J Electr Eng Technol Vol. 8, No.

5: 1012 1020, 2013 ISSN(Print) 1975 0102

http://dx.doi.org/10.5370/JEET.2013.8.5.1012 ISSN(Online) 2093 7423

! "
#

" – In this work, Self adaptive Differential Evolutionary (SaDE) algorithm is proposed to
solve Optimal Location and Size of Capacitor (OLSC) problem in radial distribution networks. To
obtain the SaDE algorithm, two improvements have been applied on control parameters of mutation
and crossover operators. To expand the study, three load conditions have been considered, i.e.,
constant, varying and effective loads. Objective function is introduced for the load conditions. The
annual cost is fitness of problem, in addition to this cost, CPU time, voltage profile, active power loss
and total installed capacitor banks and their related costs have been used for comparisons. To confirm
the ability of each improvements of SaDE, the improvements are studied both in separate and
simultaneous conditions. To verify the effectiveness of the proposed algorithm, it is tested on IEEE 10
bus and 34 bus radial distribution networks and compared with other approaches.

# $ : Annual cost, Differential evolutionary algorithm, Optimal capacitor allocation, Radial

distribution networks

%& ' out similar work in [2]. Jabr has proposed a two stage
technique to minimize OLSC problem in the presence of
Majority of loads in power systems use reactive power. fixed and switching capacitors [3]. In first and second
In a power system, active power in several MWs is only stages, OLSC problem is formulated as a conic program
generated by synchronous generators, while reactive power and a mixed integer linear program (MILP) based on
is produced not only by synchronous generator but also is minimizing the L1 norm, respectively.
injected by the other devices such as: Static VAr The heuristic methods are in the other category of
Compensator (SVC), Synchronous Condenser (SC), and methods which have been suggested for solving OLSC
Capacitor. Among these equipments, capacitor has the problem in [4, 5]. The proposed technique in [4] uses the
slowest and stepped speed response while installation and solution from the mathematical model after relaxing the
operating costs of capacitor are considerably lower than the integrality of the discrete variables to elect candidate buses
other reactive power sources. Despite technical limits of for installation capacitor banks. Da Silva et al. have
capacitors, a capacitor could be a better option to generate proposed a technique by nonlinear mixed integer
reactive power at least for the economic advantages. The optimization to solve OLSC problem. For this, sigmoid
Optimal Location and Size of Capacitor (OSLC) problem function was used to determine capacitor location and then
has been solved by many techniques. In this paper, these the problem is formulated using the primal dual interior
techniques have been categorized in three classes; point method [5]. In addition to heuristic methods, meta
numerical and mathematical methods, heuristic and heuristic approaches have been proposed in [6, 7]. Also, in
artificial intelligence techniques. [8], Memetic Algorithm (MA) was proposed to solve
In [1 3], a set of numerical programming approaches OLSC problem in large distribution networks. The MAs
have been proposed. To solve OLSC problem a are population based methods which can be taken as an
computational method were suggested in [1], in first step extension of Genetic Algorithms (GAs).
of this method the candidate buses for capacitor installation, The artificial intelligence is third category of approaches
optimal size and proper type of capacitors (fixed or for solving OLSC problem. The most used artificial
switching) are selected. In 2008, Khodr and et al. carried intelligences to solve OLSC problem are Evolutionary
† Corresponding Author: Young Researchers Club, Ardabil Branch, Algorithm (EA), Swarm Intelligence Algorithm (SIA),
Islamic Azad University, Ardabil, Iran. ([email protected]) Neural Networks (NNs) and fuzzy sets. Also, in [9 11], GA
* Young Researchers Club, Ardabil Branch, Islamic Azad University, or its improved branches were suggested to solve optimal
Ardabil, Iran.
** Department of Electrical Engineering, Ahar branch, Islamic Azad capacitor placement. Main problem of GA is low
University, Ahar, Iran. (mb ahmadi@iau ahar.ac.ir) convergence velocity, which authors of [12] have claimed
*** Young Researchers Club, Parsabad Moghan Branch, Islamic Azad that this problem is declined by methods based on the
University, Parsabad Moghan, Iran. ([email protected])
Received: July 15, 2012; Accepted: May 13, 2013 reduction of the search space of GAs or based on micro
 Mohammadhosein Dideban, Noradin Ghadimi, Mohammad Bagher Ahmadi and Mohammmad Karimi

GAs. Abu Elanien and Salama in [13] have suggested (&%

Discrete Wavelet Transform (DWT) integrated with a feed
forward artificial neural network (FFANN) for PQ In this load pattern, it is assumed that load of system is
improvement and solving OLSC problem, in simultaneous constant. This condition is the simplest pattern. First term
manner. Main problems of NNs are: network training is of objective function is capacity of installed capacitor
very difficult, obtained solutions’ accuracy is strongly banks multiplied by corresponding cost. The second term is
depend on the size of training set, and finally predicting the total power loss of network multiplied by related cost.
future performance of the network (by popularity) is not
possible.  NC  NB −1 
Particle Swarm Optimization (PSO) is the most famous Min  ∑ Coper × QCi  + CPloss × ∑ PLoss  (1)
member of SIA. In [14], similar to [13], this work was  i =1  i =1 
performed using PSO. Hybrid PSO (HPSO) has been used
to solve OLSC problem in unbalanced distribution in the where, Coper and CPLoss are costs of power loss, in $/kW/
presence of harmonics in [15]. This PSO obtained by year, and operation of each capacitor bank, in $/kVAr,
combining PSO and radial distribution power flow respectively. QCi and PLoss are capacity of capacitor bank, in
algorithm. Main disadvantages of PSO are: high possibility kVAr, and the total active power loss of network, in kW,
of lying on local optimum point, especially in problems respectively. NC and NB are the total the number of
with large size and dimensions. Ant colony is another capacitors and bus, respectively.
approach among SIA used by Chang for reconfiguration,
capacitor placement and for loss reduction. Theoretical (&( *
discussion of ant colony is difficult and is time consuming
for convergence. The second load pattern is varying load. In this load
Fuzzy is free of problem structure and can be combined condition, the load of network changes in duration year.
with other algorithms, then, in [16, 17], fuzzy set has been For this, several load levels and related durations are
composed with ant colony, Immune Algorithm (IA) and defined. The difference between objective function of
GA, respectively. Creating membership function of fuzzy constant and varying load pattern is the second term of
sets is difficult and in most cases is not viable. objective function. In varying load, to apply duration of
In this paper, a novel algorithm is used to solve OLSC each load level, energy loss is inserted in objective function.
problem. The proposed SaDE algorithm is obtained by
applying two improvements on original DE algorithm.  NC  NLL NB −1 
Main goal of these improvements is self adapting of two ( )
Min  ∑ Coper × QCi  + ∑ CEloss ∑ PLoss ,Th × Th  (2)
 i =1  h =1 i =1 
important control parameters of mutation and crossover
operators. The fitness is a function of annual cost which is
where, PLoss,T and CEloss are power loss of any load level, in
presented for two scenarios; constant and varying load
kW, and cost per energy loss, in $/kWh/year. Th is the
conditions. The system load has been modeled in three
duration of hth load level. NLL is yearly total number of
patterns; constant, varying and effective patterns. In addition
load levels&
to objective function, in case studies, voltage profile and
power loss and its related cost and total installed capacitor
banks and its corresponding cost and CPU time have been (&+ , -
used for comparison criteria. To illustrate effect of each
improvement, the results of these improvements have been The load levels of varying load condition has an
presented separately and compared with SaDE. Simulations effective level which its value is calculated by Eq. (3) and
have been implemented on IEEE 10 bus and 34 bus radial applied in Eq. (1),
distribution networks. NLL
Ti × Si
S Eff (i ) = ∑ NLL
(3)
(&
i =1
∑T
i =1
i

)
where, Ti and Si are duration of the ith load level, in hour,
The optimal location and sizing of capacitor problem has and the ith load level, in pu. Several constrains should be
been formulated with different goals. In majority studies, considered to solve OLSC which are visible in [18]
main target of capacitor installation is minimizing annual
cost. In this paper, OLSC problem has been formulated as
function of annual cost. In this study, to model different +& ,- "
load conditions, three load patterns used; constant, varying
and effective load. The Differential Evolutionary (DE) algorithm was
 Optimal Location and Sizing of Shunt Capacitors in Distribution Systems by Considering Different Load Scenarios

proposed by Storn and Price in 1997 [19]. To start acceptable solution is obtained, this termination criterion is
algorithm, first an initial population is generated and then used when optimal solution of problem is given. (ii) The
to find global optimal point using mutation and crossover number of iteration reaches to the predetermined iteration
operators, the population are changed. In following quintet number. In all optimization problem solution this criterion
steps have been presented. is used.
.' To generate initial matrix, thee lower
and upper limitations are selected. The difference of these
limitations multiplied to a random value in range [0, 1] and 0& 1 - ,-
then summed with upper boundary. "

uiG, k = uk min + rand1[0,1] × (uk max − uk min ) , i ∈ [1, PN ] , (4) In many studies, better solutions have been extracted
k ∈ [1, VN ] from original DE algorithm by applying improvements on
simple DE. In general, applied improvements on original
In Eq. (9) ukmin and ukmax are lower and upper boundaries DE have two categories; adaptive approaches and structure
of the j component, respectively, which are selected based change. Main problem of Evolutionary Algorithms (EAs)
on the type of problem. rand1 is a random value between is proper value allocation for control parameters. The DE
[0,1]. PN and VN are the number of population and has three control parameters: SF, CR and population. In
variables, respectively. adaptive approach these parameters are selected dynamically
. After generation of initial population by and not by trial and error technique. The SaDE has two
mutation operator, vectors of population are changed and improvement steps. The SaDE has two improvement steps.
modified randomly. In this process, three random vectors, These improvements are applied on two control parameters;
Ua, Ub and Uc, are placed in Eq. (5), i.e. SF and CR. The capability and reliability of SaDE is
confirmed by test on 21 test function [20].
U i((Gmut
)
) = Ua
(G )
+ SF (U b(G ) − U c(G ) ) , i = 1,..., PN (5)
0&% ,1 2 1 3
where, a, b and c are randomly selected from the set
In second step of original DE, SF is multiplied by the
{1,…,PN}, that a≠b≠c≠i, and also SF is a real and constant
difference of two selected vectors. Value of SF is selected
factor ϵ[0,2].
from range [0, 2], randomly. The more the value of SF is
. - All population has not recombination
lower (close to zero), the larger is the effect of the first
ability because corresponding fitness to them are far away
selected vector, Ua, and the lesser is the search space. But
from the global optimum, then crossover operator is
if SF was large (close to 2), search space was larger. If the
applied on population. If crossover rate is more than
search space is too large maybe the algorithm go away
random number in range [0, 1], vectors produced from
from the global optimum solution and if it is too small the
mutation step are selected, otherwise, selection is
mutation step is useless. The SF is defined in range [0,2].
performed from initial population. CR is real value in the
While, in practical issue, to reach the optimal point it
range [0, 1].
varies from 0.4 to 1[21]. Thus, in Eq. (8) a novel approach
is proposed for SF,
 U ji ( mut ) if ρ j ≤ CR or j = q
(G )

U (jiG( cross ) =  (G )
)
, (6)
U ji otherwise  SF + rand 2 [0,1] × SFmax if rand 2 < τ 1
SFi G +1 =  min (8)
i = 1,..., PN ; j = 1,..., VN  SFi G otherwise

where, ρj and q are chosen randomly from [1,…,VN] where, SFmin and SFmax are lower and upper limits of SF. To
-. Selection operator selects vectors corres obtain optimal value, SFmin and SFmax are adjusted on 0.4
ponding to the best solution for next generation. and 1.0, respectively. The selection criteria between a new
value for SF and old value of SF are τ1. If τ1 was lower
U ) ) ≤ f (U i
than a random value in range [0,1], a new value is
(G )
if f ( U i((Gcross
) (G )
)
U i( G +1) =  i ( cross )
(G ) , (7) generated by first scenario of Eq. (8), otherwise value of
 U i otherwise
SF is maintained fixed for next case without any changes.
i = 1,..., PN
0&( ,1 2 1 4
where, f(Ui(cross)) and f(Ui) are fitness corresponding to
vectors of crossover and initialization steps, respectively& The CR is main control parameter. This parameter is
-. / This algorithm terminates if selected in range [0, 1] by operator. For the next step, CR
one of two termination criteria is satisfied; i.e., (i) an determines which vectors of initial or mutation is selected.
 Mohammadhosein Dideban, Noradin Ghadimi, Mohammad Bagher Ahmadi and Mohammmad Karimi

In original DE, the CR is selected experimentally or by trial algorithm is repeated until the maximum number of
and error method. In SaDE, CR is generated by Eq. (9), iteration is reached. Flowchart of OLSC problem solution
by SaDE algorithm has been illustrated in Fig. 1.
rand3 [0,1] if rand3 < τ 2
CRiG +1 =  G (9)
 CRi otherwise
6&
The role of τ2 is same of role τ1 in SaDE i.
In this section, to confirm robustness of SaDE in solving
0&+ " 5 OLSC problem, three load patterns are tested on two test
systems. The constant load is applied on IEEE 10 bus, and
In this work, self adaptive DE algorithm has been both varying and effective loads are applied on IEEE 34
proposed to solve optimal capacitor placement. The bus standard radial network. A backward forward load
algorithm has two improvement phases; in first and second flow approach used in this paper is same as in [22]. Cost
phases, novel equations for SF and CR has been used, per power loss and cost per energy loss are 168 $/kW/year
respectively. and 0.06 $/kWh/year, respectively [23]. τ1 and τ2 are equal
The solution of OLSC problem using SaDE algorithm, at to 0.1 [20]. The capacity and related cost of capacitor
first the size of population, generation, buses, capacitor banks have been presented in [23].
type, values for base voltage and power, consumed active The SaDE has two improvement steps. To show the
and reactive powers and lines impedance are applied. By capability of each improvement steps, simulations are
load flow program, voltage profile and power loss of test carried out for each step separately. Then, SaDE i and
system have been extracted. Then SaDE algorithm is SaDE ii show first and second improvements (Eq.(7) and
initialized by Eq. (4) and objective function (OF) is Eq.(8)), respectively. It should be noted again that SaDE is
computed based on initial values. After, mutation and composed of SaDE I and SaDE ii. To compare different
crossover operators are applied on initial population based methods, seven parameters have been introduced which
on Eqs. (5 6). Load flow is recomputed and OF is are: annual cost, in $, total installed capacitor bank, in kVAr,
calculated again by new values. The selection operator CPU time, in sec, power loss, in kW, and their related costs,
selects best solution between two obtained OFs (after in $, minimum voltage, in pu, and annual cost, in $.
initialization and crossover). In this work, termination
criterion is ending the number of iteration, and then this 6&%

The constant load is applied on IEEE 10 bus standard

radial network. Topology of IEEE 10 bus has been
illustrated in Fig. 2 [29]. Table 1 shows results of OSLC
problem in IEEE 10 bus radial network with constant load.

/ %. Results of capacitor placement on 10 bus with

constant load
Cost
Meth.

CPU Power
Min.Volt. Loss+Cap.
Time Loss
=Annual
1 783.8 0.8375 0+131675=131675
2 681.28 0.9001 1865+114455=116320
3 26.450 675.43 0.9001 1932+113472=115404
4 22.442 675.37 7&877( 1930+1134629%%6+8+
5 (7&(:8 ;<6&+; 7&877( %8+7=%%+0;79%%6+8+

3 & %& Flowchart of OLSC problem solution by SaDE

algorithm 3 & (& IEEE 10 bus distribution network
 Optimal Location and Sizing of Shunt Capacitors in Distribution Systems by Considering Different Load Scenarios

3 & +& Location and size of installed capacitors on 34 bus with Varying load

In this table, results of various SaDE algorithms have been level with 0.8 pu load for duration 6780 h and 3rd load level
compared with results of hybrid method which is created with 0.5 pu for 1000 h. The IEEE 34 bus radial distribution
by compositing fuzzy and genetic algorithm [19]. Methods network is test case of varying load condition (see Fig. 3)
1, 2, 3, 4 and 5 are without capacitor, hybrid, SeDE i, [29]. The results of varying load simulation have been
SaDE ii and SaDE, respectively. listed in Table 2. In all cases, the best solution has been
By considering results of Table 1, SaDE algorithm has bolded, while the worst are crossed with a line.
the best solution. The CPU time for SaDE is the minimum According to results of Tables 2, the minimum voltage
among SaDE family and is less 6.1613 and 2.1529 sec of SaDE approach, in all load levels, are best solution. For
respect to the related parameter of SaDE i and SaDE ii power loss and its related cost and total installed capacitor
algorithms, respectively. The active power of SaDE is less banks and its related cost, SaDE ii and SaDE i algorithms
than hybrid, SaDE i, and SaDE ii methods; the differences give relatively better options, while annual cost of SaDE ii
are 5.92, 0.07 and 0.01 kW, respectively. The minimum technique is the worst solution. In first level, CPU time of
voltage of hybrid and SaDE i approaches are close to each SaDE algorithm is minimum amount among SaDE family
other and also less than minimum voltage of SaDE and and this value is less for 10.7442 and 5.7873 sec less than
SaDE ii algorithms. The power loss cost of SaDE algorithm SaDE i and SaDE ii techniques, respectively. SaDE i
is less than hybrid, SaDE i and SaDE ii algorithms; the
related amounts are 995, 12, and 2 in $, respectively. The / (. Results of capacitor placement on 34 bus with
annual cost of SaDE is equal to SaDE ii, the values are 972 varying load
and 11 $ less than hybrid and SaDE i algorithms,
Method W/O Cap. Hybrid SaDE i SaDE ii SaDE

respectively. Fig. 2 shows optimal location/size of installed Cost($)

Level

CPU Power Min.

capacitor banks in IEEE 10 bus radial network. In this Time Loss Volt. Loss+ Cap.= Annual
figure, presented values for capacitor banks is in term of 1 221.72 0.9492 13303+0=13303
kVAr. 2 139.16 0.9609 56443+0=56443
The presented optimal location/size of SaDE and SaDE 3 52.855 0.9783 3171+0=3171
ii algorithms are same and in most cases similar to SaDE i 1 160.5 0.9486 9630+611=10241
algorithm. The number of capacitor banks of hybrid 2 101.18 0.9593 41041+497=41538
method is less than the number of capacitor banks of each 3 39.276 0.9749 2357+320=2677
1 64.387 160.49 0.9501 9629+6%%=10140
SaDE family. 2 56.439 100.05 0.9608 40580+692=41272
3 6+&;06 +8&((8 7&8<6+ (+60=(0;9(;77
6&( * 1 59.427 %6:&8+ 0.9499 86+;=9629%708:
2 66&:(6 88&::6 0.9606 076%+=1005941518
3 54.449 39.233 0.9746 (+60=52592879
In practical cases, the network load is changing daily. To 1 6+&;+8 160.11 7&867( 9607+524=%7%+%
study varying load, three load levels are defined; i.e. 1st 2 55.905 100.04 7&8;78 40576+692=0%(;:
load level with 1.0 pu load for duration 1000 h, 2nd load 3 53.957 39.3 7&8<6+ (+6:=(0;92604
 Mohammadhosein Dideban, Noradin Ghadimi, Mohammad Bagher Ahmadi and Mohammmad Karimi

algorithm presents best solution for capacitor banks cost. 0.389, and 0.356 kW less than hybrid, SaDE i, as well as
From viewpoint of annual cost SaDE ii is the worst case SaDE, respectively. The least CPU time of effective load is
among SaDE family. In first and second load levels, SaDE obtained by SaDE i algorithm, CPU time of SaDE i
presents best solution. Proposed optimal location/size of algorithm is 0.5348 and 1.6033 second less than thoes of
installed capacitor banks by SaDE family and hybrid SaDE and SaDE algorithms, respectively.
method have been presented in Fig. 3. The minimum voltage of SaDE ii and SaDE are the most
In all load levels, the number of capacitor banks and least minimum voltages among SaDE family. The
proposed by SaDE ii algorithm is more than other SaDE ii algorithm presents optimal power loss cost being
approaches. In first load level, the number of capacitor 1372, 66, and 60 $ less than hybrid, SaDE i, and SaDE
banks proposed by SaDE and SaDE i is equal to each other, approaches, respectively. The total cost of SaDE algorithm
and is 1 and 5 capacitor banks less than hybrid and SaDE ii is optimal value among four methods and is 1158, 107, and
algorithms, respectively. In 2nd load level, the number of 303 $ less than annual cost of hybrid, SaDE i, and SaDE ii,
installed capacitor banks of SaDE ii algorithm is 3 banks respectively.
more than related parameter of SaDE i and SaDE
techniques. In third load level, this deference is same for 6&0
second load level.
In this part, to analyze performance of these three
6&+ , - approaches to solve OLSC problem, two criteria are used.
First criterion is the number of best or worst solution for
The varying load has an effective level which is each technique among solutions. The criterion shows
calculated by Eq. (3). For the introduced three load levels, reliability of any approach, the more the number of best
effective load level is equal to 0.78858 pu. Tables 3 and Fig. solution, the more is the reliability of methods. The number
4 show solution results of OLSC problem for effective load of best/worst for SaDE family has been illustrated in Fig. 4.
and their optimal location/size of installed capacitor banks Results of SaDE family are better than hybrid method; this
of in IEEE 34 bus radial distribution network, respectively. fact confirms capability of SaDE family to solve OFLSC
Methods 1, 2, 3 and 4 are hybrid, SaDE i, SaDE ii and problem. Thus, in Fig. 5, the obtained results by hybrid
SaDE approaches, respectively. technique have been ignored.
By considering results of Table 3, it is obvious that the Focusing on Fig. 5 reveals that SaDE ii, and SaDE i
power loss of SaDE ii has the least value, and is 8.16, algorithms have better solution. The SaDE is the best
option among these four algorithms; this algorithm has the
/ +. Results of capacitor placement on 34 bus with minimum value in worst solution, and the maximum value
effective load
Meth.

CPU Power Cost

Min.Volt.
Time Loss Loss+Cap.=Annual
1 105.08 0.96015 17653+474=18127
2 6%&6<7 97.304 0.96131 16347+729=17076
3 52.105 8;&8% 7&8;%0( %;(:%=991917272
4 53.173 97.271 0.96103 16341+;(<=%;8;8

3 & 6& The number of best/worst solution of SaDE family

!" #$ ‑ ‑

3 & 0& Compression of location/size of installed capacitor

on 34 bus with effective load 3 & ;& Error percentage of minimum voltage
 Optimal Location and Sizing of Shunt Capacitors in Distribution Systems by Considering Different Load Scenarios

!" #$ ‑ ‑
SaDE ii approach presents the best value which its value is
0.0835, 0.0780 and 0.059 more than hybrid, SaDE i, and
'% SaDE algorithms, respectively. These increments in level 2
'
%
are 0.0658, 0.0083 and 0.0078. The SaDE I algorithm in
3rd level of varying and effective loads has best value. In
&% 3rd level of varying load, the difference between error
&
%
percentage of SaDE ii and three other algorithms; i.e.
hybrid, SaDE ii and SaDE techniques are 0.0024, 0.0002
and 0.0036, respectively. Finally, in effective load level,
SaDE ii algorithm has maximum power loss error
3 & <& Error percentage of power loss percentage and its value is 0.4083, 0.0195 and 0.0178 more
than hybrid, SaDE i, SaDE algorithms, respectively.
for best solutions. Among SaDE family, SaDE i technique
is the worst option. 6&6
The other criterion is error percentage of base value
respect to computed value. The base value of minimum In this study, to solve optimal capacitor allocation in
voltage and power loss are 1 pu and 2000 kW, respectively. radial distribution network a novel algorithm based on
Fig. 5 shows error percentage of minimum voltage of simple DE algorithm has been proposed. The proposed
constant and varying as well as effective loads. algorithm, SaDE, has two improvement steps: self
Regarding results of Fig. 5, for minimum voltages, adapting SF and CR called SaDE i and SaDE ii,
SaDE algorithm reaches better solution in most cases; this respectively. From results of simulation and comparison,
algorithm except effective load level has optimal value. followings have been extracted:
The SaDE ii algorithm has better solution among four 4 > . In addition to capacity of installed capacitor
approaches only in constant load and effective load level. banks, dispatch manner of capacitor banks has
In this case, SaDE i has the worst solution among SaDE considerably effects on cost. This fact has been extracted
family and only in 3rd level of varying load presents by comparing cost of capacitor among SaDE family, in
optimal value which is equal to related parameter of SaDE level 2 of varying load between SaDE i and hybrid, in level
algorithm. In constant load, minimum voltage error 3 of varying load between hybrid and SaDE ii. Thus less
percentage of SaDE ii is equal to corresponding parameter installed capacitor always does not result in lower cost.
of SaDE which both is 0.011 and 0.013 less than hybrid 4 > . The numbers of buses have impact on CPU
and SaDE i algorithms, respectively. In first level of time more than total demand of network. In varying load,
varying load, the error percentage of SaDE is 0.159, 0.004 demand of 1st load level is twice of 3rd load level demand
and 0.029 less than related values of hybrid, SaDE i, SaDE while CPU time changes only about 20%. This fact could
techniques, respectively. In level 2, these reduction values be derived by comparing between 10 bus and 34 bus radial
are 0.159, 0.006 and 0.028. In third load level, minimum networks. The number of buses are one of initial matrix
voltage error percentage of SaDE i and SaDE algorithms is dimension, then the more the number of buses, the lower is
equal and is 0.037 and 0.064 less than hybrid and SaDE ii convergence velocity.
techniques, respectively. Finally, in effective load level, 4 > . Form the view point of voltage profile
error percentage of minimum voltage of SaDE ii is 0.127, improvement, SaDE ii is better than SaDE i algorithm. The
0.011, and 0.039 less than error percentage of hybrid, SaDE i has the worst solution in SaDE family. This fact
SaDE i, SaDE approaches, respectively. confirms that among control parameters of DE algorithm;
After computation of minimum voltage error percentage, crossover rate has the maximum impact on voltage profile.
error percentage of power loss is computed by the Then self adapting CR helps to extract better solution from
approach which was used for minimum voltage. The base algorithm respect to adjust a constant value for CR in
value of power is 2 MW. Fig. 6 shows error percentage of original DE. The better results of SaDE ii is compared to
power loss for three load conditions. SaDE illustrates that self adapting SF does not affect
It should be mentioned that, in this case, the more the remarkably on voltage profile.
power loss error percentage, the better is the solution. 4 > -. The capability of SaDE i algorithm is
Among five cases, the SaDE ii algorithm in most cases confirmed in less installed capacitor banks presented by
presents better solution. In error percentage of power loss this algorithm. In constant load, total installed capacitor
similar to error percentage of minimum voltage, SaDE is banks of SaDE family is equal, in effective load this
the worst option among SaDE family with only one best parameter of SaDE and SaDE i algorithms are 300 kVAr
solution. less than SaDE ii algorithm. Unlike SaDE i, SaDE ii in the
The error percentage of SaDE, in constant load, is 0.292, most cases has largest installed capacity; the total installed
0.0035, and 0.0005 more than hybrid, SaDE i, and SaDE ii capacitor banks in second and third level of varying load
algorithms, respectively. In first level of varying load, and effective load confirm this extraction. Thus, SF
 Mohammadhosein Dideban, Noradin Ghadimi, Mohammad Bagher Ahmadi and Mohammmad Karimi

improvement reduces installed capacitor banks more than distribution systems, IEEE Transaction on Power
CR improvement. Systems Vol. 23, No. 4, pp. 1619 1626, 2008.
[6] Srinivasa Rao R. Capacitor placement in radial
distribution system for loss reduction using artificial
;& bee colony algorithm, International Journal of
Engineering and Applied Sciences, pp.359 363, 2010.
In this work, an improved DE algorithm, named SaDE, [7] Sirjani R., Mohamed A., Shareef H. Optimal
was used to solve OLSC problem. The proposed algorithm capacitor placement in radial distribution system
was obtained by self adapting control parameter of using harmony search algorithm, journal of Applied
mutation and crossover operators; i.e. CR and SF, Sciences, Vol. 10, No. 23, pp. 2998 3006, 2010.
respectively. Three load conditions, constant and varying [8] Mendes A., Fran ca P.M., Lyra C., Pissarra C.,
as well as effective, were tested on 10 bus and 34 bus Cavellucci C. Capacitor placement in large sized
radial distribution networks. Fitness is function of annual radial distribution Networks, IEE Proc.>Gener.
cost, in addition to cost, six other parameters used for Transm. Distrib., Vol. 152, No. 4, pp. 496 502, 2005.
comparison were installed capacitor banks and related cost, [9] Sundhararajan S., Pahwa A. Optimal selection of
CPU time, minimum voltage, active power loss and related capacitor for radial distribution systems using a
cost. To compare results, the number of best/worst genetic algorithm, IEEE Transactions on Power
solutions and error percentage of minimum voltage and Systems, Vol. 9, No.3, pp.1499 1507, 1994.
power loss were used. [10] Masoum M. A. S., Ladjevardi M., Jafarian A., Fuchs
From simulation results of case studies it can be claimed E. F. Optimal placement, replacement and sizing of
that: In general, self adapting CR is more effective than capacitor banks in distorted distribution networks by
self adapting SF, latter has better solution for the number of genetic algorithms, IEEE Transaction on power
installed capacitor banks and related cost. Sum of these delivery, Vol. 19, No.4, pp.1794 1801, 2004.
two improvements has better overlapping and give an [11] Taher S. A., Hasani M., Karimian A. A novel method
optimal solution. Cost of total installed capacitor banks is for optimal capacitor placement and sizing in
less than 10% of annual cost, remaining cost (90%) is distribution systems with nonlinear loads and DG
power loss cost. Therefore, it is better to focus on power using GA, Commun Nonlinear Sci Numer Simulat,
loss reduction. Load decline has less impact on the speed Vol. 16, pp. 851 862, 2011.
of algorithm running, and decreasing the load amount only [12] Carpinelli G., Proto D., Noce C., Russo A., Varilone P.
20% decrease the speed of algorithm running. In addition Optimal allocation of capacitors in unbalanced multi
to capacitor banks amount, the number of installed converter distribution systems, A comparison of some
capacitor banks also have remarkable impact on cost. fast techniques based on genetic algorithms, Electric
Power Systems Research, Vol. 80, pp. 642 650, 2010.
[13] Abu Elanien A. E. B., Salama M. M. A. A wavelet
4 ANN technique for locating switched capacitors in
distribution systems, IEEE Transaction on Power
[1] Khodr H. M., Vale Z. A., Ramos C. Optimal cost Delivery, Vol. 24, No. 1, pp. 400 409, 2009.
benefit for the location of capacitors in radial distri [14] Taher S. A., Hasani M., Karimian A. A new method
bution systems, IEEE Transaction on Power Delivery, for optimal location and sizing of capacitors in
Vol. 24, No. 2, pp. 787 796. 2009. distorted distribution networks using PSO algorithm,
[2] Khodr H.M., Olsina F.G., De Oliveira De Jesus P.M., Simulation Modelling Practice and Theory, Vol. 19,
Yusta J.M. Maximum savings approach for location No. 2, pp. 662 672, 2011.
and sizing of capacitors in distribution systems, [15] Eajal A. A., El Hawary M. E. Optimal capacitor
Electric Power Systems Research, Vol. 78, No. 7, pp. placement and sizing in unbalanced distribution
1192 1203, 2008. systems with harmonics consideration using particle
[3] Jabr R.A. Optimal placement of capacitors in a radial swarm optimization, IEEE Transaction on Power
network using conic and mixed integer linear pro Delivery, Vol. 25, No. 3, pp. 1734 1741, 2010.
gramming, Electric Power Systems Research,.Vol. 78, [16] Gasbaoui B., Chaker A., Laoufi A., Abderrahmani, A.,
No. 6, pp. 941 948, 2008. Alloaoua B. Optimal placement and sizing of
[4] Segura S., Romero R., Rider M. J. Efficient heuristic capacitor banks using fuzzy ant approach in electrical
algorithm used for optimal capacitor placement in distribution systems, Leonardo Electronic Journal of
distribution systems, Electrical Power and Energy Practices and Technologies, Vol. 16, pp. 75 88, 2010.
Systems, Vol. 32, pp. 71 78, 2010. [17] Chang G. W., Chang W.Ch., Chuang Ch.Sh., Shih
[5] Da Silva I. Ch., Carneiro S., De Oliveira E. J., de D.Y. Fuzzy logic and immune based algorithm for
Souza Costa J., JRezende Pereira. L. A Heuristic placement and sizing of shunt capacitor banks in a
constructive algorithm for capacitor placement on distorted power network, IEEE Transaction on power
 Optimal Location and Sizing of Shunt Capacitors in Distribution Systems by Considering Different Load Scenarios

delivery, Vol. 26, No. 4, pp. 2145 2153, 2011. ! " was
[18] Seifi A., Hesamzadeh M.R. A hybrid optimization born in 1978 in Tabriz, Iran. He
approach for distribution capacitor allocation con received the B.Sc and M.S.c degrees
sidering varying load conditions, Electrical Power both in electrical engineering from
and Energy Systems, Vol. 31, No. 10, pp. 589 595, I.A.U, Ardebil and Ahar, Iran, in 2006
2009. and 2012, respectively. Mr. Ahmadi has
[19] Storn R., Price Kh. Differential evolution a simple been working in Benz diesel industry
and efficient heuristic for global optimization over of iran (I.D.E.M) as the manager of
continuous spaces, Journal of Global Optimization, electrical repairman since 2004. And since 2011 he has
Vol. 11, pp. 341 359, 1997. been working as the instructor in I.A.U. He has been
[20] Brest J., Greiner S., Boškovic B., Mernik M., umer V. involved with the smart grid, wind and solar energy
Self adapting control parameters in differential (renewable energy), distribution systems planning and
evolution: A comparative study on numerical bench optimization.
mark problems, Transaction on Evolutionary Com>
putation, Vol. 10. No. 6, pp.646 657, 2006.
[21] Rahnamayan Sh., Tizhoosh H. R., Salama M. M. A. # was born in 1983
Opposition based differential evolution, IEEE Trans> in Tehran, Iran. He received the B.Sc
actions on Evolutionary Computation, Vol. 12, No. 1, and M.S.c degrees both in electrical
pp. 64 79, 2008. engineering from I.A.U, Ardebil and
[22] Teng J.H. A direct approach for distribution system Ahar, Iran, in 2008 and 2011, re
load flow solutions, IEEE Transactions on Power spectively. He has been involved with
Delivery, Vol. 18, No. 3, pp. 882 887, 2003. the FACTS devices (control and
[23] Mekhamer S.F., Soliman S.A., Moustafa M.A., El placement), distribution systems plan
Hawary M.E. Application of fuzzy logic for reactive ing and optimization and was the author or co author in
power compensation of radial distribution feeders, more than 50 articles published in Journals and conference
IEEE Transactions on Power Systems, Vol. 18, No. 1, Proceedings.
pp. 206 213, 2003.

was born
in 1964 in Ardabil, Iran. He received
the B.Sc degrees in electrical engine
eering from Tabriz University in 1996.
Mr. Dideban joined to Ardabil province
electricity distribution Co from 1996.
His areas of interest in research are
distributed generation planning, pro
tective relying and distribution systems operating.

was born in ardabil

Iran in 1985, and received the B.Sc.
degree in electrical engineering from
the Islamic Azad University, Ardabil
Branch, Ardabil, Iran, in 2009 and the
M.Sc. degree in electrical engineering
from the Islamic Azad University Ahar
Branch, Ahar, Iran, in 2011. His research
interests include Power System Protection, modeling and
analysis of Distributed Generations, renewable energy and
communications systems.