Electrical Power and Energy Systems 54 (2014) 198–210

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Faster evolutionary algorithm based optimal power flow using

incremental variables
S. Surender Reddy ⇑, P.R. Bijwe, A.R. Abhyankar
Department of Electrical Engineering, Indian Institute of Technology Delhi, New Delhi 110 016, India

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

Article history: This paper proposes an efficient approach for evolutionary algorithm based Optimal Power Flow (OPF).
Received 19 February 2013 The main drawback of evolutionary based OPF is the excessive execution time due to large number of
Received in revised form 22 June 2013 power flows required in the solution process. The proposed Efficient Evolutionary Algorithm (EEA) uses
Accepted 13 July 2013
the concept of incremental power flow model, based on sensitivities. With this, the number of power
flows are reduced substantially, resulting in solution speed up. The original advantages of the evolution-
ary algorithms, like: the ability to handle discontinuities, complex non-linearities in the objective func-
Keywords:
tion, discrete variables, and multi-objective optimization, are still available in the proposed approach. The
Enhanced genetic algorithm
Evolutionary algorithms
OPF solution is obtained with single objectives (fuel cost, loss, voltage stability index) and multiple objec-
Linear programming tive (fuel cost and voltage stability index). The potential of the proposed approach is tested on IEEE 30,
Multi-objective optimization 118 and 300 bus systems, and the results obtained with proposed EEA are compared with other evolu-
Optimal power flow tionary algorithms. The proposed approach is generic one and can be used with any evolutionary algo-
rithm based OPF.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction Foraging and Simulated Annealing (SA) methods, etc. GA, EGA, Im-
proved GA have been successfully applied for solution of OPF prob-
The Optimal Power Flow (OPF) has been commonly used as an lem [3–5]. In [6], PSO was used to solve the OPF problem. PSO has a
important tool in the power system planning and operation for flexible and well-balanced mechanism to enhance and adapt the
many years. It is also an important tool in modern Energy Manage- global and local exploration abilities. Different types of PSOs’ are
ment System (EMS). It plays an important role in maintaining the proposed in [6–11]. Ref. [12] presents an improved PSO for the
security and economy of the power system. The OPF optimizes multi-objective optimization problem. In [13], an effective and reli-
the power system operating objective function, while satisfying a able algorithm, based on Shuffle Frog Leaping Algorithm (SFLA) and
set of equality and inequality constraints. The equality constraints SA is proposed for solving OPF problem with non-smooth and non-
are power flow equations, and inequality constraints are the limits convex generator fuel cost characteristics.
on power system variables/functions (control variables and func- Evolutionary Algorithms (EAs) differ from classical search and
tional operating constraints). The problem of OPF was originally optimization algorithm [14] in many ways. Classical search tech-
formulated in 1962 by Carpentier [1]. The OPF problem in [2] is a niques use a single solution updates in every iteration, and mainly
highly nonlinear, non-convex, large scale, static optimization prob- use some deterministic transition rules for approaching the opti-
lem with both continuous (generator voltage magnitudes and ac- mum solution. Such algorithms start from a random guess solu-
tive powers) and discrete (transformer taps and switchable shunt tion, and based on some pre-specified transition rule, the
devices) control variables. algorithm suggests a search direction which is arrived at by consid-
The OPF solution based on mathematical programming ap- ering local information. A unidirectional search is performed along
proaches are not guaranteed to converge to the global optimum the search direction, to find a best solution. The best solution be-
of the general non-convex OPF problem, and discrete variables. Re- comes the new solution, and the search is continued for a number
cent attempts to overcome the limitations of mathematical pro- of times.
gramming approaches include the application of Genetic Classical optimization methods can be direct methods and gra-
Algorithms (GA), Enhanced Genetic Algorithms (EGA), Particle dient based methods. In direct search methods, only the objective
Swarm Optimization (PSO), Differential Evolution (DE), Bacterial function and the constraint values are used to guide the search
process, whereas gradient based methods use first order or second
order derivative of objectives and constraints to guide the search
⇑ Corresponding author. Tel.: +91 11 26591095; fax: +91 11 26581606.
process. Direct search methods are usually very slow requiring
E-mail address: [email protected] (S. Surender Reddy).

0142-0615/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2013.07.019
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 199

many function evolutions for convergence. Without major change where the minimization function ‘F’ (objective function) can take
in the algorithm, they can be applied to solve many problems. Gra- different forms.
dient based methods quickly converge to an optimal solution, but g is the equality constraints, represent the non-linear power
are not efficient for non-differentiable or discontinuous problems flow equations.
[14]. h is the system operating constraints that include functional
All meta-heuristic/ evolutionary algorithms perform a separate operating constraints and limits on control variables.‘x’ is the vec-
power flow for each chromosome. This may be used for objective tor of dependent variables consisting of load bus voltage magni-
function evaluation and/or constraint feasibility check. Penalties tude limits, reactive capabilities of generators, slack bus active
are added as per the extent of infeasibility. This exercise is re- power and branch flow limits.
peated for every chromosome. Hence, the number of power flows
xT ¼ ½V L1 ; . . . ; V LNL ; Q G1 ; . . . ; Q GNG ; P GSlack ; SL1 ; . . . ; SLnl  ð2Þ
to be run is enormous. This is the main reason for excessive com-
putational burden in these algorithms. Other parts of the algorithm where NL, NG and nl are number of load buses, number of generator
require comparatively insignificant time. buses and number of transmission lines, respectively.‘u’ Is the vec-
It has been recognized that EAs perform much better, if we can tor of control or independent variables consisting of generator bus
make use of domain specific knowledge of the problem at hand in voltage magnitudes, active power generations, transformer tap set-
the computational process. In view of this, the present paper ex- tings and reactive shunt compensators.
plores a possibility of reducing computational burden by perform-
ing much lesser power flows. This is possible because the effects of uT ¼ ½PG2 ; . . . ; PGNG ; V G1 ; . . . ; V NG ; T 1 ; . . . ; T NT ; Q C1 ; . . . ; Q C NC  ð3Þ
the control changes on the network are not likely to be too large. In where NT and NC are number of transfer taps and number of
that case, it should be possible to use one power flow for a chromo- switchable VAR sources, respectively.
some and evaluate the changes due to chromosome differences, by In this paper, three objectives: (i) real power generation cost,
using incremental power flow model and sensitivities, using a non- (ii) system transmission loss, and (iii) voltage stability index of
linear approximation. Objective functions with discontinuities, the system are considered. First, these three objectives are opti-
complex non-linearities and discrete variables can be handled as mized independently, and then the best combination of objectives
easily as in the original evolutionary algorithms. The main aim of are optimized simultaneously using Strength Pareto Evolutionary
the proposed approach is to reduce the time of execution, and is Algorithm 2+ (SPEA 2+).
suitable for any type of complex problem involving any objective. The proposed Efficient Evolutionary Algorithm (EEA) is the hybrid
It may be noted that sensitivities of control variables have also approach considering Enhanced Genetic Algorithm (EGA) and some
been used extensively in Linear Programming (LP), Quadratic Pro- concepts borrowed from the Successive Approximation using Linear
gramming (QP) and other gradient based approaches. However, it Programming (SALP) or Quadratic Programming (SAQP). Many of
is well known that all these lead to a local optima. The proposed these models use incremental variable modeling approach. Moreover,
EEA uses sensitivity information in an approach which has near the popular compact model does not use original equality constraints
global optimization capability. Most of the real world problems in- as part of main OPF. Power flow is solved outside the main OPF itera-
volve simultaneous optimization of several, often mutually concur- tion. It also generates sensitivity relations for expressing dependent
rent objectives. Multi-objective evolutionary algorithms are used variable changes in terms of control variable changes.
to find the optimal trade-off solution. In multi-objective optimiza- Usual meta-heuristic OPF algorithms performs a separate load
tion, gradient based methods are often impossible to apply. How- flow for every chromosome, in every iteration. Hence, the number
ever, the proposed EEA approach can easily be applied to any of load flows required is equal to number of chromosomes  number
multi-objective based OPF. of iterations. Whereas, proposed EEA approach, performs only one
The present work chooses EGA [4] for implementing the pro- load flow for best-fit chromosome, and evaluates load flow infor-
posed Efficient Evolutionary Algorithm (EEA), but the modeling mation for other chromosomes using sensitivities from this load
methodology can be easily extended to include any type of evolu- flow, in a non-linear approximation of functions/variables, in every
tionary algorithms such as PSO, DE, Bacterial Foraging, and SA. In iteration. Therefore, the number of load flows required is equal to
this paper, some case studies are also performed by applying the number of iterations/generations.
proposed EEA approach to the PSO.
The rest of the paper is organized as follows: Section 2 presents 2.1. Real and reactive power OPF for real power generation cost
incremental variable modeling for OPF, real power generation cost, minimization using EEA
real power loss and system voltage stability index minimizations
using proposed EEA. Section 3 presents brief description of multi- In this OPF formulation, generator fuel cost minimization is
objective optimization. Section 4 presents results and discussion. considered as an objective function and it is formulated as follows,
Finally, Section 5 brings out contributions with concluding remarks.
X
NG
minimize F T ¼ F i ðPGi Þ ð4Þ
2. OPF problem formulation i¼1

The OPF problem optimizes the steady state performance of the The proposed approach can be applicable for any type of objective
power system in terms of the objective function while satisfying a function like discontinuous (prohibited operating zones), discrete,
set of equality and inequality constraints. OPF is a highly non-lin- valve point loading, etc. The control variables considered in this
ear, non-convex, large scale static optimization problem due to optimization problem are changes in active power generations
large number of variables and constraints. In general, OPF is formu- (DPG), generator bus voltage magnitudes (DVG), transformer tap set-
lated as a constrained optimization problem, and can be mathe- tings (Dt) and bus shunt admittances (DBsh), from the base case. All
matically stated as follows: other variable changes are also with respect to the base case load
flow.
minimize Fðx; uÞ The quadratic fuel cost function of ith generator is given by
Subject to gðx; uÞ ¼ 0 [15,16],

hðx; uÞ 6 0 ð1Þ F i ðPGi Þ ¼ ai þ bi PGi þ ci P2Gi ð5Þ

200 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

The feasible operating zones of ith generating unit is described

as follows [19]:
8 min l
>
< PGi 6 PGi 6 PGi;1
PGi  PGi;k1 6 PGi 6 P lGi;k
u
ðk ¼ 2; . . . ; zi Þ ð9Þ
>
: Pu 6 P 6 Pmax
Gi;zi Gi Gi

where zi is the number of prohibited zones of unit i, k is the index of

prohibited zones of a unit i; PlGi;k and PuGi;k are lower and upper
bounds of the kth prohibited zone of unit i.

2.1.1. Constraints
Constraints on control variables:
The outputs of generators are restricted by their lower and
upper limits. In terms of incremental changes we have,

Fig. 1. Fuel cost function with and without valve point loading (VPL) effect. Pmin 0 max
Gi 6 ðP Gi þ DP Gi Þ 6 P Gi ð10Þ
If we include the generator ramp rate constraints, then the outputs
where ai, bi and ci are fuel cost coefficients of ith generator. PGi is real of generators are restricted as follows,
power generation of ith generator. up
maxðP min 0 down
Gi ; P Gi  RGi Þ 6 ðP0Gi þ DP Gi Þ 6 minðPmax 0
Gi ; P Gi þ RGi Þ ð11Þ
The above objective function with quadratic fuel cost function,
in terms of incremental variables is given by, where Rdown
Gi and Rup
Gi are the ramp down and ramp up limits of gen-

NG
erators in megawatts per hour.
X 2
min: F T ¼ ½ai þ bi ðP0Gi þ DPGi Þ þ ci ðP0Gi þ DPGi Þ  ð6Þ The changes in active power generations (DPGi), generator bus
i¼1 voltage magnitudes (DVGi), transformer tap settings (Dti), and
bus shunt admittances (DBsh,i) are restricted by their upper and
where P 0Gi and DPGi are the initial power and change in power of ith lower permissible limits.
generator, respectively.
maxðP min 0 down
Gi ; P Gi  RGi Þ  P0Gi 6 DPGi
Non-smooth cost function with Valve Point Loading (VPL) 6 minðPmax 0 up 0
Gi ; P Gi þ RGi Þ  P Gi ð12Þ
effect:

ðV min 0 max 0
Gi  V Gi Þ 6 DV Gi 6 ðV Gi  V Gi Þ ð13Þ
Fig. 1 [13] shows the cost-output characteristics for a generat-
ing unit with steam admission valves for with and without valve
point loading (VPL) effect. The VPL effect introduce ripple in the ðt min
i  t0i Þ 6 Dti 6 ðt max
i  t0i Þ ð14Þ
heat rate function and make the fuel cost function highly non-lin-
ear, discontinuous and multiple local optimum [17]. ðBmin 0 max 0
sh;i  Bsh;i Þ 6 DBsh;i 6 ðBsh;i  Bsh;i Þ ð15Þ
The VPL effect is taken in consideration by adding a sine compo-
where P min max
Gi ; P Gi are the minimum and maximum generation capac-
nent to the fuel cost of the generating units. The fuel cost function min max
ities; V Gi ; V Gi are minimum and maximum values of generator bus
of ith generation unit with VPL [13,18] is given by,
voltage magnitudes; t min i ; tmax
i are minimum and maximum values of
F i ðPGi Þ ¼ ai þ bi PGi þ ci P2Gi þ jdi  sinðei  ðPmin transformer taps; Bsh;i ; Bmax
min
sh;i are minimum and maximum values of
Gi  P Gi ÞÞj ð7Þ
the bus shunt susceptances.
where di and ei are the cost coefficients of the unit with valve-point
effects. Constraint on reactive power generations:
The objective function in terms of incremental variables with
VPL is The reactive power generation (QGi) is restricted by lower and
X
NG upper limit as
2
min: F T ¼ ½ai þ bi ðP0Gi þ DPGi Þ þ ci ðP0Gi þ DPGi Þ þ jdi  sinðei
i¼1

 ðPmin 0
Gi  ðP Gi þ DP Gi ÞÞÞj ð8Þ

Observe that the objective function is still non-linear and non-

convex.

Non-smooth cost function with Prohibited Operation Zones:

The cost curves of practical generators have prohibited operat-

ing zones (POZs) due to some faults in the shaft bearing or vibra-
tion of machines or their accessories such as pumps or boilers
[13]. A generating unit with prohibited operating zone (POZ) has
discontinuous input–output characteristics, since it is difficult to
determine the actual POZ by real performance testing or operating
records, so normally the best economy is achieved by avoiding
operation in areas that are in actual operation. Thus, the input–out-
put performance curve for a typical thermal unit can be repre-
sented as shown in Fig. 2 [13]. Fig. 2. Fuel cost function with prohibited operating zones (POZs).
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 201

Q min 0 max
Gi 6 ðQ Gi þ DQ Gi Þ 6 Q Gi ð16Þ Let L = J1.
Further, [DX] can be modified as,
Hence, the changes in reactive power (DQGi) are restricted as  
follows: Dd
½DX ¼ ¼ ½LMDU pq  LADV G  LBDt NT  LC DBsh  ð26Þ
DV
ðQ min 0 max 0
Gi  Q Gi Þ 6 DQ Gi 6 ðQ Gi  Q Gi Þ ð17Þ
Constraints on Line Flow Limits: ½DX ¼ ½W DU pq  DDV G  EDt NT  F DBsh  ð27Þ
The power flow through each line (Pij) is limited by where W = [LM], D = [LA], E = [LB] and F = [LC].
ðPmax  P0ij Þ 6 DPij 6 ðP max  P0ij Þ ð18Þ From Eq. (27), the changes in voltage magnitudes and angles
ij ij
can be written as
where P max
ij is the power flow limit of line between the buses i and j. N sh
X
N X
N PV X
NT X
Ddi ¼ ½W 0ij DU j   ½D0ij DV Gj   ½E0ij Dtj   ½F 0ij DBsh;j  ð28Þ
The power balance inside the GA optimization: j¼1 j¼1 j¼1 j¼1
The power balance inside the GA optimization is
X
2N X
NPV X
NT Nsh
X
X
NG
PGi  PD  Ploss ¼ 0 ð19Þ DV i ¼ ½W 00ij DU j   ½D00ij DV Gj   ½E00ij Dt j   ½F 00ij DBsh;j  ð29Þ
i¼1 j¼Nþ1 j¼1 j¼1 j¼1

In terms of incremental variables it can be written as follows, where W0 , W00 are sub-matrices of W; D0 , D00 are sub-matrices of D; E0 ,
E00 are sub-matrices of E; and F0 , F00 are sub-matrices of F consisting
X
NG
terms relating Ddi and DVi, respectively.
DPGi  DPloss ¼ 0 ð20Þ
i¼1
Updating the control variables:
Nodal power balance constraints:
The OPF equality constraints reflect the physics of the power tþ1
PGi ¼ PtGi þ DPGi ð30Þ
system [13]. Load flow (equality constraints) is performed outside
each OPF iteration. The nodal power balance constraints includes
V tþ1 t
Gi ¼ V Gi þ DV Gi ð31Þ
real and reactive power balances.
X
n
ttþ1 ¼ t ti þ Dt i ð32Þ
PGi  PDi ¼ ðV i V j Y ij cosðhij þ dj  di ÞÞ ð21Þ i

j¼1
tþ1
Bsh;i ¼ Btsh;i þ DBsh;i ð33Þ
X
n
Q Gi  Q Di ¼  ðV i V j Y ij sinðhij þ dj  di ÞÞ ð22Þ
j¼1
Loss Evaluation:
In Eqs. (21) and (22), i = 1, . . . , n. Where n is the number of buses in
the system. PGi, QGi are active, reactive power generations at bus i; With the updated X, the total transmission loss (Ploss) can be cal-
PDi, QDi are corresponding active and reactive load demands; Vi is culated using
the magnitude of voltage, di is angle of voltage and hij is load angle.
X
n X
n  2  2
2.1.2. Incremental variable/function evaluation Ploss ¼ 0:5 Gij ½ V 0i þ DV i þ V 0j þ DV j
After the converged load flow solution, the solution for DX i j–i

(dependent voltages and angles changes) can be evaluated from    h i

the following: 2 V 0i þ DV i V 0j þ DV j cos ðd0i þ Ddi Þ  ðd0j þ Ddj Þ  ð34Þ
½DX ¼ ½Dd DVT ð23Þ Hence,
At this stage, we have the following linearized relation [1]
  DPloss ¼ Ploss  P0loss ð35Þ
DP
½DX ¼ ½J 1  1
¼ J ½DU pq  ð24Þ where P0loss
is the power loss of the best fit chromosome, DPloss is
DQ
change in power loss for the selected chromosome from the best
where J is Jacobian matrix in Newton–Raphson load flow. The fol- fit chromosome.
lowing modified set of linearized equations are used to get the de-
sired sensitivities [20]. In this the effect of changes in control on the Evaluation of Reactive Power Generation:
mismatches are reflected.
  In terms of incremental state variables, the reactive power gen-
½DX ¼ J 1 MDU pq  ADV G  BDt  C DBsh ð25Þ eration (QGi) can be written as
It is noted that for a given OPF, all controls may not be present [20]. "
Where M, A, B and C are the sensitivity matrices with respect to Upq, 2 X
n
Q Gi ¼  Bii ðV 0i þ DV i Þ þ ðV 0i þ DV 0 ÞðV 0j þ DV j ÞY ij
VG, t and Bsh respectively, and they are
j–i
 T  T #
@P @Q @P @Q h i
M¼ ; A¼ 0 0
sin hij þ ðdj þ Ddj Þ  ðdi þ Ddi Þ ð36Þ
@U pq @U pq @V G @V G
 T  T
@P @Q @Q
;B ¼ and C ¼ 0 : Hence,
@t @t @Bsh
202 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

 Step 7: Using the updated control variables, evaluate the objec-

tive function. A penalty function [5] is added to the objective
function, if the functional operating constraints violates any of
the limits. Calculate the fitness of each chromosome. The
description of fitness function evaluation is presented in [21].
 Step 8: Check the convergence criteria, i.e., if fitness (first chro-
mosome) = fitness (last chromosome). If yes, calculate optimal
objective function value (e.g. fuel cost or loss) and STOP. Else,
apply basic GA operators, advanced Enhanced GA (EGA) opera-
tors, and generate new population from old one.
 Step 9: Update generation count.
 Step 10: Check if generation count > maximum number of gen-
erations (convergence criteria). If yes, STOP. Else repeat from
Step 4.

2.2. OPF with real power loss minimization using EEA

For reactive power optimization, system transmission loss min-

imization is considered as the objective function. OPF with real
power loss minimization is formulated as follows:
X
n X
n
Ploss ¼ 0:5 Gij ½V 2i þ V 2j  2V i V j cosðdi  dj Þ ð40Þ
i¼1 j–i

where Ploss represent the real power losses in the system. Gij is the
conductance of the line between bus i and bus j in p.u. for i – j,Vi is
the bus voltage at node i in p.u and di is the phase angle in radians.
The real power loss is nothing but a mismatch in power which oc-
Fig. 3. Flow Chart of OPF using Proposed Efficient Evolutionary Algorithm (EEA) curs while solving any power flow. The mismatch is compensated
considering Enhanced Genetic Algorithm (EGA).
by the slack generator. So, the slack bus real power injection is used
as the objective function in real power loss minimization instead of
actual real power loss expression.
DQ Gi ¼ Q Gi  Q 0Gi ð37Þ
X
n
min:Ploss ¼ min:Pk ¼ Gkk V 2k þ V k V j Y kj cosðhkj þ dj  dk Þ ð41Þ
j–k
Line Flow Evaluation:
where index ‘k’ in Eq. (41) represents the slack bus number.
Converting the objective function (Eq. (41)) in terms of incre-
With the updated X, the power flow (Pij) can be calculated using
mental state variables and the latest values of state variables
2
Pij ¼ Gij ðV 0i þ DV i Þ þ ½ðV 0i þ DV i ÞðV 0j þ DV j ÞY ij cos½hij þ ðd0j Ploss ¼ Pk
þ Ddj Þ  ðd0i þ Ddi Þ ð38Þ  2 X
n
¼ Gkk V 0k þ DV k þ ðV 0k þ DV k ÞðV 0j
Hence, j–k
h i
DPij ¼ Pij  P0ij ð39Þ þ DV j ÞY kj cos hkj þ ðd0j þ Ddj Þ  ðd0k þ Ddk Þ ð42Þ

Observe that the incremental relations for FT, DPloss, DQGi and DPij It can be observed that, the objective function is still a non-linear
are still non-linear. equation. The equality and inequality constraints are similar as de-
The flow chart of the proposed Efficient Evolutionary Algorithm scribed in Sections 2.1.1 and 2.1.2.
(EEA) using Enhanced Genetic Algorithm (EGA) is shown in Fig. 3.
2.3. OPF with Voltage Stability Enhancement Index (VSEI) using EEA
2.1.3. Brief algorithm
To monitor the voltage stability in power system, L-index [22]
 Step 1: Neglecting losses and network constraints, get economic of the load buses is considered. This L-index uses the information
dispatch solution (P0Gi ) using Genetic Algorithm (GA). from a normal load flow, and is in the range of 0 (no load) to 1
 Step 2: Run base case load flow with P0Gi s in Step 1, and initial (voltage collapse). The control against voltage collapse is based
guess for reactive controls. on minimizing the sum of squared L-indices for a given system
 Step 3: Randomly generate the population of chromosomes. operating condition. The L-index is a measure for the estimation
 Step 4: From the second generation onwards, run load flow only of the distance of the actual state of the system to the stability limit
for the best-fit chromosome. Evaluate the sensitivities after the [23]. The L-index describes the stability of the complete system.
converged load flow solution. Here, system voltage stability index (i.e., L-index) has been formu-
 Step 5: Evaluate the load flow information for other chromo- lated as an objective, and sum squared L-indices is minimized to
somes using sensitivities from this load flow, in a non-linear improve the stability margin of the system [12].
approximation of functions/variables, in every iteration.  
 XNG
Ei 
 Step 6: Evaluate the changes in dependent voltages and angles 
Lj ¼ 1  F ji  ð43Þ
using sensitivities in Step 4.  i¼1
Ej 
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 203

X
n
how effective neighborhood crossover is in the multi-objective ge-
L  index ¼ VSEI ¼ L2j ð44Þ
netic algorithm. The detailed description of neighborhood cross-
j¼NG þ1
over has been presented in [28]. Neighborhood crossover
where Ei, Ej are the complex voltages of generator and load buses, implicates crossing over between individuals located close to each
and j = NG + 1, . . . , n. The values of Fji are obtained from the Y-Bus other in the objective space.
matrix [24]. The L-indices for the given load condition are computed The algorithm of SEPA 2+ [30] is described next:
for all load buses and the maximum of L-indices gives the proximity
of the system to voltage collapse.  Step 1: Generate initial population (P0) and empty objective
The proposed incremental variable model based OPF can also be archive population (OA0), design variable archive population
applied for the trade-off solution similar to other multi-objective (VA0). Set the generation count k = 0.
evolutionary algorithms. The brief description of multi-objective  Step 2: Fitness values of all individuals Pk, OAk and VAk are eval-
optimization is presented next. uated using SPEA 2’s fitness assignment method [26].
 Step 3: All non-dominated individuals in Pk, OAk and VAk are
copied to OAk+1 and VAk+1. If the number of individuals of
3. Multi-objective optimization OAk+1 and VAk+1 have exceeded the archive size, archive trunca-
tion in objective space is applied to the individuals in OAk+1, and
Real-world problems naturally involve multiple and conflicting archive truncation in variable space is made VAk+1 to reduce the
objectives to be optimized simultaneously. Defining multiple number of individuals. If the number of individuals of OAk+1 or
objectives often gives better idea of the problem. A single attribute VAk+1 is less than archive size, individuals with good fitness
that is most important and appropriate for a particular operating from Pk, OAk and VAk are used to fill OAk+1 and VAk+1.
condition, has been used as an objective for optimization. The  Step 4: If maximum number of generations are exceeded or
other important attributes are included in the mathematical model other termination conditions are met, then stop the search
as constraints with carefully chosen limits. Most of the times process.
choosing these limits is also not that simple. This is a rigid ap-  Step 5: Pk+1 is generated by copying them into OAk+1. The neigh-
proach and does not provide room for trade-off between various borhood crossover and mutation operations are performed.
attributes, which can be beneficial for system operation. Such a Count k is incremented by 1, and control sent back to Step 2.
flexibility is provided by the multi-objective optimization [21].
The multi-objective optimization problem is minimization or
3.2. Best compromise solution
maximization of multiple evaluation criteria having conflict with
each other. The solution which is an optimum for one criterion
After determining the Pareto optimal set of non-dominated
may not be optimal for multi-objective optimization, because the
solutions (Pareto optimal front) from SPEA 2+, fuzzy approach
multiple criteria have trade-off relationships with each other.
[27] provides the best compromise solution to the decision maker.
Strength Pareto Evolutionary Algorithm (SPEA) is a multi-objective
Considering the imprecise nature of the decision maker’s judg-
genetic algorithm, which maintains an external population at
ment, it is natural to assume that the decision maker may have fuz-
every generation, storing all non-dominated solutions obtained
zy or imprecise goals for each objective functions. The fuzzy sets
so far. At each generation, external population is mixed with the
are defined by equations called membership functions. These func-
current population. All non-dominated solutions in the mixed pop-
tions represent the degree of membership in some fuzzy sets using
ulation are assigned fitness based on the number of solutions they
values from 0 to 1. The membership value 0, indicates incompati-
dominate. Dominated solutions are assigned fitness worse than the
bility with the sets, while 1 means full compatibility. By taking ac-
worst fitness of any non-dominated solutions. A deterministic clus-
count of the minimum and maximum values of each objective
tering technique is used to ensure diversity among non-dominated
function together with the rate of increase of membership satisfac-
solutions. More details of SPEA can be found in [14,25]. SPEA 2 is an
tion, the decision maker must detect membership function l(Fi) in
improved version of SPEA, which incorporates a fine-grained fit-
a subjective manner. Here, it is assumed that l(Fi) is a strictly
ness assignment strategy, a density estimation technique, and an
monotonic decreasing and continuous function defined as
enhanced archive truncation method [26]. SPEA 2+ is a new mul-
8
ti-objective genetic algorithm that improves the search perfor- if F i 6 F min
< 1 max
> i
mance of SPEA 2. In this paper SPEA 2+ is used. After finding the Fi F i
lðF i Þ ¼ max min if F min
i < F i < F max
i ði ¼ 1; 2; . . . ; Nobj Þ ð45Þ
Pareto optimal set of non-dominated solutions using SPEA 2+, >
: F i F i
the fuzzy min–max approach [27] provides the best compromise 0 if F i P F max
i
solution to the decision maker. The brief description of SPEA 2+
where F max
i and F min
i are the maximum and minimum values of the
is presented next:
ith objective function among all the ‘K’ non-dominated solutions.
The value of membership function suggests how far (in the
3.1. Strength Pareto Evolutionary Algorithm 2+ (SPEA 2+) scale from 0 to 1) a non-inferior (non-dominated) solution has sat-
isfied the Fi objective. The sum of membership function values l(Fi)
SPEA 2+ is SPEA 2 with the addition of the following three for all the objectives can be computed in order to measure the
mechanisms [28]: accomplishment of each solution in satisfying the objectives. The
accomplishment of each non-dominated solution can be rated with
1. Mating selection which reflects all good solutions preserved in respect to all the ‘K’ non-dominated solutions by normalizing its
the archive. accomplishment over the sum of the accomplishments of ‘K’ non-
2. Neighborhood crossover to allow crossing over between indi- dominated solutions as follows [31]:
viduals located close to each other in the objective space. P
Nobj
3. Application of two archives to maintain diverse solutions in the lðF ki Þ
objective space and the design variable space.
lkD ¼ PK i¼1
PNobj k
ð46Þ
k¼1 i¼1 lðF i Þ

Mating selection is the selection of next generation’s search The function lD in Eq. (46) can be treated as a membership function
population from the archive population [29]. Ref. [29] reports for non-dominated solutions, in a fuzzy set and represented as fuz-
204 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

zy cardinal priority ranking of the non-dominated solutions. The and proposed EEA. The optimum cost obtained with EGA is
solution that attains the maximum membership lkD , in the fuzzy 799.56 $/h. As explained earlier, in this case the EGA is stopped,
set so obtained can be chosen as the ‘best’ solution or the one hav- when all the chromosomes assume similar fitness values (converg-
ing the highest cardinal priority ranking. ing criteria), which is 76 iterations. The convergence time with EGA
is 30.5933 s, and it is shown in Table 2.
MaxflkD : k ¼ 1; 2; . . . ; Kg ð47Þ By applying the proposed EEA approach, the optimum cost ob-
tained is 800.0831 $/h, in 109 iterations, which is very close to that
with the EGA. It can be seen that even with a slightly different set
4. Results and discussion
of controls in EGA and EEA, the objective function values are nearly
the same. In this case, the objective function (i.e., fuel cost) value
The effectiveness of the proposed algorithm is tested on IEEE 30,
obtained with EEA is 0.0654% higher than EGA. These things hap-
118 and 300 bus test systems. As stated earlier, the proposed Effi-
pen when we do not have a very sharp minima. This is obviously
cient Evolutionary Algorithm (EEA) approach can be used with any
because of the non-linear approximation of functions being very
evolutionary algorithm based OPF. In this paper, Enhanced Genetic
good, but still not a perfect one. Table 1 also shows, the results pre-
Algorithm (EGA) is used to develop the proposed EEA. For one case
sented in the literature with other algorithms like Biogeography
study, the proposed EEA is also developed using PSO. The parame-
Base Optimization (BBO) [15], Differential Evolution (DE) [34],
ters used in the EGA are: Population size: 60, maximum number of
PSO [6], GA, Improved GA [5], Modified Differential Evolution
generations/iterations: 200, uniform crossover probability: 0.95,
(MDE) [35], gradient method and General Algebraic Modeling Sys-
mutation probability: 0.001 and elitism index: 0.15. Here, the Rou-
tem (GAMS). From the results in Table 1, it is clear that the results
lette wheel parent selection technique is used. In EGA, in addition
obtained from proposed EEA are better than some of the reported
to the basic genetic operators, problem specific (EGA) operators [4]
results. However, it should be noted that the proposed EEA ap-
are applied with a probability of 0.5. It implies that problem spe-
proach is a generic one, and it can be applied to any better evolu-
cific operators are enforced only if the generated random number
tionary based algorithm. For example, in Table 1, the minimum
is less than 0.5. The parameters used in PSO are: Swarm size: 60,
cost obtained using BBO is 799.1116 $/h, and the proposed EEA
maximum number of generations: 100, acceleration constants
can easily applied to this BBO to get the optimum solution near
c1 = c2 = 2.05, inertia weight (w): 1.2, and constriction factor:
to BBO, with substantial reduction in execution time.
0.7295. The algorithm is stopped when all chromosomes/particles
Table 2 shows that, the number of load flows performed with
assume similar fitness values or maximum number of generations
EGA for fuel cost minimization objective are 4560 (i.e., 60  76),
reached. The optimization programs are coded in MATLAB, and
and the convergence time is 30.5933 s. Whereas, the number of
implemented on a PC-Core2 Quad computer with 3.24 GB of
load flows performed in the proposed EEA are 109 (i.e., equal to
RAM. Real and reactive power OPF results for generation cost, loss,
number of iterations). Hence, there is drastic reduction in the num-
voltage stability index objectives and multi-objective optimization
ber of load flows performed. Therefore, the convergence time with
on best combination of conflicting objectives have been obtained
EEA is 5.9694 s, i.e., the execution speed is approximately 5.1 times
with EGA and proposed EEA.
faster than the EGA.
It may be noted that non-professional, MATLAB based programs
4.1. Results for IEEE 30 bus system have been used by the authors. Hence, the computational/conver-
gence time comparison is just illustrative of the trend. However,
For IEEE 30 bus, 41 branch system, the network parameters are because of the drastic reduction in the number of load flows re-
taken from [32]. The network consists of 6 generator buses, 21 load quired in EEA as compared to that in EGA should lead to substantial
buses and 41 branches, of which 4 branches are tap setting trans- speed up, even with the professional packages, even after account-
former branches. Buses 10, 12, 15, 17, 20, 21, 23, 24 and 29 have ing for the time required for sensitivity calculations. Therefore, the
been [5] selected as shunt compensation buses. proposed EEA approach has better speed-accuracy trade-off.
GA encoding is performed using different gene lengths for each
set of control variables, depending on the desired accuracy. The 4.1.1.2. Solution quality. The comparison of best, worst and average
gene length for generator real power output is 12 bits, generator solutions for quadratic fuel cost objective function is shown in Ta-
voltage magnitude is 8 bits, and these are treated as continuous ble 3, for the proposed EEA, EGA and other evolutionary algo-
controls. The transformer tap settings can take 17 discrete values. rithms. The average value of the objective function obtained from
Each one is encoded, using 5 bits. The lower and upper limits are proposed EEA approach is very close to minimum value. This indi-
0.9 p.u. and 1.1 p.u, respectively, and the step size is 0.0125 p.u. cate the ability of the proposed EEA approach to reach either to
The bus shunt admittances can take 6 discrete values. Each one optimum value or very nearer to it in run. The obtained standard
is encoded using 3 bits; the lower and upper limits being 0.0 p.u. deviations with EGA and proposed EEA are 0.0406, 0.0407, respec-
and 0.05 p.u., respectively, with the step size of 0.01 p.u. (on sys- tively. The fuel cost difference between highest (worst) and lowest
tem MVA basis). The lower and upper limits of load bus voltages (best) cost for EGA is 0.1289$/h, and for proposed EEA is 0.1292$/h,
are 0.9 p.u. and 1.05 p.u. respectively. The generator bus voltage this confirms the effectiveness of the proposed approach.
magnitude limits considered are 0.95 p.u and 1.1 p.u. [33].
4.1.1.3. Case 2: OPF with real power loss minimization. Table 4 shows
4.1.1. Results for IEEE 30 bus system at base load condition with the control variables and objective function values with real power
quadratic fuel cost function loss minimization objective at base load condition. The minimum
In IEEE 30 bus system, at base load condition, the total demand loss obtained with EGA is 3.244 MW, whereas by using proposed
in the system is 283.4 MW, and the quadratic fuel cost coefficients EEA approach is 3.2823 MW, which is very close to the solution ob-
data is taken from [34]. Here, three different cases have been sim- tained with EGA. Table 3 also shows the optimum loss obtained
ulated, and they are described next: with GA and PSO. From this table, it can be seen that the optimum
loss obtained with EEA is very close to loss obtained with other
4.1.1.1. Case 1: OPF with real power generation cost minimiza- evolutionary algorithms, i.e., the solution obtained with EEA is near
tion. Table 1 shows the control variables and objective function to global optimum. Table 2 shows the convergence time and num-
values for quadratic fuel cost minimization objective using EGA ber of load flows performed for EGA and proposed EEA. In this case
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 205

Table 1
Control variables settings for IEEE 30 bus system with quadratic fuel cost minimization objective at base load condition.

Control variables and objective fun. Proposed EGA BBO [15] DE [34] PSO Improved GA MDE Gradient method GAMS
values EEA [6] [5] [35] [34]
PG1 (MW) 173.4593 177.285 177.0177 176.2592 176.96 177.594 175.974 187.219 177.1
PG2 (MW) 47.7363 48.93 48.641 48.5602 48.98 48.722 48.884 53.781 48.8
PG5 (MW) 23.7692 21.29 21.239 21.3402 21.30 21.454 21.51 16.955 21.4
PG8 (MW) 23.2234 20.49 21.136 22.0553 21.19 20.954 22.24 11.288 21.5
PG11 (MW) 11.3724 11.93 11.944 11.7785 11.97 11.768 12.251 11.287 12
PG13 (MW) 12.2530 12.23 12.054 12.0217 12 12.0520 12 13.355 12
V1 (p.u) 1.0994 1.098 1.1 1.0999 1.0855 1.0810 1.0500 1.1000 1.08
V2 (p.u) 1.0853 1.080 1.0876 1.089 1.0653 1.063 1.0382 1.08 1.061
V5 (p.u) 1.0506 1.053 1.0614 1.0659 1.0333 1.034 1.0113 1.03 1.032
V8 (p.u) 1.0700 1.062 1.0695 1.0697 1.0386 1.038 1.0191 1.04 1.039
V11 (p.u) 1.0735 1.08 1.0982 1.0965 1.0848 1.1 1.0951 1.08 1.018
V13 (p.u) 1.0976 1.078 1.0998 1.0996 1.0512 1.055 1.0837 1.08 1.047
T6,9 (p.u) 0.9875 0.975 1.05 1.0429 1.0233 1 0.9866 1.072 NA
T6,10 (p.u) 0.9250 1.05 0.9 0.9179 0.9557 0.975 0.9714 1.07 NA
T4,12 (p.u) 1.0375 1.0125 0.99 1.0190 0.9724 0.975 0.9972 1.032 NA
T28,27 (p.u) 1.0250 1.0125 0.97 0.9836 0.9728 1 0.9413 1.068 NA
Bsh,10 (p.u) 0.04 0.05 0.05 0.0455 0.0335 0.001 NA 0.0069 NA
Bsh,12 (p.u) 0.01 0.01 0.05 0.0442 0.0220 0.007 NA 0.0005 NA
Bsh,15 (p.u) 0.05 0.04 0.05 0.0417 0.0198 0.019 NA 0.0029 NA
Bsh,17 (p.u) 0.03 0.02 0.05 0.0252 0.0315 0.024 NA 0.0029 NA
Bsh,20 (p.u) 0.04 0.03 0.05 0.0209 0.0454 0.015 NA 0.0021 NA
Bsh,21 (p.u) 0.01 0.01 0.05 0.0420 0.0381 0.022 NA 0.0000 NA
Bsh,23 (p.u) 0.05 0.03 0.04 0.0255 0.0398 0.047 NA 0.0033 NA
Bsh,24 (p.u) 0.03 0.02 0.05 0.0438 0.0500 0.047 NA 0.0094 NA
Bsh,29 (p.u) 0.05 0.01 0.03 0.0275 0.0251 0.024 NA 0.0027 NA
Fuel Cost ($/h) 800.0831 799.56 799.1116 799.2891 800.41 800.805 802.376 804.853 801.5198
Power Loss (MW) 8.4137 8.755 8.63 8.6150 9.2169 – 9.459 10.486 9.4
VSEI (i.e., L-index) 0.1303 0.12 – 0.1226 – – – – –

Table 2
CPU time of EGA and proposed EEA algorithms for fuel cost, loss and VSEI minimization objectives.

Case 1: fuel cost min. Case 2: loss min. Case 3: VSEI min.
EGA Proposed EEA EGA Proposed EEA EGA Proposed EEA
Convergence time (s) 30.5933 5.9694 29.7118 5.7167 31.6297 6.1895
No. of iterations 76 109 75 94 79 116
No. of load flows performed 4560 109 4500 94 4740 116

Table 3 and objective function values, while optimizing VSEI minimization

Comparison among different methods for IEEE 30 bus system after 50 trials/runs (for with GA, PSO, BBO, Multi-Objective Differential Evolution (MODE),
Case 1: Quadratic fuel cost minimization).
EGA and proposed EEA approaches. The optimum value of VSEI
Algorithms Generation cost ($/h) Standard (i.e., L-index) obtained with EGA is 0.1048, whereas the optimum
deviation VSEI obtained with proposed EEA is 0.1052, which is very close
Maximum Minimum Average
(worst) (best)
to the value obtained from EGA. From Table 2, it can be seen that
the OPF with VSEI minimization considering EGA is converged in
Proposed 800.2123 800.0831 800.1730 0.0407
EEA
79 iterations, hence the number of load flows performed are
EGA 799.6889 799.5600 799.6497 0.0406 4740 (i.e., 60  79), and the execution time is 31.6297 s. But, the
BBO [15] 799.2042 799.1116 799.1985 NA proposed EEA approach is converged in 116 iterations, hence the
EP [37] 805.6100 802.6200 803.5100 NA number of load flows performed are 116 (i.e., equal to number of
SA [38] NA 799.4500 NA NA
iterations), and the execution time is 6.1895 s, i.e., the proposed
IEP [39] 802.5810 802.4650 802.5210 NA
DE [34] NA 799.2891 NA NA EEA approach is approximately 5.1 times faster than EGA.
PSO [6] NA 800.4100 NA NA
MDE [35] 802.4040 802.3760 802.3820 NA
4.1.2. Results for IEEE 30 bus system at base load condition considering
generating units with Valve Point Loading (VPL) effect
(i.e., real power loss minimization), the number of load flows per- In this subsection, the OPF problem is solved by considering fuel
formed with EGA are 4500 (i.e., 60  75), and the convergence time cost function with VPL effect. All the data related to the generating
is 29.7118 s. But, the number of load flows performed for proposed units with valve points and prohibited zones is presented in
EEA is 94 (i.e., equal to number of iterations), and the convergence Appendix (Table 15). Table 6 presents the scheduled power out-
time is 5.7167 s. Therefore, the execution speed is approximately puts and objective function values at base load condition consider-
5.2 times faster than EGA. ing generating units with VPL effect. From this table, it can be seen
that the optimum total cost obtained with proposed EEA approach
is 826.8492 $/h, which is very close (0.0643% higher) to the cost
4.1.1.4. Case 3: OPF with system Voltage Stability Enhancement Index obtained from EGA, i.e., 826.3176 $/h (here, the results of EGA
(VSEI) minimization. Table 5 shows the control variables settings are compared with EEA, as we have developed proposed EEA using
206 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

Table 4 (SFLA-SA) algorithm to get the solution near to the solution ob-
Control variables settings for IEEE 30 bus system with real power loss minimization tained from Hybrid SFLA-SA).
objective at base load condition.
From Table 7, it can be seen that the execution time required for
Control variables and objective Proposed EGA GA PSO EGA and EEA are 34.9211 s and 6.8002 s, respectively. Therefore,
fun. values EEA the proposed EEA approach is approximately 5.1 times faster than
PG1 (MW) 59.3216 51.674 56.1602 56.6613 EGA.
PG2 (MW) 74.8132 79.97 77.82 78.9597
PG5 (MW) 49.8547 50 49.94 49.1795
PG8 (MW) 34.9084 35 34.75 35
PG11 (MW) 28.1099 30 29.897 29.8242 4.1.3. Results for IEEE 30 bus system at base load condition considering
PG13 (MW) 39.7538 40 38.11 37.094 generating units with VPL effect and Prohibited Operating Zones
V1 (p.u) 1.0547 1.0518 1.058 1.0694
V2 (p.u) 1.0418 1.0488 1.051 1.0729
(POZs)
V5 (p.u) 1.0247 1.027 1.034 1.05 In this subsection, the OPF problem (fuel cost minimization) is
V8 (p.u) 1.0335 1.0306 1.042 1.0476 solved by considering VPL effect and prohibited operating zones
V11 (p.u) 1.0229 1.0612 1.089 1.0176 (POZs), simultaneously. Table 8 shows the scheduled generations
V13 (p.u) 1.0776 1.0382 1.042 1.0576
and objective function value for GA, PSO, Simulated Annealing
T6,9 (p.u) 1.0125 1.0750 1.0625 0.95
T6,10 (p.u) 0.9125 0.9500 1.0125 1.0125 (SA), Shuffle Frog Leaping Algorithm (SFLA), Hybrid SFLA-SA, EGA
T4,12 (p.u) 1.0125 0.9875 1.025 0.9875 and proposed EEA approaches. As explained earlier, here EEA is
T28,27 (p.u) 1.0125 1.0125 1.0125 1.0375 implemented with EGA, hence results obtained from proposed
Bsh,10 (p.u) 0.04 0.04 0.04 0.05 EEA and EGA should be compared. The optimum cost obtained
Bsh,12 (p.u) 0.02 0.04 0.03 0.05
Bsh,15 (p.u) 0.05 0.02 0.05 0.05
with EGA is 835.1917 $/h, and with proposed EEA is 835.7822 $/
Bsh,17 (p.u) 0.01 0.05 0.02 0.03 h, which is 0.0707% higher than cost obtained from EGA. Table 7
Bsh,20 (p.u) 0.05 0.03 0.04 0.04 shows the convergence time, number of load flows performed in
Bsh,21 (p.u) 0.00 0.01 0.02 0.05 EGA and EEA approaches. From this table, it can be seen that the
Bsh,23 (p.u) 0.02 0.00 0.05 0.02
execution speed with proposed EEA approach is approximately
Bsh,24 (p.u) 0.05 0.03 0.00 0.00
Bsh,29 (p.u) 0.02 0.04 0.00 0.01 5.1 times faster than EGA. From Table 8, it can be seen that the to-
Fuel cost ($/h) 952.3785 967.93 957.814 954.3483 tal cost obtained with Hybrid SFLA-SA approach is 834.6339 $/h,
Power loss (MW) 3.2823 3.244 3.2772 3.318 which is less than all other algorithms. The proposed EEA approach
VSEI 0.1533 0.1510 0.1638 0.1246 is generic one, and can also be applied to Hybrid SFLA-SA algo-
rithm, to get the optimum solution near to the solution obtained
from Hybrid SFLA-SA.
The important feature of EAs, i.e., multi-objective optimization
EGA). From the Table 6, it is clear that the optimum cost obtained
can also be applied to the proposed EEA approach to get the best
with proposed EEA is much better than some of the evolutionary
compromise solution. The choice of appropriate objectives is very
algorithms. However, it should be noted that, the proposed EEA ap-
critical for multi-objective optimization. Generator fuel cost is
proach is generic one, and can be implemented with any better
important objective under all situations. Under stressed loading
evolutionary algorithm (For example, it can be implemented with
condition, saving the system from voltage collapse is more impor-
Hybrid Shuffle Frog Leaping Algorithm – Simulated Annealing
tant. Therefore, under stressed loading condition, we consider gen-

Table 5
Control variables settings for IEEE 30 bus system with VSEI minimization objective at base load condition.

Control variables and objective fun. values Proposed EEA EGA GA PSO BBO [15] MODE [36]
PG1 (MW) 106.8846 102.91 117.971 133.8369 99.415 127.12
PG2 (MW) 75.033 79.824 76.13 55 34.794 38.85
PG5 (MW) 48.9231 50 30.99 37.858 49.901 44.76
PG8 (MW) 34.0904 35 33.43 29.02 34.831 34
PG11 (MW) 11.6361 10 19.099 19.586 29.569 26.69
PG13 (MW) 12.9846 12.4380 13.832 16.92 39.995 17.38
V1 (p.u) 1.0718 1.0340 1.04 1.00 1.0995 1.07
V2 (p.u) 1.0965 1.0430 1.0570 1.034 1.0822 1.052
V5 (p.u) 1.0953 1.0670 1.0718 1.046 1.0738 1.061
V8 (p.u) 1.0994 1.0212 1.0223 1.02 1.0499 1.040
V11 (p.u) 1.1000 1.0253 1.0248 1.0117 1.0837 1.098
V13 (p.u) 1.0906 1.0447 1.0450 1.0528 0.9640 1.052
T6,9 (p.u) 0.9125 0.9125 0.9250 0.9000 1.0999 1.039
T6,10 (p.u) 0.9000 0.9000 0.9125 0.9500 1.0999 0.959
T4,12 (p.u) 0.9000 0.9000 0.9000 0.9250 1.1000 0.996
T28,27 (p.u) 1.0125 0.9250 1.0750 0.9250 0.9025 0.982
Bsh,10 (p.u) 0.03 0.00 0.00 0.04 0.0477 0.0405
Bsh,12 (p.u) 0.04 0.05 0.05 0.04 0.0495 0.0442
Bsh,15 (p.u) 0.03 0.01 0.05 0.03 0.0475 0.0419
Bsh,17 (p.u) 0.05 0.02 0.03 0.05 0.0471 0.0498
Bsh,20 (p.u) 0.03 0.04 0.02 0.01 0.0294 0.0486
Bsh,21 (p.u) 0.05 0.05 0.03 0.04 0.0495 0.049
Bsh,23 (p.u) 0.05 0.04 0.04 0.02 0.0494 0.0496
Bsh,24 (p.u) 0.03 0.02 0.05 0.04 0.0483 0.049
Bsh,29 (p.u) 0.02 0.04 0.05 0.01 0.0481 0.049
Fuel cost ($/h) 886.9053 900.64 844.473 837.06 917.3597 856.9
Power loss (MW) 6.1507 6.7720 8.052 8.8209 4.95 5.4
VSEI 0.1052 0.1048 0.1133 0.1106 0.098 0.1246
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 207

Table 6
Scheduled power outputs with different algorithms for generation cost minimization objective considering valve point loading (VPL) effect.

Algorithm PG1 (MW) PG2 (MW) PG3 (MW) PG4 (MW) PG5 (MW) PG6 (MW) Fuel cost ($/h)
RCGA [40] 198.81 38.96 19.16 10.64 13.56 12.03 831.04
DE [40] 199.13 38.32 20.17 11.43 10.43 12.66 826.54
GA [13] 222.164 22.9946 15 10 10 17.1568 829.4493
PSO [13] 227.0399 20 16.707 10 10.3476 12 826.5898
SA [13] 216.8178 25.5491 15.9877 15.9876 10 12.017 827.8263
SFLA [13] 223.4773 22.1001 17.1164 10.9885 10.1779 12.4697 825.9906
Hybrid SFLA-SA [13] 223.2516 24.3121 17.0479 10 10 12.0294 825.6922
EGA 219.8096 26.1538 16.0427 11.3675 10.0098 12.1094 826.3176
Proposed EEA 219.501 26.3297 17.4701 10.0855 10.2442 12.0615 826.8492

Table 7 4.2.2. Case 2: OPF with system VSEI minimization

CPU time of EGA and proposed EEA algorithms for fuel cost minimization objective Table 9 also shows the scheduled power generations and objec-
considering VPL effect and POZs.
tive function values when VSEI is optimized independently. In this
Generating units with Generating units with case, the optimum value of VSEI obtained by using EGA is 0.4749,
VPL effect VPL effect and POZs which is converged in 105.3531 s. By using the proposed EEA
EGA Proposed EGA Proposed approach, the obtained VSEI is 0.4774, which is quite close to that
EEA EEA with the EGA, but the execution speed is approximately 5.2 times
Convergence time (s) 34.9211 6.8002 36.8067 7.1784 faster than the EGA.
No. of iterations 87 131 94 140 From the above cases 1 and 2, it can be seen that when real
No. of load flows 5220 131 5640 140
power generation cost is optimized independently, then the fuel
performed
cost is optimum, but VSEI is near to voltage collapse point. Simi-
larly, when VSEI is optimized independently, the obtained VSEI va-
lue is minimum, but fuel cost is more. Hence, there exists a conflict
eration cost and VSEI are the best combination of objectives to be between objective function values, when one objective is opti-
optimized simultaneously. mized independently. Therefore, real power generation cost and
VSEI need to be optimized simultaneously. For the multi-objective
optimization we employ SPEA 2+ approach, and to get the best
compromise solution, we employ fuzzy min–max approach, as de-
4.2. Results on IEEE 30 bus system at stressed load condition
scribed in Section 3.
Here, the stressed case (near critical loading) condition is ob-
tained by taking line 36 out (connecting buses 27 and 28), which 4.2.3. Case 3: real power generation cost and VSEI minimization using
is found to be worst contingency in IEEE 30 bus system and by SPEA 2+
increasing the base case load. At stressed loading condition, the to- Table 9 also shows the scheduled power generations and objec-
tal demand in the system is 357.084 MW. In this subsection, gen- tive function values for Case 3. The best compromise solution ob-
erating units with VPL effects are considered. Here, three tained with SPEA 2+ applied to EGA has generation cost of
different cases have been simulated, and they are described next: 1175.5252 $/h and VSEI of 0.6140. The convergence time for SPEA
2+ with EGA is 162.2885 s. The best compromise solution obtained
with SPEA 2+ applied to EEA has generation cost of 1178.0309 $/h
and VSEI of 0.6128. The convergence time for SPEA 2+ with EEA is
4.2.1. Case 1: OPF with real power generation cost minimization 31.3504 s. From this it can be seen that, the best compromise solu-
Table 9 shows the scheduled power generations and objective tion obtained using two approaches are almost same, and the exe-
function values for fuel cost minimization objective considering cution speed obtained with EEA is approximately 5.2 times faster
generating units with valve point effects using EGA and proposed than EEA. Hence, the proposed EEA has better speed-accuracy
EEA. The minimum cost obtained with EGA is 1130.8185 $/h. By trade-off.
applying the proposed EEA approach, the obtained minimum cost
is 1131.5169 $/h, which is 0.062% more than the EGA, but the exe- 4.3. Application of proposed EEA approach to Particle Swarm
cution speed is approximately 5.1 times faster than the EGA. As Optimization (PSO)
discussed earlier, in this case both the algorithms are stopped
when maximum number of iterations are reached, i.e., 200. When As explained earlier, the proposed EEA approach is generic one,
generation cost is optimized independently, the value obtained for and can be applied to any evolutionary algorithm based OPF. Here,
VSEI is 0.8938, i.e., system is near to voltage collapse point. the proposed EEA is implemented using PSO. The flow chart of pro-

Table 8
Scheduled power outputs with different algorithms for generation cost minimization objective considering valve point loading (VPL) effect and prohibited operating zones (POZs).

Algorithm PG1 (MW) PG2 (MW) PG3 (MW) PG4 (MW) PG5 (MW) PG6 (MW) Fuel cost ($/h)
GA [13] 226.1783 24.3344 16.9393 10 10.1525 12 838.1727
PSO [13] 221.1471 30.8367 15 10 10 12 835.4786
SA [13] 217.6117 32.7708 16.5177 10 10 12 836.5364
SFLA [13] 219.4201 31.1173 15.8535 10.0415 10 12 834.8166
Hybrid SFLA-SA [13] 219.816 29.7707 16.667 10 10 12 834.6339
EGA 219.3148 24.7033 17.7778 13.3211 10.337 12 835.1917
Proposed EEA 199.3421 44 15.5556 10.1465 12.7057 12.9709 835.7822
208 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

Table 9
Results on IEEE 30 bus system at stressed load condition.

Objective function values Case 1: cost min. Case 2: VSEI min. Case 3: cost and VSEI min.
EGA EEA EGA EEA EGA EEA
PG1 (MW) 219.994 219.7904 166.5046 156.3465 186.539 190.7915
PG2 (MW) 72.381 74.7106 71.8095 75.9707 74.652 79.956
PG5 (MW) 22.0855 21.8291 49.8205 45.1453 19.3162 36.2564
PG8 (MW) 31.2515 30.0488 34.9206 34.0415 34.1148 26.5934
PG11 (MW) 14.8205 15.6264 25.9219 24.1978 29.9121 24.7985
PG13 (MW) 13.935 12.6427 20.1778 33.1009 28.6291 24.7985
Fuel cost ($/h) 1130.8185 1131.5169 1215.127 1212.6803 1175.5252 1178.0309
Loss (MW) 17.3833 17.5638 12.0688 11.7165 16.0777 16.5785
VSEI 0.8938 0.8837 0.4749 0.4774 0.6140 0.6128
Convergence time (s) 95.2582 18.5892 105.3531 20.0952 162.2885 31.3504
No. of iterations 200 200 200 200 200 200
No. of load flows performed 12,000 200 12,000 200 12,000 200

posed EEA approach using PSO is shown in Fig. 4. The results for
IEEE 30 bus system using PSO and proposed EEA applied to PSO
are presented next:

4.3.1. Case 1: OPF with real power generation cost minimization

Table 10 shows the control variables and objective function val-
ues for fuel cost minimization objective with PSO and proposed
EEA approach applied to PSO. The optimum cost obtained with
PSO is 800.41 $/h, where as optimum cost obtained with the EEA
applied to PSO is 800.917 $/h, which is very close (0.0633% higher)
to the solution obtained from PSO. The number of load flows per-
formed in OPF with PSO is 4440 (i.e., 60  74), and the execution
time is 28.9625 s. Whereas, by using EEA applied to PSO, the num-
ber of load flows are reduced to 106 (i.e., equal to number of iter-
ations), and the execution speed has decreased to 5.7802 s.
Therefore, the execution speed with EEA is approximately 5.1
times faster than the EGA. Hence, the proposed EEA approach has
better speed-accuracy tradeoff.

4.3.2. Case 2: OPF with real power loss minimization

Table 10 also shows the control variables and objective function
values for real power loss minimization objective with PSO and
proposed EEA approach applied to PSO. The optimum loss obtained
with PSO is 3.318 MW, and this OPF is converged in 28.3051 s.
Whereas, the optimum loss obtained with EEA applied to PSO is
3.3541 MW, which is very close to the solution obtained from
EGA, and the execution time has decreased substantially to
5.4419 s. Hence, the execution speed with EEA is approximately Fig. 4. Flow Chart of OPF using Proposed Efficient Evolutionary Algorithm (EEA)
5.2 times faster than the EGA. considering Particle Swarm Optimization (PSO).
From the above case studies on IEEE 30 bus system it is clear
that, the solution obtained by EGA and proposed EEA are very close
to each other. But, the execution speed with proposed EEA is mum cost obtained is 49558.3605 $/h, which is 0.0551% higher
approximately 5.1 times faster than EGA. than the EGA. The execution time for the EEA is 40.2761 s. Hence,
the execution speed with EEA is approximately 4.9 times faster
4.4. Results for IEEE 118 Bus system than the EGA. Therefore, the proposed EEA approach has better
speed-accuracy trade-off.
IEEE 118 bus test system [41] consists of 54 generator buses,
and 186 branches, of which 9 branches are tap setting transformer
branches, and 13 buses have been selected as shunt compensation
buses. OPF results for IEEE 118 bus system considering real power
generation cost and loss minimization objectives are presented
next: 4.4.2. OPF with real power loss minimization
Table 12 shows the objective function values for IEEE 118 bus
4.4.1. OPF with real power generation cost minimization system with real power loss minimization as objective using EGA
Table 11 shows the objective function value for IEEE 118 bus and proposed EEA approaches. The minimum real power loss ob-
system with real power generation cost minimization as objective tained with the EGA is 31.3519 MW. By applying the proposed
using EGA and proposed EEA approaches. The minimum cost ob- EEA approach, the minimum loss obtained is 31.9724 MW, which
tained with the EGA is 49531.0613 $/h, and the convergence time is quite close to that with the EGA. But, the execution speed is
of 195.6183 s. By applying the proposed EEA approach, the mini- approximately 4.9 times faster than the EGA.
 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210 209

Table 10
Control variables settings for IEEE 30 bus system using PSO and proposed EEA approach applied to PSO.

Control variables and objective function values Case 1: fuel cost min. Case 2: loss min.
PSO [6] Proposed EEA with PSO PSO Proposed EEA with PSO
PG1 (MW) 176.96 176.4511 56.6613 59.3216
PG2 (MW) 48.98 48.4249 78.9597 74.8132
PG5 (MW) 21.3 21.1624 49.1795 49.8547
PG8 (MW) 21.19 17.3871 35 34.9084
PG11 (MW) 11.97 15.7631 29.8242 28.1099
PG13 (MW) 12 13.2581 37.094 39.7538
V1 (p.u) 1.0855 1.0965 1.0694 1.0547
V2 (p.u) 1.0653 1.0794 1.0729 1.0418
V5 (p.u) 1.0333 1.0329 1.0500 1.0247
V8 (p.u) 1.0386 1.0582 1.0476 1.0335
V11 (p.u) 1.0848 1.0512 1.0176 1.0229
V13 (p.u) 1.0512 1.023 1.0576 1.0776
T6,9 (p.u) 1.0233 0.975 0.95 1.0125
T6,10 (p.u) 0.9557 1.0750 1.0125 0.9125
T4,12 (p.u) 0.9724 1.0375 0.9875 1.0125
T28,27 (p.u) 0.9728 1.0125 1.0375 1.0125
Bsh,10 (p.u) 0.0335 0.04 0.05 0.04
Bsh,12 (p.u) 0.0220 0.02 0.05 0.02
Bsh,15 (p.u) 0.0198 0.02 0.05 0.05
Bsh,17 (p.u) 0.0315 0.05 0.03 0.01
Bsh,20 (p.u) 0.0454 0.04 0.04 0.05
Bsh,21 (p.u) 0.0381 0.03 0.05 0.00
Bsh,23 (p.u) 0.0398 0.01 0.02 0.02
Bsh,24 (p.u) 0.0500 0.00 0.00 0.05
Bsh,29 (p.u) 0.0251 0.02 0.01 0.02
Fuel cost ($/h) 800.41 800.917 954.3483 952.3785
Power loss (MW) 9.2169 9.047 3.318 3.3541
Convergence time (s) 28.9625 5.7802 28.3051 5.4419
No. of iterations 74 106 72 102
No. of load flows performed 4440 106 4320 102

4.5. Results for IEEE 300 bus system

Table 14
Real power loss minimization with EGA and Proposed EEA for IEEE 300 bus system.
IEEE 300 bus test system [41] consists of 69 generator buses,
and 411 branches, of which 62 branches are tap setting trans- Objective function value EGA Proposed EEA
former branches, and 12 buses have been selected as shunt com- Total system losses (MW) 646.2998 650.6027
pensation buses. OPF results for IEEE 300 bus system considering Convergence time (s) 894.1152 167.5691
No. of iterations 200 200
No. of load flows performed 12,000 200

Table 11
Real power generation cost minimization with EGA and proposed EEA for IEEE 118
bus system. Table 15
Generator fuel cost coefficients with valve point effects and prohibited zones for IEEE
Objective function value EGA Proposed EEA 30 bus system.
Fuel cost ($/h) 49531.0613 49558.3605
Generator a b c d e Pmin Pmax Prohibited
Convergence time (s) 195.6183 40.2761
no. zones
No. of iterations 200 200
No. of load flows performed 12,000 200 1 0 2 0.00375 18 0.037 50 250 [55–66],
[80–120]
2 0 1.75 0.01750 16 0.038 20 80 [21–24],
[45–55]
Table 12 5 0 1 0.06250 14 0.040 15 50 [30–36]
Real power loss minimization with EGA and Proposed EEA for IEEE 118 bus system. 8 0 3.25 0.00830 12 0.045 10 35 [25–30]
11 0 3 0.02500 13 0.042 10 30 [25–28]
Objective function value EGA Proposed EEA
13 0 3 0.02500 13.5 0.041 12 40 [24–30]
Total system losses (MW) 31.3519 31.9724
Convergence time (s) 236.3497 48.5099
No. of iterations 200 200
No. of load flows performed 12,000 200 real power generation cost and loss minimization objectives are
presented next:

Table 13 4.5.1. OPF with real power generation cost minimization

Real power generation cost minimization with EGA and proposed EEA for IEEE 300
Table 13 shows the objective function value for IEEE 300 bus
bus system.
system with real power cost minimization as objective using EGA
Objective Function Value EGA Proposed EEA and proposed EEA approaches. The minimum cost obtained with
Fuel cost ($/h) 834442.6782 834986.7348 EGA is 834442.6782 $/h, and the convergence time of 823.0461 s.
Convergence time (s) 823.0461 154.9357 By applying the proposed EEA approach, the minimum cost ob-
No. of iterations 200 200
tained is 834986.7348 $/h, which is 0.0652% higher than EGA.
No. of load flows performed 12,000 200
The execution time for EEA is approximately 5.3 times faster than
210 S. Surender Reddy et al. / Electrical Power and Energy Systems 54 (2014) 198–210

EGA. Therefore, the proposed EEA approach has better speed-accu- [11] Saber AY, Senjyu T, Yona A, Funabashi T. Unit commitment computation by
fuzzy adaptive particle swarm optimisation. IET Gener Transm Distrib
racy trade-off.
2007;1:456–65.
[12] Niknam T, Narimani MR, Aghaei J, Azizipanah-Abarghooee R. Improved
4.5.2. OPF with real power loss minimization particle swarm optimisation for multi-objective optimal power flow
Table 14 shows the objective function value for IEEE 300 bus considering the cost, loss, emission and voltage stability index. IET Gener
Transm Distrib 2012;6:515–27.
system with real power loss minimization as objective function [13] Niknam T, Narimani MR, Azizipanah-Abarghooee R. A new hybrid algorithm
using EGA and proposed EEA approaches. The minimum real power for optimal power flow considering prohibited zones and valve point effect.
loss obtained with EGA is 646.2998 MW. By applying the proposed Energy Convers Manage 2012;58:197–206.
[14] Deb K. Multi-objective optimization using evolutionary algorithms. John Wiley
EEA approach, the minimum loss obtained is 650.6027 MW, which and Sons; 2001.
is quite close to that with the EGA. But, the execution speed is [15] Bhattacharya A, Chattopadhyay PK. Application of biogeography-based
approximately 5.3 times faster than EGA. optimisation to solve different optimal power flow problems. IET Gener
Transm Distrib 2011;5:70–80.
All the case studies on the three test systems indicate that the [16] Varadarajan M, Swarup KS. Solving multi-objective optimal power flow using
proposed Efficient Evolutionary Algorithm (EEA) overcomes the differential evolution. IET Gener Transm Distrib 2008;2:720–30.
main drawback (execution time) of evolutionary algorithms, i.e., [17] Chaturvedi KT, Pandit M, Srivastava L. Particle swarm optimization with time
varying acceleration coefficients for non-convex economic power dispatch.
the proposed EEA is approximately 5 times faster than any Electr Power Energy Syst 2009;31:249–57.
conventional evolutionary algorithm. The minor difference in the [18] Labbi Y, Ben Attous D. Big bang-big crunch optimization algorithm for
objective function value is the only price to be paid for the economic dispatch with valve-point effect. J Theor Appl Inf Technol
2010:48–56.
speed-accuracy trade-off.
[19] Mahdad B, Srairi K, Bouktir T, Benbouzid ME. Optimal power flow with
discontinuous fuel cost functions using decomposed GA coordinated with
5. Conclusions shunt FACTS. J Electr Eng Technol 2009;4:457–66.
[20] IEEE tutorial course on optimal power flow: solution techniques, requirements
and challenges; 1996.
This paper proposes a new Efficient Evolutionary Algorithm [21] Reddy SS, Abhyankar AR, Bijwe PR. Reactive power price clearing using multi-
(EEA) to solve the Optimal Power Flow (OPF) problem. It uses objective optimization. Energy 2011;36:3579–89.
[22] Kessel P, Glavitsch H. Estimating the voltage stability of a power system. IEEE
incremental power flow model and sensitivities. The proposed ap- Trans Power Delivery 1986;1:346–55.
proach uses the concepts from evolutionary algorithms and classi- [23] Bansilal, Thukaram D, Parthasarathy. Optimal reactive power dispatch
cal optimization algorithms. Here, the OPF is performed taking into algorithm for voltage stability improvement. Electr Power Energy Syst
1996;18:461–8.
consideration the minimization of real power generation cost, sys-
[24] Thukaram D, Jenkins L, Visakha K. Load modeling at electric power distribution
tem losses and voltage stability index while satisfying all the substations using dynamic load parameters estimation. IEE Proc Gener Transm
equality and inequality constraints. OPF problem using proposed Distrib 2006;153:237–46.
[25] Abido MA. Environmental/economic power dispatch using multi-objective
EEA approach applied to Enhanced Genetic Algorithm (EGA) and
evolutionary algorithms. IEEE Trans Power Syst 2003;18:1529–37.
Particle Swarm Optimization (PSO) have been solved and com- [26] Zitzler E, Laumanns M, Thiele L. SPEA 2: Improving the Performance of the
pared with conventional evolutionary algorithms. IEEE 30, 118 Strength Pareto Evolutionary Algorithm. Technical Report 103, computer
and 300 bus systems are selected to demonstrate the effectiveness engineering and communication networks lab (TLK). Swiss Federal Institute of
Technology (ETH), Zurich; 2001.
of the proposed EEA approach. All the investigations indicate that [27] Abido MA, Bakhashwain JM. Optimal VAR dispatch using a multi-objective
the proposed EEA is approximately 5 times faster than any conven- evolutionary algorithms. Electr Power Energy Syst 2005;27:13–20.
tional evolutionary algorithm, with insignificant difference in the [28] Hiroyasu T, Nakayama S, Miki M. Comparison Study of SPEA 2+, SPEA 2, and
NSGA-II in diesel engine emissions and fuel economy problem. In: IEEE
objective function value. congress on evolutionary computation; 2005. p. 236–42.
[29] Watanabe S, Hiroyasu T, Miki M. Neighborhood cultivation genetic algorithm
for multi-objective optimization problems. In: Proc of the 4th Asia-Pacific
Appendix A. IEEE 30 bus cost coefficients
conference on Simulated Evolution And Learning (SEAL-2002); 2002. p. 198–
202.
See Table 15. [30] Kim M, Hiroyasu T, Miki M. SPEA 2+: improving the performance of the
strength Pareto evolutionary Algorithm 2. Parallel Problem Solving from
Nature-PPSN VIII; 2004. p. 742–51.
References [31] Kothari DP, Dhillon JS. Power system optimization. New Delhi: PHI Learning
Pvt. Ltd.; 2011.
[1] Wood AJ, Bruce F. Power generation operation and control. John Wiley and [32] Alsac O, Scott B. Optimal power flow with steady state security. IEEE Trans
Sons; 1996. Power Ap Syst 1974;93:745–51.
[2] Carpentier J. Contibution a. ‘l’etude du dispatching economique. Bull Soc [33] Reddy SS, Abhyankar AR, Bijwe PR. Multi-objective day-ahead real power
Francaise Elect 1962;3:431–47. market clearing with voltage dependent load models. Int J Emerg Electric
[3] Osman MS, Abo-Sinna MA, Mousa AA. A solution to the optimal power flow Power Syst 2011;12:1–22.
using genetic algorithm. Appl Math Comput 2004;155:391–405. [34] Abou El Ela AA, Abido MA, Spea SR. Optimal power flow using differential
[4] Bakirtzis AG, Biskas PN, Zoumas CE, Petridis V. Optimal power flow by evolution algorithm. Electric Power Syst Res 2010;80:875–85.
enhanced genetic algorithm. IEEE Trans Power Syst 2002;17:229–36. [35] Sayah S, Zehar K. Modified differential evolution algorithm for optimal power
[5] Lai LL, Ma JT, Yokoyama R, Zhao M. Improved genetic algorithms for optimal flow with non-smooth cost functions. Energy Convers Manage
power flow under both normal and contingent operation states. Electr Power 2008;49:3036–42.
Energy Syst 1997;19:287–92. [36] Abido MA, Al-Ali NA. Multi-objective differential evolution for optimal power
[6] Abido MA. Optimal power flow using particle swarm optimization. Electr flow. In: Power Engineering, Energy and Electrical Drives (POWERENG 2009).
Power Energy Syst 2002;24:563–71. Lisbon, Portugal; 2009. p. 101–06.
[7] Kennedy J, Eberhart R. Particle swarm optimization. In: IEEE international [37] Yuryevich J, Wong KP. Evolutionary programming based optimal power flow
conference on neural networks; 1995. p. 1942–48. algorithm. IEEE Trans Power Syst 1999;14:1245–50.
[8] Kennedy J. Stereotyping: Improving particle swarm performance with cluster [38] Roa-Sepulveda CA, Pavez-Lazo BJ. A solution to the optimal power flow using
analysis. In: Cong. on evolutionary computation. LA Jolla, CA; 2000. p. 1507– simulated annealing. Electr Power Energy Syst 2003;25:47–57.
11. [39] Ongsakul W, Tantimaporn T. Optimal powers flow by improved evolutionary
[9] Dutta S, Singh SP. Optimal rescheduling of generators for congestion programming. Electr Power Comp Syst 2006;34:79–95.
management based on particle swarm optimization. IEEE Trans Power Syst [40] Basu M. Multi-objective optimal power flow with FACTS devices. Energy
2008;23:1560–9. Convers Manage 2011;52:903–10.
[10] Gnanadass R, Ajjarapu V. Assessment of dynamic available transfer capability [41] http://www.ee.washington.edu/research/pstca.
using FDRPSO algorithm. Elektrika 2008;10:20–5.