Multiobjective Optimal Power Flow Using a Fuzzy Based Grenade

Explosion Method

(1, 2,
H.R.E.H. Bouchekara *), A.E. Chaib(1), M.A. Abido (3)

(1) Constantine Electrical Engineering Laboratory, LEC, Department of Electrical Engineering,

University of Freres Mentouri Constantine, 25000 Constantine, Algeria.

(2) Laboratory of Electrical Engineering of Constantine, LGEC, Department of Electrical Engineering,

University of Freres Mentouri Constantine, 25000 Constantine, Algeria.

(3) Electrical Engineering Department, King Fahd University of Petroleum and Minerals, Dhahran

31261, Saudi Arabia

*Corresponding author: bouchekara.houssem@gmail.com, Tel: (+213)666605628, Fax: (+213)31908113

Abstract ─

The aim of this paper is to solve the Multiobjective Optimal Power Flow (MOPF) problem using

a new metaheuristic that is the Grenade Explosion Method (GEM). The MOPF problem is

formulated by assuming that the decision maker may have a fuzzy goal for each of the objective

functions. Six objectives are considered which are: the minimization of generation fuel cost, the

improvement of voltage profile, the enhancement of voltage stability, the reduction of emission and

the minimization of active and reactive transmission losses. The proposed approach has been tested

on the IEEE 30-bus test system. The obtained results show the effectiveness of the proposed

method.

Keywords- Multiobjective optimal power flow; Power system optimization; Grenade explosion

method, Fuzzy logic.

1
1 Introduction

Over the past half-century, Optimal Power Flow (OPF) has been one of the foremost research

works for the power system community. It has become the heart of economically efficient and

reliable independent system operator power markets [1].

The conventional power flow problem can be stated by specifying loads at PQ buses, generated

powers and voltage magnitudes at PV buses in addition to a complete topological description of the

network. The objective of such problem is to determine voltages and then to compute all the other

quantities like currents, line flows and losses [2]. However, in the OPF problem, the generated

powers have to be adjusted according to some criteria, for instance the minimum generating cost

and under some constraints like power flow constraints. Therefore, the main purpose of the OPF

problem is to give the optimal settings of a power system (generated powers, voltages at PV

buses,..) by optimizing an objective function (minimum cost, minimum loses,…) while satisfying

some equality and inequality constraints (power flow, operating limits of system components,

security constraints…) [3].

Even 50 years after the problem was first formulated, researchers worldwide are still thirst for

solving the OPF problem. Because it is complex economically, electrically and computationally [1].

The first methods that have been used to solve the OPF problem are traditional ones. However,

these methods have shown some limitations as reported in the literature [4]-[5]. Even tough, many

works are still conducted in the fled of traditional methods, progressively metaheuristics are

becoming a serious and a reliable alternative for solving the OPF problem. Some examples of

metaheuristics used to solve the OPF problem are: Evolutionary Programming (EP) [6], Genetic

Algorithm (GA) [7], Particle Swarm Optimization (PSO) [8], Tabu Search (TS) [9], Simulated

Annealing (SA) [10], Differential Evolution (DE) [11], Biogeography-based Optimization (BBO)

[12], Imperialist Competitive Algorithm (ICA) [13], League Championship Algorithm (LCA) [14],

Black-Hole-Based Optimization (BHBO) [15], Teaching-Learning-Based Optimization (TLBO)

2
[16], Differential Search Algorithm (DSA) [17], Colliding Bodies Optimization (CBO) [18],

Electromagnetism-like Mechanism (EM) [19] and Backtracking Search Optimization (BSA) [20]. A

review of many metaheuristics applied to solve the OPF problem is given in [4] and [5].

However, due to the varied formulations and objectives of the OPF problem, no algorithm is the

best in solving all OPF problems. Therefore, there is always a need for a new algorithm which can

solve some of the OPF problems efficiently. Furthermore, in recent years, the field of

metaheuristics has witnessed a true tsunami of novel methods [21]. These new metaheuristics are

very efficient since they are tested on hard benchmarks and compared with many other methods.

Therefore, this paper aims to apply a new metaheuristic, which has not received yet much attention

in the power systems community that is the Grenade Explosion Method (GEM) in order to solve the

OPF problem.

The GEM is new metaheuristic based on the mechanism of grenade explosion [22], [23]. Ahrari

et al., demonstrate in [22] and [23] the superiority of the GEM method compared with other popular

metaheuristics.

In this paper, the OPF problem is formulated as a multiobjective optimization problem. Six

objectives are investigated that are: the minimization of generation fuel cost, the improvement of

voltage profile, the enhancement of voltage stability, the reduction of emission and the

minimization of active and reactive transmission losses. The objective functions are replaced by

membership functions using fuzzification.

The remainder of this paper is organized as follows. First, the problem formulation is presented

in section 2. Then, the GEM method is presented in section 3. Next, the results after solving

different cases of OPF (single objective and multiobjective cases) using GEM are discussed in

section 4. Finally, the conclusions are drawn in the last section of this paper.

3
2 Problem Formulation

The OPF is a power flow problem which gives the optimal settings of the control variables for a

given settings of load by minimizing a predefined objective function such as the cost of active

power generation. The OPF considers the operating limits of the system and it can be formulated as

a nonlinear constrained optimization problem as follows:

Minimize f ( x , u)
¿ ¿ g ( x ,u )=0 (1)
¿ h ( x , u) ≤ 0

Where:
u: vector of independent variables or control variables.
x: vector of dependent variables or state variables.
f ( x , u ): objective function.
g ( x , u ) : set of equality constraints.
h ( x ,u ): set of inequality constraints.

In the Multiobjective Optimal Power Flow (MOPF) problem, instead of only one objective

function a vector of objective functions is optimized. Therefore, the MOPF can be formulated as

follows:

T
Minimize F ( x , u )= [ f 1 ( x , u ) , f 2 ( x , u ) , … , f k ( x ,u ) ]
¿ ¿ g ( x , u )=0 (2)
¿ h ( x ,u ) ≤ 0

where: F(x, u) is the vector of objective functions and k is the number of objective functions.

2.1 Design variables

2.1.1 Control Variables

These are the set of variables which can be modified to satisfy the load flow equations. The set

of control variables in the OPF problem formulation are:

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

VG: voltage magnitudes at PV buses.
T: tap settings of transformers.
4
 QC: shunt VAR compensation.
Hence, u can be expressed as:

u =[ PG ⋯ P G , V G ⋯V G ,Q C ⋯ QC ,T 1 ⋯ T NT ]
T
2 NG 1 NG 1 NC
(3)

where NG, NT and NC are the number of generators, the number of regulating transformers and

the number of VAR compensators, respectively.

2.1.2 State Variables

These are the set of variables which describe any unique state of the system. The set of state

variables for the OPF problem formulation are:

PG1: active power generation at slack bus.

VL: voltage magnitudes at PQ buses or load buses.
QG: reactive power output of all generator units.
Sl: transmission line loadings (or line flow).

Hence, x can be expressed as:

T
x =[ PG ,V L ⋯ V L ,QG ⋯ Q❑ , Sl ⋯ Sl ] N
(4)
1 1 NL 1 1 nl

where, NL, and nl are the number of load buses, and the number of transmission lines,

respectively.

2.2 Objective functions

Many objective functions can be found in the literature for the OPF problem. In this paper, six

objective functions are studied. These objectives are discussed in the following sections.

2.2.1 Objective function # 1: generation fuel cost

The generation fuel cost is usually expressed by a quadratic function as follows:

NG
C=∑ a i+ bi P Gi+ c i PGi2 (5)
i=1

5
 It is worth to mention here that the cost function given by (3) is convex. However, in some cases

the cost function can be non-convex for instance when multi-fuels options are used as in [16], [18].

2.2.2 Objective function # 2: voltage deviation

Bus voltage is one of the most important and significant indication of safety and service quality.

The voltage profile can be improved by minimizing the load bus voltage deviations (VD) from 1.0

p.u, which is given by:

NL
VD=∑ |V i−1.0| (6)
i=1

2.2.3 Objective function # 3: voltage stability index

Prediction of voltage instability is an issue of paramount importance. In [24] Kessel and Glavitch

have developed a voltage stability index referred to as Lmax which is defined based on local

indicators Lj and it is given by:

Lmax=max ( L j ) j=1 ,2 , … , NL (7)

where: Lj is the local indicator of bus j and it is represented as follows:

| |
NG
Vi
L j = 1−∑ H LG ji j=1 , 2 , … , NL (8)
i=1 Vj
where: H matrix is generated by the partial inversion of Ybus. More details can be found in [24].

The Lmax varies between 0 and 1 and the lower this indicator is the more the system is stable.

Hence, in order to enhance the voltage stability, Lmax has to be minimized.

2.2.4 Objective function # 4: emission

There is an increasing worldwide emphasizes on environmental pollution reduction and control

[25]. Therefore, it is desirable to adjust the OPF to account for emission which will drastically

affects the whole system operation [25] [26].

6
 The total ton/h emission (E) of the atmospheric pollutants such as sulfur oxides SO x and nitrogen

oxides NOx caused by fossil-fueled thermal units can be expressed by [26] [27]:

NG
E=∑ 10−2 ( α i + β i PGi + γ i P2G )+ ( ωi e ( )
μi PGi )
i
(9)
i=1

Where: αi, βi, γi, ωi and μi are coefficients of the ith generator emission characteristics.

2.2.5 Objective function # 5: active power transmission losses

The active power losses can be expressed by:

NB NB NB
PL=∑ Pi=∑ P Gi−∑ PDi (10)
i=1 i=1 i=1

where NB is the number of busses and PD is the active load demand.

2.2.6 Objective function # 6: reactive power transmission losses

The reactive power losses can be expressed by:

NB NB NB
QL=∑ Qi=∑ Q Gi−∑ QDi (11)
i=1 i=1 i=1

where QD is the reactive load demand.

2.3 Constraints

OPF constraints can be classified into equality and inequality constraints, as detailed in the

following sections.

2.3.1 Equality constraints

a) Real Power Constraints:

NB
PGi−P Di−V i ∑ V j [ Gij cos ( θij ) +B ij sin ( θij ) ] =0 (12)
j =i

b) Reactive Power Constraints:

7
 NB
QGi−Q Di−V i ∑ V j [ Gij sin ( θ ij ) −B ij cos ( θ ij ) ]=0 (13)
j =i

where θij =θ i−θ j, Gij and Bij are the elements of the admittance matrix ( Y ij =Gij + j B ij )

representing the conductance and susceptance between bus i and bus j, respectively.

2.3.2 Inequality constraints

a) Generator constraints

min
V G ≤ V G ≤V G ,i=1 , … , NG
max
(14)
i i i

min
PG ≤ PG ≤ PG ,i=1 , … , NG
max
(15)
i i i

min
QG ≤ QG ≤QG ,i=1 , … , NG
max
(16)
i i i

b) Transformer constraints

min
T i ≤T i ≤T i , i=1 , … , NT
max
(17)

c) Shunt VAR compensator constraints

Qmin
max
C ≤ Q C ≤ Q C , i=1 , … , N
(18)
i i i

d) Security constraints

V min max
L ≤ V L ≤ V L , i=1 , … , NL
(19)
i i i

Sl ≤ Slimax ,i=1 , … , nl (20)

i

2.3.3 Constraints handling

It is worth mentioning that control variables are self-constrained. The inequality constraints of

dependent variables which contain load bus voltage magnitude; real power generation output at

slack bus, reactive power generation output and line loading can be included into the objective

function as quadratic penalty terms. In these terms, a penalty factor multiplied with the square of the

disregard value of dependent variable is added to the objective function [16], [17].

8
3 Grenade Explosion Method (GEM)

3.1 Overview

The GEM is a novel metaheuristic for optimizing real-valued bounded black-box optimization

problems inspired by the mechanism of grenade exposition [22], [23]. In the GEM, once the

grenades explode, the resulting shrapnel hit objects that are located within a neighborhood radius

called Le. The damages caused by shrapnel on objects are calculated. The damage-per-shrapnel

value indicates the value of objects in that area. In order to cause more damage, the next grenade is

thrown in the location of the greatest damage that has been caused. The overall damage caused by

the hit is considered as the fitness of the solution at the object’s location.

Furthermore, the GEM has a unique feature, which is the concept of agent’s territory radius (R t)

[22], [23]. Each agent (a grenade here), does not allow other agents to come closer than a certain

distance that is Rt. Therefore, when several grenades expose in the search space, a high value of R t

guarantees that grenades are spared quit uniformly in the search space while a small value of R t

allows the grenades to search local regions together [22], [23].

3.2 Algorithm

The pseudo code of the GEM is given in Algorithm 1. The GEM starts by scaling all

independent variables within the interval [-1,1]. Then, the problem parameters like the number of

grenades (Ng) and the maximum number of iterations (Max_iterations) are selected and L e and Rt

are initialized. After that, Ng grenades, distant by R t from each other, are randomly generated in the

n-dimensional scaled space. These grenades are ranked in a descending order based on their fitness.

For each grenade Nq pieces of shrapnel are generated using the following expression:

X j={ X m + sign(r m )×|r m| × L e } , j=1 ,2 , … … N q

' p
(21)

9
 where: X mis the location of the grenade, r m is a uniformly distributed random number in [−1,1]

and p is a constant used to tune the intensity of the exploration. The value of p is updated using the

probability of territory search (Tw ) as follows:

{ }
Rt
log ⁡(
)
p=max 1 , n × ¿ (22)
log ⁡( T w )

While generating the Nq pieces of shrapnel, some produced shrapnel my collide to objects

outside the scaled feasible space and they have to be transported to the [-1,1]n interval.

Then, the damage caused by every piece of shrapnel around a grenade is computed. If the fitness

of the best generated point is better than the fitness of the current location of the grenade, the

position of the grenade is updated and the grenade moves to the location of the best point.

To increase the global search ability, L e and Rt are adjusted during the iterations. High values of

these parameters are necessary to cover the whole search space in initial iterations, however, they

have to be reduced over iterations with taking fitness value into account. The territory radius is

updated as follows:

R t−initial
Rt = (23)
¿¿

where Rrd represents the ratio of the value of Rt in the first iteration to its value in the last

iteration and it has to be set before the algorithm starts.

Likewise, Le is decreased over iterations as follows:

m 1−m
Le =( L¿ ¿ e−initial) (R¿¿ t ) , 0 ≤ m≤ 1¿ ¿ (24)

where m is calculated using the following expression:

m=mmax − ( Iterationnumber
Max −iterations )
(mmax −mmin ) (25)

10
 In order to save global search ability, the rate of decrease of L e during the iterations is slower

than the one of Rt [22], [23]. Once Le and Rt are adjusted the value of p has to be updated

accordingly. Finally, the termination criterion used in the GEM is the number of iterations i.e., if the

number of iterations exceeds the maximum value the algorithm stops.

Algorithm 1: the pseudo code of the grenade explosion method [22].

1 Scale the independent variables’ ranges to [−1,1].

2 Select problem parameters : (Ng, Nq, Le-initial, Rt-initial, Rrd, Max_iterations, mmax, mmin, Tw)
3 Let (Le=Le-initial, Rt=Rt-initial, Iteration_ number=1)
4 Generate Ng grenades in random locations in an n-dimension space ⃗
X i ∈¿ which are at a distance of
at least Rt apart from each other.
5 While Iteration number ≤ Max iterations.
6 Arrange the grenades based on their fitness (the best grenade is the first).
7 Let: i=1.
8 While Nq accepted points are generated around ith grenade.
9 Generate a point ( X ' ) around ith grenade according to Eq. (21).
1 If ( X ' ) is outside [−1,1]n, transport it into [−1,1]n.
0
1 If ¿) is a distance of at least R t apart from the location of grenades 1, 2, . . ., i−1,
1 then ¿) is accepted.
1 End while % Nq pieces of shrapnel are thrown.
2
1 Calculate the fitness of the new generated points around the ith grenade. If the fitness of the
3 best point is better than current location of the ith grenade, move the grenade to the location
of the best point
1 Let: i = i+1.
4
1 If Ng ≥ i then go to 8.
5
1 Reduce Rt according to Eq.(23).
6
1 Calculate exponent m according to Eq. (25)
7
1 Update Le according to Eq. (24).
8

11
 1 Update exponent p according to Eq. (22).
9
2 End While % Number of iterations
0

4 The proposed multiobjective optimization approach

A complete optimal solution that simultaneously optimizes all the objective functions of a

Multiobjective Optimization (MOO) problem usually does not exist especially when the objective

functions present a conflicting behavior like in MOPF. Therefore, the task is to find a Pareto

optimal solution. A point x0X is said to be a Pareto optimal solution to an optimization problem P

if there is no xX such that F(x)  F(x0) [28].

For the MOPF problem, considering the vague or fuzzy nature of human judgment, it is quite

logic to assume that the decision maker (DM) may have a fuzzy goal for each of the objective

functions. Therefore, objective functions can be replaced by fuzzy membership functions.

The following steps describe the proposed MOPF approach in this paper:

Step 1: Minimization and maximization of each objective function separately in order

to calculate the individual minimum and maximum values of each objective
function under constraints.
Step 2: Computation of the membership functions μ1 , μ 2 , … μk by taking into account
the calculated individual maximum and maximum values of each objective
function in step 1.
Step 3: Computation of the aggregation function μ D ( μ1 , μ2 , … μ k ).
Step 4: Maximization of μ D using the GEM method.
Step 5: If the DM is satisfied with the current results i.e., the values of the
membership functions, go to step 6. Otherwise, modify the aggregation
function then go to step 4.
Step 6: Stop and print optimal results.

The first step of the of the proposed approach consists of minimizing then maximizing each

objective function separately in order to compute the maximum and minimum values of each

objective function. These two values correspond to the best and worst values that can each objective
12
function reach, respectively. Then, using these values, fuzzy membership functions are calculated.

The membership function of the ith objective is defined by [28]:

{
min
1 fi≤fi
max
f i −f i min max
μi = max min
f i <f i<f i (26)
f i −f i
max
0 fi≥fi

where: f max
i and f min
i are the maximum and minimum values of the ith objective function,

respectively. Therefore, a membership function of a given objective is equal to 1 if this objective is

fully satisfied (Figure 1).

Figure 1: Linear fuzzy membership function for fuzzy goal.

Once the membership functions are calculated, they are aggregated in one objective function as

shown in Figure 2. The aggregation function can be seen as the satisfaction degree of the objective

defined by the user [29].

13
 Figure 2: Flow chart of fuzzy aggregation for the six objective functions.

There are many aggregation methods that can be used. In this paper four methods are

investigated.

The weighted sum method: The first aggregation function investigated in this paper is the most

evident one that is the weighted sum method where the membership functions are multiplied by

weights then summed to give an aggregation function as follows:

k
μd = ∑ ω i μi (27)
i=1

where: ω 1,ω 1 , ω 2 , … ω k are weighting factors or weights. Usually, the weights are greater than or

equal to 0 and their sum equals 1.

The exponential weighted method: The second aggregation function considered in this study is

the exponential weighted method. This method has been proposed due to the inability of the

weighted sum method to capture points on non-convex portions of the Pareto optimal surface

[30].Therefore, μ D can be expressed by:

k
μd =∑ (e p ω −1)e p μ
i i
(28)
i=1

where p=1 in this study.

14
 The minimum operator : For the third aggregation function, we consider the well-known fuzzy

decision of Belmann and Zadeh or the minimum operator as follows [29]:

μd =min ( μ1 , μ 2 , … μk ) (29)

The ε-constraint method: In many cases, all objective functions do not have the same

importance, for instance reducing the generation fuel cost is usually more important than reducing

losses or emission. In fact, the fuel cost minimization is the most important objective to achieve.

However, at the same time the remaining objectives have to be taken into consideration in a way or

another. In this case using a method like the ε-constraint method is very interesting. This method

(also called the e-constraint or trade-off method) optimizes one of the objective functions while

imposing constraints on the remaining objective functions. In other words, it minimizes one

objective function (the most important one) and at the same time it maintains acceptable levels for

the remaining objective functions [31]. Thus, the aggregation function can be expressed by:

μ1
μd = μ2 ≤ ε 2 (30)
⋮
μk ≤ ε k

where : { ε 2 , ε 3 , … , ε k } are the vector of constraints to be chosen by the DM or sometimes it can be

imposed by some regulations.

In order to circumvent the numerical problem posed by the constraints they are augmented in

into μd using the following expression:

μd =μd Augmented=μ 1−∑ λ j ( μ j−ε j )

2
(31)
j

where: j are penalty factors and j takes the value of all the violated constraints.

5 Application and results

The proposed GEM method has been applied to the IEEE 30-bus test system. This test system

has a total generation capacity of 900.2 MW( in matpower data the sum of PMax of generators is

900.2MW( in our case is 435MW), ; the Total active power demand is 283.400MW,) and its main
15
characteristics are given in Table 1. Detailed data about these test systems can be derived from [32].

Cost and emission coefficients used in this paper are given in Table 2.

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

System Characteristics Value Details
Buses 30 -
Branches 41 -
Generators 6 Buses: 1, 2, 5, 8, 11 and 13.
Shunts 9 Buses: 10, 12, 15, 17, 20, 21, 23, 24 and 29.
Transformers 4 Branches: 11, 12, 15 and 36.
Control variables 24 -
Table 2: Cost and emission coefficients for the IEEE 30-bus test system.
Bus a b c d β γ ω μ
1 0 2 0.00375 18 -5.554 6.49 2.00E-04 2.857
2 0 1.75 0.0175 16 -6.047 5.638 5.00E-04 3.333
5 0 1 0.0625 14 -5.094 4.586 1.00E-06 8
8 0 3.25 0.00834 12 -3.55 3.38 2.00E-03 2
11 0 3 0.025 13 -5.094 4.586 1.00E-06 8
13 0 3 0.025 13.5 -5.555 5.151 1.00E-05 6.667
It is worth to mention that, the developed software program is written in the commercial

MATLAB computing environment and applied on a 2.20 GHz i7 personal computer with 8.00 GB-

RAM using parallel processing to run the 30 different runs.

This paper investigates 12 cases. The first six cases are single objective OPF cases whilst the

remaining cases are MOPF cases.

The control parameters of the GEM method are given Table 3. These parameters have been

selected after several tests.

Table 3: control parameters of the GEM method.

Name Description Value
Le Neighborhood raduis 10
Max_iterations Maximum No. Of iterations 150
Mmin Exploitation ratio 0.3
Ng No. Of grenades 4
Nq No. Of shrapnel 50
Psin Weight of optimal search direction 0.1
Rrd Reduction of territory radius 100
Rt Territory radius 5

16
5.1 Single objective OPF

The single objective OPF cases are:

CASE 1: minimization of generation fuel cost.

CASE 2: improvement of voltage profile by the minimization of voltage deviation.

CASE 3: enhancement of the voltage stability by the minimization of voltage stability index.

CASE 4: reduction of emission.

CASE 5: minimization of active losses.

CASE 6: minimization of reactive losses.

The proposed GEM has been applied to solve CASE 1 through CASE 6 and the obtained

optimal settings are given in Table 4.

Table 4: Optimal settings of control variables for single objective OPF cases using GEM.

Name CASE 1 CASE 2 CASE 3 CASE 4 CASE 5 CASE 6

PG1 177.1995 176.6211 139.7908 63.9687 51.4565 51.3560
PG2 48.6683 23.0226 45.5509 67.4512 79.8979 79.9903
PG5 21.3889 41.6352 23.3529 49.9982 49.9960 49.9980
PG8 20.8814 24.4240 34.9950 34.9986 34.9991 35.0000
PG11 11.8876 13.9965 29.9416 29.9997 29.9973 29.9864
PG13 12.0001 12.0107 16.0002 39.9996 39.9395 39.9822
VG1 1.1000 1.0464 1.0992 1.0919 1.0895 1.0870
VG2 1.0803 1.0276 1.0810 1.0834 1.0828 1.0798
VG5 1.0542 1.0162 1.0670 1.0661 1.0640 1.0626
VG8 1.0636 1.0022 1.0717 1.0743 1.0718 1.0702
VG11 1.1000 1.0140 1.0932 1.0993 1.0987 1.0994
VG13 1.0998 1.0207 1.0924 1.0993 1.0996 1.0995
T11(6-9) 1.0296 1.0172 0.9985 0.9699 1.0179 0.9720
T12(6-10) 0.9006 0.9003 0.9011 0.9758 0.9095 0.9723
T15(4-12) 0.9731 0.9867 1.0056 0.9766 0.9715 0.9767
T36(28-27) 0.9557 0.9682 0.9583 0.9678 0.9619 0.9730
QC10 4.9963 0.9632 3.1034 3.0313 3.4362 4.9911
QC12 4.9931 0.0116 0.5164 4.0171 4.7518 4.9779
QC15 4.9931 2.3100 4.9002 3.7516 3.5556 4.9819
QC17 4.9977 0.0579 3.0580 4.9944 4.9976 4.9994
17
 QC20 4.3337 4.9897 0.0565 3.7166 3.7225 4.9842
QC21 4.9997 4.9920 1.6048 4.9989 4.8917 4.9873
QC23 2.6505 4.9956 3.4331 2.6582 3.2080 4.4496
QC24 4.9865 4.9859 4.9470 4.9530 4.9583 4.9912
QC29 2.2507 2.5850 4.7145 3.0087 2.1188 3.8284
C ($/h) 799.0463 840.6457 816.9095 943.6358 966.7473 967.1100
VD 1.9312 0.1163 1.8320 1.9504 1.9755 2.0020
Lmax 0.1264 0.1487 0.1257 0.1269 0.1265 0.1272
E (ton/h) 0.3665 0.3629 0.2802 0.2048 0.2072 0.2072
PL (MW) 8.6257 8.3101 6.2313 3.0160 2.8863 2.9129
QL (MVar) 36.7420 35.4534 28.1011 14.8163 16.3060 14.3674

In order to assess the effectiveness of the proposed method we have compared it to some other

well-known methods. The results of this comparison are displayed in Table 5. For more

comparison, and since CASE 1 has been widely investigated in the literature, the results found for

this case are compared to those found in the literature in Table 6. It appears that the GEM

outperforms many other optimization methods.

Table 5: Comparisons of GEM with other well-known methods using our codes.

Method GEM GA PSO ABC BBO

C ($/h) 799.0463 800.3108 799.9409 799.0563 799.0860
VD 0.1163 0.1580 0.1451 0.1073 0.1085
Lmax 0.1257 0.1261 0.1262 0.1256 0.1252
E (ton/h) 0.2048 0.2052 0.2050 0.2047 0.2048
PL (MW) 2.8863 3.1493 3.1437 2.8496 2.8732
QL (MVar) 14.3674 15.8595 15.0959 14.2155 14.3183

Table 6: Comparison of simulation results for CASE 1 with those found in the literature.
Method Cost Method description Reference
GEM 799.0463 Grenade Explosion Method
BBO 799.1116 Biogeography-Based Optimization [12]
LCA 799.1974 League Championship Algorithm [14]
DE 799.2891 Differential Evolution [11]
SA 799.45 Simulated Annealing [10]
BHBO 799.9217 Black-Hole-Based Optimization [15]
EADHDE 800.1579 Genetic Evolving Ant Direction HDE [33]
EADDE 800.2041 Evolving Ant Direction Differential Evolution [34]
PSO 800.41 Particle Swarm Optimization [8]
FPSO 800.72 Fuzzy Particle Swarm Optimization [35]
IGA 800.805 Improved Genetic Algorithms [36]
PSO 800.96 Particle Swarm Optimization [35]
18
 GAF 801.21 Fuzzy Genetic Algorithm [35]
ICA 801.843 Imperialist Competitive Algorithm [13]
EGA 802.06 Enhanced Genetic Algorithm [37]
TS 802.2900 Tabu Search [9]
MDE 802.376 Modified Differential Evolution Algorithm [38]
IEP 802.465 Improved Evolutionary Programming [39]
EP 802.62 Evolutionary Programming [6]
RGA 804.02 Refined Genetic Algorithm [7]
GM 804.853 Gradient Method [40]
GA 805.94 Genetic Algorithm [7]

Figure 3 shows the evolution of cost and penalty term versus iterations for CASE 1. In Figure 4

the evolution of the voltage deviation versus iterations for CASE 2 is shown. Furthermore, the

voltage profile obtained for CASE 2 is sketched in Figure 5. The evolution of voltage stability index

versus iterations for CASE 3 is given in Figure 6.

8
10
900 2
Cost
880 Penalty

860
Cost ($/h)

Penalty

840 1

820

800

780 0
1 36 71 106 142
Iterations

Figure 3: Evolution of the objective function versus iterations for CASE 1.

19
 1.4

1.2

VD (p.u.)
0.8

0.6

0.4

0.2

0
0 50 100 150
Iterations
Figure 4: Evolution of the objective function versus iterations for CASE 2.

1.1

1.05
Voltage Profile (p.u.)

0.95

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

Figure 5: Voltage profile for CASE 2.

0.16

0.155

0.15

0.145
Lmax

0.14

0.135

0.13

0.125
0 50 100 150
Iterations

Figure 6: Evolution of the objective function versus iterations for CASE 3.

20
 Before starting solving the MOPF cases, let us start by analyzing the results found when only

one objective is considered at the time. In Table 7, the fuzzy membership functions of all objectives

are calculated. We can see from this table that, when an objective function is optimized its

membership function equals to 1 i.e., the objective is fully achieved. However, the remaining

membership functions lie between 0 and 1. For instance, when the cost is minimized μ E=0.3061 and

μQL =0.4605 while μ Lmax =0.9790. These values indicate the nature and intensity of the conflict

between these objective functions. The same analysis can be made on the remaining objectives.

Furthermore, a statistical analysis has been made in order to compute the correlation coefficients

between the six objectives (Table 8). Note that the correlation coefficient (r) is a dimensionless

measure of the degree of linear association or dependence between two sets of data X and Y that

varies between [-1,1] where: r=0 indicates that there is no linear dependence between X and Y, r=1

indicates that there is a total dependence with X and Y varying in the same direction and r=-1

indicates that there is a total dependence with X and Y varying in the opposite direction [41].

Therefore, The closer the coefficient is to either −1 or 1, the stronger the correlation between the

sets of data. The correlation coefficients can be arranged, into a symmetrical correlation matrix,

where each element is the correlation coefficient of the respective column and row variables. Since

the matrix is symmetrical, we show only the upper triangular part of this matrix. From Table 8, it

appears for example that there is a strong correlation between cost and emission in opposite

directions (r=-0.8948). the coefficient of correlation computed between emission and active losses

(r=0.9938) shows that there is a strong correlation in the same direction between these two objectives.

The same analysis can be made on the remaining results.

Table 7: Membership functions for the six objectives.

C VD Lmax E PL QL
C 1 0.0376 0.9790 0.3061 0.4494 0.4605
VD 0.7617 1 0.3028 0.3215 0.4797 0.4916
Lmax 0.8977 0.0901 1 0.6762 0.6791 0.6689
E 0.1719 0.0273 0.9639 1 0.9876 0.9892

21
 PL 0.0395 0.0141 0.9762 0.9895 1 0.9533
QL 0.0374 0.0000 0.9544 0.9893 0.9975 1

Table 8: The matrix of correlation coefficients.

C VD Lmax E PL QL
C 1 0.3587 -0.2619 -0.8948 -0.9383 -0.9389
VD 1 -0.9914 -0.6056 -0.5681 -0.5608
Lmax 1 0.5517 0.5021 0.4913
E 1 0.9938 0.9902
PL 1 0.9974
QL 1

5.2 Multiobjective OPF

The six MOPF cases investigated in this paper are summarized below:

CASE 7: the weighted sum method is used as an aggregation function with

1
ω 1=ω 2=ω3 =ω 4=ω 5=ω 6=ω= .
6

CASE 8: in this case, the minimum operator is used to generate the aggregation function.

CASE 9: exponential weighted method is used here to generate the aggregation function with

ω 1=ω 2=ω3 =ω 4=ω 5=ω 6=ω=1.

CASE 10: the ε-constraint method is explored where the objective to be optimized is the cost

and ε 2=0.75 , ε 3=0.75 , ε 4=0.75 , ε 5=0.75 andε 6=0.75 are the constraints that are imposed on the

remaining objectives.

CASE 11: the ε-constraint method is used where the objective to be optimized is the cost and

ε 2=0.75 , ε 3=0.75 , ε 4=0.5 , ε 5=0.5 andε 6=0.5 are the constraints that are imposed on the remaining

objectives.

22
 CASE 12: the ε-constraint method is tested where the objective to be optimized is the cost and

ε 2=0.5 , ε 3=0.5 , ε 4 =0.5 , ε 5=0.5 andε 6=0.5 are the constraints that are imposed on the remaining

objectives.

In order to compare the relative efficiencies of the different MOO techniques applied to solve

MOPF problems, we use the Euclidean distance N, which is defined as follows:

{∑ [ }
k 1
2 2
N= μi −1 ] (32)
i=1

N gives an indication on how much a solution is close to the utopia point. Therefore, the lower

the value of N the better the MOO is [42].

The GEM has been applied to solve the abovementioned MOPF cases and the optimal results are

given in Table 9. The N index is given in the last row of Table 9.

From Table 9, we can make the following comments:

The weighted sum method gives excellent results as shown in CASE 7. Furthermore, it is

probably the simplest way to solve the MOPF problem. However, the weights associated with each

objective functions have to be carefully selected, which is a cumbersome task for the DM.

The minimum operator method applied in CASE 8 achieves good results (N=0.5690). However,

there is no control on the objectives. In other words, this method treats all objectives as they have

the same importance which not the case in MOPF problems.

The exponential weighted method used in CASE 9 gives bad results as shown by the higher

value of the Euclidian distance (N=1.8).

The ε-constraint method used in CASE 10, CASE 11 and CASE 12 is very interesting in solving

MOPF problems for many reasons. As previously mentioned, in power system planning,

minimizing the generating fuel cost is for sure the most important case however; other issues and/or

objectives have to be taken into consideration as additional constraints rather than to be included as

objectives. This is more true in complex systems where the DM does not know exactly the values of
23
objective functions and then he cannot tune or select weights like in weighted methods. However,

he can impose constraints as percentages of the objective functions even without knowing their

values.

Table 9: Optimal settings of control variables for MOPF cases.

Name CASE 7 CASE 8 CASE 9 CASE 10 CASE 11 CASE 12

PG1 93.8656 113.6484 198.4283 115.6002 160.1093 159.4414
PG2 58.9613 58.3347 49.4983 50.8268 44.9976 51.6691
PG5 38.7369 35.3086 15.0000 31.7100 23.2081 23.2037
PG8 34.9952 33.8173 10.0480 34.9778 30.1999 26.5259
PG11 29.9714 26.8159 10.1489 27.8367 16.2299 14.9584
PG13 31.8635 20.9409 12.0755 27.9397 16.6367 15.3136
VG1 1.0334 1.0732 1.0768 1.0856 1.0808 1.0928
VG2 1.0210 1.0595 1.0564 1.0712 1.0618 1.0732
VG5 0.9757 1.0331 1.0252 1.0440 1.0331 1.0442
VG8 0.9957 1.0412 1.0252 1.0515 1.0362 1.0581
VG11 1.0539 1.0613 1.0999 1.0130 1.0378 1.0745
VG13 0.9867 1.0132 1.0924 1.0223 1.0454 1.0642
T11(6-9) 0.9361 1.0270 0.9004 1.0054 1.0578 1.0499
T12(6-10) 0.9849 1.0268 0.9009 1.0288 0.9689 0.9509
T15(4-12) 1.0437 1.0100 0.9588 1.0388 1.0236 1.0365
T36(28-27) 1.0123 0.9204 1.0758 0.9274 0.9169 0.9973
QC10 1.1937 3.4534 2.6908 3.8701 4.3804 0.5522
QC12 3.4795 2.6089 1.3943 2.0659 0.6287 2.3136
QC15 2.8760 4.9907 1.8208 2.4456 4.8076 3.3500
QC17 4.2058 3.5272 2.9409 2.9075 4.1373 3.2765
QC20 3.9844 3.1345 4.9883 3.6315 1.0483 1.9456
QC21 4.9836 2.9383 2.5457 4.5462 2.7822 2.2040
QC23 3.7470 2.8899 2.7516 0.4454 2.2670 3.9495
QC24 3.1526 2.1449 0.0213 3.6411 0.9005 3.5588
QC29 4.7621 2.3354 2.7536 2.5171 1.5840 3.5527
C ($/h) 874.6042 842.2508 809.4609 840.1226 805.2611 802.5583
VD 0.5006 0.5829 1.2236 0.5877 0.5875 0.8264
Lmax 0.1588 0.1339 0.1527 0.1340 0.1339 0.1390
E (ton/h) 0.2209 0.2432 0.4378 0.2427 0.3212 0.3213
PL (MW) 4.9939 5.4659 11.7990 5.4912 7.9814 7.7120
QL (MVar) 23.5780 24.6318 45.5608 24.1269 33.7012 33.5910
N 0.56908 0.57735 1.82 0.56964 0.91163 0.99128

6 Conclusion

In this paper, a recently developed metaheuristic, which is the grenade explosion method, is used

to solve multiobjective optimal power flow problems. Six objective functions are considered

24
namely: the minimization of generation fuel cost, the improvement of voltage profile, the

enhancement of voltage stability, the reduction of emission and the minimization of active and

reactive transmission losses. The fuzzy decision making approach is utilized to handle the

multiobjective OPF problem. The correlation between different objectives is identified. Many

multiobjective approaches are investigated which are the weighted sum method, the exponential

weighted method, the minimum operator and the ε-constraint method. The obtained results show

the efficiency of the GEM method when solving the OPF problem for single objective and

multiobjective cases.

References

[1] M. Cain, R. O’neill, A. Castillo, History of optimal power flow and formulations, FERC Staff Tech.
Pap. (2012) 1–36.
[2] H. Glavitsch, R. Bacher, Optimal power flow algorithms, in: Anal. Control Syst. Tech. Electr. Power
Syst. Acad. Press Inc, 1991.
[3] S. Sivasubramani, K.S. Swarup, Multiagent based differential evolution approach to optimal power
flow, Appl. Soft Comput. J. 12 (2012) 735–740. doi:10.1016/j.asoc.2011.09.016.
[4] 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.
doi:10.1016/j.epsr.2008.10.004.
[5] S. Frank, I. Steponavice, Optimal power flow : a bibliographic survey II Non-deterministic and
hybrid methods, Energy Syst. (2012) 259–289. doi:10.1007/s12667-012-0057-x.
[6] J. Yuryevich, Evolutionary programming based optimal power flow algorithm, IEEE Trans. Power
Syst. 14 (1999) 1245–1250. doi:10.1109/59.801880.
[7] S.R. Paranjothi, K. Anburaja, Optimal Power Flow Using Refined Genetic Algorithm, Electr. Power
Components Syst. 30 (2002) 1055–1063. doi:10.1080/15325000290085343.
[8] M.A. Abido, Optimal power flow using particle swarm optimization, Int. J. Electr. Power Energy
Syst. 24 (2002) 563–571. doi:10.1016/S0142-0615(01)00067-9.
[9] M.A. Abido, Optimal Power Flow Using Tabu Search Algorithm, Electr. Power Components Syst.
(2002) 469–483.
[10] C. a. 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. doi:10.1016/S0142-0615(02)00020-0.
[11] 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. doi:10.1016/j.epsr.2009.12.018.
[12] A.B.P.K. Chattopadhyay, Application of biogeography-based optimisation to solve different optimal
power flow problems, 5 (2011) 70–80. doi:10.1049/iet-gtd.2010.0237.
[13] 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.
[14] 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. doi:10.1016/j.enconman.2014.06.088.
25
[15] H.R.E.H. Bouchekara, Optimal power flow using black-hole-based optimization approach, Appl.
Soft Comput. (2014). doi:10.1016/j.asoc.2014.08.056.
[16] 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.
doi:10.1016/j.epsr.2014.03.032.
[17] H.R.E.-H. Bouchekara, M.A. Abido, Optimal Power Flow Using Differential Search Algorithm,
Electr. Power Components Syst. 42 (2014) 1683–1699. doi:10.1080/15325008.2014.949912.
[18] H.R.E.H. Bouchekara, A.E. Chaib, M.A. Abido, R.A. El-Sehiemy, Optimal power flow using an
Improved Colliding Bodies Optimization algorithm, Appl. Soft Comput. 42 (2016) 119–131.
doi:10.1016/j.asoc.2016.01.041.
[19] H.R. El-Hana Bouchekara, M.A. Abido, A.E. Chaib, Optimal Power Flow Using an Improved
Electromagnetism-like Mechanism Method, Electr. Power Components Syst. 44 (2016) 434–449.
doi:10.1080/15325008.2015.1115919.
[20] A.E. Chaib, H.R.E.H. Bouchekara, R. Mehasni, M.A. Abido, Optimal power flow with emission and
non-smooth cost functions using backtracking search optimization algorithm, Int. J. Electr. Power
Energy Syst. 81 (2016) 64–77. doi:10.1016/j.ijepes.2016.02.004.
[21] S. Kenneth, Metaheuristics – the Metaphor Exposed, (2012) 1–20.
[22] A. Ahrari, A. a. Atai, Grenade Explosion Method - A novel tool for optimization of multimodal
functions, Appl. Soft Comput. J. 10 (2010) 1132–1140. doi:10.1016/j.asoc.2009.11.032.
[23] a. Ahrari, M. Shariat-Panahi, a. a. Atai, GEM: A novel evolutionary optimization method with
improved neighborhood search, Appl. Math. Comput. 210 (2009) 376–386.
doi:10.1016/j.amc.2009.01.009.
[24] P. Kessel, H. Glavitsch, Estimating the Voltage Stability of a Power System, IEEE Trans. Power
Deliv. 1 (1986) 346–354. doi:10.1109/TPWRD.1986.4308013.
[25] M. Behrangrad, H. Sugihara, T. Funaki, Effect of optimal spinning reserve requirement on system
pollution emission considering reserve supplying demand response in the electricity market, Appl.
Energy. 88 (2011) 2548–2558. doi:10.1016/j.apenergy.2011.01.034.
[26] M. a. Abido, Environmental/economic power dispatch using multiobjective evolutionary algorithms,
IEEE Trans. Power Syst. 18 (2003) 1529–1537. doi:10.1109/TPWRS.2003.818693.
[27] a. M. Jubril, O. a. Olaniyan, O. a. Komolafe, P.O. Ogunbona, Economic-emission dispatch problem:
A semi-definite programming approach, Appl. Energy. 134 (2014) 446–455.
doi:10.1016/j.apenergy.2014.08.024.
[28] Y. Collette, P. Siarry, Optimisation multiobjectif, 2002.
[29] M. Sakawa, Genetic Algorithms And Fuzzy Multiobjective Optimization, Springer, 2002.
[30] R.T. Marler, J.S. Arora, Survey of multi-objective optimization methods for engineering, Struct.
Multidiscip. Optim. 26 (2004) 369–395. doi:10.1007/s00158-003-0368-6.
[31] T.J. Engin, E. Sci, A Comparative Study of Multiobjective Optimization Methods in ¸ ok Objektifli
Optimizasyon Tekniklerinin Mukayeseli Bir, Chem. Eng. 25 (2001) 69–78.
[32] C.E.M.-S.& D. (David) G. Ray D. Zimmerman, MATPOWER,
Http://www.pserc.cornell.edu/matpower/#docs. (n.d.).
[33] K. Vaisakh, L.R. Srinivas, K. Meah, A genetic evolving ant direction DE for OPF with non-smooth
cost functions and statistical analysis, Energy. (2010) 3155–3171. doi:10.1007/s00202-012-0251-9.
[34] K. Vaisakh, L.R. Srinivas, Evolving ant direction differential evolution for OPF with non-smooth
cost functions, Eng. Appl. Artif. Intell. 24 (2011) 426–436. doi:10.1016/j.engappai.2010.10.019.
[35] R.-H. Liang, S.-R. Tsai, Y.-T. Chen, W.-T. Tseng, Optimal power flow by a fuzzy based hybrid
particle swarm optimization approach, Electr. Power Syst. Res. 81 (2011) 1466–1474.
doi:10.1016/j.epsr.2011.02.011.
26
[36] 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.
[37] A.G. Bakirtzis, P.N. Biskas, C.E. Zoumas, V. Petridis, Optimal power flow by enhanced genetic
algorithm, IEEE Trans. Power Syst. 17 (2002) 229–236. doi:10.1109/TPWRS.2002.1007886.
[38] S. Sayah, K. Zehar, Modified differential evolution algorithm for optimal power flow with non-
smooth cost functions, 49 (2008) 3036–3042. doi:10.1016/j.enconman.2008.06.014.
[39] W. Ongsakul, T. Tantimaporn, Optimal Power Flow by Improved Evolutionary Programming,
Electr. Power Components Syst. 34 (2006) 79–95. doi:10.1080/15325000691001458.
[40] K.Y. Lee, Y.M. Park, J.L. Ortiz, A United Approach to Optimal Real and Reactive Power Dispatch,
IEEE Trans. Power Appar. Syst. PAS-104 (1985) 1147–1153. doi:10.1109/TPAS.1985.323466.
[41] J.P. Marques De Sá, Applied statistics using SPSS, STATISTICA, MATLAB and R, 2007.
doi:10.1007/978-3-540-71972-4.
[42] H.R.E.H. Bouchekara, a. Kedous-Lebouc, J.P. Yonnet, C. Chillet, Multiobjective optimization of
AMR systems, Int. J. Refrig. 37 (2014) 63–71. doi:10.1016/j.ijrefrig.2013.09.009.

27