You are on page 1of 7

Optimal Reconfiguration and Distributed Generator

allocation in Distribution Network using an advanced


Adaptive Differential Evolution
Partha P Biswas1, R. Mallipeddi2 P.N. Suganthan1, Gehan A J Amaratunga3
1 3
Nanyang Technological University, Singapore University of Cambridge, United Kingdom
2
Kyungpook National University, South Korea epnsugan@ntu.edu.sg, gaja1@hermes.cam.ac.uk
parthapr001@e.ntu.edu.sg, mallipeddi.ram@gmail.com

Abstract—Power loss in an electrical network is unavoidable Loss minimization by only reconfiguration of the networks
due to its inherent resistance. However, for economical and is performed in many literatures. Some of the recent studies that
efficient operation, the network loss shall be minimized to the implement evolutionary algorithms (EAs) for the optimization
extent possible. Construction of the distribution network is usually task are with fireworks algorithm (FWA) [2], cuckoo search
closed loop, though operation is radial with opening of a suitable
algorithm (CSA) [3], modified bacterial foraging optimization
sectionalizing switch that disconnects a branch in the loop. This
process of reconfiguration i.e. selection of sectionalizing switch is algorithm (MBFOA) [4] and modified particle swarm
done in a way such that system loss in minimized. Another optimization (MPSO) [5]. Further, several works concentrate on
effective way of reducing power loss is addition of distributed optimization of DGs only to minimize system loss. Handpicked
generators (DGs) locally to the system nodes (buses). DG size and studies in recent times focus only on allotment of DGs are ref.
location also need to be optimized for minimization of real power [6] where analytical approach is taken, ref. [7] where genetic
loss. This paper presents an application of a metaheuristic to algorithm (GA) is adopted and ref. [8] where hybrid method
simultaneously allocate DGs and perform reconfiguration of a with ant colony optimization (ACO) and artificial bee colony
couple of standard radial distribution networks. Both location and (ABC) is applied. A limited number of studies have been
placement of the DG are optimized by the algorithm. As an
performed for simultaneous system reconfiguration and
obvious fact, the optimization problem is a combination of discrete
(location) and continuous (rating) variables. Linear population allocation of DGs. Incidentally maximum reduction in loss can
size reduction technique of success history based adaptive be achieved with this approach. Reference [9] employed
differential evolution (L-SHADE) is implemented to perform the fireworks algorithm (FWA) to perform the optimization task.
optimization task with objective of minimizing network real power Adaptive cuckoo search algorithm (ACSA) [10] improved the
loss. The algorithm is tested on standard IEEE 33-bus and 69-bus past results by FWA. A heuristic method named uniform voltage
radial distribution networks. The simulation results are found to distribution based constructive reconfiguration algorithm
be promising and highly competitive when compared with results (UVDA) [11] reported some of the best results on the problem
of other equivalent algorithms. of network reconfiguration accompanied by DG allocation.
Keywords—Distribution network, real power loss, distributed
generator, optimal reconfiguration, L-SHADE algorithm. Our research presented in this paper uses L-SHADE
algorithm, a well-established optimization algorithm for the
constrained, multimodal non-linear problems. L-SHADE
I. INTRODUCTION
improves the performance of success history based adaptive
The distribution system network accounts for a significant differential evolution (SHADE) with progressive linear
amount of power loss, sometimes in the order of 10-13% of the reduction of population size. The algorithm performed best
generated output [1]. High distribution loss means inefficiency among non-hybrid algorithms for the optimization of real
and poor voltage regulation of the network. To increase parameter single-objective function in CEC 2014 competition
efficiency and enhance voltage regulation of the system, [12]. L-SHADE has successfully been applied to the problems
network is to be properly reconfigured. In addition, locally of discrete location optimization of wind turbines in a windfarm
added distributed generations (DGs) in the form of a diesel [13], total harmonic distortion minimization in multilevel
generator, solar photovoltaics (PV), a wind turbine etc. can inverters [14] and hybrid active power filter parameter
reduce losses, improve voltage profile and boost capacity of the optimization [15]. Motivated by the growing application and
system. However, reconfiguration of the network or allotment encouraging performance of the algorithm in power domain, we
of the DG cannot be random as it might result in inefficient further investigate its usefulness on the problem of distribution
performance and unintended operation of the network. Radial network. Minimization of power loss in the distribution network
arrangements of the feeders in the network must be maintained is set as the objective with optimal reconfiguration and DG
after reconfiguration. Size and placement of the DG shall be allocation. The algorithm simultaneously optimizes DG size and
optimized such that power loss is minimized satisfying system location (bus no.) and reconfigures the network. IEEE standard
constraints on power balance, line capacity and bus voltage. 33-bus and 69-bus radial distribution networks are studied under

978-1-5386-2726-6/17/$31.00 ©2017 IEEE


the purview of this literature and simulation results are found to normally open tie switches. Nodes (buses) are numbered within
be among the best. the circles. As a first step to apply the algorithm for optimal
reconfiguration, all sectionalizing and tie switches of the
In rest of the paper, problem formulation for simultaneous network are closed. This process creates certain loops in the
network reconfiguration and power flow are included in section network marked as Loop1 to Loop5 in the diagrams. Thereafter,
II. Section III provides detail of the study cases performed in this any one switch (either tie or sectionalizing) in the loop is opened
work. The algorithm and its application are explained in section to maintain the radial nature of the network. The switch must be
IV. Simulation results and comparisons are included in section unique for each loop and opening of the switch should not isolate
V, finally ending with conclusion in section VI. any node (bus) in the network. The algorithm performs several
iterations to come up with the switch numbers, opening of which
II. MATHEMATICAL FORMULATION minimizes the power loss. It must be noted that this operation
The aim of this paper is to decide optimal reconfiguration and allocation of DG are simultaneous. Output solutions of the
alongwith DG allocation in the radial distribution network. In algorithm are the suggested switches for opening together with
following sub-sections, the approach to maintain the radial optimal sizes and locations of DGs that minimize network loss.
nature of the network after reconfiguration and power flow
equations are described. B. Power flow formulation
Fig. 3 represents single line diagram of a radial feeder
A. Network reconfiguration configuration. Computation of power flow is performed using a
Fig. 1 and Fig. 2 show the base configurations for 33-bus and set of recursive equations on active power, reactive power and
69-bus systems respectively with dotted lines representing bus voltages [16]:

Fig. 1: Base configuration of IEEE-33 bus distribution network

Fig. 2: Base configuration of IEEE-69 bus distribution network


0 1 k-1 k k+1 N
P0 ,Q0 P1 ,Q1 Pk-1 ,Qk-1 Pk ,Qk Pk+1 ,Qk+1 PN ,QN

PL1 ,QL1 PLk-1 ,QLk-1 PLk ,QLk PLk+1 ,QLk+1 PLN ,QLN

Fig. 3: Single line diagram of a radial feeder

ܲ௞ଶ ൅ ܳ௞ଶ the currents in base case. Therefore, it is not essential to check
ܲ௞ାଵ ൌ ܲ௞ െ ܲ௅௞ାଵ െ ܴ௞ǡ௞ାଵ Ǥ (1) this constraint during optimization.
ȁܸ௞ ȁଶ
ܲ௞ଶ ൅ ܳ௞ଶ III. STUDY CASES
ܳ௞ାଵ ൌ ܳ௞ െ ܳ௅௞ାଵ െ ܺ௞ǡ௞ାଵ Ǥ (2)
ȁܸ௞ ȁଶ The study cases performed for the two distribution networks
are discussed in this section. Scenarios of only reconfiguration,
ȁܸ௞ାଵ ȁଶ ൌ ȁܸ௞ ȁଶ െ ʹ൫ܴ௞ǡ௞ାଵ Ǥ ܲ௞ ൅ ܺ௞ǡ௞ାଵ Ǥ ܳ௞ ൯ only DG allocation and simultaneous reconfiguration and DG
ܲ௞ଶ ൅ ܳ௞ଶ (3) allocation are considered in case studies. It is worthwhile to note
൅ ൫ܴ௞ǡ௞ାଵ ଶ ൅ ܺ௞ǡ௞ାଵ ଶ ൯Ǥ
ȁܸ௞ ȁଶ that number and total power of the added DGs play vital roles in
where, ܲ௞ and ܳ௞ are the real and reactive power flowing out of the obtained final loss of the network. Increase in number of
bus ݇ ; ܲ௅௞ାଵ and ܳ௅௞ାଵ are the real load and reactive load at DGs beyond a certain value may not be technically and/or
bus ݇ ൅ ͳ. The line section between buses ݇ and ݇ ൅ ͳ has a commercially feasible. Again, DG penetration cannot be too
high; else the network might turn into an active network. Too
resistance of ܴ௞ǡ௞ାଵ and a reactance of ܺ௞ǡ௞ାଵ . Magnitude of
large cumulative rating of added DGs might increase short
voltage of bus݇ is ȁܸ௞ ȁ. The power loss (ܲ௅௢௦௦ ) in the line
circuit levels of the components connected to the system.
section connecting buses ݇ and ݇ ൅ ͳ is computed by:
Therefore, we judiciously limit the number and penetration of
ܲ௞ଶ ൅ ܳ௞ଶ the DGs in line with some past literatures for fair and valid
ܲ௅௢௦௦ ሺ݇ǡ ݇ ൅ ͳሻ ൌ ܴ௞ǡ௞ାଵ Ǥ (4) comparison of current results with previously reported results.
ȁܸ௞ ȁଶ
Total power loss (ܶܲ௅௢௦௦ ሻ in the network is summation of all the TABLE I. SUMMARY OF STUDY CASES
line losses calculated by:
Network Case no. Case description No. of Max DG
ேିଵ DGs penetration
ܶܲ௅௢௦௦ ൌ ෍ ܲ௅௢௦௦ ሺ݇ǡ ݇ ൅ ͳሻ (5) 33-bus Case 1 Only reconfiguration - -
Case 2 Only DG allocation 3 2.5 MW
௞ୀ଴ Case 3 Simultaneous reconfiguration 3 2.5 MW
This study considers that the DG units supply only real power and DG allocation
with unity power factor (e.g. solar PV). Hence, a DG delivering 69-bus Case 4 Only reconfiguration - -
Case 5 Only DG allocation 3 2.5 MW
output power ܲ஽ீ when added to ݇-th bus, the loading of that Case 6 Simultaneous reconfiguration 3 2.5 MW
bus changes from ܲ௅௞ to ሺܲ௅௞ െ ܲ஽ீ ሻ. The algorithm checks all and DG allocation
possible locations with all probable ratings of the DGs during
the search process to find the most optimum allocation. Total load of 33-bus network is 3.72 MW and 2.30 MVAr,
while that of 69-bus network is 3.80 MW and 2.69 MVAr. Line
During the process of network reconfiguration and DG addition, parameters and detailed data on load demands for the networks
following constraints of the network must be satisfied: are provided in [17]. A summary of the case studies is presented
ܸ௠௜௡ ൑ ȁܸ௞ ȁ ൑ ܸ௠௔௫ (6) in Table I, where it can be observed that the number of DGs is
limited to 3. Minimum rating of DG is set to 200 kW for
practicability of application with cumulative rating of all DGs in
ห‫ܫ‬௞ǡ௞ାଵ ห ൑ ‫ܫ‬௞ǡ௞ାଵሺ୫ୟ୶ሻ (7)
the network not to exceed 2.5 MW.
where, equation (6) is for allowable voltage magnitude range of
IV. APPLICATION OF L-SHADE ALGORITHM
bus ݇ between ܸ௠௜௡ and ܸ௠௔௫ . In this study, we consider
ܸ௠௜௡ ൌ ͲǤͻͲp.u. and ܸ௠௔௫ ൌ ͳǤͲͷp.u. Equation (7) defines the During the last decade, Differential evolution (DE) evolved
constraint on current carrying capacity of the line between as an efficient and powerful population-based stochastic search
buses ݇ and ݇ ൅ ͳ. Branch current limits of the standard IEEE technique for solving optimization problems over continuous
bus systems are not definite and explicit. Further, as voltage search spaces [18]. As the performance of conventional DE is
profile is improved throughout the network with network highly dependent on the settings of the scaling factor (F), the
reconfiguration and DG allocation, branch currents reduce from crossover rate (CR), the population size (Np), and the chosen
mutation/crossover strategies [19], different ideas have been 0.1 signifies variance and scale parameter for the corresponding
proposed. In SHADE [20], the control parameters F and CR ሺ௧ሻ ሺ௧ሻ
distributions. Ɋ‫ܨ‬௥  & Ɋ‫ܴܥ‬௥ are randomly chosen from the
are adapted during evolution based on success history of the memory where the scale factors and crossover rates of
parameters. As a further improvement, in addition to F and CR, successful candidates of previous generations are stored. These
control parameter population size (Np) is reduced linearly over two values are initialized to 0.5 and subsequently modified by
successive generations in L-SHADE [21]. A brief description weighted Lehmer mean, the detail is in reference [20,21].
of L-SHADE and its application to distribution network
problem are presented in this section. D. Crossover
ሺ௧ሻ
In the crossover operation, donor vector ‫ݒ‬௜ combines its
A. Initialization ሺ௧ሻ
elements with the target vector ‫ݔ‬௜ to form the trial/offspring
The process of DE starts with an initial population (Np ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ
vectors with d dimensions) of randomly generated candidate vector ‫ݑ‬௜ ൌ ሺ‫ݑ‬௜ǡଵ ǡ ‫ݑ‬௜ǡଶ ǡ ǥ Ǥ Ǥ ǡ ‫ݑ‬௜ǡௗ ሻ . Binomial crossover is
solutions within the search space specified by lower and upper commonly adopted to operate on each variable. The binomial
ሺ௧ሻ
bounds. In the population, the jth component of the ith decision crossover with crossover rate ‫ܴܥ‬௜ is expressed as:
vector is initialized as:
ሺ௧ሻ ሺ௧ሻ
ሺ଴ሻ
‫ݔ‬௜ǡ௝ ൌ ‫ݔ‬௠௜௡ǡ௝ ൅ ‫݀݊ܽݎ‬௜௝ ሾͲǡͳሿሺ‫ݔ‬௠௔௫ǡ௝ െ ‫ݔ‬௠௜௡ǡ௝ ሻ (8) ሺ௧ሻ ‫ݒ‬௜ǡ௝ ‹ˆ݆ ൌ ‫݀݊ܽݎ”‘ܭ‬௜ǡ௝ ሾͲǡͳሿ ൑ ‫ܴܥ‬௜ ǡ
‫ݑ‬௜ǡ௝ ൌ ൝ ሺ௧ሻ
(13)
‫ݔ‬௜ǡ௝ ‘–Š‡”™‹•‡
where i = 1 to Np and j = 1 to d. ‫݀݊ܽݎ‬௜௝ ሾͲǡͳሿ is a random
number between 0 and 1 and superscript ‘0’ signifies where K is any randomly chosen natural number in the range
initialization. ሾͳǡ ݀ሿ, where d is the dimension of the problem.

B. Mutation E. Selection
After initialization, corresponding to each member in the Selection process determines if the trial (offspring) vector is
ሺ௧ሻ
population referred to as target vector ‫ݔ‬௜ , a donor/mutant going to replace the target vector at next generation t+1 by
ሺ௧ሻ
vector ‫ݒ‬௜ is generated at current generation t through mutation performing the following comparison:
operation. In the current work, the mutation strategy adopted is ሺ௧ሻ
referred as ‘current-to-pbest/1’: ሺ௧ାଵሻ —௜ ሺ௧ሻ ‹ˆ݂൫—௜ ሺ௧ሻ ൯ ൑ ݂൫‫ݔ‬௜ ൯ǡ
‫ݔ‬௜ ൌ൝ ሺ௧ሻ
(14)
‫ݔ‬௜ ‘–Š‡”™‹•‡
ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ ሺ௧ሻ
‫ݒ‬௜ ൌ ‫ݔ‬௜ ൅ ‫ܨ‬௜ Ǥ ൫‫ݔ‬௣௕௘௦௧ െ ‫ݔ‬௜ ൯ ൅ ‫ܨ‬௜ Ǥ ሺ‫ݔ‬ െ‫ݔ‬ ሻ (9)
ோభ೔ ோమ೔
where f(.) is the objective function to be minimized.
The indices ܴଵ௜
& ܴଶ௜
are randomly selected from the range
ሺ௧ሻ F. Linear population size reduction (LPSR)
[1, Np] and are mutually exclusive; ‫ݔ‬௣௕௘௦௧ is chosen from the
The achievement of SHADE that employs success history
top ܰ‫ ݌‬ൈ ‫݌‬ሺ‫ ג ݌‬ሾͲǡͳሿሻ individuals of current generation. The
ሺ௧ሻ based adaptation technique of scaling factor F and crossover
scaling factor ‫ܨ‬௜ is a positive control parameter for scaling the rate CR has been demonstrated in [20]. In addition, it has been
difference vectors corresponding to the ith individual at the tth also discovered that dynamic reduction of population size (Np)
ሺ௧ሻ
generation. If an element ‫ݒ‬௜ǡ௝ of the mutant vector fails to accelerates the performance of SHADE. Therefore, L-SHADE
satisfy the boundary conditions specified by ሾ‫ݔ‬௠௜௡ǡ௝ ǡ ‫ݔ‬௠௔௫ǡ௝ ሿ, it was proposed where the population is reduced over generation
and the size of population reductions is determined by a linear
is corrected as:
ሺ௧ሻ ሺ௧ሻ
function given in equation (15). After each generation t, the
ሺ௧ሻ ሺ‫ݔ‬௠௜௡ǡ௝ ൅ ‫ݔ‬௜ǡ௝ ሻȀʹ‹ˆ‫ݒ‬௜ǡ௝ ൏ ‫ݔ‬௠௜௡ǡ௝ population size in the next generation t+1 is calculated by –
‫ݒ‬௜ǡ௝ ൌ ൝ ሺ௧ሻ ሺ௧ሻ
(10)
ሺ‫ݔ‬௠௔௫ǡ௝ ൅ ‫ݔ‬௜ǡ௝ ሻȀʹ‹ˆ‫ݒ‬௜ǡ௝ ൐ ‫ݔ‬௠௔௫ǡ௝
ே௣೘೔೙ ିே௣೔೙೔
ܰ‫݌‬ሺ‫ ݐ‬൅ ͳሻ  ൌ ‫ ݀݊ݑ݋ݎ‬ቂቀ ቁ Ǥ ܰ‫ ܧܨ‬൅ ܰ‫݌‬௜௡௜ ቃ (15)
ேிா೘ೌೣ
C. Parameter Adaptation
ܰ‫݌‬௠௜௡ is set to 4 because the selected mutation strategy
At a given generation t, each individual generates a new trial
ሺ௧ሻ ሺ௧ሻ requires minimum 4 individuals. ܰ‫݌‬௜௡௜ is the initially selected
(offspring) vector with its associated parameters ‫ܨ‬௜ and ‫ܴܥ‬௜ population size while ܰ‫ܧܨ‬௠௔௫ and NFE are the maximum
which are adapted using – number of fitness evaluations and current number of fitness
ሺ௧ሻ
‫ܨ‬௜ ൌ ‫ܿ݀݊ܽݎ‬ሺɊ‫ܨ‬௥ ǡ ͲǤͳሻ
ሺ௧ሻ
(11) evaluations, respectively. If ܰ‫݌‬ሺ‫ ݐ‬൅ ͳሻ ൏ ܰ‫݌‬ሺ‫ݐ‬ሻ , a total of
ሺ௧ሻ ሺ௧ሻ [ܰ‫݌‬ሺ‫ݐ‬ሻ െ ܰ‫݌‬ሺ‫ ݐ‬൅ ͳሻ] worst individuals are deleted from the
‫ܴܥ‬௜ ൌ ‫݊݀݊ܽݎ‬ሺɊ‫ܴܥ‬௥ ǡ ͲǤͳሻ (12)
population [21].
ሺ௧ሻ ሺ௧ሻ
where ‫݊݀݊ܽݎ‬ሺɊ‫ܴܥ‬௥ ǡ ͲǤͳሻ & ‫ܿ݀݊ܽݎ‬ሺɊ‫ܨ‬௥ ǡ ͲǤͳሻ are sampled
values from Normal and Cauchy distributions with mean A summary of steps involved in the algorithm is described
ሺ௧ሻ ሺ௧ሻ herein -
Ɋ‫ܴܥ‬௥ and location parameter Ɋ‫ܨ‬௥ , respectively. The value
A. Input and initialization: the 3-DGs. As cases 3 and 6 are for both reconfiguration and
1. Input ܰ‫݌‬௜௡௜ , ܰ‫ܧܨ‬௠௔௫ DG allocation, the number of decision variables in these cases
2. Define vector ‫ ݔ‬and range of all its elements. is 11. Due to higher number of variables in cases 3 and 6, initial
3. Create random initial population of ܰ‫݌‬௜௡௜ such vectors population sizes are selected a little larger than the remaining
defined as ‫ݔ‬௜ as per equation (8). cases. Increasing the population size beyond 120 does not have
4. Set generation counter t = 0, dynamic population size notable impact in the outcome as number of fitness evaluations
ܰ‫݌‬ሺ‫ݐ‬ሻ ൌ ܰ‫݌‬௜௡௜ , evaluation counter ܰ‫ ܧܨ‬ൌ ͳ and performed for these cases is also higher. It may be noted that
ሺ଴ሻ ሺ଴ሻ
control parameters Ɋ‫ܨ‬௥ ൌ Ɋ‫ܴܥ‬௥ ൌ ͲǤͷ. initial population size and maximum number of fitness
evaluations are selected after several trials of the algorithm.
B. Algorithm loop:
ሺ௧ሻ V. RESULTS AND COMPARISONS
1. Evaluate ݂൫‫ݔ‬௜ ൯ , i.e. ܶܲ௅௢௦௦ as per equation (5) for
ሺ௧ሻ Simulation results with application of L-SHADE algorithm
‫ݔ‬௜ where i = 1 to Np. Increase counter NFE by Np i.e.
NFE = NFE + Np. are analyzed in this section. Each study case is run 5 times and
2. while termination criteria ܰ‫ ܧܨ‬൏ ܰ‫ܧܨ‬௠௔௫ do results are found to be consistent with negligible variations
3. for i = 1 to Np do among different runs. Table III summarizes the results together
--------------- with comparison of present study with similar past studies. The
ሺ௧ሻ ሺ௧ሻ best loss values among comparable algorithms are highlighted
4. Adapt control parameters ‫ܨ‬௜ and ‫ܴܥ‬௜ as per
in bold fonts. Selected DG ratings are included in the tables
equations (11) & (12).
ሺ௧ሻ
with bus nos. to allocate the DGs are provided in parentheses
5. Perform mutation to generate vector ‫ݒ‬௜ as per next to the selected optimal ratings.
equation (9).
ሺ௧ሻ
6. Perform crossover to generate element ‫ݑ‬௜ǡ௝ as per In case 1 of network reconfiguration for 33-bus system,
equation (13). algorithm L-SHADE achieves lowest loss value alongwith
ሺ௧ሻ UVDA [11]. Results of similar study case 4 for 69-bus system
7. Evaluate ݂൫‫ݑ‬௜ ൯ i.e. ܶܲ௅௢௦௦ as per equation (5) for
ሺ௧ሻ are same for all the algorithms. Little variation in decimal
‫ݑ‬௜ . Increase evaluation counter NFE by 1 i.e. NFE = places is observed possibly due to rounding off some numerical
NFE+1. values of network parameters. Change in opening of any switch
8. Select best fit individuals for next generation. If, among 55, 56, 57 and 58 in corresponding loop does not affect
ሺ௧ሻ ሺ௧ାଵሻ ሺ௧ሻ ሺ௧ାଵሻ ሺ௧ሻ
݂൫—௜ ሺ௧ሻ ൯ ൑ ݂൫‫ݔ‬௜ ൯, ‫ݔ‬௜ ൌ ‫ݑ‬௜ . Else ‫ݔ‬௜ ൌ ‫ݔ‬௜ . the final output results.
End for loop.
------------------ In study case 2 of DG siting and sizing for 33-bus system,
9. Update population size for next generation ܰ‫݌‬ሺ‫ ݐ‬൅ ͳሻ algorithm L-SHADE attains real power loss value of 72.90 kW,
as per LPSR strategy in equation (15). least among all. Another important aspect to note on this study
10. Increase generation counter t = t+1. Go to step 2 of result is cumulative rating for the DGs. While L-SHADE limits
algorithm loop. the total rating to 2.5 MW, ACSA [10] selected more than 3.2
MW. The difference is significant as larger DG rating incurs
TABLE II. L-SHADE ALGORITHM PARAMETER higher cost of installation. Output results of UVDA [11] with a
total cumulative DG rating of about 2.7 MW are better than
Parameter Case no. Value
ACSA [10]. On voltage regulation front, high DG rating helps
Dimension of optimization problem, d Case 1 & 4 5
to maintain slightly better voltage profile as can be observed
Case 2 & 5 6
from results of ACSA [10]. However, voltage regulation is not
Case 3 & 6 11
the primary objective of this optimization. Study case of DG
Initial population size, ܰ‫݌‬௜௡௜ Case 1, 2, 4 & 5 100 allotment for 69-bus system is case 5 where remarkably low
Case 3 & 6 120 loss figure of 69.60 kW is realized in present study. Like case
Maximum no. of fitness evaluations, Case 1, 2, 4 & 5 20,000 2, ACSA [10] selects largest total rating for the DGs.
ܰ‫ܧܨ‬୫ୟ୶ Case 3 & 6 25,000
In case 3 of simultaneous reconfiguration and DG allocation
Table II lists the dimensions of the optimization problem for in 33-bus network, present study reports a loss value of 53.15
various study cases and also applied parameters for L-SHADE kW, almost same as reported by ACSA [10]. However, total
algorithm. Cases 1 and 4 are for network reconfiguration only. selected DG rating by ACSA [10] is about 3.3 MW, close to
As mentioned before, 5 loops are recognized for each network. 90% loading of the network. L-SHADE optimally distributes a
A unique sectionalizing switch is to be identified in each loop total of 2.5 MW of DG power and suitably reconfigures the
for opening. Therefore, a total of 5 decision variables are to be network such that lowest loss can be achieved with minimum
optimized in these cases. Cases 2 and 5 are for allotment of rating of the DG. It may be noted that reference [9] adopted
suitably rated 3-DGs. The candidate bus for allocation of a DG sequential approach in placing and sizing the DGs. After
can be any bus in the network. 3 variables represent 3 locations network reconfiguration, candidate buses are identified for DG
(buses) in the network while remaining 3 signify ratings of all allocation followed by DG capacity optimization. However, as
TABLE III. SIMULATION RESULTS AND COMPARISON WITH PAST STUDIES
Network Case Case Parameter Various optimization algorithm
no. description L-SHADE FWA [9] ACSA [10] UVDA [11]
33-bus Case 1 Only Open switches 7, 9, 14, 32, 37 7, 9, 14, 28, 32 7, 9, 14, 28, 32 7, 9, 14, 32, 37
reconfiguration Real power loss (kW) 139.55 139.98 139.98 139.55
Min. bus voltage, p.u. (bus no.) 0.9378 (32) 0.9413 (32) 0.9413 (32) 0.9378 (32)
Case 2 Only DG Open switches 33, 34, 35, 36, 37 33, 34, 35, 36, 37 33, 34, 35, 36, 37 33, 34, 35, 36, 37
allocation Real power loss (kW) 72.90 88.68 74.26 74.21
DG sizes in kW (bus no.) 733.9 (14), 589.7 (14), 779.8 (14), 875 (11),
733.5 (25), 189.5 (18), 1125.1 (24), 931 (24),
1032.6 (30) 1014.6 (32) 1349.6 (30) 925 (29)
Min. bus voltage, p.u. (bus no.) 0.9658 (33) 0.9680 (30) 0.9778 (33) 0.962 (33)
Case 3 Simultaneous Open switches 7, 9, 14, 27, 30 7, 11, 14, 28, 32 11, 28, 31, 33, 34 7, 10, 13, 27, 32
reconfiguration Real power loss (kW) 53.15 67.11a 53.21 57.29
and DG DG sizes in kW (bus no.) 523.8 (12), 531.5 (18), 964.6 (7), 649 (15),
allocation 779.6 (18), 615.8 (29), 896.8 (18), 486 (21),
1196.6 (25) 536.7 (32) 1438.1 (25) 1554 (29)
Min. bus voltage, p.u. (bus no.) 0.9688 (31) 0.9713 (14) 0.9806 (31) 0.976 (32)
69-bus Case 4 Only Open switches 14, 55, 61, 69, 70 14, 56, 61, 69, 70 14, 57, 61, 69, 70 14, 58, 61, 69, 70
reconfiguration Real power loss (kW) 98.60 98.59 98.59 98.58
Min. bus voltage, p.u. (bus no.) 0.9495 (61) 0.9495 (61) 0.9495 (61) 0.9495 (61)
Case 5 Only DG Open switches 69, 70, 71, 72, 73 69, 70, 71, 72, 73 69, 70, 71, 72, 73 69, 70, 71, 72, 73
allocation Real power loss (kW) 69.60 77.85 72.44 72.63
DG sizes in kW (bus no.) 423.1 (11), 225.8 (27), 602.2 (11), 604 (11),
380.4 (18), 1198.6 (61), 380.4 (18), 417 (17),
1696.5 (61) 408.5 (65) 2000 (61) 1410 (61)
Min. bus voltage, p.u. (bus no.) 0.9776 (65) 0.9740 (62) 0.9890 (65) 0.9688 (65)
Case 6 Simultaneous Open switches 14, 58, 61, 69, 70 13, 55, 63, 69, 70 14, 58, 61, 69, 70 14, 58, 63, 69, 70
reconfiguration Real power loss (kW) 35.54 39.25a 37.02 37.11
and DG DG sizes in kW (bus no.) 517.6 (11), 1127.2 (61), 541.3 (11), 538 (11),
allocation 559.4 (27) 275.0 (62), 1724.0 (61), 673 (17),
1423.0 (61) 415.9 (65) 553.6 (65) 1472 (61)
Min. bus voltage, p.u. (bus no.) 0.9811 (61) 0.9796 (61) 0.9869 (50) 0.9816 (63)
a. DG location and size are optimized sequentially

observed from results of other methods, simultaneous siting and 1.01


sizing of DGs are more effective in reducing system losses. In
1
case 6 for 69-bus system, the minimum real power loss of 35.54
kW is achieved by L-SHADE algorithm with total added 0.99
capacity for the DGs is capped at 2.5 MW. When compared
Voltage (in p.u.)

with UVDA [11], the loss is about 1.5 kW lower with about 200 0.98
kW less total proposed capacity of the DGs. 0.97

Fig. 4 shows voltage profiles of the network buses for various 0.96
study cases performed for IEEE 33-bus system. Voltage profile Case-4 Case-5
0.95
is the best when network reconfiguration is accompanied by Case-6
optimal sizing and placement of the DGs. 0.94
1 7 13 19 25 31 37 43 49 55 61 67
1.01 Bus no.
1 Fig. 5: Bus voltage profiles of 69-bus system for various case studies
0.99
Fig. 5 is for bus voltage profiles of various study cases for
Voltage (in p.u.)

0.98
IEEE 69-bus system. Justifiably best profile is observed in case
0.97
6 of simultaneous reconfiguration and DG allocation. In both
0.96 the diagrams of voltage profiles, clearly voltage limits (0.90
0.95 Case-1 p.u. – 1.05 p.u.) for the network buses are duly compiled in all
Case-2 study cases.
0.94
Case-3
0.93 VI. CONCLUSION
1 5 9 13 17 21 25 29 33
Bus no. This paper presents a useful and effective application of L-
SHADE algorithm for optimization of simultaneous discrete
Fig. 4: Bus voltage profiles of 33-bus system for various case studies and continuous variables in a real-world problem of distribution
network. With selection of more appropriate network switches [9] Imran, A. Mohamed, M. Kowsalya, and D. P. Kothari. "A novel
integration technique for optimal network reconfiguration and distributed
for opening, with little reshuffle of DG location, and with more generation placement in power distribution networks." International
accurate distribution of DG capacity, L-SHADE achieves the Journal of Electrical Power & Energy Systems 63 (2014): 461-472.
lowest real power loss in any study case among equivalent [10] Nguyen, Thuan Thanh, Anh Viet Truong, and Tuan Anh Phung. "A novel
algorithms. Lowering loss by any margin is both technically method based on adaptive cuckoo search for optimal network
and commercially beneficial. The proposed algorithm is highly reconfiguration and distributed generation allocation in distribution
network." International Journal of Electrical Power & Energy
efficient and productive in finding most optimal solutions for Systems 78 (2016): 801-815.
the single objective optimization problem discussed in this [11] Bayat, A., A. Bagheri, and R. Noroozian. "Optimal siting and sizing of
literature. Application of the algorithm on a larger network with distributed generation accompanied by reconfiguration of distribution
higher number of buses and on a network with added VAR networks for maximum loss reduction by using a new UVDA-based
compensators remains the topic for future study. heuristic method." International Journal of Electrical Power & Energy
Systems 77 (2016): 360-371.
[12] Liang, J. J., B. Y. Qu, and P. N. Suganthan "Problem definitions and
evaluation criteria for the CEC 2014 special session and competition on
ACKNOWLEDGMENT single objective real-parameter numerical optimization." Computational
This project is funded by the National Research Foundation Intelligence Laboratory, Zhengzhou University, Zhengzhou China and
Technical Report, Nanyang Technological University, Singapore (2013).
Singapore under its Campus for Research Excellence and
Technological Enterprise (CREATE) program. [13] Partha P. Biswas, P. N. Suganthan, and Gehan AJ Amaratunga. "Optimal
placement of wind turbines in a windfarm using L-SHADE algorithm."
In Evolutionary Computation (CEC), 2017 IEEE Congress on, pp. 83-88.
REFERENCES IEEE, 2017.
[14] Partha P. Biswas, Noor H. Awad, P. N. Suganthan, Mostafa Z. Ali, and
[1] Ng, H. N., M. M. A. Salama, and A. Y. Chikhani. "Classification of
Gehan AJ Amaratunga. "Minimizing THD of multilevel inverters with
capacitor allocation techniques." IEEE Transactions on power
optimal values of DC voltages and switching angles using LSHADE-
delivery 15.1 (2000): 387-392.
EpSin algorithm." In Evolutionary Computation (CEC), 2017 IEEE
[2] Imran, A. Mohamed, and M. Kowsalya. "A new power system Congress on, pp. 77-82. IEEE, 2017.
reconfiguration scheme for power loss minimization and voltage profile
[15] Partha P. Biswas, P. N. Suganthan, and Gehan AJ. Amaratunga.
enhancement using Fireworks Algorithm." International Journal of
"Minimizing Harmonic Distortion in Power System with Optimal Design
Electrical Power & Energy Systems 62 (2014): 312-322.
of Hybrid Active Power Filter using Differential Evolution." Applied Soft
[3] Nguyen, Thuan Thanh, and Anh Viet Truong. "Distribution network Computing 61C (2017): 486-496.
reconfiguration for power loss minimization and voltage profile
[16] Partha P. Biswas, R. Mallipeddi, P. N. Suganthan, and Gehan AJ.
improvement using cuckoo search algorithm." International Journal of
Amaratunga. "A multiobjective approach for optimal placement and
Electrical Power & Energy Systems 68 (2015): 233-242.
sizing of distributed generators and capacitors in distribution
[4] Naveen, S., K. Sathish Kumar, and K. Rajalakshmi. "Distribution system network." Applied Soft Computing 60 (2017): 268-280.
reconfiguration for loss minimization using modified bacterial foraging
[17] Aman, M. M., Jasmon, G. B., Bakar, A. H. A., & Mokhlis, H. (2014).
optimization algorithm." International Journal of Electrical Power &
Optimum network reconfiguration based on maximization of system
Energy Systems 69 (2015): 90-97.
loadability using continuation power flow theorem. International journal
[5] Flaih, F. M., Xiangning, L., Dawoud, S. M., & Mohammed, M. A. (2016, of electrical power & energy systems, 54, 123-133.
October). Distribution system reconfiguration for power loss
[18] Qin A. Kai, V. L. Huang, and P. N. Suganthan. "Differential evolution
minimization and voltage profile improvement using Modified particle
algorithm with strategy adaptation for global numerical optimization."
swarm optimization. In Power and Energy Engineering Conference
IEEE transactions on Evolutionary Computation 13.2 (2009): 398-417.
(APPEEC), 2016 IEEE PES Asia-Pacific (pp. 120-124). IEEE.
[19] R. Mallipeddi, P. N. Suganthan, Q. K. Pan, and M. F. Tasgetiren.
[6] Viral, Rajkumar, and D. K. Khatod. "An analytical approach for sizing
"Differential evolution algorithm with ensemble of parameters and
and siting of DGs in balanced radial distribution networks for loss
mutation strategies." Applied Soft Computing 11, no. 2 (2011): 1679-
minimization." International Journal of Electrical Power & Energy
1696.
Systems 67 (2015): 191-201.
[20] Tanabe, Ryoji, and Alex Fukunaga. "Success-history based parameter
[7] Ayodele, T. R., A. S. O. Ogunjuyigbe, and O. O. Akinola. "Optimal
adaptation for differential evolution." 2013 IEEE Congress on
Location, Sizing, and Appropriate Technology Selection of Distributed
Evolutionary Computation (CEC). IEEE, 2013.
Generators for Minimizing Power Loss Using Genetic Algorithm."
Journal of Renewable Energy 2015 (2015). [21] Tanabe, Ryoji, and Alex S. Fukunaga. "Improving the search performance
of SHADE using linear population size reduction." 2014 IEEE Congress
[8] Kefayat, M., A. Lashkar Ara, and SA Nabavi Niaki. "A hybrid of ant
on Evolutionary Computation (CEC). IEEE, 2014.
colony optimization and artificial bee colony algorithm for probabilistic
optimal placement and sizing of distributed energy resources." Energy
Conversion and Management 92 (2015): 149-161.

You might also like