You are on page 1of 36

Accepted Manuscript

Title: Optimal reactive power dispatch by improved


GSA-based algorithm with the novel strategies to handle
constraints

Author: Gonggui Chen Lilan Liu Zhizhong Zhang Shanwai


Huang

PII: S1568-4946(16)30577-4
DOI: http://dx.doi.org/doi:10.1016/j.asoc.2016.11.008
Reference: ASOC 3898

To appear in: Applied Soft Computing

Received date: 5-3-2016


Revised date: 7-10-2016
Accepted date: 8-11-2016

Please cite this article as: Gonggui Chen, Lilan Liu, Zhizhong Zhang, Shanwai
Huang, Optimal reactive power dispatch by improved GSA-based algorithm
with the novel strategies to handle constraints, Applied Soft Computing Journal
http://dx.doi.org/10.1016/j.asoc.2016.11.008

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Optimal reactive power dispatch by improved GSA-based algorithm
with the novel strategies to handle constraints

Gonggui Chen a,b, Lilan Liu a, Zhizhong Zhang c, Shanwai Huang a


a Key Laboratory of Industrial Internet of Things and Networked Control, Ministry of Education, Chongqing University of Posts and
Telecommunications, Chongqing 400065, China
b Research Center on Complex Power System Analysis and Control, Chongqing University of Posts and Telecommunications,
Chongqing 400065, China
c Key Laboratory of Communication Network and Testing Technology, Chongqing University of Posts and Telecommunications,
Chongqing 400065, China

Graphical abstract

Initial population B New initial population

Order based on fitness


Concept of OL
Form new population

A
The first
Condition & Selection
The second
Condition & Selection
The third
Condition & Selection
vid (k  1)  Rand  vid (k )  aid (k ) C
vid (k  1)  Rand  vid (k )  aid (k )
New population
 c1  Rand  ( Pibest
d
 X id (k ))
c2  Rand  ( Pgbest
d
 X id (k ))

Highlights:
 The novel constraints-handling method is proposed, which is named conditional selection strategies (CSS).

 The improved GSA-based algorithms including GSA-CSS and IGSA-CSS are proposed.

 Three test systems with different sizes are used to test the proposed algorithms.
 According to simulation results, IGSA-CSS outperforms basic gravitational search algorithm (GSA), particle
swarm optimization (PSO) and GSA-CSS; and GSA-CSS also outperforms GSA.
ABSTRACT

Optimal reactive power dispatch (ORPD) is well known as a complex mixed integer nonlinear optimization
problem where many constraints are required to handle. In the last decades, many artificial intelligence-based
optimization methods have been used to solve ORPD problem. But, these optimization methods lack an effective
means to handle constraints on state variables. Thus, in this paper, the novel and feasible conditional selection
strategies (CSS) are devised to handle constraints efficiently in the proposed improved gravitational search
algorithm (GSA-CSS). In addition, considering the weakness of GSA itself, the improved GSA-CSS (IGSA-CSS)
is presented which employs the memory property of particle swarm optimization (PSO) to enhance global
searching ability and utilizes the concept of opposition-based learning (OBL) for optimizing initial population.
The presented GSA-CSS and IGSA-CSS methods are applied to ORPD problem on IEEE14-bus, IEEE30-bus and
IEEE57-bus test systems for minimization of power transmission losses (Ploss) and voltage deviation (Vd),
respectively. The comparisons of simulation results reveal that IGSA-CSS provides better results and the
improvements of algorithm in this work are feasible and effective.

Keywords: Optimal reactive power dispatch (ORPD); Conditional selection strategies (CSS); Gravitational search
algorithm (GSA); Improved GSA (GSA-CSS); Improved GSA-CSS (IGSA-CSS)

1. Introduction

Optimal reactive power dispatch (ORPD), as one of a sub-problem of the optimal power flow (OPF)
calculation, plays a significant role in power system operation. The main purpose of ORPD is to identify optimal
setting of control variables for minimization of the given objective functions while satisfying a series of system
constraints over the entire dispatch period. Control variables contain discrete variables such as tap positions of
transformers and reactive compensation capacity and continuous variables like generator voltages. Besides,
system constraints are composed of two equality constraints and a set of inequality constraints. Hence, when
regarded as an optimization problem, ORPD is a complex mixed integer nonlinear optimization problem and has
gotten much attention over the last few decades [1]. From the point of view of improving economy and security of
power system, the main goals of optimal reactive power control are to minimize total active power transmission
losses and voltage deviation. Thus, minimizing the total active power transmission losses (Ploss) and the voltage
deviation (Vd) can be considered as two different optimal objective functions in the problem of ORPD.
Numerous classical methods with excellent convergence characteristic have been developed and applied to
solve ORPD problem in the past decades[2], such as interior point method (IPM) [3], linear programming (LP) [4]
and quadratic programming (QP) [5]. However, there exist some limitations and challenges for these methods to
handle non-differentiable objective functions and discrete variables in ORPD problem, which has been mentioned
in [6-9] and got a further discussion in [10]. Of late, a number of artificial intelligence algorithms appear to solve
ORPD problems efficiently without any requirements on the attributes of objective functions and variables, which
aim to search for the global optimal solution of problems; despite the fact that they cannot always guarantee the
global optimal solution, a suboptimal solution that is near the global optimal solution may also been found. There
are three categories for artificial intelligence algorithms listed as follows [11]: evolutionary computation
techniques like differential evolution (DE) [12], swarm intelligence algorithms such as particle swarm
optimization (PSO) [7] and physical heuristic methods like harmony search algorithm (HSA) [13]. The artificial
intelligence algorithms are numerous, but, it's worth noting that different method has its peculiar strength and
weakness. So, there is an increasing number of exertions to merge two and more algorithms or mechanisms for
solving ORPD problem effectively now. For example, the authors in [14] presented the Chaotic KHA (CKHA)
which considered the logistic map based krill herd algorithm (KHA), and the proposed CKHA approach has been
proved to yield better optimization efficacy over KHA for ORPD problem. In [15], combining the modified
imperialist competitive algorithm (ICA) and invasive weed optimization (IWO), the hybrid MICA-IWO method
can provide better results compared the original ICA and IWO when being used for handling ORPD problem. A
modified PSO which is introduced the aging leader and challenges mechanism on the basis of original PSO is
successfully applied for the solution of ORPD problem in [16]. In [17], an efficient local search method called
Nelder Mead (NM) simplex subroutine is introduced in the internal architecture of the Firefly Algorithm (FA)
algorithm to solve ORPD problem, and the results show that the hybrid method has better convergence
characteristics and robustness than the original version of FA. The applications of modified and hybrid methods
seem like a good mode of learning from one's strong points to make up others' weaknesses, however, it is still not
an easy task to guarantee the state variables within limits when using these methods to solve complex
optimization problems. Thus, for solving ORPD problem, what is greatly significant is not only to improve the
current methods but also to handle constraints on state variables well simultaneously.
Gravitational search algorithm (GSA) proposed by Esmat Rashedi etc. in 2009 is a physical heuristic
optimization algorithm [18]. This method gradually stands out and gets widely applied to solve many optimization
problems depending on the easy-to-understand and effective searching characteristic that is inspired by the
thought of Newton's law [19-21]. Based on the above description, the improvement for basic GSA need to be
taken into consideration and an effective constraint-handling technique is also required when GSA is applied. As
things stand today, the modifications for GSA are prevailing, such as, the modified GSA-Kepler algorithm in [22]
and the novel hybrid Genetic Algorithm–Gravitational Search Algorithm (hGA–GSA) in [23]. But, there is no
literature which does not use penalty function method [15, 24] to handle constraints in ORPD problem though the
setting and adjustment of penalty coefficients cost a lot of time. In this paper, a new improved algorithm
(GSA-CSS) is proposed for solving ORPD problem, which focuses on a new constraint-handling method in the
search process without using conventional penalty function. The proposed novel constraint-handling technique—
conditional selection strategies (CSS) guide the search towards the feasible region to handle the inequality
constraints on state variables depending on the reasonable selections under three different possible conditions.
CSS is not required to cost a lot of time to set penalty coefficients and adjust them repeatedly like penalty function
method, which can be completed along with the iterations of algorithms. GSA merges CSS to settle the difficulty
of conventional GSA in handling the inequality constraints on state variables. Then, GSA-CSS seems to be a good
technique to solve optimization problem, but GSA itself is a memory-less algorithm which cannot record the
optimal solution. As for the obvious defect, particle swarm optimization (PSO) [25] is considered to be used to
make up the shortage. PSO is known for numerous absorbing aspects, such as simple thought, convenient
implementation, high efficiency, powerful search ability etc., but, its most prominent feature is the memory
property that aims at recording global optimum of every generation and individual optimum for all agents during
the search process. Thus, the memory property of PSO is incorporated in GSA-CSS method to solve ORPD
problem, which fills in the gap of GSA. Moreover, to further improve the optimization performance, a concept of
opposition-based learning (OBL) [26] presented by Tizhoosh is considered as a mechanism for population
initialization. The main idea behind OBL is the consideration of an estimate and its corresponding opposite
estimate in order to achieve a better approximation for the current candidate solution. Ref. [27] reviews a lot of
applications of OBL based on optimization methods from 2005 to 2012. In this paper, the proposed new method is
named IGSA-CSS, which is combined memory property of PSO and introduced the concept of OBL on the basis
of GSA-CSS.
In this work, CSS are presented as the novel constraint-handling technique to handle the inequality constraints
on state variables, which are applied in the proposed GSA-CSS and IGSA-CSS algorithms. IGSA-CSS is
considered to combine the memory property of PSO and a concept of OBL based on GSA-CSS. Finally, the
proposed GSA-CSS and IGSA-CSS are studied to solve ORPD problem considering two different objectives
respectively on IEEE14-bus, IEEE30-bus and IEEE57-bus test systems. Simulation results reveal the proposed
IGSA-CSS method is more feasible and effective than the compared algorithms.
This rest of the paper is organized as follows. Section 2 introduces the formulation of ORPD problem. GSA is
described in detail in Section 3. Section 4 presents the calculation process of GSA-CSS algorithm for ORPD
problem, and especially, the details of CSS are stated in this section. Section 5 outlines the memory property of
PSO and the concept of OBL. Section 6 presents the solution technique using IGSA-CSS method for ORPD
problem. Some simulation experiments are shown in Section 7 and Section 8 gives the conclusions.

2. Mathematical problem formulation

The ORPD problem is formulated a complex nonlinear constrained optimization problem which can be
expressed as follows:

min f1  min Ploss &min f 2  min Vd (1)

s.t.
G (c, s )  0
H (c, s )  0
(2)
c  [cmin , cmax ]
s  [ smin , smax ]
where Eq. (1) represents two different objective functions in this paper for minimization of the total active power
transmission losses (Ploss) and voltage deviation (Vd), respectively; "s.t." is the abbreviation of "subject to"; Eq. (2)
are the definitions for system constraints and variables; G(c,s) and H(c,s) respectively represent the equality
constraints and inequality constraints of system; c and s denote the vector of control variables and state variables,
respectively. c and s are expressed in this paper as below:

cT  [Vg1 , Vgi , ,VgNPV , T1 , Ti , , TNT , Qc1, , Qci , , QcNC ] (3)

sT  [Vl1 , ,Vli , ,VlNPQ , Q g1 , , Q gi , , QgNPV , Smn1 , , Smni , , Smn NP ] (4)

where NPV , NT, NC, NPQ and NP are the number of PV buses, transformer branches, banks of capacitor or inductor,
PQ buses and network branches; Vgi is the voltages at the ith generator node; Ti is the tap positions of the ith
transformer; Qci denotes the reactive compensation capacity of the ith bank of capacitor or inductor; Vli is the
voltage at the ith load node; Qgi is the reactive power outputs at the generator i; Smn is the apparent power of the
transmission line between bus m and n; and "T" represents transposition.
The detailed discussion about objective functions and system constraints are shown later.

2.1. Objective function

2.1.1 Minimization of Ploss


For minimization of total active power transmission losses, the objective function in reactive power
optimization is expressed as follows:

 
min f1  min  Ploss   g k (Vi 2  V j2  2VV
i j cos  ij )  (5)
 kN P 
where gk is the conductance of the kth branch which connects the bus i and bus j ; Vi and Vj respectively denote the
voltage magnitude of the ith and jth bus; δij is the voltage phase between bus i and j.
2.1.2 Minimization of Vd
In power system, the voltage is one of the significant security indexes to evaluate power quality. As for ORPD
problem, minimization of total voltage deviations in all the load buses is one of ways to improve voltage quality.
The objective function can be expressed as:

 N PQ

min f 2  min Vd   Vi  V REF  (6)
 i 1 
where Vi is the bus voltage at the ith load node; VREF which is equal to 1.0 p.u. represents the desired voltage
magnitude; and Vd denotes the total deviations of bus voltages and the desired voltage magnitude.

2.2. System constraints

The abovementioned objective functions are subject to the system constraints which contain the equality and
inequality constraints.

2.2.1. Equality constraints


There are two equality constraints describing the active and reactive power balances as follows:
ni
Pgi  Pli  Vi V j (Gij cos  ij  Bij sin  ij )  0 in (7)
j 1

ni
Qgi  Qli  Vi V j (Gij cos  ij  Bij sin  ij )  0 i  N PQ (8)
j 1

where ni is the number of the buses adjacent to the ith bus (including bus i); n is the number of total buses except
for slack bus; Pgi is the injected active power at bus i; Pli is the demanded active power at bus i; Qgi is the injected
reactive power at bus i; Qli is the demanded reactive power at bus i; Gij and Bij respectively represent the
conductance and susceptance between the ith bus and the jth bus, respectively.

2.2.2. Inequality constraints


According to the difference of variables' types, inequality constraints can be classified as inequality constraints
of control variables and inequality constraints of state variables. Besides, another reason of distinguishing the
inequality constraints based on variables' type lies in the fact that CSS technique used to handle inequality
constraints on state variables is one of the innovations in this paper.
i. Inequality constraints of control variables
• Generator voltages limits

Vgimin  Vgi  Vgimax , i  N PV (9)

• Tap positions of transformers limits

Ti min  Ti  Ti max , i  NT (10)

• Amount of reactive compensation limits

Qcimin  Qci  Qcimax , i  NC (11)

ii. Inequality constraints of state variables


• Voltages limits at PQ bus

Vlimin  Vli  Vlimax , i  N PQ (12)


• Reactive power outputs limits of generators

Qgimin  Qgi  Qgimax , i  N PV (13)

• Apparent power limits

Smni  Smni
max
, i  NP (14)

3. Overview of GSA

GSA, as a newly developed heuristic search algorithm, was proposed by Rashedi etc. based on the law of
gravity and law of motion [18]. In GSA, a group of agents are considered as a series of objects with masses which
are used to measure their performance. These objects attract each other by the gravity force, while this force
guides objects with lower masses towards the objects with heavier masses, which is a global movement towards
the position of object with the highest mass. The detailed description of GSA on how to solve problem is as
follows.
Assumed there are N agents distributed in space and the dimension of search space is D. Eq. (15) represents the
position of the ith agent.

X i  ( xi1 , , xid , xiD ) i  1, 2, , N (15)

where xid denotes the position of agent i in the dth dimension.


The position of every agent is updated at each iteration. And the velocity, position and acceleration of agent i at
the next iteration are computed as follows:

vid (k  1)  Rand  vid (k )  aid (k ) (16)

xid (k  1)  vid (k  1)  xid (k ) (17)

aid (k )  Fi d (k ) M i (k ) (18)
where vid(k), xid(k) and aid(k) are the velocity , position and acceleration of agent i at the kth iteration in the dth
dimension, respectively; Rand denotes a random number distributed on (0,1); Fid(k) is the total force acting on
agent i from the other agents; and Mi(k) represents the mass of the ith agent at iteration k which is computed
according to the current agents' fitness. The expressions of Fid(k) and Mi(k) are as follows:
N
Fi d (k )  
jkbest , j i
Rand j  Fijd (k ) (19)

M i (k )  M j (k )
Fijd (k )  G(k )   ( x dj (k )  xid (k )) (20)
Rij (k )  

G(k )  G0  e k kmax (21)


where kbest denotes the set of the first K agents with the best fitness value, which is a function and reduces with
iterations form the initial value K0; Fijd(k) is the force acting on agent i by agent j which is computed based on the
law of gravity; ε is a small constant used to avoid denominator equal to zero; Rij(k) represents the Euclidian
distance between agent i and agent j, which is defined as Rij(k)=‖xid(k), xjd(k)‖2; G(k) is the gravitational constant at
the kth iteration which is reduced from an initial value with iteration k.
mi (k )
M i (k )  (22)
 j 1 m j (k )
N

Fiti (k )  Fworst (k )
mi (k )  (23)
Fbest (k )  Fworst (k )

where Fiti(k) denotes the fitness value of the ith agent at iteration k; Fbest(k) and Fworst(k) respectively represent the
best and worst fitness value among the N agents at iteration k, which are defined as follows:
Fbest (k )  min Fiti (k )
i{1, , N }
(24)
Fworst (k )  max Fiti (k )
i{1, , N }

4. Implementation of GSA-CSS algorithm for ORPD problem

This section describes the procedures of GSA-CSS to solve ORPD problem in detail. Especially, conditional
selection strategies (CSS) will be introduced on how to handle the constraints on state variables when altering
each agent's position in search space in GSA. The main steps of GSA-CSS method for solving ORPD problem are
elaborated as follows.

4.1. Generate initial population

The initial population for ORPD problem can be shown by a matrix as Eq. (25).
 p11 p12 p1N 
p p22 p2 N 
P   21 (25)
 
 
 pD1 pD 2 pDN 
where N is the size of population, D is the dimension of control variables for ORPD problem.
Each column in matrix P represents an agent in GSA, which is also a potential solution for ORPD problem. The
N agents are created randomly within the feasible range while satisfying the constraints given in Eqs. (9)-(11), and
the jth variable in the dth dimension is generated based on Eq. (26).

p j ,d  pmin,d  Rand  ( pmax,d  pmin,d ) j [1, N ], d [1, D] (26)

4.2. Optimize individuals

After obtaining the initial population, load flow calculation can be performed to get many results including the
total real power losses, bus voltage amplitudes, constraints of state variables and so on. The initial individuals are
regard as the current own best at the initialization period. And according to the each agent's objective value, the
global best agent also can be picked out.
According to Eqs. (18)-(24) in GSA, update G, Fbest, Fworst and M, and compute F and a. Then, update velocity
and obtain new position for all agents by Eq. (16) and Eq. (17). At last, record corresponding objective values and
constraints of state variables for all agents by load flow calculation.

4.3. Select optimal individual and handle constraints using CSS

In Eq. (1) and Eq. (2), G(c,s) is satisfied because it's considered as the termination conditions when calculating
Jacobian matrix in Newton Raphson load flow calculation. The limits of control variables can be given within the
limits by Eq. (26) at the beginning, which is a self-restricted mode.
However, never can the limits for state variables adopt this mode. Thus, it's a key issue to handle constraints of
state variables. The most popular strategy for handling constraints at present is to use penalty function, such as
Refs. [24, 28, 29]. The penalty function can transform a constrained problem into an unconstrained one. In the use
of the penalty function, the solutions which violate constraints are punished and are regarded as the infeasible
solutions, thus protecting the feasible solutions during selection process. However, there exists a main weakness:
the penalty coefficients which are used to precisely size up the degree of penalization need to get a careful fine
tuning, which may cost much time.
In order to overcome this weakness, the novel and feasible conditional selection strategies (CSS) are devised
to handle constraints on state variables when we use GSA-CSS to solve ORPD problem. CSS give an instruction
for deciding the better individual to form a new population as the next generation in the selection process. But,
different instructions depend on three different conditions in CSS. No time is spent in tuning the penalty
coefficients so as to improve the efficiency of optimization. Meanwhile, compared to original GSA, no extra
parameters are required to set in GSA-CSS.
Xi(k) is the current position of ith agent at kth iteration, Oi(k) is the recorded best position of ith agent until the
kth iteration, and f is the fitness function. vio is a constraints evaluation function used to denote whether the agent
violate constraints. When the value of vio(Xi(k)) is equal to zero, the agent i at kth iteration is indicated within the
limits; but, if the value of vio(Xi(k)) is not equal to zero, the agent i at kth iteration is certified to be without the
limits, and the bigger the value of vio(Xi(k)) is, the greater of the degree of constraints violation is. Since the
objective functions of ORPD require minimization, the procedure of CSS is as follows:
i. When both vio(Oi(k)) and vio(Xi(k+1)) are equal to zero, f(Oi(k)) and f(Xi(k+1)) are compared. The individual
with smaller fitness function value will be selected as the best individual of agent i at iteration k+1.
ii. When either vio(Oi(k)) or vio(Xi(k+1)) is equal to zero, it's no doubt that the individual within the limits is
preferred to the other one to be the best individual at next generation.
iii. When neither vio(Oi(k)) nor vio(Xi(k+1)) is equal to zero, vio(Oi(k)) and vio(Xi(k+1)) are compared. The
individual with lower constraints violation value will be selected as the best individual of agent i at iteration
k+1.
Based on the three conditional selections, selecting the optimal individual and solving the constraints violation
problem are performed concurrently. In short, the situations and degrees of constraints violation are considered as
the conditions of selecting optimal individual, which guides the agents to move to the more feasible region for
obtaining the better solution.
In this paper, fitness function is calculated by Eq. (5) or Eq. (6). The constraints violation value can be
calculated by constraint evaluation function which is the sum function expressed as follows.
N PQ N PV NP
vio( X )   vioVli ( X )   vioQgj ( X )   vioSmnh ( X ) (27)
i 1 j 1 h 1

Where X is the position of the current individual; vioVl, vioQg, and vioSmn are expressed as Eq. (28) which are
used to calculate the constraints violation values of voltages at PQ bus, reactive power outputs of generators, and
apparent power of transmission line, respectively.

 Vl min  Vl ,Vl  Vl min  Qgmin  Qg , Qg  Qgmin


 
 S  S , S  S
max max
 
vioVl   Vl  Vl max ,Vl  Vl max vioQg   Qg  Qgmax , Qg  Qgmax vioS   (28)
   0, otherwise
 0, otherswise  0, otherswise
Finally, the position of the best agent at the last iteration is the global best solution.

5. Opposition-based learning mechanism and memory property of PSO

5.1. Opposition-based learning

The initial population is generated at random without a priori information about the solution in evolutionary
optimization algorithms. The initial solutions and CPU time are related to the initial positions of agents. If the
initial agent is not far away from the optimal solution, the convergence will be fast, however, it's highly possible
that the initial agent is far away from the optimal solution, let alone in worst case it's in the opposite location [26].
Thus, the opposition-based learning (OBL) proposed by Tizhoosh is employed in this paper for solving the
difficulty of distributing and optimizing initial population. The description and application of OBL are as follows.
Let x[A,B] be a real number. The opposite number is defined as follows:

x  A B  x (29)

Similarly, the application for higher dimensions is as follows.


Let X={x1,…, xi,…, xd} be a point in d-dimensional space with xi[0,1], if A and B are equal to 0 and 1,
respectively, the corresponding opposite point can be defined by its components expressed in Eq. (30).

xi  1  xi (30)
In initialization, X={x1,…, xi,…, xd} represents the position of an initial agent generated at random, and g is the
fitness function used to evaluate initial agent's fitness. According to the abovementioned definition, the position of
corresponding opposite agent can be expressed as below.

X  {x1 ,..., xi ,..., xd } (31)


N agents in initial population mean N corresponding opposite agents can be generated. Thus, there are 2N
agents in the space. Every agent has a fitness value based on function g, and we can order the 2N fitness values
from smallest to largest, then the first N agents are selected out to be the new population.

5.2. Memory property of PSO

Schools of fish and flocks of birds always can find foods, which attributes to the communication between
individuals and the memory ability for individual best direction and global best direction. In PSO algorithm, these
behaviors of animals are simulated, and the update of velocity is defined as follows:

vid (k  1)  w  vid (k )  c1  Rand  ( Pibest


d
 X id (k ))  c2  Rand  ( Pgbest
d
 X id (k )) (32)

where w is inertia weight; c1 and c2 are acceleration factors; Pdibest and Pdgd respectively denote the best position of
particle i and the best particle in swarm in the dth dimension at iteration k.

6. The proposed IGSA-CSS algorithm for ORPD problem

GSA-CSS described in section 4 can improve the efficiency of search to some extent, but there exists an
intrinsic defect for GSA itself: GSA only considering the current position when updating the position of agents
without taking the individual memorability into consideration. Besides, like the majority of stochastic intelligent
algorithms, GSA also has difficult in optimizing and selecting initial population. Given that GSA-CSS still owns
the abovementioned weaknesses, the OBL mechanism and memory property of PSO are introduced to optimize
the initial population and improve the search quality, respectively.
The steps of the proposed IGSA-CSS method for solving ORPD problem are as below.
Step1: Set the parameters of systems and algorithms.
Step2: Initialize the opposition-based population. Generate uniformly distributed random agent Pj, where j[1,N],
the corresponding opposite agent is Pj . P and OP are the initial and corresponding opposite population,
respectively, and OPP represents the collection of P and OP.
for i=1: 2N % N: Population size
FV(i)=g(OPP(i)); % FV: a matrix containing the fitness value of 2N agents
end
[Sort_FV, Index_FV]=sort(FV);
for i=1: N
NewP(:, i)= OPP(:, Index_FV(i));
end
NewP is the new initial population generated by OBL mechanism.
Step3: Compute the fitness of every agent, and record Pdjbest and Pdgd.
Step4: Update G, Fbest, Fworst and M of the population based on Eq. (21)-(24).
Step5: Compute F and a of every agent according to Eqs. (18)-(20).
Step6: Update the velocity and position according to Eq. (33) and Eq. (17).
vid (k  1)  Rand  vid (k )  aid (k )  c1  Rand  ( Pibest
d
 X id (k ))
(33)
c2  Rand  ( Pgbest
d
 X id (k ))
where c1 and c2 are acceleration factors which can balance the effect of gravitation and memorability information
for search.
Step7: Check stop criterion. Go to the next step if the number of iterations reaches the maximum of iteration,
otherwise apply CSS technique, update the Pdjbest and Pdgd and continue the Step 4.
Step8: Select the best solution as the global best solution which is the best solution for ORPD problem.
And the computational flow of IGSA-CSS algorithm for ORPD problem is shown in Fig. 1.

7. Case study

In order to verify the effectiveness and feasibility of the proposed CSS and the improvement for GSA, the
proposed GSA-CSS, IGSA-CSS and traditional GSA and PSO algorithms have been examined and tested in IEEE
14-bus, IEEE 30-bus and IEEE 57-bus test systems to solve the ORPD problem of power systems with the
objectives minimizing total active power transmission losses (Ploss) and voltage deviation (Vd) respectively. All the
code of the algorithms are written by MATLAB R2013b programming language and run on PC with Intel(R)
Core(TM)2 Duo CPU E7500 @ 2.93GHz with 2GB RAM. To show the advantages of the proposed CSS and the
improvement for GSA, simulation results obtained from IGSA-CSS and GSA-CSS algorithms are compared with
the basic GSA and PSO algorithms. In this paper, the compared GSA and PSO algorithms employ the penalty
function to handle the constraints on state variables.

7.1 Description of cases

7.1.1 IEEE14-bus system.

The detailed data of IEEE14-bus test system is taken from [30],and its single line diagram is shown in Fig. 2.
Table 2 lists the reactive power' limits of generators. This system contains 20 branches, 3 transformers which are
connected to the branches 4-7, 4-9 and 5-6 and 1 capacitor bank set at bus 9. In the 14 buses of this network, bus 1
is regarded as the slack bus, 2, 3, 6 and 8 are the PV buses and the rest 9 are the PQ buses.

7.1.2 IEEE30-bus system.

The IEEE30-bus test system is taken as the second test system, whose detailed data is given in [31]. Table 3
lists the reactive power' limits of generators. As shown in Fig. 3, this network consists of 41 branches, 4
transformers and 9 capacitor banks. The capacitor banks are set at buses 10, 12, 15, 17, 20, 21, 23, 24 and 29. The
4 transformers are respectively connected to branches 6-9, 6-10, 4-12, and 28-27. In the 30 buses of this system,
bus 1 is the slack bus, 2, 5, 8, 11 and 13 are taken as PV buses and the rest 24 are the PQ buses.

7.1.3 IEEE57-bus system.

The IEEE57-bus system is the third test system in the paper, as a larger scale system which can verify the wide
application of the proposed algorithms. And its detailed data is taken from [30]. As shown in Fig. 4, the network
has 80 branches, 17 transformers and 3 capacitor banks which are set at buses 8, 25 and 53. The 17 transformers
are connected to the branches 4-18, 4-18, 21-20, 24-25, 24-25, 24-26, 7-29, 34-32, 11-41, 15-45, 14-46, 10-51,
13-49, 11-43, 40-56, 39-57 and 9-55. This test system contains 57 buses. Bus 1 is the slack bus, 2, 3, 6, 8, 9 and 12
are taken as PV buses and the rest 50 are PQ buses. The modified reactive power' limits of generators are listed in
Table 4.

7.2 Parameter settings

After a lot of repeated trials, considering various factors, we find the parameter settings have an important
impact on performance of algorithms. For example, the smaller population size means a high computational
efficiency but it cannot guarantee the diversity of agents, and the larger population size can improve the diversity
of agents but the computational complexity will be increased. Therefore, by a carefully tuning, the parameter
settings of different algorithms are listed in Table 5.

7.2 Comparison of results

GSA, PSO, GSA-CSS and IGSA-CSS methods run 30 times in the abovementioned three test systems for
minimization Ploss and Vd,respectively. The comparisons of average convergence curves of Ploss are seen in Figs.
5-7. Figs. 8-10 show the compared algorithms' average convergence curves of Vd. The distributions of the
obtained results of the compared algorithms with two different objectives for the test systems are shown in Figs.
11-16, respectively, and the obtained optimal solutions, best objective values and average CPU times are seen in
Tables 6-8. Table 9 lists the statistical details in three different test systems for four algorithms, including the best
objective values, worst objective values, mean objective values and standard deviation for two different objectives.
What's more, to clarify, Figs. 5, 7, 8 and 10 contain two pictures respectively, because the applications of penalty
function in conventional PSO and GSA algorithms cause the much larger original values compared to those of
GSA-CSS and IGSA-CSS, which makes the convergence curves not easy to distinguish in the late stage of
optimization. The picture (b) is given as the partial amplification of picture (a).
According to the obtained results, the comparisons of all algorithms are analyzed as follows:
i. The quality of solution
According to Tables 6-9, it's obvious that both the minimum Ploss and Vd obtained by proposed IGSA-CSS
method are the best optimization values among all the compared algorithms in the three test systems with different
sizes. For example, in IEEE 30-bus test system, the value of Ploss (4.76601 MW) yielded IGSA-CSS is less by
0.24353, 0.14977 and 0.027 MW compared to GSA, PSO and GSA-CSS algorithms, respectively. Though these
values are small, they cannot be neglected owing to the numerous power plants all over the world as well as the
continuous operations for power net dispatch. Besides, based on Figs. 11-16, in 30 independent simulation
experiments, the range of results' distribution of IGSA-CSS is relatively smaller than the other compared methods,
which demonstrates the proposed IGSA-CSS has a better robustness; besides, in general, small standard deviation
of IGSA-CSS in Table 9 also can prove that. Furthermore, we even find that the worst values of IGSA-CSS get
close to the best values of compared algorithms to some extent. Thus, IGSA-CSS method yields higher quality
solutions.
ii. Computation efficiency
As seen in Tables 6-8, the CPU times of the compared algorithms are almost synchronized. The CPU time of
IGSA-CSS is a little bit more than that of GSA-CSS because the IGSA-CSS employs the OBL to optimize initial
population, and uses a more complex formula to update agents' velocity at every iteration. Besides, the application
of CSS rather than conventional penalty function method can avoid spending time in tuning the penalty
coefficients repeatedly, and this tuning time is not included in the CPU times, which contributes to increase
computation efficiency to some extent.
iii. The effectiveness of opposition-based learning
From Figs. 5-10, it's easy to find that all the initial values of GSA and PSO algorithms is bigger than those of
GSA-CSS and IGSA-CSS methods, which results from the tunings of penalty coefficients. Besides, the initial
values of IGSA-CSS are always lower than those of GSA-CSS, for example, in the simulation experiment of
minimizing Ploss for IEEE 30-bus test system (result is shown in Fig. 6), the initial value of GSA-CSS is over than
6 MW, and it's obviously bigger than that of IGSA-CSS, which is ascribed to the application of OBL for
optimizing the initial population.
iv.The effectiveness of CSS
What most reflects the effectiveness of the proposed constraint-handling method is the difference of
optimization effect between GSA and GSA-CSS methods. No matter Ploss minimization or Vd minimization for
three test systems with different sizes, all the optimal results yielded by GSA-CSS are better than those of GSA,
which directly reveals the good effect of the novel proposed constraint-handling method. Moreover, the
distribution ranges of results of GSA-CSS are smaller than those of GSA generally according to Figs. 11-16.
Hence, GSA-CSS has a better optimization effect than GSA, which proves the effectiveness of CSS. In a word,
compared with the frequently-used penalty function method, CSS neither requires nor tunes the penalty
coefficients and can guide the agents to the feasible range quickly.

8. Conclusion

This paper researches the minimization of active power transmission losses and minimization of the voltage
deviation in power system by optimal reactive power control. The proposed GSA-CSS and IGSA-CSS algorithms
have been successfully applied for solving the ORPD problem. By simulation experiments in IEEE 14-bus, IEEE
30-bus and IEEE 57-bus test systems, CSS is proved to be greatly effective, and OBL mechanism is also verified
to be a good selection to optimize initial population. The results demonstrate the effectiveness and potential of
proposed IGSA-CSS methods to solve ORPD problem and proposed CSS to handle constraints on state variables.
Especially compared to basic GSA and PSO algorithms, the improved GSA-based algorithm with the novel
strategies to handle constraints (IGSA-CSS) is much more effective and can obtain the optimal solution with
higher quality.

Acknowledgments

The authors would like to thank the editors and the reviewers for their constructive comments. This work was
supported by Chongqing University Innovation Team under Grant KJTD201312 and the National Natural Science
Foundation of China (Nos. 51207064 and 61263030).

References

[1] G.G. Chen, L.L. Liu, P.Z. Song, Y.W. Du, Chaotic improved PSO-based multi-objective optimization for minimization of
power losses and L index in power systems, Energy Conver. Manage. 86(2014) 560-584.
[2] B.S. Raoa, K. Vaisakh, Multi-objective adaptive clonal selection algorithm for solving optimal power flow considering
multi-type FACTS devices and load uncertainty, Appl. Soft Comput. 23(2014) 286-297.
[3] S. Granville, Optimal reactive dispatch through interior point methods, IEEE Trans. Power Syst. 9(1994) 136-146.
[4] D.S. Kirschen, H.P. Van Meeteren, MW/Voltage Control in a Linear Programming Based Optimal Power Flow, IEEE Trans.
Power Syst. 3(1998) 481-489.
[5] V.H. Quintana, M. Santos-Nieto, reactive power-dispatch by successive quadratic programming, IEEE Trans. Energy. Conv.
4(1989) 425-435.
[6] M. Basu, Multi-objective optimal reactive power dispatch using multi-objective differential evolution, Elec. Power Energy Syst.
82(2016) 213-224.
[7] B. Kanna, S.N. Singh, Towards reactive power dispatch within a wind farm using hybrid PSO, Elec. Power Energy Syst.
69(2015) 232-240.
[8] G.J. Xiong, Y.H. Li, J.F. Chen, D.Y. Shi, et al, Polyphyletic migration operator and orthogonal learning aided
biogeography-based optimization for dynamic economic dispatch with valve-point effects, Energy Conver. Manage. 80(2014)
457-468.
[9] A. Rajan, T. Malakar, Exchange market algorithm based optimum reactive power dispatch, Appl. Soft Comput. 43(2016)
320-336.
[10] G.A. Bakare, G.K. Venayagamoorthy, U.O. Aliyu, Reactive Power and Voltage Control of the Nigerian Grid System Using
Micro-Genetic Algorithm, in: IEEE Power Engineering Society General Meeting, 2005, pp. 1-7.
[11] M.H. Sulaiman, Z. Mustaffa, M.R. Mohamed, O. Aliman, Using the gray wolf optimizer for solving optimal reactive power,
Appl. Soft Comput. 32(2015) 286-292.
[12] A.A.A.E. Ela, M.A. Abido, S.R. Spea, Differential evolution algorithm for optimal reactive power dispatch, Electric Pow Syst
Res. 81(2011) 458-464.
[13] A.H. Khazali, M. Kalantar, Optimal reactive power dispatch based on harmony search algorithm, Elec. Power Energy Syst.
33(2011) 684-692.
[14] A. Mukherjee, V. Mukherjee, Chaotic krill herd algorithm for optimal reactive power dispatch considering FACTS devices,
Appl. Soft Comput. 44(2016) 163-190.
[15] M. Ghasemi, S. Ghavidel, M.M. Ghanbarian, A. Habibi, A new hybrid algorithm for optimal reactive power dispatch problem
with discrete and continuous control variables, Appl. Soft Comput. 22(2014) 126-140.
[16] R.P. Singha, V. Mukherjee, S.P. Ghoshal, Optimal reactive power dispatch by particle swarm optimization with an aging leader
and challengers, Appl. Soft Comput. 29(2015) 298-309.
[17] A. Rajan, T. Malakar, Optimal reactive power dispatch using hybrid Nelder–Mead simplex based firefly algorithm, Elec.
Power Energy Syst. 66(2015) 9-24.
[18] E. Rashedi, H. Nezamabadi-pour, S. Saryazdi, GSA A Gravitational Search Algorithm, Information Sciences. 179(2009)
2232-2248.
[19] S. Duman, U. Güvenç, Y. Sönmez, N. Yörükeren, Optimal power flow using gravitational search algorithm, Energy Conver.
Manage. 59(2012) 86-95.
[20] S. Duman, Y. Sönmez, U. Güvenç, N. Yörükeren, Optimal reactive power using a gravitational search algorithm, Generation,
Transmission & Distribution, IET. 6(2012) 563-576.
[21] C. Li, J. Zhou, Parameters identification of hydraulic turbine governing system using improved gravitational search algorithm,
Energy Conver. Manage. 52(2011) 374-381.
[22] S. Sarafrazi, H. Nezamabadi-pour, S.R. Seydnejad, A novel hybrid algorithm of GSA with Kepler algorithm for numerical
optimization, Journal of King Saud University–Computer and Information Sciences. 27(2015) 288-296.
[23] R.K. Khadanga, J.K. Satapathy, A new hybrid GA–GSA algorithm for tuning damping controller parameters for a unified
power flow controller, Elec. Power Energy Syst. 73(2015) 1060-1069.
[24] A. Ketabi, A. Alibabaee, R. Feuillet, Application of the ant colony search algorithm to reactive power pricing in an open
electricity market, Elec. Power Energy Syst. 32(2010) 622-628.
[25] J. Kennedy, R. Eberhart, Particle swarm optimization, in: IEEE International Conference on Meural Networks, 1995, pp.
1942-1948.
[26] H.R. Tizhoosh, Opposition-Based Learning: A New Scheme for Machine Intelligence, in: International Conference on
Computational Intelligence for Modelling, Control and Automation, and International Conference on Intelligent Agents, 2005,
pp. 695-701.
[27] N. Qingzheng Xu A, L. Wang, N. Wang, X. Hei, et al, A review of opposition-based learning from 2005 to 2012, Engineering
Applications of Artificial Intelligence. 29(2014) 1-2.
[28] Y. Zhang, Z. Ren, Optimal Reactive Power Dispatch Considering Costs of Adjusting the Control Devices, IEEE Trans. Power
Syst. 20(2005) 1349-1356.
[29] C.H. Dai, W.R. Chen, Y.F. Zhu, X.X. Zhang, Reactive power dispatch considering voltage stability with seeker optimization
algorithm, Electric Pow Syst Res. 79(2009) 1462-1471.
[30] R.D. Zimmerman, C.E. Murillo-Sánchez, D. Gan, MATPOWER: a matlab power system simulation package.
http://www.pserc.cornell.edu/matpower/, 2007 (accessed 16.02.23).
[31] G.G. Chen, L.L. Liu, Y.Y. Guo, S.W. Huang, Multi-objective enhanced PSO algorithm for optimizing power losses and
voltage deviation in power systems, COMPEL. 35(2015) 350-372.
Figures:

Start

Set parameters

Generate initial population


Get corresponding opposite population
Select out new initial population
Initialization

Evaluate fitness of agents


Record current Pdibest and Pdgd

Update G, Fbest, Fworst and M of the population

Apply CSS
Compute F and a for agents
Update Pdibest and Pdgd

k=k+1 Update the velocity and position

No
Is stopping criteria met ?

Yes

Select the global solution

Stop

Fig. 1. Computational flow of IGSA-CSS algorithm for ORPD problem


G G
G
3
2

1
4
5
G
8
G 7

11 10 9
6
12

13 14

Fig. 2. The single line diagram of IEEE14-bus test system.


G G G

1
2 5
7
G
3
4
6

13 9
12 G
10 28
G
11
16
27
14 17
15
29

21
23
18 20
19 30
22

24
25
26

Fig. 3. The single line diagram of IEEE30-bus test system.


G G G
5 4 3 2 1 16

45 15 17
G
6 18 19 20
14 13 12
G
21
46

47
44 48 50

26 24 49

23 22 38
39 57
37
25
40 56 41 11
36
27
30 35 42 43
33

28 31 32 34

7 29 52 53 54 55

8 9 10 51
G G

Fig. 4. The single line diagram of IEEE57-bus test system


Fig. 5. Average convergence curves of active power transmission losses for IEEE14-bus system
Fig. 6. Average convergence curves of active power transmission losses for IEEE30-bus system
Fig. 7. Average convergence curves of active power transmission losses for IEEE57-bus system
Fig. 8. Average convergence curves of total voltage deviation for IEEE14-bus system
Fig. 9. Average convergence curves of total voltage deviation for IEEE30-bus system
Fig. 10. Average convergence curves of total voltage deviation for IEEE57-bus system
Fig. 11. The results' distribution of active power transmission losses for IEEE14-bus system
Fig. 12. The results' distribution of active power transmission losses for IEEE30-bus system
Fig. 13. The results' distribution of active power transmission losses for IEEE57-bus system
Fig. 14. The results' distribution of total voltage deviation for IEEE14-bus system
Fig. 15. The results' distribution of total voltage deviation for IEEE30-bus system
Fig. 16. The results' distribution of total voltage deviation for IEEE57-bus system
(Note: Figs. 5, 7, 8 and 10 are composed of picture (a) and picture (b). And the reason has been given in this revised
manuscript.)
Tables:

Table 1 System data of test systems


System data IEEE14-bus system IEEE30-bus system IEEE57-bus system
Active power of generators (MW) 272.39 289.23 1278.66
Reactive power of generators (MVAr) 82.44 139.10 321.08
Active power demands (MW) 259.00 283.40 1250.80
Reactive power demands (MVAr) 73.50 126.20 336.40
Active power losses (MW) 13.393 5.832 27.864
Reactive power losses (MVAr) 54.54 30.23 121.67

Table 2 Limits of reactive power of generators in IEEE14-bus test system


Bus number Qgimax (MVAr) Qgimin (MVAr)
1 10 0
2 50 -40
3 40 0
6 24 -6
8 24 -6

Table 3 Limits of reactive power of generators in IEEE30-bus test system


Bus number Qimax (MVAr) Qimin (MVAr)
1 200 -20
2 100 -20
5 80 -15
8 60 -15
11 50 -10
13 60 -15

Table 4 Limits of reactive power of generators in IEEE57-bus test system


Bus number Qimax (MVAr) Qimin (MVAr)
1 200 -140
2 130 -17
3 120 -10
6 55 -8
8 200 -170
9 70 -3
12 240 -150

Table 5 Parameter settings of algorithms


Parameter GSA PSO GSA-CSS IGSA-CSS
Population size: N 30 30 30 30
Maximum iterations: kmax 1000 1000 1000 1000
Penalty coefficients: min/max 10/100 10/100 - -
Inertia weight: wmin/wmax - 0.4/0.9 - -
Initial gravitational constant: G0 100 - 100 100
Constant:  20 - 20 20
Acceleration factors: c1/c2 - 2/2 - 2/2
Table 6 Comparisons of simulation results for IEEE14-bus test system with minimizing Ploss and Vd
Control minimization of Ploss (MW) Minimization of Vd (p.u.)
Variables GSA PSO GSA-CSS IGSA-CSS GSA PSO GSA-CSS IGSA-CSS
Vg1 (p.u.) 1.100000 1.100000 1.100000 1.10000 1.061589 1.061683 1.060821 1.060879
Vg2 (p.u.) 1.076398 1.077022 1.077774 1.076578 1.035651 1.042381 1.041352 1.040856
Vg3 (p.u.) 1.052355 1.046782 1.042583 1.046787 0.990180 1.013994 1.012131 1.011222
Vg6 (p.u.) 1.008185 1.020621 1.014102 1.062305 1.024779 1.023954 1.026130 1.016776
Vg8 (p.u.) 1.049006 1.071699 1.041648 1.097861 1.0309559 1.018293 1.031092 1.035129
T1 1.04 1.02 1.06 1.02 1.04 1.10 1.02 1.04
T2 1.02 1.00 0.94 0.94 0.94 0.90 0.92 0.9
T3 1.00 1.04 1.04 1.00 0.96 0.90 0.92 0.92
Qc1 (p.u.) 0.035 0.000 0.030 0.050 0.030 0.050 0.045 0.050
BOV 12.64782 12.46588 12.50985 12.39706 0.06727 0.08808 0.03829 0.03390
CPU time (s) 120.2959 124.3201 123.1497 123.8657 136.7222 138.0274 135.8793 136.8736
BOV: best objective value including minimum power losses and minimum voltage deviation
Table 7 Comparisons of simulation results for IEEE30-bus test system with minimizing Ploss and Vd
Control minimization of power losses (Ploss) Minimization of voltage deviation (Vd)
Variables GSA PSO GSA-CSS IGSA-CSS GSA PSO GSA-CSS IGSA-CSS
Vg1 (p.u.) 1.085932 1.074271 1.080987 1.081281 1.038617 1.016542 1.021714 1.008481
Vg2 (p.u.) 1.077857 1.065482 1.071832 1.072177 1.030072 1.015542 1.021245 1.005722
Vg5 (p.u.) 1.043181 1.043299 1.049583 1.050142 1.015308 1.016575 1.020442 1.019090
Vg8 (p.u.) 1.051025 1.043186 1.049744 1.050234 1.009714 1.011846 1.007064 1.010263
Vg11 (p.u.) 1.071938 1.030075 1.087238 1.100000 1.015652 1.035743 1.022921 1.018422
Vg13 (p.u.) 1.021988 1.076433 1.073330 1.068826 1.027194 1.014986 1.007145 1.007997
T1 0.980 1.100 1.040 1.080 0.990 1.042 0.984 1.034
T2 1.000 0.900 0.964 0.902 0.980 0.900 0.928 0.900
T3 1.020 1.100 1.020 0.990 0.986 0.962 0.968 0.984
T4 0.980 1.000 0.972 0.976 0.958 0.948 0.962 0.978
Qc1 (p.u.) 0.0300 0.0500 0.0255 0.0000 0.0255 0.0000 0.0245 0.0500
Qc2 (p.u.) 0.0350 0.0000 0.0335 0.0000 0.0235 0.0000 0.0250 0.0500
Qc3 (p.u.) 0.0300 0.0500 0.0315 0.0380 0.0245 0.0000 0.0365 0.0500
Qc4 (p.u.) 0.0050 0.0500 0.0350 0.0490 0.0278 0.0000 0.0270 0.0000
Qc5 (p.u.) 0.0200 0.0000 0.0260 0.0395 0.0390 0.0500 0.0440 0.0500
Qc6 (p.u.) 0.0400 0.0500 0.0300 0.0500 0.0255 0.0500 0.0195 0.0500
Qc7 (p.u.) 0.0300 0.0500 0.0350 0.0275 0.0260 0.0500 0.0355 0.0500
Qc8 (p.u.) 0.0250 0.0500 0.0360 0.0500 0.0430 0.0500 0.0290 0.0500
Qc9 (p.u.) 0.0500 0.0250 0.0300 0.0240 0.0185 0.0000 0.0285 0.0495
BOV 5.00954 4.91578 4.79301 4.76601 0.17241 0.10462 0.12394 0.08968
CPU time (s) 149.7679 150.9430 151.9015 151.7146 180.1832 1832.6396 179.5372 179.7369
BOV: best objective value including minimum power losses and minimum voltage deviation

Table 8 Comparisons of simulation results for IEEE57-bus test system with minimizing Ploss and Vd
Control minimization of power losses (Ploss) Minimization of voltage deviation (Vd)
Variables GSA PSO GSA-CSS IGSA-CSS GSA PSO GSA-CSS IGSA-CSS
Vg1 (p.u.) 1.100000 1.100000 1.100000 1.100000 1.100000 1.100000 1.008490 1.014306
Vg2 (p.u.) 1.100000 1.100000 1.100000 1.097588 1.100000 1.100000 0.998396 1.022312
Vg3 (p.u.) 1.089813 1.100000 1.090813 1.083212 1.073797 1.100000 1.005008 1.022256
Vg6 (p.u.) 1.084215 1.100000 1.084225 1.075434 1.042273 1.100000 1.005364 1.000151
Vg8 (p.u.) 1.100000 1.100000 1.100000 1.094606 1.052395 1.100000 1.038139 1.010769
Vg9 (p.u.) 1.084676 1.100000 1.084676 1.077254 1.045513 1.100000 1.022633 1.005348
Vg12 (p.u.) 1.080065 1.100000 1.080065 1.073357 1.046829 1.100000 1.022720 1.012448
T1 1.10 1.10 1.10 0.90 1.01 1.10 1.10 0.90
T2 1.01 1.10 1.01 1.10 1.01 0.90 0.99 1.04
T3 1.10 1.10 1.10 1.09 1.03 1.10 0.98 0.98
T4 1.10 0.90 1.10 1.10 0.98 1.02 1.00 1.10
T5 0.97 1.10 0.97 0.98 0.98 1.10 0.95 0.98
T6 1.10 0.98 1.10 1.00 1.02 1.10 1.01 1.00
T7 1.10 0.99 1.10 0.98 1.00 1.03 0.99 0.99
T8 0.90 0.94 0.92 0.96 1.00 0.93 0.93 0.92
T9 0.90 1.10 0.90 0.90 0.98 1.05 0.90 0.90
T10 0.98 1.01 0.97 0.97 1.02 0.96 0.90 0.93
T11 0.97 1.02 0.97 0.97 1.00 1.10 0.97 0.96
T12 0.98 1.10 0.98 0.97 0.99 1.10 1.00 1.00
T13 0.94 1.02 0.94 0.96 1.01 1.03 0.91 0.90
T14 1.09 1.10 1.09 0.97 0.99 1.10 0.98 0.96
T15 1.03 0.99 1.02 0.99 1.02 0.90 0.98 1.00
T16 1.10 0.97 1.10 0.95 0.98 0.90 0.93 0.90
T17 1.10 1.10 1.10 0.98 1.00 1.10 0.99 0.97
Qc1 (p.u.) 0.000 0.200 0.000 0.000 0.080 0.000 0.105 0.000
Qc2 (p.u.) 0.156 0.180 0.156 0.174 0.108 0.180 0.090 0.168
Qc3 (p.u.) 0.150 0.180 0.156 0.120 0.078 0.180 0.108 0.180
BOV 22.76482 24.22017 22.69912 22.27179 1.35090 1.77537 0.66286 0.60347
CPU time (s) 253.080 255.977 255.373 256.272 354.7317 357.1274 354.6700 355.1832
BOV: best objective value including minimum power losses and minimum voltage deviation
Table 9 Statistical details
Test BOV (p.u.) WOV (p.u.) MOV (p.u.) Standard Deviation
Algorithm
system Ploss Vd Ploss Vd Ploss Vd Ploss Vd
GSA 12.64782 0.06727 14.36926 0.30376 13.21897 0.17910 0.520 0.066
PSO 12.46588 0.08808 13.67714 0.27049 12.78373 0.18294 0.380 0.0603
IEEE14-bus
GSA-CSS 12.50985 0.03829 13.53354 0.23449 12.97324 0.12422 0.305 0.061
IGSA-CSS 12.39706 0.03390 12.90281 0.09056 12.46443 0.04583 0.094 0.017
GSA 5.00954 0.17241 5.91261 0.26975 5.39037 0.22439 0.260 0.027
PSO 4.91578 0.10462 5.51813 0.28832 5.08901 0.17280 0.199 0.054
IEEE30-bus
GSA-CSS 4.79301 0.12394 4.89061 0.14797 4.82697 0.13517 0.028 0.006
IGSA-CSS 4.76601 0.08968 4.79140 0.11672 4.77271 0.09827 0.005 0.007
GSA 22.76482 1.35090 27.13467 2.60296 24.44780 1.89318 1.204 0.399
PSO 24.22017 1.77537 26.05227 2.20909 24.83563 2.02555 0.415 0.101
IEEE57-bus
GSA-CSS 22.69912 0.66286 25.56537 0.89999 24.20598 0.76067 0.665 0.055
IGSA-CSS 22.27179 0.60347 23.98512 0.87729 23.02431 0.72697 0.421 0.083
BOV: best objective value; WOV: worst objective value; MOV: mean objective value

You might also like