Applied Soft Computing 42 (2016) 119–131

Contents lists available at ScienceDirect

Applied Soft Computing

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

Optimal power flow using an Improved Colliding Bodies Optimization

algorithm
H.R.E.H. Bouchekara a,b,∗ , A.E. Chaib a , M.A. Abido c , R.A. El-Sehiemy d
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
d
Electrical Engineering Department, Faculty of Engineering, Kafrelsehiekh University, Egypt

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

Article history: This paper proposes Improved Colliding Bodies Optimization (ICBO) algorithm to solve efficiently the
Received 24 October 2015 optimal power flow (OPF) problem. Several objectives, constraints and formulations at normal and pre-
Received in revised form 9 January 2016 ventive operating conditions are used to model the OPF problem. Applications are carried out on three
Accepted 22 January 2016
IEEE standard test systems through 16 case studies to assess the efficiency and the robustness of the
Available online 4 February 2016
developed ICBO algorithm. A proposed performance evaluation procedure is proposed to measure the
strength and robustness of the proposed ICBO against numerous optimization algorithms. Moreover, a
Keywords:
new comparison approach is developed to compare the ICBO with the standard CBO and other well-
Colliding Bodies Optimization
Optimal power flow
known algorithms. The obtained results demonstrate the potential of the developed algorithm to solve
Security-constrained optimal power flow efficiently different OPF problems compared to the reported optimization algorithms in the literature.
Power system optimization © 2016 Elsevier B.V. All rights reserved.
Metaheuristics

1. Introduction Nevertheless, traditional methods rely on some simplification assumptions such

The optimal power flow (OPF) problem is among the tools used in operation
and planning of energy systems [1]. Since its introduction by Carpentier in 1962,
the OPF usefulness is progressively being recognized, and nowadays it becomes the
most important tool used by the system operator in power systems exploitation
and planning [2]. Several models have been developed and adopted to formulate
different kinds of OPF problems, objectives, sets of design variables and constraint
types [3].
The OPF can be defined as an optimization problem which aims to adjust two
sets of control variables (continuous and discrete) in order to optimize a predefined
objective function while satisfying operational equality and inequality constraints.
Further, the purpose of traditional OPF is mainly concerned with the minimization of
total generating cost. However, more realistic operating conditions should be inves-
tigated when solving OPF problems. Complexities and constraints like multi-fuels,
valve-point effect, security constraints and prohibited zones have to be included.
Therefore, the OPF problem is generally a highly constrained, mixed-integer, non-
linear and nonconvex optimization problem [2,4,5,6].
Initially, several traditional (deterministic) optimization techniques were
employed successfully to solve the OPF problem [7]. Surveys of various traditional
methods used to solve the OPF problem are given in [8–10].
∗ Corresponding author at: Constantine Electrical Engineering Laboratory, LEC,
Department of Electrical Engineering, University of Freres Mentouri Constantine,
25000 Constantine, Algeria. Tel.: +213 666605628; fax: +213 31908113.
E-mail address: bouchekara.houssem@gmail.com (H.R.E.H. Bouchekara).
as convexity, smoothness, continuity and differentiability. However, actual OPF
problems may have nonlinear characteristics such as valve point effects, prohibited
operating zones and piecewise quadratic cost function [11]. Therefore, traditional
methods for example quasi-Newton method or conjugate gradient method gener-
ally fail in solving such OPF problems due to their rugged search landscape.
The evolution of computational resources over the last few decades had moti-
vated the development of what is called metaheuristics. These methods can
overcome many drawbacks of the traditional methods [12]. Some of these methods
have been used to solve the OPF problem such as: Genetic Algorithm (GA) [13,14],
Tabu Search (TS) [15], Particle Swarm Optimization (PSO) [16], Simulated Annealing
(SA) [17], Differential Evolution (DE) [18], Imperialist Competitive Algorithm (ICA)
[19,20], Biogeography Based Optimization (BBO) [21,22], Gravitational Search Algo-
rithm (GSA) [23,24], Harmony Search (HS) [25], Artificial Bee Colony (ABC) [26,27],
Black Hole (BH) [28], Teaching Learning based Optimization (TLBO) [29], League
Championship Algorithm (LCA) [30], Group Search Optimization (GSO) [31] and
many others. Surveys of various metaheuristics used to solve the OPF problem are
presented in [6,32,33].
However, due to the variability of objectives where different functions can be
considered for modeling the OPF problem, no algorithm can be considered as the best
in solving all OPF problems. Therefore, there is always a need for a new algorithm
that can solve some of the OPF problems efficiently.
The Colliding Bodies Optimization (CBO) is a new nature inspired metaheuris-
tic which is based on the law of collision between two bodies. The CBO has been
developed by Kaveh and Mahdavi [34]. Moreover, Kaveh and Ghazaan [35] pro-
posed an Enhanced CBO referred to as ECBO. The ECBO uses memory to save some
best solutions and a mechanism to escape from local optima.
The aim of this paper is to develop an Improved CBO algorithm referred to
as ICBO for solving OPF problems. In order to justify the development of ICBO,

http://dx.doi.org/10.1016/j.asoc.2016.01.041
1568-4946/© 2016 Elsevier B.V. All rights reserved.
120 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

its performances are compared to CBO, ECBO and other well-known optimization Hence, u can be expressed as:
algorithms.  
The main contributions of this paper can be summarized as follows: uT = PG2 . . .PGNG , VG1 . . .VGNG , QC1 . . .QCNC , T1 . . .TNT (7)

1. Development of an improved version of the CBO algorithm. where NG, NT and NC are the number of generators, the number
2. Implementation of ICBO, CBO, ECBO and other well-known optimization algo- of regulating transformers and the number of VAR compensators,
rithms for solving realistic OPF problems. respectively.
3. Implementation of a complete set of tests in order to assess optimization algo- It is worth mentioning that, transformer tap settings and shunt
rithms using different OPF problems, test systems, objective functions and
constraints.
devices settings are discrete in nature. In many works reported
4. Resolution of the OPF problem using practical constraints like prohibited zones in literature addressing the OPF, these settings are considered as
and using non-smooth objective functions by including valve point effect and continuous variables for simplicity. Then, the discrete variables are
multi-fuels options for a more realistic OPF. set to their nearest discrete value after the optimization has been
5. Resolution the OPF problem considering security constraints for more challeng-
done. The results have shown that this approach leads to acceptable
ing conditions.
6. Implementation of a new comparison method based on best and average values. results without incurring the exponential complexity as reported
7. Utilization of nonparametric statistics for the validation of the comparative by [38]. This last approach is adopted in this paper.
method.
2.4. State variables
The remainder of this paper is organized as follows. In Section 2, the OPF problem
is formulated. In Section 3, the proposed ICBO algorithm along with the standard
and enhanced versions of the CBO are described. The applications and results are The set of state variables for the OPF problem formulation are:
presented in Section 4. Finally, the conclusions are drawn in Section 5.
PG1 : active power generation at slack bus.
2. Problem formulation VL : voltage magnitudes at PQ buses or load buses.
QG : reactive power output of all generator units.
As previously mentioned, generally, the objective of the OPF Sl : transmission line loadings (or line flow).
problem is to identify or adjust a set of control variables that opti-
mize predefined power system objectives while satisfying system Hence, x can be expressed as:
and practical constraints [36,37]. In this paper, two formulations of  
xT = PG1 , VL1 . . .VLNL , QG1 . . .QGNG , Sl1 . . .Slnl (8)
the OPF are considered. These are the classical OPF formulation and
the security constrained optimal power flow (SCOPF) formulation. where NL and nl are the number of load buses and the number of
transmission lines, respectively.
2.1. Classical OPF formulation
2.5. Constraints
The classical OPF problem can be formulated as follows [25,30]:
2.5.1. Equality constraints
Minimize F(x, u) (1)
The equality constraints of the OPF reflect the physics of the
Subject to g(x, u) = 0 (2) power system which is represented by the typical power flow equa-
tions [39,40]. The equality constrains can be represented as follows:
and h(x, u) ≤ 0 (3)

NB
where u is the vector of independent variables or control variables. PGi − PDi − Vi Vj [Gij cos (ij ) + Bij sin (ij )] = 0 (9)
x is the vector of dependent variables or state variables. F(x,u): j=1
objective function. g(x,u): set of equality constraints. h(x,u): set of
inequality constraints. 
NB
QGi − QDi − Vi Vj [Gij sin (ij ) − Bij cos (ij )] = 0 (10)
2.2. SCOPF formulation j=1

where  ij =  i −  j , which are voltage angles at bus i and j, respec-

The SCOPF (the preventive approach) problem can be formu-
tively, NB is the number of buses, PD is the active load demand,
lated as follows:
QD is the reactive load demand, Gij and Bij are the elements of the
Minimize F(x0 , u0 ) (4) admittance matrix (Yij = Gij + jBij ) representing the conductance and
susceptance between bus i and bus j, respectively.
Subject to gk (xk , u0 ) = 0 k = 0, . . ., c (5)

and h(xk , u0 ) ≤ 0 k = 0, . . ., c (6) 2.5.2. Inequality constraints

where x0 , u0 is the state and the control variables of the base case,
respectively. xk , uk : the state and the control variables of the kth
post-contingency state, respectively. c is the number of contingen-
cies considered.
The inequality constraints of the OPF reflect the limits on phys-
ical devices present in the power system as well as the limits
imposed to guarantee system security [39,40].
(a) Generator constraints

2.3. Control variables VGmin ≤ VGi ≤ VGmax , i = 1, . . ., NG (11)

i i

The set of control variables in the OPF problem formulation are: PGmin ≤ PGi ≤ PGmax , i = 1, . . ., NG (12)
i i

QGmin ≤ QGi ≤ QGmax , i = 1, . . ., NG (13)

PG : active power generation at PV buses except the slack bus. i i

VG : voltage magnitudes at PV buses. (b) Transformer constraints

T: tap settings of transformers.
QC : shunt VAR compensation. Timin ≤ Ti ≤ Timax , i = 1, . . ., NT (14)
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131 121

2.6. Objective functions

2.6.1. Basic cost function

The basic objective function considered in OPF studies and the
one which is widely used is the total generating fuel cost of the
whole system where each generating unit has its own cost curve
represented by a quadratic function. Therefore, the objective func-
tion representing the total generating fuel cost is given by:


NG
F(x, u) = ai + bi PGi + ci PG2 (19)
i
i=1

where ai , bi and ci are the cost coefficients of the ith generator.

2.6.2. Voltage profile

The voltage profile is an important safety and service quality
index. The amelioration of the voltage profile of the system consists
of minimizing the deviation of voltages of PQ buses form the unity.
This voltage deviation is referred to as VD and it can be calculated
Fig. 1. Cost function for one generator with POZ constraints. by:
 NL 

(c) Shunt VAR compensator constraints VD = |VLi − 1| (20)
i=1
QCmin ≤ QCi ≤ QCmax , i = 1, . . ., NC (15)
i i
Therefore, in order to simultaneously minimize the total gener-
(d) Security constraints ating fuel cost represented by the sum of quadratic functions and
the improve the voltage profile represented by VD, we propose the
VLmin ≤ VLi ≤ VLmax , i = 1, . . ., NL (16) following two-fold objective function:
i i
 NG 
Sli ≤ Slmax , i = 1, . . ., nl (17) 
i F(x, u) = ai + bi PGi + ci PG2 + VD × VD (21)
i
i=1
2.5.3. Prohibited zones constraints
In practice, generators have Prohibited Operating Zones (POZ) where VD is a scaling factor to balance between the objective func-
due to some faults in the shaft bearing or vibration of machines tion values and to avoid the dominance of an objective over another.
or their accessories such as pumps or boilers [41,40]. POZ lead to a In this paper VD is selected as 100.
solution space with disjoint feasible regions for each generator [41].
The cost function of a given generator with POZ is shown in Fig. 1. 2.6.3. Voltage stability
Therefore, generators have a set of allowed operating zones (i.e. In practice, voltage stability becomes a basic requirement
non-prohibited zones) and they must work in one of these zones. because power systems are heavily stressed. Kessel and Glavitch
POZ can be included as constraints as follows: [42] have developed a voltage stability index (L-index) based on the
feasibility of power flow equations for each node. It varies between
∀i ∈ NG PGmin ≤ PGi ≤ PGmax m = 1, 2, . . .Mi (18) 0 and 1 corresponding to no load and voltage collapse cases, respec-
im im
tively. In other words, L-index calculated at one bus indicates the
where PGmin and PGmax are the minimum and maximum limits for a proximity of voltage collapse condition of that bus. Hence, the max-
im im
the mth allowed operating zone and Mi is the number of allowed imum L-index referred to as Lmax has to be minimized in order to
operating zones for the ith generator. Obviously each generator has enhance the voltage stability of the system.
at least one allowed operating zone. Therefore, in order to simultaneously minimize the total gener-
Therefore, when a generator operates in one of POZ, the strategy ating fuel cost represented by the sum of quadratic functions and
is to add a penalty term to the total objective function. This penalty to enhance the voltage stability represented by Lmax , we propose
term depends on how much the operating point is far from the the following two-fold objective function:
minimum or the maximum of the POZ.  NG 

F(x, u) = ai + bi PGi + ci PG2 + Lmax × Lmax (22)
i
2.5.4. Constraints handling
i=1
It is worth mentioning that control variables are self-
constrained. The inequality constraints of dependent variables where Lmax as VD is a scaling factor to balance between the objec-
that contain load bus voltage magnitude; active power gener- tive function values and to avoid the dominance of an objective
ation at slack bus, reactive power generation and line loadings over another. In this paper Lmax is selected as 5000.
can be included into the objective using a penalizing strategy.
Consequently, the infeasible solutions are considered during the 2.6.4. Multi-fuels
optimization process. However, the unconstrained objective func- In some practical cases, thermal generating plants may have
tion is extended by a penalty function that will penalize infeasible multi-fuel sources such as natural gas and oil. Hence, the fuel cost
solutions. In other words, the amount of infeasibility or the square function or curve is divided into piecewise quadratic cost functions
of the distance to feasibility for each constraint is multiplied by a depending on the number and nature of used fuels as shown in
penalty factor and added to the objective function. Fig. 2.
122 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

Therefore, the objective function reflecting the total generating

without Multi-fuels fuel cost considering valve-point effect is given by:

with Multi-fuels f 
NG

F(x, u) = ai + bi PGi + ci PG2 + |di × sin (ei × (PGi

min
− PGi ))| (25)
i
Generation Cost ($/h)

i=1

e where di and ei are the coefficients that represent the valve-point

loading effect.
d
3. Optimization algorithms

c The optimization algorithms used in this paper to solve the OPF

problem are based on the CBO algorithm. In this section, three ver-
sions of the CBO algorithm that are the standard CBO, the enhanced
b version of the CBO (ECBO) and the improved version of CBO (ICBO)
that is proposed in this paper are detailed.
a The flowchart of the three algorithms used in this paper is given
in Fig. 4.
Output power [MW]
3.1. Colliding Bodies Optimization (CBO)
Fig. 2. Cost function for a single generator with and without multi-fuels.

As previously mentioned, the CBO is a new optimization algo-

Therefore, the generating fuel cost considering multi-fuels rep- rithm that is inspired from the law of collision between two bodies
resented by a piecewise function for one generator can be given [34,44,45]. The CBO uses a population of Colliding Bodies (CBs)
by: where each CB has a mass and a velocity; this last one varies before
and after collision.
fi (x, u) = aik + bik PGi + cik PG2 for fuelk PGmin ≤ Pi ≤ PGmax (23)
i ik ik The CBO main steps are [34,44,45]:
where k is the fuel option.
Hence, the objective function reflecting the total generating fuel Step 1: generate a population of CBs in the search space.
cost is given by: Step 2: calculate the mass of each CB using the following expres-
 NG  sion:

F(x, u) = fi (x, u) (24) 1/F(k)
mk = i=1 , k = 1, 2, . . ., n (26)
i=1 n
1/F(i)

2.6.5. Valve-point effect where F is the objective function and n is the population size.
For more realistic and precise modeling of fuel cost function, Step 3: sort the population where the best CB is ranked first. After
the valve-point effect has to be considered. The generating units that, the population is divided into two groups (with equal size).
with multi-valve steam turbines exhibit a greater variation in the The first group starts from the first (best) CB to the one of the
fuel-cost functions. The valve opening process of multi-valve steam middle and the second group starts from the middle to the end of
turbines produces a ripple-like effect as illustrated in Fig. 3 [43]. the population.
The significance of this effect is that the real cost curve function The CBs of the first group (i.e. the best ones) are considered as
of a large steam plant is not continuous but more important it is stationary whilst the CBs of the second group (the worst ones)
non-linear [43]. move toward the first group. Therefore, the velocities before col-
lision of the first group (stationary CBs) are given by:
n
vi = 0, i = 1, . . ., (27)
2
without valve point effect
and for the second group (moving CBs) they are given by:
with valve point effect
n
vi = xi − xi−(n/2) i = + 1, . . ., n (28)
2
e
Generation Cost ($/h)

where vi and xi are the velocity and the position of the ith CB,
respectively.
Step 4: calculate the new velocities after the pairwise collision
between the members of the first group and the ones of the second
d group. Therefore, the velocities after collision of the first group are
given by:

c (mi+(n/2) + εmi+(n/2) )vi+(n/2) n

vi = , i = 1, . . ., (29)
mi + mi+(n/2) 2
b and for the second group they are given by:
a
(mi − εmi−(n/2) )vi n
vi = , i= + 1, . . ., n (30)
mi + mi−(n/2) 2
Output power [MW]
where vi is the velocity after collision and ε is the Coefficient Of
Fig. 3. Cost function for a single generator with and without valve-point effect. Restitution (COR). It is worth mentioning here that the COR is
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131 123

Start

Read input data

(bus data, line data,
cost coefficients,…)

Select objecves to be opmized

Select control variables

Select constraints

Select the OPF formulaon

Select opmizaon algorithm :

1) ICBO, 2) CBO , 3)ECBO

Which
ICBO algorithm is ECBO
selected

inialize All CBs inialize All CBs

CBO

calculate the masse of

calculate the masse of each CB using (26)
each CB using (26)
inialize All CBs
Update Colliding
Memory (CM) and
Create an archive pool populaon
calculate the masse of
each CB using (26)
Select three different Sort All CBs based on
CBs their objecve funcon
Sort All CBs based on values
their objecve funcon
Collide the selected values Divide the CBs into two
three CBs groups
Divide the CBs into two
groups Collide CBs of the
Evaluate the new
second group with
posions of The three
those of the first group
CBs aer collision using
Collide CBs of the
(40), (41) and (42)
second group with
those of the first group Evaluate the new
posions of CBs aer
Diversify the populaon collision using using (32)
using the CBs of the and (33)
archive pool Evaluate the new
posions of CBs aer
collision using (32) and Regenerate some
Select the best CBs from (33) dimensions of CBs using
all CBs (populaon and the predefined
archive pool) probability Pro

Is the terminaon Is the terminaon Is the terminaon

criterion sasfied criterion sasfied criterion sasfied

Stop

Print opmal
soluon

Fig. 4. Flowchart of the proposed OPF solution using ICBO, CBO and ECBO.
124 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

Table 1 Table 6
The main characteristics of the IEEE 30-bus test system. The main characteristics of the IEEE 118-bus test system.

System characteristics Value Details System characteristics Value Details

Buses 30 [47] Buses 118 [48]

Branches 41 [47] Branches 186 [48]
Generators 6 Buses: 1, 2, 5, 8, 11 and 13. Generators 54 Buses: 1, 4, 6, 8, 10, 12, 15, 18, 19, 24, 25, 26,
Shunts 9 Buses: 10, 12, 15, 17, 20, 21, 23, 24 and 29. 27, 31, 32, 34, 36, 40, 42, 46, 49, 54, 55, 56, 59,
Transformers 4 Branches: 11, 12, 15 and 36. 61, 62, 65, 66, 69, 70, 72, 73, 74, 76, 77, 80, 85,
Control variables 24 – 87, 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
Table 2 Transformers 9 Branches: 8, 32, 36, 51, 93, 95, 102, 107 and
Cost coefficients for the IEEE 30-bus test system. 127.
Control variables 130 –
Generator Bus a b c d e

1 1 0 2 0.00375 18 0.037
2 2 0 1.75 0.0175 16 0.038 and the positions of stationary CBs are updated as follows:
3 5 0 1 0.0625 14 0.04
4 8 0 3.25 0.00834 12 0.045 n
5 11 0 3 0.025 13 0.042
xinew = xi + rand × vi , i = 1, . . ., (33)
2
6 13 0 3 0.025 13.5 0.041
where xinew is the new position of the ith CB after collision.
Steps from 2 to 5 are repeated until a termination criterion is
Table 3
met.
Cost coefficients for generators 1 and 2 for the IEEE 30-bus test system when consid-
ering multi-fuels.

Generator Bus a b c PGmin PGmax a b c PGmin PGmax 3.2. Enhanced Colliding Bodies Optimization (ECBO)
i1 i1 i2 i2

1 1 55 0.7 0.005 50 140 82.5 1.05 0.0075 140 200

As aforesaid, the main addition of the ECBO with respect to the
2 2 40 0.3 0.01 20 55 80 0.6 0.02 55 80
standard CBO is the concept of memory and a mechanism to escape
from local minima. The main steps of the ECBO are [35]:
Table 4
Prohibited zones for the IEEE 30-bus test system. Step 1: generate a population of CBs in the search space.
Step 2: calculate the masse of CBs using (26).
Generator Bus Prohibited zones
Step 3: save the best so far solutions in a Colliding Memory (CM).
Zone 1 Zone 2 Then, the CBs of CM are added to the population and the same
min max min max numbers of current worst CBs are deleted.
Step 4: sort the CBs then create two groups namely: stationary and
1 1 55 66 80 120
2 2 21 24 45 55
moving bodies. The moving bodies move toward the stationary
3 5 30 36 – – ones using the velocities calculated by (27) and (28).
4 8 25 30 – – Step 5: collide bodies two by two where the first one is selected
5 11 25 28 – – from the group of stationary CBs and the second one is selected
6 13 24 30 – –
from the moving CBs. The velocities after collisions are calculated
using (29) and (30).
Step 6: update the new positions of CBs after collision using (32)
Table 5
and (33).
The main characteristics of the IEEE 57-bus test system.
Step 7: regenerate some dimensions of CBs. In this step, for each
System characteristics Value Details CB a random number is generated if this number is lower than a
Buses 57 [48] predefined parameter called Pro within (0,1), one dimension (ran-
Branches 80 [48] domly selected) of this CB is regenerated. The goal of this step to
Generators 7 Buses: 1, 2, 3, 6, 8, 9 and 12. help the algorithm to escape from local optima, it is equivalent to
Shunts 3 Buses: 18, 25 and 53.
the mutation phase in the GA.
Transformers 17 Branches: 19, 20, 31, 35, 36, 37, 41, 46,
54, 58, 59, 65, 66, 71, 73, 76 and 80. Steps from 2 to 7 are repeated until a termination criterion is
Control variables 33 – met as in the standard CBO.

3.3. Improved Colliding Bodies Optimization (ICBO)

defined in order to have a good balance between the exploration
and the exploitation phases of the CBO and it is given by: In the standard version of the CBO the CBs are divided into two
equal groups namely, the stationary CBs (best ones) and the moving
1 − iter CBs (worst ones). Collisions occur between two objects that have
ε= (31)
itermax the same order in their respective groups. For instance, the CB # 4
from the stationary group collides with the CB # 4 from the mov-
where iter and itermax are the current iteration number and the
ing group. Therefore, the worst CBs follow the best ones and then
maximum number of iterations, respectively.
collide with them. The objectives of these two operations are: to
Step 5: evaluate the new positions of CBs after collision. For moving
improve the position of the worst CBs and to move the best CBs
CBs, their positions are updated using the following expression:
toward better positions.
n The main idea proposed here for the ICBO is to collide three
xinew = xi−(n/2) + rand × vi , i= + 1, . . ., n (32)
2 bodies instead of two bodies unlike in CBO and ECBO algorithms.
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131 125

Table 7
Summary of the studied cases.

Test system CASE # Formulation Objective Constraints Considered

N−1
contingencies
(outages of
branches)

CASE # 1 OPF Cost minimization using (19) Equality and non-equality constraints –
CASE #2 SCOPF Cost minimization using (19) Equality and non-equality constraints 1 and 6
for normal and selected N − 1
conditions.
CASE #3 OPF Cost minimization using (19) Equality, non-equality and prohibited –
IEEE 30-bus zones constraints
test system CASE #4 SCOPF Cost minimization using (19) Equality, non-equality and prohibited 1 and 6
zones constraints for normal and
selected N − 1 conditions.
CASE # 5 OPF Cost minimization and voltage profile Equality and non-equality constraints –
improvement using (21)
CASE #6 SCOPF Cost minimization and voltage profile Equality and non-equality constraints 1 and 6
improvement using (21) for normal and selected N − 1
conditions.
CASE #7 OPF Cost minimization and voltage stability Equality and non-equality constraints
enhancement using (22)
CASE #8 SCOPF Cost minimization and voltage stability Equality and non-equality constraints 1 and 6
enhancement using (22) for normal and selected N − 1
conditions.
CASE # 9 OPF Cost minimization considering Equality and non-equality constraints –
multi-fuels using (24)
CASE #10 SCOPF Cost minimization considering Equality and non-equality constraints 1 and 6
multi-fuels using (24) for normal and selected N − 1
conditions.
CASE #11 OPF Cost minimization considering valve Equality and non-equality constraints –
point effect using (25)
CASE #12 SCOPF Cost minimization considering valve Equality and non-equality constraints 1 and 6
point effect using (25) for normal and selected N − 1
conditions.

IEEE 57-bus CASE # 13 OPF Cost minimization using (19) Equality and non-equality constraints –
test system CASE # 14 SCOPF Cost minimization using (19) Equality and non-equality constraints 8 and 50
for normal and selected N − 1
conditions.

IEEE 118-bus CASE # 15 OPF Cost minimization using (19) Equality and non-equality constraints –
test system CASE # 16 SCOPF Cost minimization using (19) Equality and non-equality constraints 21 and 50
for normal and selected N − 1
conditions.

Therefore, in every iteration, three CBs are selected based on a b. calculate the new velocities of the three CBs after collision using
selection rule. The best CB among these three CBs is assumed to the following expressions:
be stationary and the remaining two CBs are assumed to be mov-
(mCB2 + εmCB2 )vCB2 (mCB3 + εmCB3 )vCB3
ing in the search space. The moving CBs collide the stationary one vCB1 = + (37)
and move in the search space with one more research direction mCB1 + mCB2 mCB1 + mCB3
than in CBO and ECBO. (mCB2 − εmCB1 )vCB2
The main steps of the proposed ICBO are: vCB2 = (38)
mCB2 + mCB1

(mCB3 − εmCB1 )vCB3

Step 1: initialize the population of CBs. vCB3 = (39)
Step 2: calculate the masses of CBs using (26). mCB3 + mCB1
Step 3: create an archive pool that has a size of 1/4 of the population c. update the position of each CB using the following expression:
size.
new 
Step 4: for i = 1 → n xCB = xCB1 + rand × xCB (40)
1 1
a. select three CBs based on a selection rule. If one of the CBs is new 
the same as another replace it. Then rank these three CBs based xCB = xCB2 + rand × xCB (41)
2 2
of their fitness. The first CB or CB1 is considered as stationary new 
xCB = xCB3 + rand × xCB (42)
while CB2 and CB3 are considered as moving. The velocities of 3 3

each CB before collision are defined by: d. check the CBs, if a CB goes beyond the bounds of the search
space it is brought back inside it.
vCB1 = 0 (34) Step 5: diversify the population by exchanging some dimensions
of the CBs from the population with the ones from the archive pool.
vCB2 = xCB1 − xCB2 (35) Step 6: group both the CBs from the population and the ones from
the archive pool and select the best n ones to keep the same num-
vCB3 = xCB1 − xCB3 (36) ber of CBs.
126 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

Table 8
Optimal results for CASE 1 through CASE 12.

CASE 1 CASE 2 CASE 3 CASE 4 CASE 5 CASE 6 CASE 7 CASE 8 CASE 9 CASE 10 CASE 11 CASE 12

PG1 177.0420 129.8879 179.1541 129.8979 176.1456 129.8709 176.6085 129.8745 139.9998 129.9232 198.9038 129.9606
PG2 48.6983 63.8534 44.9999 64.5881 48.7418 65.0360 48.4090 63.8713 54.9998 54.9872 44.0371 78.4900
PG5 21.3264 26.4452 21.4360 25.3971 21.6498 26.1814 21.4258 27.3903 24.4796 27.7688 18.6926 24.7823
PG8 21.0768 34.9958 22.1359 34.7964 22.4701 34.8561 21.6314 31.5182 34.6169 34.8721 10.0033 34.3127
PG11 11.8689 21.8744 12.2960 21.1735 12.0759 21.5512 11.9694 22.8264 18.1996 24.8975 10.0002 14.9040
PG13 12.0008 18.2776 12.0003 19.8971 12.0622 19.0480 12.0023 19.8798 17.4871 23.6521 12.0000 12.9360
VG1 1.1000 1.0997 1.1000 1.0876 1.0395 1.0462 1.0996 1.0982 1.0993 1.0636 1.1000 1.0963
VG2 1.0807 1.0845 1.0804 1.0724 1.0205 1.0299 1.0808 1.0831 1.0828 1.0475 1.0777 1.0813
VG5 1.0542 1.0600 1.0536 1.0485 1.0041 1.0065 1.0613 1.0606 1.0563 1.0214 1.0500 1.0519
VG8 1.0619 1.0503 1.0617 1.0388 1.0031 0.9991 1.0612 1.0467 1.0690 1.0202 1.0612 1.0484
VG11 1.1000 1.0631 1.1000 1.0621 1.0353 1.0320 1.0950 1.0994 1.0995 1.0597 1.0999 1.0990
VG13 1.1000 1.0635 1.1000 1.0263 1.0011 1.0018 1.1000 1.0966 1.0996 1.0566 1.1000 1.0870
T11(6–9) 1.0221 1.0760 1.0230 0.9876 1.0494 1.0402 0.9725 1.0123 0.9833 1.0201 1.0366 0.9641
T12(6–10) 0.9001 0.9119 0.9002 0.9980 0.9040 0.9007 0.9054 0.9088 0.9738 0.9046 0.9218 0.9724
T15(4–12) 0.9665 0.9448 0.9676 0.9774 0.9601 0.9683 0.9630 0.9271 0.9811 0.9183 0.9886 0.9314
T36(28–27) 0.9542 0.9799 0.9544 0.9853 0.9676 0.9671 0.9390 0.9353 0.9680 0.9688 0.9666 0.9576
QC10 4.9953 4.3870 4.9951 4.1421 4.9889 0.6189 4.4335 2.5856 4.8294 4.2266 4.9961 4.1170
QC12 4.9618 0.8513 4.9909 3.3363 0.5828 0.2966 4.8880 1.9035 2.9372 1.8160 4.9903 2.1736
QC15 4.9586 4.5079 4.7669 3.0336 4.9932 4.9827 4.7446 1.0663 4.0182 4.3858 4.9386 3.6649
QC17 4.9978 3.7012 4.9627 3.1257 0.1543 3.6664 3.9321 1.3616 4.0978 1.9859 4.9959 3.5345
QC20 4.3264 2.3512 4.3618 1.2970 4.9998 4.9831 4.2391 3.3712 4.4819 2.7092 4.4360 0.9915
QC21 4.9985 3.6837 4.9958 3.7637 4.9965 4.3942 4.9758 4.1988 3.8690 4.8731 5.0000 4.0513
QC23 2.6712 4.2058 2.6157 1.6389 4.9852 4.8205 1.7367 2.1157 3.4973 1.7431 3.0879 2.9981
QC24 4.9988 4.3439 4.9996 2.3082 4.9940 4.9650 1.7975 2.2735 4.6754 4.0602 4.9884 4.4103
QC29 2.3894 3.4305 2.4016 2.6300 2.5994 2.8754 0.8419 3.5901 2.1650 3.4776 2.5689 4.4294

Cost ($/h) 799.0353 823.5223 799.3079 824.2754 803.3978 826.3644 799.3277 824.6657 645.1668 678.5076 830.4531 868.4815
VD (pu) 1.9652 1.2261 1.9526 0.6039 0.1014 0.1187 1.9961 1.8000 1.8232 0.8595 1.7450 1.7116
Lmax 0.1261 0.1347 0.1263 0.1419 0.1490 0.1488 0.1252 0.1259 0.1282 0.1398 0.1289 0.1286
Ploss (MW) 8.6132 11.9342 8.6221 12.3501 9.7453 13.1435 8.6465 11.9605 6.3828 12.7008 10.2370 11.9856
Qloss (MVar) 36.3549 50.6337 36.3904 48.1407 42.3559 53.9316 34.8570 48.0963 26.1448 51.1102 41.9725 47.0434

Steps from 2 to 6 are repeated until a termination criterion is 57-bus test system. This system has a total generation capacity of
met. The termination criterion used for the ICBO is the same one 1975.9 MW and its main characteristics are given in Table 5. The
used for the CBO and the ECBO that is the maximum number of cost coefficients and the detailed data of this test system can be
iterations. derived from [48].

4. Applications and results 4.1.3. IEEE 118-bus test system

The proposed ICBO algorithm along with the CBO and ECBO algo-
rithms are implemented to solve the OPF problem. In this paper, 16
different case studies have been investigated using 3 test systems.
The developed programs for this paper are written using MAT-
LAB and MATPOWER [46] and applied on a 2.20 GHz i7 personal
computer with 8.00 GB-RAM using parallel processing to run the
different runs.
4.1. Test systems
Three test systems are used to evaluate the performance of the
proposed algorithm in addition to the comparison algorithms. The
main characteristics of these tests are described below.
4.1.1. IEEE 30-bus test system
The IEEE 30-bus test system has a total generation capacity of
435.0 MW and its main characteristics are given in Table 1. It is
worth mentioning that, for CASE 9 and CASE 10 generators of buses
1 and 2 have two fuel options i.e. k = 2 and for CASE 11 and CASE 12
generators of buses 1 and 2 are considered to have valve-point load-
ing effects on their characteristics. Cost coefficients and prohibited
zones used for this test system are given in Tables 2–4. Detailed
data about this test system can be derived from [47].
4.1.2. IEEE 57-bus test system
In order to test the scalability of the proposed ICBO algorithm, a
large-scale test system is considered in this paper, which is the IEEE
Another large-scale test system is considered in this paper
which is the IEEE 118-bus test system. This system has a total gener-
ation capacity of 9966.2 MW and its main characteristics are given
in Table 6. The cost coefficients and the detailed data of this test
system can be derived from [48].
4.2. Case studies
As previously mentioned, 16 cases have been investigated in this
paper. These cases are described in Table 7.
4.3. Results and discussion
The proposed ICBO has been applied to the investigated cases
and the optimal results found are given in Table 8 for the IEEE 30-
bus test system, in Table 9 for the IEEE 57-bus test system and in
Table 10 for the IEEE-118-bus test system.
For the IEEE 30-bus test system, solving the classical OPF model
in CASE 1 by the ICBO leads to a fuel cost of 799.035 $/h at transmis-
sion losses of 8.613 MW while the voltage deviation is 1.965 p.u. and
the stability index is 0.126. Applying the ICBO for the SCOPF model
in CASE 2 increases the fuel costs, active and reactive transmission
power losses by 24.487 $/h, 3.321 MW and 14.279 MVar, respec-
tively. Similar analysis can be made for the remaining cases. CASE
5 has the lowest voltage deviation level (0.101 p.u.) and CASE 7 has
the lowest Lmax value (0.125).
For the IEEE 57-bus test system, the total fuel cost in CASE
13 is 41,697.332 $/h, the voltage deviation is 1.317 p.u., the sta-
bility index is 0.277 and active and reactive power losses are
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131 127

Table 9 Table 10
Optimal settings of control variables for CASE 13 and CASE 14. Optimal settings of control variables for CASE 15 and CASE 16.

CASE 13 CASE 14 CASE 15 CASE 16

PG1 142.7369 183.7091 PG1 370.4200 351.8375

PG2 89.1299 100.0000 PG4 30.0012 30.0038
PG3 45.0609 52.0539 PG6 30.0060 30.2107
PG6 71.7544 44.1313 PG8 30.0004 30.4316
PG8 459.8216 391.7525 PG10 30.0017 30.0119
PG9 96.5248 100.0000 PG12 316.8150 286.2531
PG12 361.3184 409.9998 PG15 68.0579 58.6399
VG1 1.0331 1.0504 PG18 30.0344 72.8628
VG2 1.0312 1.0453 PG19 30.0022 87.4491
VG3 1.0257 1.0273 PG24 30.0009 33.4671
VG6 1.0387 1.0096 PG25 30.0031 30.0033
VG8 1.0546 0.9972 PG26 152.2583 137.9710
VG9 1.0225 0.9996 PG27 221.2265 191.1271
VG12 1.0203 1.0542 PG31 30.0023 30.0043
T19(4–18) 0.9700 0.9253 PG32 32.1001 32.1167
T20(4–18) 0.9386 0.9380 PG34 30.0468 30.0031
T31(21–20) 1.0091 1.0455 PG36 30.0023 70.9907
T35(24–25) 0.9348 0.9203 PG40 30.0002 63.8937
T36(24–25) 0.9653 0.9193 PG42 30.0011 30.0212
T37(24–26) 1.0263 1.0073 PG46 30.0074 30.0095
T41(7–29) 0.9688 0.9731 PG49 35.7188 35.7103
T46(34–32) 0.9382 0.9242 PG54 162.5469 154.8000
T54(11–41) 0.9003 0.9534 PG55 44.4008 44.5130
T58(15–45) 0.9485 0.9515 PG56 30.3515 30.1080
T59(14–46) 0.9362 0.9451 PG59 30.0022 30.0317
T65(10–51) 0.9465 0.9679 PG61 126.3960 120.8982
T66(13–49) 0.9093 0.9312 PG62 123.5531 116.7452
T71(11–43) 0.9427 0.9311 PG65 30.0022 30.0125
T73(40–56) 1.0000 0.9927 PG66 290.6133 273.2605
T76(39–57) 0.9713 0.9708 PG69 290.4850 278.3913
T80(9–55) 0.9659 0.9789 PG70 30.0008 30.0343
QC18 4.9036 1.6566 PG72 30.0018 30.4936
QC25 5.0000 4.9947 PG73 30.0016 30.0281
QC53 5.0000 4.9999 PG74 30.0015 30.0361
PG76 30.0023 30.0955
Cost ($/h) 41,697.3324 42,107.4039
PG77 30.0046 30.6346
VD (pu) 1.3173 1.1280
PG80 350.1090 332.8898
Lmax 0.2776 0.2825
PG85 30.0038 30.0208
Ploss (MW) 15.5470 30.8466
PG87 31.2000 31.2020
Qloss (MVar) 69.5480 116.6855
PG89 378.9986 364.1937
PG90 30.0003 30.0493
PG91 30.0006 30.0347
PG92 30.0004 30.6989
15.547 MW and 69.548 MVar, respectively. For the SCOPF model, PG99 30.0000 30.5101
PG100 176.5001 176.4264
the fuel cost, stability index, active and reactive power losses are
PG103 42.0025 42.0132
increased compared to CASE 13 by 410.072 $/h, 0.0049, 15.299 MW PG104 30.0007 30.0258
and 47.138 Mvar while the voltage deviation has decreased by PG105 30.0190 30.0035
0.189 p.u. PG107 30.0016 31.6551
PG110 30.0008 30.0071
For the IEEE 118-bus test system, the total fuel cost in CASE 15
PG111 40.8017 40.8041
is 135,121.570 $/h, the voltage deviation is 0.898 p.u., the stability PG112 30.0005 30.0866
index is 0.0691 and active and reactive power losses are 62.731 MW PG113 30.0048 30.0034
and 387.199 MVar, respectively. For the SCOPF model, the fuel cost PG116 30.0170 30.1226
is increased by 1231.7436 $/h compared with the classical OPF VG1 0.9720 1.0037
VG4 1.0000 1.0177
model.
VG6 0.9927 1.0283
Some samples of results are sketched in Figs. 5–8. These figures VG8 1.0388 0.9999
show the evolution of the objective function over iterations. VG10 1.0499 1.0280
VG12 0.9879 1.0246
VG15 0.9951 1.0057
4.4. Performance evaluation study VG18 0.9998 1.0179
VG19 0.9960 1.0120
In order to evaluate the performances of the ICBO, it has been VG24 1.0203 1.0572
compared with several optimization algorithms including CBO, VG25 1.0374 1.0441
VG26 1.0600 0.9884
ECBO, DE, PSO, ABC, GA and BBO. Control parameters of optimiza- VG27 1.0094 1.0199
tion algorithms used in this investigation are given in Table 11. VG31 1.0028 1.0199
In order to evaluate the performances of a new optimization VG32 1.0074 1.0222
algorithm, most of the approaches focus their judgments only on VG34 0.9943 1.0213
VG36 0.9901 1.0225
the best values achieved for each case, which is not fully correct. In
VG40 0.9781 0.9852
this paper, the evaluation of the developed algorithm performances VG42 0.9833 0.9840
compared with other algorithms is achieved based on the best and VG46 1.0093 1.0329
average values. Another misconception is that many researchers VG49 1.0181 1.0108
think that for an algorithm A to be better than another algorithm
128 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

Table 10 (Continued) 10 7
840 10
CASE 15 CASE 16
Cost
VG54 0.9959 0.9695
Penalty
VG55 0.9949 0.9638 830 8
V56 0.9954 0.9667
V59 1.0193 0.9803
VG61 1.0276 0.9893 820 6

Cost ($/h)

Penalty
VG62 1.0250 0.9898
VG65 1.0261 1.0040
VG66 1.0322 1.0159
810 4
VG69 1.0288 1.0264
VG70 1.0030 1.0154
VG72 1.0168 1.0291
800 2
VG73 1.0084 1.0331
VG74 0.9793 0.9840
VG76 0.9704 0.9630
VG77 1.0020 0.9885 790 0
VG80 1.0140 0.9957 1 125 250 375 500
Iterations
VG85 1.0136 0.9966
VG87 1.0355 1.0251
Fig. 5. Evolution of cost and penalty for CASE 1.
VG89 1.0248 1.0086
VG90 1.0078 0.9925
VG91 1.0115 0.9956
VG92 1.0114 0.9949 10 8
VG99 1.0126 0.9940 1000 4
VG100 1.0124 1.0026
VG103 1.0082 1.0039 Cost
VG104 1.0030 0.9985 Penalty
VG105 0.9998 0.9973
VG107 0.9932 0.9973
VG110 0.9989 1.0058
Cost ($/h)

Penalty
VG111 1.0077 1.0180
VG112 0.9892 0.9989 900 2
VG113 1.0083 1.0153
VG116 1.0236 0.9867
T8(8–5) 1.0324 0.9727
T32(26–25) 1.0741 1.0009
T36(30–17) 1.0270 1.0239
T51(38–37) 1.0134 0.9709
T93(63–59) 0.9821 1.0103
T95(64–61) 0.9947 1.0138 800 0
T102(65–66) 0.9849 0.9970 1 125 250 375 500
Iterations
T107(68–69) 0.9463 0.9296
T127(81–80) 0.9931 0.9912
QC5 3.6110 2.7316 Fig. 6. Evolution of cost and penalty for CASE 2.
QC34 3.9118 4.9990
QC37 0.2035 1.7433
QC44 4.9939 2.7843 900 0.4
QC45 4.9696 4.9801
QC46 0.4866 1.5183 Cost
QC48 4.8699 2.0490 VD
QC74 0.9253 3.3752
QC79 4.8979 1.8538
QC82 4.9221 0.8065
Cost ($/h)

QC83 4.9237 5.0000 VD (p.u.)

QC105 4.6048 2.2138 850 0.2
QC107 2.4390 2.0355
QC110 3.7956 1.7877

Cost ($/h) 135,121.5704 136,353.3140

VD (pu) 0.8976 1.0525
Lmax 0.0691 0.0669
Ploss (MW) 62.7315 61.8480
Qloss (MVar) 387.1991 358.0187
800 0
1 125 250 375 500
Iterations

B, A has to be better than B over all tested cases which is, once
again, not true. Therefore, in order to make reliable conclusions, in
this paper, a ranking procedure is applied on several algorithms to
detect the best one or more precisely to detect where the developed
ICBO stands or ranks among these algorithms.
Ranking is given to each algorithm on every case based on the
following criteria:
- Cases: Total 16 cases are tested in this paper.
- Runs/case: 30 runs per case are done.
Fig. 7. Evolution of cost and VD for CASE 5.
The developed ranking procedure is achieved as follows:
1. Rank the algorithms based on their ‘best’ values.
2. Rank the algorithms based on their ‘average’ values.
3. Calculate the overall rank, which is equal to the sum of ranks for
each algorithm for all cases.
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131 129

835 0.165 Table 11

Control parameters of the related algorithms used in the tests.
Cost
830 0.16
L max Algorithm Control parameters
825 0.155 Value Description

820 0.15 ICBO – –

Cost ($/h)

CBO – –

Lmax
815 0.145 ECBO CM = 5 Colliding memory
Pro = 0.3 Regeneration
810 0.14 probability
DE F = 0.8 Differential weight
805 0.135 Cr = 0.8905 Crossover probability
PSO w = 0.5 Inertia factor
800 0.13 c1 = 1.5 Cognitive factor
c2 = 1.5 Social factor
795 0.125 ABC FoodNumber = Np/2 Number of food
1 125 250 375 500 sources
terations GA Mutrate = .05 Mutation rate
Selection = 0.75 Fraction of population
Fig. 8. Evolution of cost and Lmax for CASE 7. kept
BBO pmutate = 0 Mutation probability
pmodify = 1 Habitat modification
4. Calculate the Average Rank using the following formula: pmutate = 0.005 probability
Initial mutation
Total Rank probability
Average Rank = (43) Common Population size = 50 for cases: 1, 2, 3, 4, 5, 6, 7, 8, 11
2 × Number of CASES
parameters and 12.
5. Rank the methods based on their Average Rank. Population size = 90 for cases: 9, 10, 13, 14, 15 and
16.
Maximum number of iterations = 500 for cases: 1,
It is worth mentioning that, while ranking algorithms based on 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 and 12.
their average values, some precautions have to be taken. Because Maximum number of iterations = 1500 for cases:
in 30 runs and for a given case an algorithm can converge (sat- 13 and 14.
Maximum number of iterations = 2500 for cases:
isfy all the constraints) only for some cases and not for all cases.
15 and 16.

Table 12
The best objective function values obtained using each algorithm.

ICBO CBO ECBO DE PSO ABC GA BBO

CASE 1 799.035 799.035 799.035 799.0344 799.9409 799.0563 800.3108 799.0860

CASE 2 823.522 824.746 825.385 823.5108 827.2236 829.6482 833.4801 829.3074
CASE 3 799.308 799.307 799.310 799.3061 804.2729 799.3356 801.5192 799.3914
CASE 4 824.275 825.035 827.611 824.4761 827.7165 827.0804 836.1545 826.0685
CASE 5 813.537 813.590 813.879 813.5170 819.9882 815.2269 823.5423 815.2450
CASE 6 838.233 838.712 842.108 841.3250 855.4596 843.7358 861.2381 848.0091
CASE 7 1425.343 1425.260 1428.285 1424.7855 1435.8177 1428.1300 1435.8668 1429.3386
CASE 8 1454.159 1460.939 1462.248 1454.5462 1481.9548 1478.6515 1498.7731 1473.1221
CASE 9 645.167 645.101 645.384 645.5585 648.5344 647.7998 653.2009 647.0433
CASE 10 678.508 679.976 678.051 677.9645 685.9116 686.1269 713.5651 694.5585
CASE 11 830.453 830.473 830.587 830.4396 832.6871 831.7708 833.1331 831.4847
CASE 12 868.482 870.216 871.085 868.1192 874.1837 871.5738 881.3757 875.0982
CASE 13 41,697.333 41,701.953 41,702.663 41,689.7303 42,386.3675 41,715.7607 41,841.8455 41,698.9307
CASE 14 42,107.404 42,142.316 42,129.602 42,096.1742 44,536.7028 42,228.8392 43,393.5257 42,302.1268
CASE 15 135,121.570 135,072.999 135,172.266 142,751.1178 – 135,145.1889 – 135,272.1959
CASE 16 136,353.314 135,297.225 135,582.150 158,920.8074 – 135,759.0150 – 135,640.6462

Table 13
The average objective function values obtained using each algorithm.

ICBO CBO ECBO DE PSO ABC GA BBO

CASE 1 799.042 799.468 800.434 799.282 NC (1) 799.723 802.723 800.986

CASE 2 825.662 827.349 834.995 NC (1) NC (8) 854.701 NC (1) NC (4)
CASE 3 808.397 806.952 800.553 807.829 NC (3) 800.109 803.871 801.338
CASE 4 839.715 NC (1) NC (2) NC (2) NC (21) NC (1) NC (25) NC (9)
CASE 5 813.971 815.761 817.251 817.187 NC (1) 817.293 834.309 819.109
CASE 6 843.913 844.329 856.080 NC (1) NC (7) 873.174 NC (1) NC (5)
CASE 7 1429.600 1435.242 1435.427 1429.360 NC (1) 1438.041 1451.715 1439.097
CASE 8 1462.752 1470.081 1491.573 NC (1) NC (7) 1544.443 NC (1) NC (3)
CASE 9 646.084 646.979 646.927 646.693 665.539 651.484 661.823 649.996
CASE 10 737.041 769.329 800.402 694.829 NC (5) 759.727 NC (1) NC (1)
CASE 11 830.550 831.744 833.239 831.026 NC (1) 834.211 840.463 834.462
CASE 12 872.369 874.754 887.878 NC (1) NC (7) 901.262 NC (1) NC (4)
CASE 13 41,713.954 41,718.610 41,737.442 NC (2) NC (25) 41,740.708 NC (14) 41,768.811
CASE 14 42,186.462 42,250.567 42,247.043 NC (6) NC (28) NC (3) NC (27) NC (4)
CASE 15 135,175.672 135,317.9243 135,407.919 NC (29) NC (30) NC (1) NC (30) NC (1)
CASE 16 136,626.021 136,002.1066 NC (1) NC (30) NC (30) 136,433.2011 NC (30) NC (10)
No. of NC 0 1 3 73 175 5 131 41
130 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131

Table 14 Table 15
Comparison of algorithms and final ranking. Wilcoxon signed ranks test results.

Algorithm Total rank Average rank Final ranking Comparison R+ R− p-Value

ICBO 63 1.97 1 ICBO versus CBO 492 36 9.71E−06

CBO 83 2.59 2 ICBO versus ECBO 503 25 4.10E−06
DE 96 3.00 3 ICBO versus DE 405 123 4.30E−03
ECBO 115 3.59 4 ICBO versus PSO 528 0 4.17E−07
ABC 144 4.50 5 ICBO versus ABC 512 16 1.85E−06
BBO 171 5.34 6 ICBO versus GA 524 4 6.10E−07
GA 222 6.94 7 ICBO versus BBO 515 13 1.41E−06
PSO 226 7.06 8

Step 3: Compute the p-value which represents an indication about

Therefore, the average values calculated for such cases are very high
and meaningless due to the high penalty factors used. To overcome
this issue, we have developed a new approach that rank algorithms
based on their average values using the following rules:
- For a given case, if two algorithms A and B converge in all runs,
they are ranked based on their calculated average values.
- For a given case, if an algorithm A converges in all runs and
another algorithm B converges only in some runs, therefore A
is ranked higher than B.
- For a given case, if an algorithm A converges in N1 out of N runs
and another algorithm B converges in N2 out of N runs, therefore
if N1 > N2, A is ranked higher than B otherwise it is the opposite.
The best objective function values obtained using the tested
algorithms for all cases are displayed in Table 12 while the
average objective function values are displayed in Table 13. In
Table 13, NC stands for Non-Convergence and the value between
brackets beside NC refers to how many times this algorithm has
not converge for this case. The final ranking of algorithms is
given in Table 14. From these tables we can make the following
comments:
1. The best algorithm i.e. the one that has the best ranking is the
ICBO while the standard CBO and the enhanced one ECBO are
ranked 2nd and 4th, respectively.
2. The ICBO is the only algorithm that converges for all cases and
over all runs.
3. The most difficult case to solve for the ICBO is CASE 10 where
the average value (normalized with respect to the best value) is
8.6% higher than the best value.
4. The ICBO is a robust algorithm because the maximum normal-
ized error between the best and the average values for the worst
case is quite low.
5. The average normalized error between the best and the average
values calculated over all cases for the ICBO is around 1.0%.
In order to validate our comparative study, a method from infer-
ential statistics that is the Wilcoxon signed ranks test is used [49].
Generally, in order to use this test, the null hypothesis H0 and the
alternative or research hypothesis H1 are stated. Then the level of
significance value associated with H0 is set. After that, the value
needed for rejection of H0 is determined and compared with the
critical value. Finally, results are interpreted based on whether H0
has been rejected or not.
In this paper, the Wilcoxon signed ranks test is implemented
in order to test whether the proposed ICBO outperforms another
algorithm A as in [49]. The procedure used is as flows:
Step 1: Collect the performances of ICBO and a given algorithm A
i.e. best and average values of the objective function over all cases.
Step 2: Compute R+ and R− which are the sum of ranks for cases
in which ICBO outperform A and the opposite, respectively.
whether a statistical test based on hypotheses is significant or not.
The smaller the p-value the stronger the evidence against H0 .
Step 4: Interpret results.
Table 15 shows the results obtained using Wilcoxon signed
ranks test for pairwise comparisons between ICBO and seven other
algorithms based on the best and average values. From these
results, we can say that ICBO shows a significant improvement over
CBO, ECBO, DE, PSO, ABC, GA and BBO with a level of significance
of ˛ = 0.01.
It is worth mentioning that, in this paper we have chosen as
comparative algorithms to the ICBO some well-known algorithms.
However, the potential of physics-based global optimization tech-
niques like those presented in [50] can be also examined for OPF
problems in our future research.
5. Conclusion
In this work, an improved version of the CBO algorithm in order
to solve the OPF problem is proposed. Two OPF formulations, three
test systems, different objectives and constraints are considered
to assess the effectiveness of the proposed algorithm. The results
of cases considered demonstrate that: (1) the high efficiency and
robustness of the proposed ICBO algorithm compared with the
standard and enhanced versions of the CBO algorithm and other
well-known algorithms, (2) the proposed method is applicable for
a variety of cases with complex objective functions, security con-
straints, prohibited zones and different test systems.
In view of the potential and superior characteristics of the
proposed ICBO, a multiobjective algorithm based on ICBO is rec-
ommended to be developed and applied to solve multiobjective
OPF problems.
Acknowledgments
Dr. M. A. Abido would like to acknowledge the support pro-
vided by King Abdulaziz City for Science and Technology (KACST)
through the Science and Technology Unit at King Fahd University
of Petroleum and Minerals (KFUPM) for funding this work through
project# 14-ENE265-04 as a part of the National Science, Technol-
ogy and Innovation Plan (NSTIP).
References
[1] A. Mukherjee, V. Mukherjee, Solution of optimal power flow using chaotic krill
herd algorithm, Chaos Solitons Fractals 78 (2015) 10–21, http://dx.doi.org/10.
1016/j.chaos.2015.06.020.
[2] J. Hazra, A.K. Sinha, A Multi-Objective Optimal Power Flow Using Particle
Swarm Optimization, 2010, http://dx.doi.org/10.1002/etep.
[3] V.H. Hinojosa, R. Araya, Modeling a mixed-integer-binary small-population
evolutionary particle swarm algorithm for solving the optimal power flow
problem in electric power systems, Appl. Soft Comput. J. 13 (2013) 3839–3852,
http://dx.doi.org/10.1016/j.asoc.2013.05.005.
[4] M. Ghasemi, S. Ghavidel, E. Akbari, A.A. Vahed, Solving non-linear, non-smooth
and non-convex optimal power flow problems using chaotic invasive weed
optimization algorithms based on chaos, Energy 73 (2014) 340–353, http://dx.
doi.org/10.1016/j.energy.2014.06.026.
 H.R.E.H. Bouchekara et al. / Applied Soft Computing 42 (2016) 119–131
[5] S. Rebennack, P.M. Pardalos, M.V.F. Pereira, N.A. Iliadis (Eds.), Handbook of
Power Systems II, Springer-Verlag, Berlin/Heidelberg, 2010, http://dx.doi.org/
10.1007/978-3-642-12686-4.
[6] M. Niu, C. Wan, Z. Xu, A review on applications of heuristic optimization algo-
rithms for optimal power flow in modern power systems, J. Mod. Power Syst.
Clean Energy 2 (2014) 289–297, http://dx.doi.org/10.1007/s40565-014-0089-
4.
[7] S. Sivasubramani, K.S. Swarup, Multiagent based differential evolution
approach to optimal power flow, Appl. Soft Comput. J. 12 (2012) 735–740,
http://dx.doi.org/10.1016/j.asoc.2011.09.016.
[8] J.A. Momoh, M. El-Hawary, R. Adapa, A review of selected optimal power flow
literature to 1993. Part II: Newton, linear programming and interior point meth-
ods, IEEE Trans. Power Syst. 14 (1999) 105–111.
[9] K.S. Pandya, S.K. Joshi, A survey of optimal power flow methods, J. Theor. Appl.
Inf. Technol. (2008) 450–458.
[10] S. Frank, I. Steponavice, Optimal power flow: a bibliographic survey
I—formulations and deterministic methods, Energy Syst. (2012) 221–258,
http://dx.doi.org/10.1007/s12667-012-0056-y.
[11] M. Ghasemi, S. Ghavidel, M. Gitizadeh, E. Akbari, An improved teaching –
learning-based optimization algorithm using Lévy mutation strategy for non-
smooth optimal power flow, Int. J. Electr. Power Energy Syst. 65 (2015)
375–384, http://dx.doi.org/10.1016/j.ijepes.2014.10.027.
[12] T. Niknam, M. rasoul Narimani, M. Jabbari, A.R. Malekpour, A modified shuf-
fle frog leaping algorithm for multi-objective optimal power flow, Energy 36
(2011) 6420–6432, http://dx.doi.org/10.1016/j.energy.2011.09.027.
[13] S.R. Paranjothi, K. Anburaja, Optimal power flow using refined genetic algo-
rithm, Electr. Power Components Syst. 30 (2002) 1055–1063, http://dx.doi.org/
10.1080/15325000290085343.
[14] M.Z.L.L. Lai, J.T. Ma, R. Yokoyama, Improved genetic algorithms for optimal
power flow under both normal and contingent operation states, Electr. Power
Energy Syst. 19 (1997) 287–292.
[15] M.A. Abido, Optimal power flow using tabu search algorithm, Electr. Power
Components Syst. (2002) 469–483.
[16] M.A. Abido, Optimal power flow using particle swarm optimization, Int.
J. Electr. Power Energy Syst. 24 (2002) 563–571, http://dx.doi.org/10.1016/
S0142-0615(01)00067-9.
[17] A.C. Roa-Sepulveda, B.J. Pavez-Lazo, A solution to the optimal power flow using
simulated annealing, Int. J. Electr. Power Energy Syst. 25 (2003) 47–57, http://
dx.doi.org/10.1016/S0142-0615(02)00020-0.
[18] A.A. Abou El Ela, M.A. Abido, S.R. Spea, Optimal power flow using differential
evolution algorithm, Electr. Power Syst. Res. 80 (2010) 878–885, http://dx.doi.
org/10.1016/j.epsr.2009.12.018.
[19] A.J. Ghanizadeh, G. Mokhtari, M. Abedi, G.B. Gharehpetian, Optimal power
flow based on imperialist competitive algorithm, Int. Rev. Electr. Eng. 6 (2011)
1847–1852.
[20] M. Ghasemi, S. Ghavidel, M.M. Ghanbarian, H.R. Massrur, M. Gharibzadeh,
Application of imperialist competitive algorithm with its modified techniques
for multi-objective optimal power flow problem: a comparative study, Inf. Sci.
(Ny.) 281 (2014) 225–247, http://dx.doi.org/10.1016/j.ins.2014.05.040.
[21] P.K. Roy, S.P. Ghoshal, S.S. Thakur, Biogeography based optimization for multi-
constraint optimal power flow with emission and non-smooth cost function,
Expert Syst. Appl. 37 (2010) 8221–8228, http://dx.doi.org/10.1016/j.eswa.
2010.05.064.
[22] A. Bhattacharya, P.K. Chattopadhyay, Application of biogeography-based opti-
misation to solve different optimal power flow problems, IET Gener. Transm.
Distrib. 5 (2011) 70–80, http://dx.doi.org/10.1049/iet-gtd.2010.0237.
[23] A. Bhattacharya, P.K. Roy, Solution of multi-objective optimal power flow using
gravitational search algorithm, IET Gener. Transm. Distrib. (2012) 751–763,
http://dx.doi.org/10.1049/iet-gtd.2011.0593.
[24] A.R. Bhowmik, A.K. Chakraborty, Solution of optimal power flow using non
dominated sorting multi objective opposition based gravitational search algo-
rithm, Int. J. Electr. Power Energy Syst. 64 (2015) 1237–1250, http://dx.doi.org/
10.1016/j.ijepes.2014.09.015.
[25] N. Sinsuphan, U. Leeton, T. Kulworawanichpong, Optimal power flow solu-
tion using improved harmony search method, Appl. Soft Comput. J. 13 (2013)
2364–2374, http://dx.doi.org/10.1016/j.asoc.2013.01.024.
[26] A. Khorsandi, S.H. Hosseinian, A. Ghazanfari, Modified artificial bee colony
algorithm based on fuzzy multi-objective technique for optimal power flow
problem, Electr. Power Syst. Res. 95 (2013) 206–213, http://dx.doi.org/10.1016/
j.epsr.2012.09.002.
[27] K. Ayan, U. Kılıç, B. Baraklı, Chaotic artificial bee colony algorithm based solution
of security and transient stability constrained optimal power flow, Int. J. Electr.
Power Energy Syst. 64 (2015) 136–147, http://dx.doi.org/10.1016/j.ijepes.2014.
07.018.
131
[28] H.R.E.H. Bouchekara, Optimal power flow using black-hole-based optimization
approach, Appl. Soft Comput. J. 24 (2014) 879–888, http://dx.doi.org/10.1016/
j.asoc.2014.08.056.
[29] H.R.E.H. Bouchekara, M.A. Abido, M. Boucherma, Optimal power flow using
teaching-learning-based optimization technique, Electr. Power Syst. Res. 114
(2014) 49–59, http://dx.doi.org/10.1016/j.epsr.2014.03.032.
[30] H.R.E.H. Bouchekara, M.A. Abido, A.E. Chaib, R. Mehasni, Optimal power flow
using the league championship algorithm: a case study of the Algerian power
system, Energy Convers. Manag. 87 (2014) 58–70, http://dx.doi.org/10.1016/j.
enconman.2014.06.088.
[31] N. Daryani, M.T. Hagh, S. Teimourzadeh, Adaptive group search optimization
algorithm for multi-objective optimal power flow problem, Appl. Soft Comput.
38 (2016) 1012–1024, http://dx.doi.org/10.1016/j.asoc.2015.10.057.
[32] M. AlRashidi, M. El-Hawary, Applications of computational intelligence
techniques for solving the revived optimal power flow problem, Electr.
Power Syst. Res. 79 (2009) 694–702, http://dx.doi.org/10.1016/j.epsr.2008.10.
004.
[33] S. Frank, I. Steponavice, Optimal power flow: a bibliographic survey II. Non-
deterministic and hybrid methods, Energy Syst. (2012) 259–289, http://dx.doi.
org/10.1007/s12667-012-0057-x.
[34] A. Kaveh, V.R. Mahdavi, Colliding bodies optimization: a novel meta-
heuristic method, Comput. Struct. 139 (2014) 18–27, http://dx.doi.org/10.
1016/j.compstruc.2014.04.005.
[35] A. Kaveh, M. Ilchi Ghazaan, Enhanced colliding bodies optimization for design
problems with continuous and discrete variables, Adv. Eng. Softw. 77 (2014)
66–75, http://dx.doi.org/10.1016/j.advengsoft.2014.08.003.
[36] M. Ghasemi, S. Ghavidel, S. Rahmani, A. Roosta, H. Falah, A novel hybrid
algorithm of imperialist competitive algorithm and teaching learning algo-
rithm for optimal power flow problem with non-smooth cost functions, Eng.
Appl. Artif. Intell. 29 (2014) 54–69, http://dx.doi.org/10.1016/j.engappai.2013.
11.003.
[37] M. Balasubbareddy, S. Sivanagaraju, C.V. Suresh, Multi-objective optimization
in the presence of practical constraints using non-dominated sorting hybrid
cuckoo search algorithm, Eng. Sci. Technol. Int. J. 18 (2015) 603–615, http://dx.
doi.org/10.1016/j.jestch.2015.04.005.
[38] B.A. Robbins, H. Zhu, A.D. Dominguez-Garcia, Optimal tap settings for volt-
age regulation transformers in distribution networks, North Am. Power Symp.
(2013) 1–6, http://dx.doi.org/10.1109/NAPS.2013.6666885.
[39] H.R.E.-H. Bouchekara, M.A. Abido, Optimal power flow using differential search
algorithm, Electr. Power Components Syst. 42 (2014) 1683–1699, http://dx.doi.
org/10.1080/15325008.2014.949912.
[40] T. Niknam, M.R. Narimani, R. Azizipanah-abarghooee, A new hybrid algo-
rithm for optimal power flow considering prohibited zones and valve point
effect, Energy Convers. Manag. 58 (2012) 197–206, http://dx.doi.org/10.1016/
j.enconman.2012.01.017.
[41] G. Xiong, D. Shi, X. Duan, Multi-strategy ensemble biogeography-based opti-
mization for economic dispatch problems, Appl. Energy 111 (2013) 801–811,
http://dx.doi.org/10.1016/j.apenergy.2013.04.095.
[42] P. Kessel, H. Glavitsch, Estimating the voltage stability of a power system, IEEE
Trans. Power Deliv. 1 (1986) 346–354, http://dx.doi.org/10.1109/TPWRD.1986.
4308013.
[43] H. Hardiansyah, A modified particle swarm optimization technique for eco-
nomic load dispatch with valve-point effect, Int. J. Intell. Syst. Appl. 5 (2013)
32–41, http://dx.doi.org/10.5815/ijisa.2013.07.05.
[44] A. Kaveh, V.R. Mahdavi, Colliding Bodies Optimization method for optimum
design of truss structures with continuous variables, Adv. Eng. Softw. 70 (2014)
1–12, http://dx.doi.org/10.1016/j.advengsoft.2014.01.002.
[45] A. Kaveh, R. Mahdavi, Colliding Bodies Optimization Extensions and Applica-
tions, Springer, 2015.
[46] R.D. Zimmerman, C.E. Murillo-Sanchez, R.J. Thomas, MATPOWER: steady-state
operations, planning, and analysis tools for power systems research and edu-
cation, IEEE Trans. Power Syst. 26 (2011) 12–19, http://dx.doi.org/10.1109/
TPWRS.2010.2051168.
[47] R.D. Zimmerman, C.E. Murillo-Sánchez & Deqiang (David) Gan, MATPOWER,
http://www.pserc.cornell.edu/matpower/#docs (n.d.).
[48] 2014 OPF PROBLEMS. https://www.uni-due.de/ieee-wgmho/competition2014
(n.d.).
[49] J. Derrac, S. García, D. Molina, F. Herrera, A practical tutorial on the use of non-
parametric statistical tests as a methodology for comparing evolutionary and
swarm intelligence algorithms, Swarm Evol. Comput. 1 (2011) 3–18, http://dx.
doi.org/10.1016/j.swevo.2011.02.002.
[50] U. Can, B. Alatas, Physics based metaheuristic algorithms for global optimiza-
tion, Am. J. Inf. Sci. Comput. Eng. 1 (2015) 94–106.