.9750
PG76 34.7917 T32(2625) 1.0125
PG77 30.4234 T36(3017) 1.0000
PG80 342.1612 T51(3837) 0.9875
PG85 30.3368 T93(6359) 1.0000
PG87 31.2143 T95(6461) 0.9875
PG89 369.3985 T102(6566) 1.0000
PG90 30.1969 T107(6869) 0.9375
PG91 30.3486 T127(8180) 0.9625
PG92 30.1950 QC5 1.3480
PG99 30.6006 QC34 2.4031
PG100 177.9280 QC37 1.6355
PG103 42.3544 QC44 2.6899
PG104 30.5564 QC45 5.0000
PG105 30.1630 QC46 3.6900
PG107 30.6484 QC48 3.2356
PG110 30.7890 QC74 4.9106
PG111 41.1484 QC79 4.9851
PG112 30.0000 QC82 3.1106
PG113 30.2901 QC83 3.4727
PG116 30.3413 QC105 0.0451
VG1 0.9787 QC107 3.2741
VG4 1.0046 QC110 2.8543
VG6 0.9955 Cost ($/h) 135333.4743
VG8 1.0034 VD (pu) 0.6773
VG10 1.0199 Lmax 0.0698
VG12 0.9937 Ploss (MW) 64.4897
VG15 0.9942 Qloss (MVar) 384.8242
VG18 0.9961
VG19 0.9932
VG24 1.0018 CASE 15: OPF by considering emission
VG25 1.0168 This case is similar to CASE 10 where the aim is to reduce the
VG26 1.0282 cost while monitoring the emission at the same time. The objective
VG27 0.9971
function is given by (30) with kEmission ¼ 1000. The BSA is run for
VG31 1.0000
VG32 0.9941 this case, the obtained control variables are tabulated in Table 8.
VG34 1.0138 It appears that, the emission has been reduced by 35.13% i.e. from
VG36 1.0131 (1.9726 ton/h) to (1.2796 ton/h) in this case, whilst the cost has
VG40 1.0098
slightly augmented from (6411.0043 $/h) to (6652.9484 $/h) i.e.
VG42 0.9996
VG46 0.9947
by 3.77%.
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 75

Table 11 capacity of 9966.2 MW and its main characteristics are given in

Control parameters of the related algorithms used in the tests. Table 9 and cost coefficients are given in [33]. Detailed data can
Algorithm Control parameters be derived from [33].
Value Description
BSA F = 1/normrnd Scale factor (where normrnd(mean, CASE 16: OPF by considering cost reduction
(0,3) standard deviation) generates random The aim here is to minimize the total generating fuel cost.
numbers from the normal distribution
Therefore, the objective function of this system is given by (18).
with mean and standard deviation
parameters) The BSA is run in order to find the optimal settings for CASE 16
F generation’s Pseudo-stable walk (levy-like) and the obtained results are given in Table 10. We can notice from
strategy this table that, the minimum cost obtained is (135333.4743 $/h).
DE F = 0.8 Differential weight
Cr = 0.8905 crossover probability
Performance evaluation study
PSO w = 0.5 Inertia factor
c1 = 1.5 Cognitive factor
c2 = 1.5 Social factor In order to evaluate the performance of the BSA, it has been
GA Mutrate = .05 Mutation rate compared to several optimization algorithms including DE, PSO,
Selection = 0.75 Fraction of population kept GA, ABC and BBO. Control parameters of optimization algorithms
ABC FoodNumber = Number of food sources used in this investigation are given in Table 11.
Np/2 In Table 12 a statistical analysis or the results is tabulated. In
BBO pmutate = 0 Mutation probability each cell that corresponds to the performance of an algorithm for
pmodify = 1 Habitat modification probability one specific case, we give five values that are: the min (or best),
pmutate = 0.005 Initial mutation probability the mean (or average), the median, the max (or worst) and the
Common Population size = 50 for cases: 1, 2, 3, 7, 8, 9 and 10 standard deviation values. Furthermore, in the last row of Table 12,
parameters Population size = 90 for cases: 4, 5, 6, 11, 12, 13, 14, 15 we calculated the feasibility rate (FR) which is defined as the num-
and 16 ber of runs where the algorithm converges to a feasible solution
Maximum number of iterations = 500 for cases: 1, 2, 3, 4,
5, 6, 7, 8, 9 and 10
over the number of all runs. The FR is computed in this paper for
Maximum number of iterations = 1500 for cases: 11, 12, 13, all cases at the same time therefore the total number of runs is
14 and 15 16 (cases) 30 (runs per case) which equals 480.
Maximum number of iterations = 2500 for cases: 16 From this table we can notice that the BSA is very robust in solv-
ing several OPF problems with different complexities. The averages
values are very near to the best ones for all cases. The maximum
IEEE 118-bus test system average value with respect to the best one is obtained for CASE
5. The normalized error between these two values is 1.48%. More-
A large-scale test system is considered in this paper, which is over, the BSA is the only algorithm that has a FR of 100% compared
the IEEE 118-bus test system. This system has a total generation with the remaining algorithms.

Table 12
comparison of the BSA with DE, PSO, GA ABC and BBO for solving different OPF problems.

Case BSA DE PSO GA ABC BBO

CASE 1 799.0760 799.0376 800.9310 800.1636 799.0541 799.1267
799.2721 799.3047 – 802.6876 799.6945 801.1927
799.2448 799.0458 – 802.2552 799.4613 801.1287
799.6240 801.5552 – 806.2791 802.6327 803.1429
0.1357 0.6624 – 1.7071 0.8145 1.0251
CASE 2 918.1038 909.0855 963.5922 965.6194 925.3049 917.2192
927.3204 935.2076 – 1008.4953 944.2215 960.8904
927.4924 917.5173 – 995.6274 943.3027 954.5105
937.9009 1068.5257 – 1193.3953 1006.1873 1080.6751
4.3840 42.4846 – 48.8007 14.7217 37.7537
CASE 3 1555.8082 1550.3449 1562.0027 1563.8450 1554.8035 1557.0275
1560.6449 1555.5616 – 1582.5913 1565.6099 1566.8208
1560.4770 1553.4312 – 1577.8640 1565.8584 1566.8000
1565.6629 1578.0121 – 1624.6329 1574.1667 1593.7834
2.6436 5.8042 – 16.4891 5.9094 6.2569
CASE 4 646.1504 645.3627 647.2879 649.9246 648.5069 647.1179
647.5781 646.7220 681.7314 659.6545 652.1451 651.0801
647.5572 646.4604 654.3286 658.2511 651.6767 651.0284
649.0638 650.7419 839.6854 671.9717 657.9807 656.9323
0.6668 1.0607 62.5562 5.9728 2.6969 2.5840
CASE 5 769.1823 767.7043 820.2628 828.0635 779.7313 760.9098
780.6187 792.0596 974.8383 895.6767 799.7922 807.1644
780.8155 781.4195 892.3997 882.9264 800.7031 788.2795
790.0544 976.7543 1531.8217 1005.1995 833.8966 1091.9235
4.4441 38.9756 181.1779 43.3107 13.3475 67.7977
CASE 6 1402.8772 1399.6596 1406.3864 1430.8548 1407.0969 1401.4215
1411.5551 1407.0634 1511.7027 1458.4391 1421.3033 1431.0379
1412.3365 1407.3550 1466.1633 1449.1097 1420.8695 1418.6145

(continued on next page)

76 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77

Table 12 (continued)

Case BSA DE PSO GA ABC BBO

1418.1184 1418.6022 1927.2722 1524.9674 1431.0045 1600.3741
3.3039 4.3221 115.7105 25.1029 5.8574 45.8851
CASE 7 830.7779 830.4425 837.5082 834.2424 831.5783 831.4581
832.0811 831.4997 – 840.9013 834.4691 835.8153
832.1224 830.5125 – 840.2634 834.1686 835.3208
834.3303 842.7195 – 854.9337 839.0831 842.5715
0.8474 3.0912 – 4.5089 1.9432 2.6118
CASE 8 956.3246 943.9238 1014.4785 1014.7138 964.6089 966.8975
966.0132 975.6403 – 1054.1639 986.1737 1000.3319
965.0669 958.2037 – 1042.2188 987.3366 989.3155
977.6529 1096.0340 – 1167.5909 1012.5744 1110.7380
5.1431 40.5799 – 36.2838 11.5230 35.4894
CASE 9 1589.6073 1583.3203 1603.9224 1603.3377 1592.8876 1592.2125
1595.7385 1590.4131 – 1625.1084 1604.9151 1607.7481
1595.9685 1589.8157 – 1622.1465 1605.4779 1607.0377
1601.3492 1623.2834 – 1666.5100 1618.1911 1624.5010
3.0705 8.0913 – 15.5846 7.3870 8.5617
CASE 10 1077.4734 1077.3561 1079.7334 1079.0457 1077.4222 1077.7629
1077.8393 1079.1245 – 1081.5680 1077.8749 1079.0579
1077.7353 1077.4365 – 1081.3841 1077.7441 1079.1963
1078.8154 1096.8296 – 1085.3204 1079.4747 1079.9783
0.3111 4.7910 – 1.3228 0.4630 0.7013
CASE 11 6411.0043 6410.1888 6748.6052 6673.7958 6411.4506 6418.5723
6411.7690 – – – 6423.8702 6450.9358
6411.2924 – – – 6422.3992 6439.3508
6414.9844 – – – 6449.4900 6639.2789
1.1107 – – – 7.9744 52.0784
CASE 12 13325.0513 13294.4903 19305.9275 16827.1927 13716.4582 13553.2580
13438.7459 – – – – –
13430.9976 – – – – –
13601.2447 – – – – –
56.5258 – – – – –
CASE 13 9175.5107 9131.5465 9947.0381 9773.5989 9242.6164 9237.9419
9197.5527 – – – 9330.4523 9525.9877
9198.5524 – – – 9325.4774 9313.0632
9217.4690 – – – 9450.7088 13538.6780
11.9426 – – – 51.7452 802.0873
CASE 14 6462.4093 6462.7525 7149.1144 6737.3653 6467.8272 6468.6389
6464.3423 – – – 6477.5392 6688.4624
6463.7772 – – – 6477.2658 6489.8889
6468.4281 – – – 6493.1610 11512.1470
1.6066 – – – 5.8875 922.5141
CASE 15 7932.5169 7932.1170 8542.6349 8349.9851 7935.4485 7942.1976
7936.6698 – – – 7944.0281 7964.3507
7935.9449 – – – 7943.6269 7961.5008
7941.1416 – – – 7952.9615 8022.0932
2.4404 – – – 4.4395 15.1495
CASE 16 135333.4743 – – – 135304.3584 135263.7289
135511.5451 – – – 135567.2697 135684.1137
135502.6493 – – – 135562.8341 135622.6172
135689.1275 – – – 135973.6155 136611.2731
93.1975 – – – 151.6905 335.0166
FR (%) 100 92.29 62.92 81.25 99.79 98.13

In order to evaluate the convergence speed of the BSA, the fol-
lowing approach is proposed in this paper. First, the values of the
objective function (average values over all runs) are recorded at
four cut-points of the search corresponding to 20%, 40%, 60% and
80% of the maximum number of iterations. Then, the percentage
that represents the value of this objective function compared with
the final value (average of final values over all runs) reached by the
algorithm is calculated. These percentages are reported in Table 13.
The final row is the average, calculated on all cases for each cut
point. From this table we can say that the BSA is very efficient in
solving several OPF problems with different complexities. For the
first cut-point, the BSA has reached already more than 85% of the
final value of the objective function for all cases except for CASE
16. For cut-point 2, cut-point 3 and cut-point 4 the BSA has reached
more than, 95%, 97%, and 99% of the final value objective function,
respectively.
Conclusion
In this paper, a newly developed algorithm called the BSA is
implemented to solve several OPF problems. Three test systems
and sixteen cases have been investigated in order to evaluate the
performance of the proposed algorithm. Furthermore, the obtained
results have been compared to those obtained using other well-
known optimization algorithms such as DE, PSO, ABC, GA and
BBO. The main conclusion that can be drawn from this paper is that
the BSA is a very effective and robust algorithm for solving OPF
problems. It has good convergence characteristics and can achieve
 A.E. Chaib et al. / Electrical Power and Energy Systems 81 (2016) 64–77 77

Table 13 [8] Bouchekara HREH, Abido Ma, Boucherma M. Optimal power flow using
Convergence speed of the BSA algorithm. teaching-learning-based optimization technique. Electr Power Syst Res
2014;114:49–59.
Case Cut-point 1 Cut-point 2 Cut-point 3 Cut-point 4 [9] Bouchekara HREH, Abido Ma, Chaib aE, Mehasni R. Optimal power flow using
the league championship algorithm: a case study of the Algerian power
CASE 1 99.12 99.66 99.86 99.95
system. Energy Convers Manag 2014;87:58–70.
CASE 2 91.81 95.89 98.25 99.38
[10] Bouchekara HRE-H, Abido MA. Optimal power flow using differential search
CASE 3 97.35 99.16 99.76 99.92 algorithm. Electr Power Components Syst 2014;42:1683–99.
CASE 4 98.19 99.37 99.72 99.89 [11] Mukherjee A, Mukherjee V. Solution of optimal power flow using chaotic krill
CASE 5 89.11 95.34 97.56 99.00 herd algorithm. Chaos, Solitons Fract 2015;78:10–21.
CASE 6 96.18 98.79 99.63 99.87 [12] Duman S, Güvenç U. Optimal power flow using gravitational search algorithm.
CASE 7 98.07 99.28 99.76 99.91 Energy Convers Manag 2012;59:86–95.
CASE 8 91.46 95.55 97.75 99.09 [13] Bhowmik AR, Chakraborty AK. Solution of optimal power flow using non
CASE 9 96.85 98.72 99.45 99.81 dominated sorting multi objective opposition based gravitational search
CASE 10 99.46 99.82 99.90 99.96 algorithm. Int J Electr Power Energy Syst 2015;64:1237–50.
CASE 11 93.65 99.70 99.93 99.98 [14] Ghasemi M, Ghavidel S, Ghanbarian MM, Massrur HR, Gharibzadeh M.
CASE 12 86.04 96.56 98.81 99.64 Application of imperialist competitive algorithm with its modified
CASE 13 94.98 99.14 99.68 99.89 techniques for multi-objective optimal power flow problem: a comparative
study. Inf Sci (Ny) 2014;281:225–47.
CASE 14 92.89 99.72 99.92 99.98
[15] Daryani N, Hagh MT, Teimourzadeh S. Adaptive group search optimization
CASE 15 93.14 99.78 99.93 99.98
algorithm for multi-objective optimal power flow problem. Appl Soft Comput
CASE 16 0.96 96.28 98.92 99.74
2016;38:1012–24.
Average 88.70 98.30 99.30 99.75 [16] AlRashidi M, El-Hawary M. Applications of computational intelligence
techniques for solving the revived optimal power flow problem. Electr
Power Syst Res 2009;79:694–702.
[17] Frank S, Steponavice I. Optimal power flow?: a bibliographic survey II non-
better performances than some well-known optimization algo- deterministic and hybrid methods. Energy Syst 2012:259–89.
rithms. This is more pronounced when dealing with large scale [18] Civicioglu P. Backtracking search optimization rithm for numerical
power systems. Finally, it is recommended that a multiobjective optimization problems. Appl Math Comput 2013;219:8121–44.
[19] Mandal S, Sinha RK, Mittal K. Comparative analysis of backtrack search
BSA algorithm based on Pareto optimal solutions has to be devel- optimization algorithm (BSA) with other evolutionary algorithms for global
oped and applied to solve OPF problems in further studies. continuous optimization. Int J Comput Sci Inf Technol 2015;6:3237–41.
[20] Civicioglu P. Circular antenna array design by using evolutionary search
algorithms. Prog Electromagn Res B 2013;54:265–84.
Acknowledgments [21] Shafiullah M, Abido MA, Coelho LS. Design of robust PSS in multimachine
power systems using backtracking search algorithm; 2015.
[22] El-Fergany A. Multi-objective allocation of multi-type distributed generators
Dr. M. A. Abido would like to acknowledge the support provided along distribution networks using backtracking search algorithm and fuzzy
by King Abdulaziz City for Science and Technology (KACST) expert rules. Electr Power Components Syst 2015;5008:1–16.
through the Science and Technology Unit at King Fahd University [23] Hendrix EMT, G.-Tóth B. Introduction to nonlinear and global optimization,
vol. 37. Springer; 2010.
of Petroleum and Minerals (KFUPM) for funding this work through [24] Ray D. Zimmerman CEM-S& D (David) G. MATPOWER. <http://www.
project# 14-ENE265-04 as a part of the National Science, Technol- pserc.cornell.edu/matpower/#docsn.d>.
ogy and Innovation Plan (NSTIP). [25] Yuryevich J. Evolutionary programming based optimal power flow algorithm.
IEEE Trans Power Syst 1999;14:1245–50.
[26] He S, Went JY, Prempaint E, Wut QH, Fitchs J, Mand S. An improved particle
swarm optimization for optimal power flow; 2004. p. 21–4.
References
[27] Kessel P, Glavitsch H. Estimating the voltage stability of a power system. IEEE
Trans Power Deliv 1986;1:346–54.
[1] Cain M, O’neill R, Castillo A. History of optimal power flow and formulations. [28] Abou El Ela AA, Abido MA, Spea SR. Optimal power flow using differential
FERC Staff Tech Pap; 2012. p. 1–36. evolution algorithm. Electr Power Syst Res 2010;80:878–85.
[2] Surender Reddy S, Bijwe PR, Abhyankar aR. Faster evolutionary algorithm [29] Hardiansyah H. A modified particle swarm optimization technique for
based optimal power flow using incremental variables. Int J Electr Power economic load dispatch with valve-point effect. Int J Intell Syst Appl
Energy Syst 2014;54:198–210. 2013;5:32–41.
[3] Pandya KS, Joshi SK. Survey Opt Power Flow Methods 2008;0:450–8. [30] Niknam T, Narimani MR, Azizipanah-abarghooee R. A new hybrid algorithm
[4] Frank S, Steponavice I. Optimal power flow?: a bibliographic survey I. for optimal power flow considering prohibited zones and valve point effect.
Formulations and deterministic methods; 2012. p. 221–58. Energy Convers Manag 2012;58:197–206.
[5] Khorsandi a, Hosseinian SH, Ghazanfari a. Modified artificial bee colony [31] Alsumait JS, Sykulski JK, Al-Othman aK. A hybrid GA–PS–SQP method to solve
algorithm based on fuzzy multi-objective technique for optimal power flow power system valve-point economic dispatch problems. Appl Energy
problem. Electr Power Syst Res 2013;95:206–13. 2010;87:1773–81.
[6] Ayan K, Kılıç U, Baraklı B. Chaotic artificial bee colony algorithm based solution [32] Abido Ma. Environmental/economic power dispatch using multiobjective
of security and transient stability constrained optimal power flow. Int J Electr evolutionary algorithms. IEEE Trans Power Syst 2003;18:1529–37.
Power Energy Syst 2015;64:136–47. [33] 2014 OPF PROBLEMS. <https://www.uni-due.de/ieee-wgmho/competition
[7] Bouchekara HREH. Optimal power flow using black-hole-based optimization 2014n.d>.
approach. Appl Soft Comput J 2014;24:879–88.