Electrical Power and Energy Systems 65 (2015) 375–384

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

An improved teaching–learning-based optimization algorithm using

Lévy mutation strategy for non-smooth optimal power flow
Mojtaba Ghasemi a, Sahand Ghavidel a,⇑, Mohsen Gitizadeh a, Ebrahim Akbari b
a
Department of Electronics and Electrical Engineering, Shiraz University of Technology, Shiraz, Iran
b
University of Isfahan, Isfahan, Iran

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

Article history: One of the major tools for power system operators is optimal power flow (OPF) which is an important tool
Received 17 September 2013 in both planning and operating stages, designed to optimize a certain objective over power network
Received in revised form 7 August 2014 variables under certain constraints. This article investigates the possibility of using recently emerged
Accepted 12 October 2014
evolutionary-based approach as a solution for the OPF problems which is based on a new teaching–
Available online 7 November 2014
learning-based optimization (TLBO) algorithm using Lévy mutation strategy for optimal settings of OPF
problem control variables. The performance of this approach is studied and evaluated on the standard
Keywords:
IEEE 30-bus and IEEE 57-bus test systems with different objective functions and is compared to methods
Optimal power flow (OPF)
Teaching–learning-based optimization
reported in the literature. At the end, the results which are extracted from implemented simulations
(TLBO) confirm Lévy mutation TLBO (LTLBO) as an effective solution for the OPF problem.
Lévy mutation Ó 2014 Elsevier Ltd. All rights reserved.

Introduction formulation optimal power flow [6]. From then, this topic has been
handled by many researchers. Generally, we can solve OPF problem
The problem of optimal power flow (OPF) has been in the focus via many traditional optimization methods such as linear program-
of wide attention over past years and has established its position as ming, non-linear programming, quadratic programming, Newton-
one of the main tools for optimal operation and planning of based techniques and interior point methods [7–12]. However,
modern power systems. Main objective of the OPF problem is to the impossibility of employing them in practical systems because
optimize chosen objective functions such as piecewise quadratic of nonlinear characteristics such as valve point effects prohibited
cost function, fuel cost with valve point effects and voltage profile operating zones and piecewise quadratic cost function remains to
improvement, by optimal adjusting the power system control be a major drawback of the mentioned approaches. With regard
variables and satisfying various system operating such as power to this, improvement of optimization methods, capable of address
flow equations and inequality constraints, simultaneously [1–4]. these shortcomings are becoming increasingly vital [5,13,14].
Nodal power balance equations and restrictions of all control or In recent decades, various population-based optimization tech-
state variables are examples of equality constraints and inequality niques have been applied to solve complex constrained optimiza-
constraints, respectively. The control variables include the tap tion problems which also include optimization problems in field
ratios of transformer, the generator real powers, the generator of power systems like economic dispatch, allocation, optimal reac-
bus voltages and the reactive power generations of VAR sources tive power flow and OPF [15]. Generally, achieving optimal or near
while state variables involve the generator reactive power outputs, optimal solution for a specific problem will require multiple trials
load bus voltages and flow of network lines. Accordingly, OPF as well as accurate adjustment of associated parameters. Some of
problem is considered a basic tool allowing electric utilities to the proposed population-based methods such as tabu search
characterize secure and cost effective operating conditions for an [16], genetic algorithm [17], improved genetic algorithm [18], par-
electric power system [5,6]. ticle swarm [19], differential evolution [20], simulated annealing
In general view, the OPF problem is described as a highly [21] and evolutionary programming [22] have proved to be suc-
constrained, large-scale nonlinear nonconvex optimization cessful in solving OPF problems.
problem. Dommel and Tinney were the first to introduce the The power system academic has made considerable efforts to
provide scientific community with simulation tools that cover dif-
⇑ Corresponding author. Tel./fax: +98 911 1480548. ferent aspects of power systems analysis. Matpower developed by
E-mail address: [email protected] (S. Ghavidel). Zimmerman et al. [23], Power Systems Analysis Toolbox (PSAT)

http://dx.doi.org/10.1016/j.ijepes.2014.10.027
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
376 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384

introduced by Milano [24] and utilized enhanced genetic algorithm Accordingly, the x vector can be illustrated as the following:
for solving OPF by Bakirtzis et al. are some examples of these
efforts. The mentioned approaches have been tested on IEEE 30- xT ¼ ½PG1 ; V L1 . . . V LNPQ ; Q G1 . . . Q GNG ; Sl1 . . . SlNTL  ð4Þ
bus system and the three areas IEEE RTS-96 which is a 73 bus, where NG defines the number of generators; NPQ and NTL depict the
120-branch system [19]. number of PQ buses and transmission lines, respectively. u is the
The presented paper offers a novel method for solving OPF vector of independent variables (control variables) consisting of:
problems based on recently developed heuristic optimization 1. Generation bus voltages VG.
called TLBO algorithm [3,25,26] with Lévy mutation operator. 2. Generator active power output PG at PV buses except at the
The mutation operator for TLBO algorithm is capable of performing slack bus PG1.
either local search or global search depending on the step size; 3. Transformer taps settings T.
especially in the early and middle stage of the evolution. The muta- 4. Shunt VAR compensation QC.
tion operator can enhance the exploration and improve the diver-
sity of the population. OPF problems – as will be thoroughly Therefore, u can be expressed as:
explained in Section ‘Problem formulation’ of the article – are for-
T
mulated as a nonlinear optimization problem with equality and u ¼ ½PG2 . . . PGNG ; V G1 . . . V GNG ; Q C1 . . . Q CNC ; T 1 . . . T NT  ð5Þ
inequality constraints in a power system. Objective functions in where NT and NC represent the number of tap regulating transform-
this problems are minimization of fuel cost such as quadratic cost ers and number of shunt VAR compensators, respectively.
function, piecewise quadratic cost function, cost function with
valve point effect and improvement of the voltage profile. Training Constraints
and learning procedure is the backbone idea of this method with
special attention to the effect of the influence of a teacher on the Equality constraints
output of learners in a class. Hence, grades of learners are In the terms below, g is the equality constraints, illustrating
considered as outputs, the teacher is generally considered as a typical load flow equations [1,2,5,6]:
highly learned person who shares his or her knowledge with the
learners and clearly, quality of teacher has direct effect on output X
NB
 
PGi  PDi  V i V j Gij cosðdi  dj Þ þ Bij sinðdi  dj Þ ¼ 0 ð6Þ
of learners. Additionally, interaction between the learners is a
j¼1
contributing factor in increasing their outcome. The practical effec-
X
NB
 
tiveness of this approach is tested and evaluated on the standard Q Gi  Q Di  V i V j Gij sinðdi  dj Þ  Bij cosðdi  dj Þ ¼ 0 ð7Þ
IEEE 30-bus and 57-bus test systems. The performance and poten- j¼1
tial capabilities of this approach are presented and the results are
presented in a comparison to above mentioned methods in this where Vi and Vj are the voltages of ith and jth bus, respectively, NB is
paper. the number of buses, PGi is the active power generation, QGi is the
The rest of the article is categorized in four sections as follows: reactive power generation, PDi is the active load demand, QDi is the
in Section ‘Problem formulation’, formulation of an optimal power reactive load demand, Gij, Bij and dij are the conductance, susceptance
flow is explained while Section ‘Teaching–learning-based and phase difference of voltages between bus i and bus j, respectively.
optimization using Lévy mutation operator (LTLBO)’ describes the
standard structure of the TLBO algorithm using Lévy mutation Inequality constraints
operator, Section ‘Numerical results of optimization 187 in h Is the inequality constraints that include:
performance comparison’ of this paper is dedicated to presenting
optimization results and comparison of the methods used to i. Generator related constraints: generation bus voltages,
solve the case studies of optimal power flow problems on IEEE active power outputs, and reactive power outputs are
30-bus and IEEE 57-bus systems and at the end, Section restricted by their lower and upper limits as:
‘Conclusions’ covers the conclusion of the implementation for the
offered method. V min max
Gi 6 V Gi 6 V Gi ; i ¼ 1; . . . ; NG
Pmin max
Gi 6 P Gi 6 P Gi ; i ¼ 1; . . . ; NG ð8Þ
Problem formulation
Q min
Gi 6 Q Gi 6 Q max
Gi ; i ¼ 1; . . . ; NG
In general view, the goal of a solution of optimal power where V min
and V max
are the minimum and maximum gener-
Gi Gi
flow problem is to optimize a selected objective function ator voltage of ith generating unit; Pmin and Pmax the mini-
Gi Gi
through optimal adjustment power system control parameters mum and maximum active power output of ith generating
while satisfying equality and inequality constraints at the same unit and Q min max
Gi and Q Gi are the minimum and maximum reac-
time. tive power output of ith generating unit.
The OPF problem can be mathematically formulated as follows:
Min Jðx; uÞ ð1Þ ii. Transformer limitations: transformer tap settings are
restricted by their lower and upper limits as:
Subject to : gðx; uÞ ¼ 0 ð2Þ T min
i 6 T i 6 T max
i ; i ¼ 1; . . . ; NT ð9Þ
hðx; uÞ 6 0 ð3Þ where T min
and
i T max
define minimum and maximum tap set-
i
tings limits of ith transformer.
In the above equation, J is the objective function to be minimized
and x is the vector of dependent variables (state vector) consisting iii. Shunt VAR compensator constraints: shunt VAR compensa-
of: tions are restricted by their limits as:
1. Generator active power output at slack bus PG1.
Q min max
Ci 6 Q Ci 6 Q Ci ; i ¼ 1; . . . ; NC ð10Þ
2. Load bus voltage VL.
3. Generator reactive power output QG. where Q min
and
Ci Q max
define minimum and maximum VAr
Ci
4. Transmission line loading (or line flow) Sl. injection limits of ith shunt compensator.
 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384 377

The value of TF can be either 1 or 2 which is again a heuristic

iv. Security constraints: include the constraints of voltages at step and decided randomly with equal probability as
load buses and transmission line loading as: T F ¼ round½1 þ randð0; 1Þf2  1g.
V min
Li 6 V Li 6 V max
Li ; i ¼ 1; . . . ; NPQ ð11Þ
Sli 6 Smax ; i ¼ 1; . . . ; NTL ð12Þ Learner phase
li
In this phase, increasing the knowledge of an individual (Xi) is
where V min
Li and V max
Li minimum and maximum load voltage of done by peer learning from an optional student (represented with
ith unit. Sli defines apparent power flow of ith branch. Smax li
Xii) and self-organized learning between the individual and other
defines maximum apparent power flow limit of ith branch. learners, where obviously, i is not equal to ii. According to knowl-
It is worth mentioning that the control variables are self-con- edge levels of these two students relative to each other, two states
strained. The hard inequalities of PG1, VL, QG, and Sl can be incorpo- can occur: if Xii is better than Xi, Xi will be moved towards Xii
rated in the objective function as quadratic penalty terms. (shown in Eq. (16)) and otherwise, it will be moved from Xii (shown
Therefore, the objective function can be augmented as follows in Eq. (17)). If student Xnew performs better by following (16) or
[2,27]: (17) Equations, he/she will be accepted into the population. The
algorithm will continue its iterations until reaching the maximum
2 X
NPQ
2 number of generations.
Jðx; uÞ ¼ F þ kP ðPG1  Plim
G1 Þ þ kV ðV Li  V lim
Li Þ
i¼1 X new ¼ X i þ rðX ii  X i Þ ð16Þ
X
NG
2 X
NTL
2 X new ¼ X i þ rðX i  X ii Þ ð17Þ
þ kQ ðQ Gi  Q lim
Gi Þ þ kS ðSli  Slim
li Þ ð13Þ
i¼1 i¼1 Besides, infeasible individuals must be appropriately handled to
where kP, kV, kQ and kS are defined as penalty factors and n is the determine whether one individual is better than another, when
number of objective functions. xlim is the limit value of the depen- applied to constrained optimization problems. The flow chart for
dent variable x and given as: the TLBO algorithm is given in Fig. 1.
8
< x;
> xmin 6 x 6 xmax Lévy mutation operator for LTLBO algorithm
xlim ¼ x ; x > xmax
max ð14Þ
>
: min
x ; x < xmin Lévy distribution [28] is stable and has probability density func-
tion that is analytically expressible. The Lévy distribution can be
expressed as below:
Teaching–learning-based optimization using Lévy mutation Z 1
1 a
operator (LTLBO) La;c ¼ ecq cosðqyÞdq ð18Þ
p 0

TLBO algorithm where y  R, c > 0 is the scaling factor, and 0 < a < 2 controls the
shape of the distribution. The Lévy mutation with c = 1 and
The TLBO algorithm is a newly introduced evolutionary algo- a = 0.8 for new learner phase of the LTLBO algorithm can be
rithm which is a population based technique, using a population described as:
of solutions to reach the global solution. The principle idea of TLBO X new ¼ X i þ Li ð0:8Þ ð19Þ
algorithm is the simulation of teaching–learning process in a tradi-
X new ¼ X ii þ Lii ð0:8Þ ð20Þ
tional classroom. In this optimization algorithm, the assumed pop-
ulation is a group of learners and parameters involved in objective where Li (0.8) indicates that the random number is generated anew
function of the optimization problem are offered to learners as for each value of i.
different subjects. At the end, learner’s grade is equivalent to the If Xii is better than Xi, Xi will be moved towards Xii (shown in Eq.
‘fitness’ value of the optimization problem. The best solution (best (19)) and otherwise, it will be moved from Xii (Eq. (20)). If student
value of objective function) in the entire population is considered Xnew performs better by following (19) or (20) Equations, he/she
as the teacher. The procedure of this method ends in two stages: will be accepted into the population.
(1) teacher phase, which is learning through teacher (2) learning
phase – which is learning through exchange of information Numerical results of optimization in performance comparison
between learners.
IEEE 30-bus test system
Teacher phase
In this phase a teacher attempts to increase the mean result of In this section, results of solving OPF problems by implementa-
the class in the taught subject. In the first step, the teaching role is tion of LTLBO algorithm, obtained by simulation runs will be pre-
assigned to the best individual (Xteacher) in the population, then the sented. In order to evaluate the performance and robustness of
algorithm improves other individuals (Xi) by shifting their posi- the proposed LTLBO algorithm based on Newtonian physical law
tions towards the position of the teacher Xteacher. The position of [6] of gravity and law of motion which is tested on standard IEEE
each individual is produced by taking into account the current 30-bus test system as shown in Fig. 2. Generator data, bus data
mean value of the individuals (Xmean) and represents the qualities and the minimum and maximum limits for the control variables
of all learners’ from the current generation. Eq. (15) simulates are presented in Appendix A.
how improvement in results of students is influenced by the differ- The IEEE 30-bus test system has six generators at the buses 1, 2,
ence between teacher’s knowledge and the qualities of all students. 5, 8, 11 and 13 and four transformers with off-nominal tap ratio at
X new ¼ X i þ rðX teacher  ðT F X mean ÞÞ ð15Þ lines 6–9, 6–10, 4–12 and 28–27. In addition, buses 10, 12, 15, 17,
20, 21, 23, 24 and 29 have been chosen as shunt VAR compensation
where TF represents a teaching factor that determines the value of buses [29]. The total system demand is 2.834 p.u. at 100 MVA base.
the mean to be changed, and r is a random number ranging from The maximum and minimum voltages of all load buses are consid-
0 to 1. ered to be 1.05–0.95 in p.u. The LTLBO algorithm has been applied
378 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384

Initialize number of population

Calculate the mean of each design

variable

Identify the best solution (teacher)

Modify solution based on best solution

X new = X i + r ( X teacher − (T F X mean ))

No Yes
Is new solution
Reject Accept
better than existing?

Select any two solutions

randomly
X i and X ii

Yes No
Is X i better than X ii ?

X new = X i + r ( X i − X ii ) X new = X i + r ( X ii − X i )

No Yes
Is new solution
Reject Accept
better than existing?

Is termination No
criteria satisfied?

Yes

Final value of solutions

Fig. 1. Flow chart showing the working of TLBO algorithm.

to solve the OPF problem for different cases with different objec- and The optimal settings of control variables, obtained by the pro-
tive functions. In each case, 30 test runs were performed for solv- posed LTLBO algorithm, are presented in Table 1 with minimum
ing the OPF problem using the LTLBO algorithm. fuel cost obtained from the proposed approach equal to
To demonstrate the effectiveness of the proposed algorithm, 799.4369 $/h. The steep convergence of total fuel cost is shown in
different cases with various objectives are considered as follow: Fig. 3. In order to assess the potential of the proposed algorithm,
a comparison between the results of fuel cost obtained by the pro-
Case 1: minimization of fuel cost posed LTLBO algorithm and those reported in the literature has
The objective function of this case is minimization of fuel cost been carried out with results shown in Table 2. The average cost
for all generators which can be formulated as follows: obtained from the LTLBO algorithm for this case is 799.7186 $/h
with a maximum cost of 800.2578 $/h.
X
NG X
NG
J¼ F i ðP Gi Þ ¼ ðai þ bi PGi þ ci P2Gi Þ ð21Þ
i¼1 i¼1
Case 2: improvement of voltage profile
where Fi and PGi are the fuel cost of the ith generator and the output Bus voltage is considered one of the most important security
of the ith generator, respectively. ai, bi and ci represent the cost and service indexes. Taking into account only a cost-based
coefficients of the ith generator and NG is the number of total objective in an OPF problem will result in a feasible solution with
generator. The values of cost coefficients are shown in Table A1 an undesirable voltage profile. So, in this case a twofold objective
 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384 379

29

28
Generator 27

Transformer 30
26 25

Load
23
24
Bus

15 19
18
17 20 21

14 16
22

12
13
10

11 9

1
3 4 6 8

2 5

Fig. 2. Single line diagram of IEEE 30-bus test system.

function is needed in order to minimize the fuel cost and improve Case 3: piecewise quadratic fuel cost functions
voltage profile by minimizing the load bus voltage deviations From practical point of view, in a power system many thermal
from 1.0 per unit. The objective function can be expressed as generating units may be supplied with several fuel sources like oil,
below [1]: coal or natural gas. The fuel cost functions of these units may be
dissevered as piecewise quadratic fuel cost functions for different
X
NG X
NPQ
2
fuel types [1]. The cost coefficients for these units are shown in
J¼ ðai þ bi PGi þ ci P2Gi Þ þ g jV i  1:0j þ kP ðPG1  Plim
G1 Þ Table A2. Clearly, the fuel cost coefficients of remaining single fuel
i¼1 i¼1 source generators have the same values as case 1 condition. The
X
NPQ
2 X
NG
2 fuel cost characteristics for the generating units connected at num-
þ kV ðV Li  V lim
Li Þ þ kQ ðQ Gi  Q lim
Gi Þ ber 1 and 2 buses are now represented by a piecewise quadratic
i¼1 i¼1
function to model different fuels given by:
X
NTL
2 8
þ kS ðSli  Slim
li Þ ð22Þ >
> ai1 þ bi1 PGi þ ci1 P 2Gi Pmin
Gi 6 P Gi 6 P Gi1
>
>
i¼1 <
FðPGi Þ ¼
ai2 þ bi2 PGi þ ci2 P 2Gi PGi1 6 PGi 6 PGi2
ð23Þ
>
> ...
where g is a suitable weighting factor in above equation which >
>
:
will be selected by the user. In this case the weighting factor is aik þ bik PGi þ cik P2Gi PGik1 6 P Gi 6 P max
Gi
selected as 100 [1,30], and proposed approach has been applied where aik, bik and cik represent cost coefficients of the ith generator
to search for the optimal solution of the problem. The resulted for fuel type k. Objective function can be described as:
optimal setting for control variables obtained by appliance of ! !
X
2 X
NG
the LTLBO algorithm is shown in Table 1. Additionally, compari- J¼ aik þ bik PGi þ cik P 2Gi þ ai þ bi PGi þ ci P2Gi
son of the resulted voltage profile in this case to the voltage pro- i¼1 i¼3
file of case 1 – as presented in Fig. 4 – shows a considerable X
NPQ
2 2
improvement of voltage profile in this case relative to previous þ kP ðPG1  P lim
G1 Þ þ kV ðV Li  V lim
Li Þ
case result, decreasing from 1.0829 p.u in case 1 to 0.0974 p.u i¼1
in case 2 which is equivalent to 91.01% reduction as a result of X
NG
2 X
NTL
2
employing LTLBO algorithm. The comparison results are gathered þ kQ ðQ Gi  Q lim
Gi Þ þ kS ðSli  Slim
li Þ ð24Þ
in Table 3. i¼1 i¼1
380 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384

Table 1 Table 2
Best control variables settings for different test cases. Comparison of the simulation results for case 1 ($/h).

Variables Case 1 Case 2 Case 3 Case 4 Case 5 Algorithms Min Average Max Time (sec)
PG1 (MW) 177.46 176.1 140 199.6 63.99 LTLBO 799.4369 799.7186 800.2578 22.14
PG2 (MW) 48.6837 48.8439 55.0000 20.0000 67.3975 TLBO 800.4212 800.8661 801.0586 24.27
PG5 (MW) 21.3146 21.6468 25.1536 22.1234 50.0000 NPSO [36] 800.6815 800.9024 801.37 20.45
PG8 (MW) 20.8867 22.3444 34.9909 25.0928 35.0000 Fuzzy-GA [37] 801.0554 801.627 802.1158 22.07
PG11 (MW) 11.8086 12.3067 17.7999 13.3536 30.0000 DE-PS [38] 800.1475 800.7962 801.2417 27.33
PG13 (MW) 12.0000 12.0000 17.3905 12.5664 40.0000 MOTLBO [39] 800.6797 NA NA NA
VG1 (p.u.) 1.1000 1.0393 1.0500 1.1000 1.1000 ABC [40] 800.66 800.8715 801.8674 NA
VG2 (p.u.) 1.0817 1.0230 1.0388 1.0784 1.0955 EGA-DQLF [41] 799.56 NA NA NA
VG5 (p.u.) 1.0509 1.0135 1.0118 1.0492 1.0775 ACO [30] 802.58 NA NA NA
VG8 (p.u.) 1.0555 1.0034 1.0211 1.0558 1.0847 MSFLA [1] 802.287 NA NA NA
VG11 (p.u.) 1.0826 1.0355 1.0961 1.0922 1.1000 PSO [29] 800.41 NA NA NA
VG13 (p.u.) 1.0574 0.9968 1.0679 1.0642 1.1000 Improved GA [18] 800.805 NA NA NA
T6–9 (p.u.) 1.0461 1.0518 1.0139 1.0431 1.0822 MDE [2] 802.376 802.382 802.404 23.25
T6–10 (p.u.) 0.9583 0.9000 0.9305 0.9599 0.9000 Gradient method [33] 804.853 NA NA 4.324
T4–12 (p.u.) 0.9996 0.9587 0.9883 0.9979 1.0033 EADDE [31] 800.2041 NA NA 3.32
T28–27 (p.u.) 0.9891 0.9625 0.9517 0.9897 0.9808 SFLA [42] 802.21 NA NA 20.75
QC10 (MVAR) 5.0000 5.0000 3.0000 1.0000 5.0000 PSO [42] 801.89 NA NA 20.19
QC12 (MVAR) 5.0000 2.0000 5.0000 0.0 5.0000 MPSO-SFLA [42] 801.75 NA NA 18.17
QC15 (MVAR) 5.0000 5.0000 4.0000 3.0000 5.0000 Enhanced GA [19] 802.06 NA 802.14 76
QC17 (MVAR) 5.0000 0.0 5.0000 5.0000 5.0000
QC20 (MVAR) 4.0000 5.0000 5.0000 5.0000 4.0000
QC21 (MVAR) 5.0000 5.0000 5.0000 5.0000 5.0000
QC23 (MVAR) 3.0000 5.0000 3.0000 4.0000 3.0000
QC24 (MVAR) 5.0000 5.0000 5.0000 5.0000 5.0000 1.1 Case 1
QC29 (MVAR) 3.0000 2.0000 2.0000 3.0000 2.0000 Case 2
Fuel cost ($/h) 799.4369 803.7431 647.4315 917.6259 943.4885 1.08
Emission (ton/h) 0.3672 0.3631 0.2835 0.4402 0.2047
Power loss (MW) 8.7558 9.8421 6.9347 9.3359 2.987
Voltage (p.u.)

1.06
VD (p.u.) 1.0829 0.0974 0.8896 1.0724 1.8212

1.04

1.02

814 1
TLBO
812 LTLBO
NPSO 0.98
810 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
fuzzy-GA
DE-PS Bus no.
Fuel cost ($/h)

808
Fig. 4. System voltage profile for case 2.
806

804

802
adding a sine component to cost curves of bus number 1 and 2
800 generators which simulates valve point loading effects on their
798 characteristics. The cost coefficients for these units are presented
0 10 20 30 40 50 60 70 80 90 100
Iteration in Table A3, obtained from [31] and fuel cost coefficient of other
generators remains intact and equal to the same values assumed
Fig. 3. Fuel cost variations of case 1. in case 1. Cost characteristics of generators 1 and 2 are defined
as follows:
   
 
F i ðPGi Þ ¼ ai þ bi PGi þ ci P2Gi þ di sin ei Pmin
Gi  P Gi 
The LTLBO algorithm is applied to this case with taking into account
the conditions of voltage magnitude upper limit set at bus 1 as 1.05 i ¼ 1 and 2: ð25Þ
and no shunt VAR compensation buses [1,29]. The obtained optimal
settings of control variables for this case study are shown in Table 1, where ai, bi, ci, di and ei are cost coefficients of the ith generating
confirming LTLBO algorithm as the most appropriate solution for unit.
minimization of fuel cost in the OPF problem. The best fuel cost
result calculated by LTLBO algorithm is compared to the results
obtained by other techniques in Table 4. Judging from Table 4, it Table 3
Comparison of the simulation results for case 2 (p.u.).
can be seen that the minimum fuel cost is 647.4315 $/h, with an
average cost of 647.4725 $/h and a maximum cost of 647.8638 $/ Algorithms Min Average Max Time (sec)
h which are less in comparison to the reported results in the liter- LTLBO 0.0974 0.0983 0.1006 20.17
ature. Also the convergence of algorithms for the OPF problem with TLBO 0.0986 0.0994 0.1015 19.63
minimum fuel cost is shown in Fig. 5. NPSO [36] 0.09815 0.1006 0.1025 19.61
Fuzzy-GA [37] 0.1052 0.1268 0.139 18.75
DE-PS [38] 0.0978 0.0997 0.1022 22.12
Case 4: quadratic cost curve with valve point loadings BBO [43] 0.102 0.1105 0.1207 13.23
DE [44] 0.1357 NA NA NA
In the final case on the IEEE 30-bus test system, we will include
PSO [29] 0.0891 NA NA NA
loading effect on generating units performance, this is done by
 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384 381

Table 4
940 LTLBO
Comparison of the simulation results for case 3 ($/h). TLBO
NPSO
Algorithms Min Average Max Time (sec)
935 fuzzy-GA
LTLBO 647.4315 647.4725 647.8638 22.78 DE-PS

Fuel cost ($/h)

TLBO 647.8125 647.8335 647.8415 24.37
930
NPSO [36] 647.907 647.9255 648.106 20.41
Fuzzy-GA [37] 648.2309 648.441 648.7682 21.95
DE-PS [38] 647.55 647.6029 647.9253 24.82 925
ABC [40] 649.0855 654.0784 659.7708 NA
BBO [43] 647.7437 647.7645 647.7928 11.94
920
DE [44] 650.8224 NA NA NA
MPSO-SFLA [42] 647.55 647.55 647.55 19.06
PSO [29] 647.69 647.73 647.87 NA 915
0 10 20 30 40 50 60 70 80 90 100
MDE [2] 647.846 648.356 650.664 37.05 Iteration

Fig. 6. Convergence of TLBO for case 4.

670
TLBO
LTLBO
665 NPSO accurate OPF solutions in this case study is reinstated by the
fuzzy-GA gathered results.
DE-PS
Fuel cost ($/h)

660
Case 5: minimization of fuel cost considering emission
655
The minimization of emission function for OPF problem can be
presented as the sum of all types of emission considered, such as
SOX, and NOX, with suitable pricing or weighting on each pollutant
650
emitted. In the present paper, two important types of emission
gases are taken into account. The emission function can be
645
0 10 20 30 40 50 60 70 80 90 100 expressed as below [1]:
Iteration
X
NG NG 
X 
Fig. 5. Convergence of TLBO for case 3. F E ðPG Þ ¼ F Ei ðPGi Þ ¼ ai þ bi P Gi þ ci P 2Gi þ ni expðki PGi Þ ð27Þ
i¼1 i¼1

where FEi represents the emission of the ith generator. ci, bi, ni and ki
Objective function can be defined as:
are the emission coefficients of ith generator, while ci (ton/h MW2),
!
X
2     bi (ton/h MW), ai (ton/h) are related to SOX, and ni (ton/h), ki (1/
J¼ ai þ bi PGi þ ci P 2Gi þ di sin ei Pmin
Gi  P Gi

 MW) are related to NOX, respectively [1].
i¼1
! The objective function for case 5 can be expressed as below:
X
NG
2
þ ai þ bi PGi þ ci P 2Gi þ kP ðPG1  Plim
G1 Þ X
NG

i¼3 J¼ ðai þ bi PGi þ ci P2Gi Þ

i¼1
X
NPQ
2 X
NG
2
þ kV ðV Li  V lim ðQ Gi  Q lim XNG  
Li Þ þ kQ Gi Þ
i¼1 i¼1 þs ai þ bi PGi þ ci P2Gi þ ni expðki PGi Þ
i¼1
X
NTL
2
þ kS ðSli  Slim
li Þ ð26Þ 2 X
NPQ
2
i¼1 þ kP ðPG1  P lim
G1 Þ þ kV ðV Li  V lim
Li Þ
i¼1
The obtained optimal settings for adjusting control variables from X
NG X
NTL
2 2
the proposed method are given in Tables 1 and 5 presents the com- þ kQ ðQ Gi  Q lim
Gi Þ þ kS ðSli  Slim
li Þ ð28Þ
parison of the proposed LTLBO algorithm other heuristic techniques i¼1 i¼1
previously mentioned in the preliminary section of paper.
In case 5, the weighting factor (s) is selected as 1000. The obtained
According to the obtained simulation data, the minimum fuel
optimal settings of control variables for case 5 study are shown in
cost obtained from the proposed approach is 917.6259 $/h with
Table 1, confirming LTLBO algorithm as the most appropriate
an average cost of 917.9524 $/h and a maximum cost of
solution for minimization of objective function in the OPF problem.
918.3766 $/h, which is less comparing to result of other algo-
rithms. The variation of the total fuel cost is shown in Fig. 6. As
can been seen, The ability of the offered LTLBO algorithm to find Table 6
Comparison of the simulation results for case 5 (ton/h).
Table 5
Comparison of the simulation results for case 4 ($/h). Algorithms Min Average Max Time (sec)
LTLBO 0.2047 0.2065 0.2104 21.55
Algorithms Min Average Max Time (sec)
TLBO 0.2124 0.2141 0.2156 23.64
LTLBO 917.6259 917.9524 918.3766 23.46 NPSO [36] 0.2098 0.2138 0.2153 22.2
TLBO 919.3943 919.471 919.6483 23.18 Fuzzy-GA [37] 0.2125 0.2155 0.2169 21.66
NPSO [36] 919.3785 919.565 920.3861 22.95 DE-PS [38] 0.2052 0.2071 0.2177 25.42
Fuzzy-GA [37] 921.5022 921.8836 922.25 23.11 ABC [40] 0.204826 NA NA NA
DE-PS [38] 919.0175 919.475 919.9842 25.07 MPSO-SFLA [42] 0.2246 NA NA NA
ABC [40] 945.4495 960.5647 973.5995 NA MSFLA [1] 0.2056 NA NA NA
BBO [43] 919.7647 919.8389 919.8876 11.15 SFLA [1] 0.2063 NA NA NA
MDE [2] 930.793 942.501 954.073 41.85 PSO [1] 0.2096 NA NA NA
EADDE [31] 930.7459 NA NA 3.98 GA [1] 0.2117 NA NA NA
382 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384

Table 7
Best control variables settings for IEEE 57-bus test system.

Variables TLBO LTLBO Variables TLBO LTLBO

PG1 (MW) 143.09 142.87 T24–25 (p.u.) 0.9980 1.0339
PG2 (MW) 87.7492 88.9529 T24–25 (p.u.) 1.0445 1.0248
PG3 (MW) 45.8006 44.7876 T24–26 (p.u.) 1.0595 1.0258
PG6 (MW) 71.0559 72.5774 T7–29 (p.u.) 1.0175 0.9841
PG8 (MW) 465.7858 460.9057 T34–32 (p.u.) 0.9864 0.9708
PG9 (MW) 94.2292 94.1794 T11–41 (p.u.) 0.9221 0.9000
PG12 (MW) 358.8398 361.6874 T15–45 (p.u.) 0.9726 0.9543
VG1 (p.u.) 1.0521 1.0380 T14–46 (p.u.) 0.9734 0.9395
VG2 (p.u.) 1.0482 1.0419 T10–51 (p.u.) 0.9877 0.9504
VG3 (p.u.) 1.0431 1.0305 T13–49 (p.u.) 0.9476 0.9119
VG6 (p.u.) 1.0533 1.0467 T11–43 (p.u.) 0.9763 0.9475
VG8 (p.u.) 1.0595 1.0600 T40–56 (p.u.) 0.9862 0.9929
VG9 (p.u.) 1.0352 1.0283 T39–57 (p.u.) 0.9810 0.9703
VG12 (p.u.) 1.0411 1.0222 T9–55 (p.u.) 1.0334 0.9724
T4–18 (p.u.) 0.9758 0.9014 QC18 (MVAR) 20.48 17.59
T4–18 (p.u.) 1.0364 1.0800 QC25 (MVAR) 15.38 17.41
T21–20 (p.u.) 0.9720 1.0193 QC53 (MVAR) 11.97 15.08
Fuel cost ($/h) 41695.6626 41679.5451
Power loss (MW) 15.7469 15.1589

x 10
4
the buses 1, 2, 3, 6, 8, 9 and 12, 80 transmission lines and 15
4.24
TLBO branches under load tap setting transformer branches. The shunt
4.23 LTLBO reactive power sources are considered at buses 18, 25 and 53
NPSO
fuzzy-GA
and the total load demand of system is 1250.8 MW and 336.4
4.22
DE-PS MVAR. The bus data, the line data and the cost coefficients and
Fuel cost ($/h)

4.21 minimum and maximum limits of real power generations are

4.2
taken from [32]. The maximum and minimum values of voltages
of all generator buses and tap setting transformer control variables
4.19 are set at 1.1–0.9 in p.u. The maximum and minimum values for
4.18 voltages of all load buses are defined as 1.06 and 0.94 in p.u form
[31], respectively and the maximum and minimum values of shunt
4.17
reactive power sources are 0.0 and 0.3 in p.u.
0 15 30 45 60 75 90 105 120 135 150 In this simulation process of LTLBO algorithm performance
Iteration
evaluation, objective function is minimization of quadratic cost
Fig. 7. Convergence of TLBO for IEEE 57-bus test system. function. During this simulation, 30 test runs were carried out
for solving OPF problem using the LTLBO algorithm. The objective
function is described as follow:
Table 8
Comparison of the simulation results for IEEE 57-bus test system. X
NG

Algorithms Fuel cost ($/h)

J¼ F i ðPGi Þ
i¼1
LTLBO 41679.5451
NG 
X 
TLBO 41695.6626 2
NPSO [36] 41699.5163
¼ ai þ bi PGi þ ci P2Gi þ kP ðPG1  Plim
G1 Þ
i¼1
Fuzzy-GA [37] 41716.2808
DE-PS [38] 41685.295 X
NPQ
2 X
NG
2
ABC [40] 41693.9589 þ kV ðV Li  V lim
Li Þ þ kQ ðQ Gi  Q lim
Gi Þ
LDI-PSO [40] 41815.5035 i¼1 i¼1
GSA [40] 52819.7052 X
NTL
2
EADDE [31] 41713.62 þ kS ðSli  Slim
li Þ ð29Þ
i¼1

where Fi and PGi represent the fuel cost of the ith generator and the
output of the ith generator, respectively. ai, bi and ci are the cost
The best fuel cost result calculated by LTLBO algorithm is compared
coefficients of the ith generator and NG is the number of total
to the results obtained by other techniques in Table 6. Judging from
generator.
Table 6, it can be seen that the minimum emission is 0.2047 ton/h,
with an average emission of 647.4725 $/h and a maximum emission Table A1
of 647.8638 $/h which are less in comparison to the reported results Generator cost coefficients for case 1 [33,34].
in the literature. Bus No. Cost coefficients
a b c
IEEE 57-bus test system 1 0.00 2.00 0.00375
2 0.00 1.75 0.01750
In order to evaluate the effectiveness and performance of LTLBO 5 0.00 1.00 0.06250
algorithm in larger scale power systems, a standard IEEE 57-bus 8 0.00 3.25 0.00834
11 0.00 3.00 0.02500
test system is introduced as the test bed for phase 2 of the simula-
13 0.00 3.00 0.02500
tions. The IEEE 57-bus test system is includes seven generators at
 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384 383

Table A2 Table A6
Generator cost coefficients for case 3 [36]. The limits of the control variables [33].

Bus No. Form MW To MW Cost coefficients Control variables Min Max

a b c PG1 (MW) 50 200
PG2 (MW) 20 80
1 50 140 55.00 0.70 0.0050
PG5 (MW) 15 50
140 200 82.50 1.05 0.0075
PG8 (MW) 10 35
2 20 55 40.00 0.30 0.0100 PG11 (MW) 10 30
55 80 80.00 0.60 0.0200 PG13 (MW) 12 40
VG1 (p.u.) 0.95 1.1
VG2 (p.u.) 0.95 1.1
VG5 (p.u.) 0.95 1.1
Table A3 VG8 (p.u.) 0.95 1.1
Generator cost coefficients for case 4 [36]. VG11 (p.u.) 0.95 1.1
VG13 (p.u.) 0.95 1.1
Bus P min P max Cost coefficients
Gi Gi T6–9 (p.u.) 0.90 1.1
No. (MW) (MW)
a b c d e T6–10 (p.u.) 0.90 1.1
T4–12 (p.u.) 0.90 1.1
1 50 200 150.00 2.00 0.0016 50.00 0.0630
T28–27 (p.u.) 0.90 1.1
2 20 80 25.00 2.50 0.0100 40.00 0.0980
QC10 (MVAR) 0.00 5.0
QC12 (MVAR) 0.00 5.0
QC15 (MVAR) 0.00 5.0
QC17 (MVAR) 0.00 5.0
Table A4
QC20 (MVAR) 0.00 5.0
Load data [33,34].
QC21 (MVAR) 0.00 5.0
Bus Load Bus Load Bus Load QC23 (MVAR) 0.00 5.0
No. No. No. QC24 (MVAR) 0.00 5.0
P Q P Q P Q
QC29 (MVAR) 0.00 5.0
(p.u.) (p.u.) (p.u.) (p.u.) (p.u.) (p.u.)
1 0.000 0.000 11 0.000 0.000 21 0.175 0.112
2 0.217 0.127 12 0.112 0.075 22 0.000 0.000 that the best fuel cost calculated by LTLBO algorithm is quite less
3 0.024 0.012 13 0.000 0.000 23 0.032 0.016
in comparison to reported best result the literature.
4 0.076 0.016 14 0.062 0.016 24 0.087 0.067
5 0.942 0.190 15 0.082 0.025 25 0.000 0.000
6 0.000 0.000 16 0.035 0.018 26 0.035 0.023
Conclusions
7 0.228 0.109 17 0.090 0.058 27 0.000 0.000
8 0.300 0.300 18 0.032 0.009 28 0.000 0.000
9 0.000 0.000 19 0.095 0.034 29 0.024 0.009 In this paper, a LTLBO algorithm has been offered as a novel
10 0.058 0.020 20 0.022 0.007 30 0.106 0.019 solution for solving OPF problems. At the beginning, the optimal
power flow problem was described as a nonlinear optimization
problem with equality and inequality constraints in power sys-
tems. During the study, different objective functions were consid-
The obtained optimal setting from simulation for adjustment of ered to enhance the voltage profile and also to minimize the fuel
control variables are shown in Table 7. Fig. 7 of the paper illustrates cost such as quadratic cost function, piecewise quadratic cost func-
convergence characteristic of the best fuel cost result obtained tion, cost function with valve point effect. As the simulation results
from the algorithms and comparison of the proposed LTLBO algo- indicated, the proposed method was successfully implemented the
rithms with heuristic technique reported in the literature is given goal to find global or near global optimal settings of the control
in Table 8. variables of an IEEE 30-bus and the IEEE 57-bus test systems with
The obtained best fuel cost from applying LTLBO algorithm is superiority relative to other approaches, as the comparison of
41679.5451 $/h. By examining the results in Table 8, it turns out results obtained from LTLBO and other mentioned methods con-

Table A5
Line data [33,34].

Line No. Form bus To bus R (p.u.) X (p.u.) Tap settings Line No. Form bus To bus R (p.u.) X (p.u.) Tap settings
1 1 2 0.0192 0.0575 – 22 15 18 0.1070 0.2185 –
2 1 3 0.0452 0.1852 – 23 18 19 0.0639 0.1292 –
3 2 4 0.0570 0.1737 – 24 19 20 0.0340 0.0680 –
4 3 4 0.0132 0.0379 – 25 10 20 0.0936 0.2090 –
5 2 5 0.0472 0.1983 – 26 10 17 0.0324 0.0845 –
6 2 6 0.0581 0.1763 – 27 10 21 0.0348 0.0749 –
7 4 6 0.0119 0.0414 – 28 10 22 0.0727 0.1499 –
8 5 7 0.0460 0.1160 – 29 21 22 0.0116 0.0236 –
9 6 7 0.0267 0.0820 – 30 15 23 0.1000 0.2020 –
10 6 8 0.0120 0.0420 – 31 22 24 0.1150 0.1790 –
11 6 9 0.0000 0.2080 1.078 32 23 24 0.1320 0.2700 –
12 6 10 0.0000 0.5560 1.069 33 24 25 0.1885 0.3292 –
13 9 11 0.0000 0.2080 – 34 25 26 0.2544 0.3800 –
14 9 10 0.0000 0.1100 – 35 25 27 0.1093 0.2087 –
15 4 12 0.0000 0.2560 1.032 36 28 27 0.0000 0.3960 1.068
16 12 13 0.0000 0.1400 – 37 27 29 0.2198 0.4153 –
17 12 14 0.1231 0.2559 – 38 27 30 0.3202 0.6027 –
18 12 15 0.0662 0.1304 – 39 29 30 0.2399 0.4533 –
19 12 16 0.0945 0.1987 – 40 8 28 0.0636 0.2000 –
20 14 15 0.2210 0.1997 – 41 6 28 0.0169 0.0599 –
21 16 17 0.0824 0.1932 –
384 M. Ghasemi et al. / Electrical Power and Energy Systems 65 (2015) 375–384

firms the superiority of the proposed LTLBO algorithm over sto- [19] Bakirtzis AG, Biskas PN, Zoumas CE, Petridis V. Optimal power flow by
enhanced genetic algorithm. Power Syst, IEEE Trans 2002;17:229–36.
chastic techniques in terms of solution quality for the OPF
[20] Varadarajan M, Swarup K. Solving multi-objective optimal power flow using
problems. differential evolution. Generat, Transm Distrib, IET 2008;2:720–30.
[21] Roa-Sepulveda C, Pavez-Lazo B. A solution to the optimal power flow using
simulated annealing. Int J Electri Power Energy Syst 2003;25:47–57.
Appendix A. [22] Somasundaram P, Kuppusamy K, Kumudini Devi R. Evolutionary programming
based security constrained optimal power flow. Electric Power Syst Res
Data of IEEE 30-bus test system (100MVA base) (Tables A1–A6). 2004;72:137–45.
[23] Zimmerman R, Carlos E, Gan D. MATPOWER: A MATLAB Power System
Simulation Package, Version 3.1 b2, User’s Manual. Power Systems
References Engineering Research Center, Tech Rep; 2006.
[24] Milano F. An open source power system analysis toolbox. Power Syst, IEEE
[1] Niknam T, Jabbari M, Malekpour AR. A modified shuffle frog leaping algorithm Trans 2005;20:1199–206.
for multi-objective optimal power flow. Energy 2011;36:6420–32. [25] Rao R, Savsani V, Vakharia D. Teaching–learning-based optimization: a novel
[2] Sayah S, Zehar K. Modified differential evolution algorithm for optimal power method for constrained mechanical design optimization problems. Computer-
flow with non-smooth cost functions. Energy Convers Manage 2008;49: Aided Des 2011;43:303–15.
3036–42. [26] Rao R, Savsani V, Vakharia D. Teaching–learning-based optimization: an
[3] Ghasemi M, Ghavidel S, Rahmani S, Roosta A, Falah H. A novel hybrid algorithm optimization method for continuous non-linear large scale problems. Informat
of imperialist competitive algorithm and teaching learning algorithm for Sci 2012;183:1–15.
optimal power flow problem with non-smooth cost functions. Eng Appl Artif [27] Mallipeddi R, Jeyadevi S, Suganthan PN, Baskar S. Efficient constraint handling
Intell 2014;29:54–69. for optimal reactive power dispatch problems. Swarm Evolut Comput
[4] Ghasemi M, Ghavidel S, Ghanbarian MM, Massrur HR, Gharibzadeh M. 2012;5:28–36.
Application of imperialist competitive algorithm with its modified [28] Gnedenko BV, Kolmogorov AN, Chung KL, Doob JL. Limit distributions for sums
techniques for multi-objective optimal power flow problem: a comparative of independent random variables. Springer; 1968.
study. Inform Sci 2014. [29] Abido M. Optimal power flow using particle swarm optimization. Int J Electric
[5] Roy P, Ghoshal S, Thakur S. Biogeography based optimization for multi- Power Energy Syst 2002;24:563–71.
constraint optimal power flow with emission and non-smooth cost function. [30] Slimani L, Bouktir T. Economic power dispatch of power system with pollution
Expert Syst Appl 2010;37:8221–8. control using multiobjective ant colony optimization. Int J Computat Intell Res
[6] Ghasemi M, Ghavidel S, Akbari E, Vahed AA. Solving non-linear, non-smooth (IJCIR) 2007;3:145–53.
and non-convex optimal power flow problems using chaotic invasive weed [31] Vaisakh K, Srinivas L. Evolving ant direction differential evolution for OPF with
optimization algorithms based on chaos. Energy 2014;73:340–53. non-smooth cost functions. Eng Appl Artif Intell 2011;24:426–36.
[7] Yan X, Quintana VH. Improving an interior-point-based OPF by dynamic [32] Zhang J, Tse C, Wang W, Chung C. Voltage stability analysis based on
adjustments of step sizes and tolerances. Power Syst, IEEE Trans 1999;14: probabilistic power flow and maximum entropy. IET Generat, Transm Distrib
709–17. 2010;4:530–7.
[8] Habibollahzadeh H, Luo G-X, Semlyen A. Hydrothermal optimal power flow [33] Lee K, Park Y, Ortiz J. A united approach to optimal real and reactive power
based on a combined linear and nonlinear programming methodology. Power dispatch. Power Apparatus Syst, IEEE Trans 1985:1147–53.
Syst, IEEE Trans 1989;4:530–7. [34] Alsac O, Stott B. Optimal load flow with steady-state security. Power
[9] Ghasemi M, Ghavidel S, Ghanbarian MM, Gharibzadeh M, Vahed AA. Multi- Apparatus Syst, IEEE Trans 1974:745–51.
objective optimal power flow considering the cost, emission, voltage deviation [35] Ghasemi M, Ghavidel S, Aghaei J, Gitizadeh M, Falah H. Application of chaos-
and power losses using multi-objective modified imperialist competitive based chaotic invasive weed optimization techniques for environmental OPF
algorithm. Energy 2014. http://dx.doi.org/10.1016/j.energy.2014.10.007. problems in the power system. Chaos Soliton Fract 2014. http://dx.doi.org/
[10] Momoh JA, Adapa R, El-Hawary M. A review of selected optimal power flow 10.1016/j.chaos.2014.10.007.
literature. Nonlinear and quadratic programming approaches to 1993. I. Power [36] Selvakumar AI, Thanushkodi K. A new particle swarm optimization solution to
Syst, IEEE Trans 1999;14:96–104. nonconvex economic dispatch problems. Power Syst, IEEE Trans 2007;22:
[11] Momoh JA, El-Hawary M, Adapa R. A review of selected optimal power flow 42–51.
literature to 1993. II. Newton, linear programming and interior point methods. [37] Hsiao Y-T, Chen C-H, Chien C-C. Optimal capacitor placement in distribution
Power Syst, IEEE Trans 1999;14:105–11. systems using a combination fuzzy-GA method. Int J Electric Power Energy
[12] Huneault M, Galiana F. A survey of the optimal power flow literature. Power Syst 2004;26:501–8.
Syst, IEEE Trans 1991;6:762–70. [38] Gitizadeh M, Ghavidel S, Aghaei J. Using SVC to economically improve
[13] Ghasemi M, Ghavidel S, Ghanbarian MM, Habibi A. A new hybrid algorithm for transient stability in long transmission lines. IETE J Res 2014;60:319–27.
optimal reactive power dispatch problem with discrete and continuous [39] Nayak M, Nayak C, Rout P. Application of multi-objective teaching learning
control variables. Appl Soft Comput 2014. based optimization algorithm to optimal power flow problem. Procedia
[14] Ghasemi M, Ghanbarian MM, Ghavidel S, Rahmani S, Mahboubi Moghaddam E. Technol 2012;6:255–64.
Modified teaching learning algorithm and double differential evolution [40] Rezaei Adaryani M, Karami A. Artificial bee colony algorithm for solving multi-
algorithm for optimal reactive power dispatch problem: a comparative objective optimal power flow problem. Int J Electric Power Energy Syst 2013;
study. Informat Sci 2014;278:231–49. 53:219–30.
[15] Gitizadeh M, Ghavidel S. Improving transient stability with multi-objective [41] Kumari MS, Maheswarapu S. Enhanced genetic algorithm based computation
allocation and parameter setting of SVC in a multi-machine power system. technique for multi-objective optimal power flow solution. Int J Electric Power
IETE J Res 2014;60:33–41. Energy Syst 2010;32:736–42.
[16] Abido M. Optimal power flow using tabu search algorithm. Electric Power [42] Ghasemi M, Ghavidel S, Ghanbarian MM, Gitizadeh M. Multi-objective optimal
Componen Syst 2002;30:469–83. electric power planning in the power system using Gaussian bare-bones
[17] Devaraj D, Yegnanarayana B. Genetic-algorithm-based optimal power flow imperialist competitive algorithm. Inform Sci 2015;294:286–304.
for security enhancement. IEE Proc-Generat, Transm Distribut 2005;152: [43] Bhattacharya A, Chattopadhyay P. Application of biogeography-based
899–905. optimisation to solve different optimal power flow problems. IET Generat,
[18] Lai LL, Ma J, Yokoyama R, Zhao M. Improved genetic algorithms for optimal Transm Distrib 2011;5:70–80.
power flow under both normal and contingent operation states. Int J Electric [44] Abou El Ela A, Abido M, Spea S. Optimal power flow using differential evolution
Power Energy Syst 1997;19:287–92. algorithm. Electric Power Syst Res 2010;80:878–85.