Article in Press

G Model
ASOC 3714 1–22 ARTICLE IN PRESS

Applied Soft Computing xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Applied Soft Computing

journal homepage: www.elsevier.com/locate/asoc
1 Integrated approach of network reconfiguration with distributed

2 generation and shunt capacitors placement for power loss
3 minimization in radial distribution networks
4 K. Muthukumar (Assistant Professor) a,∗ , S. Jayalalitha (Associate Dean) b
5 Q2 a
EEE/SEEE, SASTRA University, Tirumalaisamudram, Thanjavur, Tamilnadu, India
6 Q4
Q3 b
EIE/SEEE, SASTRA University, Tirumalaisamudram, Thanjavur, Tamilnadu, India
7
8
23 a r t i c l e i n f o a b s t r a c t
9
10 Article history: This article presents the significance of efficient hybrid heuristic search algorithm(HS-PABC) based on
11 Received 5 January 2016 Harmony search algorithm (HSA) and particle artificial bee colony algorithm (PABC) in the context of
12 Received in revised form 15 July 2016 distribution network reconfiguration along with optimal allocation of distributed generators and shunt
13 Accepted 17 July 2016
capacitors. The premature and slow convergence over multi model fitness landscape is the main limitation
14 Available online xxx
in standard HSA. In the proposed hybrid algorithm the harmony memory vector of HSA are intelligently
15
enhanced through PABC algorithm during the optimization process to reach the optimal solution within
16 Keywords:
the search space. In hybrid approach, the exploration ability of HSA and the exploitation ability of PABC
17 Radial distribution network (RDN)
18 Particle artificial bee colony algorithm
algorithm are integrated to blend the potency of both algorithms. The box plot and Wilcoxon rank sum
19 (PABC) test are used to show the quality of the solution obtained by hybrid HS-PABC with respect to HSA.The
20 Distributed generator (DG) computational results prove the integrated approach of the network reconfiguration problem along with
21 Shunt capacitors optimal placement and sizing of DG units and shunt capacitors as an efficient approach towards the
22 Harmony search algorithm (HSA) objective. The results obtained on 69 and 118 node network by proposed method and the standard HSA
reveals the powerfulness of the proposed approach which guarantees to achieve global optimal solution
with less iteration.
© 2016 Elsevier B.V. All rights reserved.
24 1. Introduction must be implemented to get the benefits of integrating the DG units 39
and shunt capacitors into the distribution networks. 40
25Q5 The locality of electrical energy generation is far away from the The network reconfiguration of the RDN is the method of chang- 41
26 consumer loads connected with lengthy feeder lines leads to addi- ing the topological structure of the network by opening and closing 42
27 tional power loss in the transmission and distribution network. of sectionalizing and tie switches to achieve optimal topology with 43
28 Power loss reduction is achieved by reconfiguring the existing net- minimum power loss. During the reconfiguration process, the sys- 44
29 work topology and by installing fixed/switched shunt capacitor tem radiality should be maintained with all loads connected to the 45
30 banks and distributed generation units (DG) in close proximity to network. In recent years, a noticeable research work has been car- 46
31 the consumer loads in transmission and distribution networks. The ried out for loss reduction using network reconfiguration problem. 47
32 allocation of such sources has numerous advantages such as post- Since there are numerous candidate switching combinations to find 48
33 ponement for investing new transmission and distribution network the optimal topology of the network, the reconfiguration prob- 49
34 construction, reduction in power loss, bus voltage profile enhance- lem is modeled as combinatorial, non-differentiable, constrained 50
35 ment. Prior to the implementation of loss reduction techniques in optimization problem. The discrete nature of sectionalizing and tie 51
36 the distribution network; there is a necessary to investigate their switches along with radiality constraints avert the application of 52
37 consequence, such as power loss, bus voltage magnitude, harmonic classical optimization methodologies. So there has been a growing 53
38 distortion and system voltage stability. A suitable planning method interest in various population based heuristic search algorithms 54
such as Artificial Immune (AIS) Systems [1], Modified particle 55
swarm optimization [2], Binary group search optimization [3], 56
Adapted ant colony [4], Fireworks algorithm [5], Harmony search 57
[6] algorithm. A fuzzy multi objective network reconfiguration 58

∗ Corresponding author. methodology for radial distribution systems has been proposed in 59
E-mail address: muthupoy75@yahoo.com (M. K.).
http://dx.doi.org/10.1016/j.asoc.2016.07.031
1568-4946/© 2016 Elsevier B.V. All rights reserved.
Please cite this article in press as: M. K., J. S., Integrated approach of network reconfiguration with distributed generation
and shunt capacitors placement for power loss minimization in radial distribution networks, Appl. Soft Comput. J. (2016),
G Model
2 M. K., J. S. / Applied Soft Computing xxx (2016) xxx–xxx
HS algorithm and the simulation results are compared with the har- 95
Nomenclature mony search algorithm (HSA), improved harmony search algorithm 96
(IHSA), global harmony search algorithm (GHSA) and self adaptive 97

QDG i Reactive power injection of ith DG unit global harmony search (SGHS) algorithm. The parameters of hybrid 98
DG
p.f max Maximum allowable power factor of DG unit algorithm and its impact has been studied with uniform design 99
p.f DG min Minimum allowable power factor of DG unit experiments optimization problems. In [24,25] the optimal design 100
p.f DG i Power factor of ith DG unit of water distribution networks is addressed using HSA incorpo- 101
nbus Total number of nodes in the RDN rated particle swarm algorithm (PSO) and the result outcomes are 102
nb Total branches in the RDN better than the genetic algorithm, simulated annealing and Tabu 103
nc Total number of capacitors to be installed in RDN search technique. In [26] a self-adaptive global best harmony search 104
Ii,i + 1 Current flow between ith branch and i + 1th branch algorithm (SGHS) is proposed which utilize a new harmony mem- 105
Ii,i + 1max Allowable maximum permissible current at branch ory enhancement strategy by dynamic adaptation of HMCR and 106
i+1 PAR, distance bandwidth (BW) as learning mechanism to balance 107
IBmax(i) Maximum allowable branch current the exploration and exploitation ability. In [27] a hybrid heuris- 108
Wcomp
IBmax(i) Branch current flow in ith branch with compensa- tic algorithm is proposed which makes use of sequential quadratic 109
tion programming technique (HSA-SQL) to accelerate the local search 110

w/comp
Ploss(i,i+1) Real ower loss in the branch between nodes i and ability and to get better accuracy in HSA solutions. To exhibit 111
i + 1 with DG units, shunt capacitor the effectiveness and sturdiness of the proposed hybrid HSA-SQL 112
w/comp algorithm, various benchmark engineering optimization problems
PTloss Total power loss in the RDN with DG units, shunt 113
capacitor are taken into consideration. The intelligent honey bee foraging 114
|Vi | Bus voltage magnitude of ith bus behavior of bee swarm is utilized in [28–30] to obtain the optimal 115
spec
Vmin Specified lower bound of bus voltage of the RDN solution in multi-dimensional numerical optimization problems. 116
spec The simulation results outperform the results obtained with the 117
Vmax Specified upper bound of bus voltage of RDN
other meta-heuristic algorithms such as a particle swarm opti-
PDG
118
max Maximum size of DG unit in kilowatts
mization algorithm, differential evolution algorithm and genetic
PDG
119
min Minimum size of DG unit in kilowatts
algorithms. In calculus based methods, the optimal solution is
QcL
120
Sum of total reactive demand of the RDN
obtained by using derivatives which is suitable only for continuous- 121
Qcj Reactive power injection by the jth shunt capacitor
valued functions rather than discrete-valued functions. In [31,32]
Pi DG
122
Size of ith DG unit in kilowatts
a new HS algorithm is proposed for solving engineering optimiza- 123
tion problems with continuous as well as discrete design variables. 124
In [33,34] the overview of the recent applications of HSA, which 125
60 [7]. Tabu search approach has been proposed for network reconfig- uses a ‘probabilistic-gradient’ to select the neighboring values of 126
61 uration in [8–10]. Mathematical illustration of radiality constraints decision variables is addressed. It is an efficient meta-heuristic 127
62 associated with the radial distribution network reconfiguration optimization tool for practitioners to solve complex optimization 128
63 problem has been addressed in [11]. Optimal network reconfigu- paradigms such as construction, telecommunications, engineering, 129
64 ration along with the placement of distribution generation units health and energy, and robotics. The HSA provides probabilistic- 130
65 for power loss minimization has been proposed in [12,13]. As gradient based search to get the local or global optimal solution 131
66 compared with classical optimization techniques, meta-heuristic instead of mathematical gradient as in conventional optimization 132
67 algorithms are capable to achieve a near optimal solution for DG techniques. In [35–38] a HSA based optimal solution is obtained for 133
68 units and shunt capacitors sizing and placement problem in dis- complex optimization problems like scheduling of multiple dam 134
69 tribution networks. Artificial Bee Colony (ABC) algorithm which systems, broadcast scheduling in packet radio networks, estima- 135
70 mimics the foraging nature of honey bees swarm has been proposed tion of the success of companies and vehicle routing. In [39] a hybrid 136
71 in [14] to find the best possible sizing of static capacitors. Various grouping HSA is proposed for the multiple-type access node loca- 137
72 models and methods applied for the optimal DG allocation, impact tion problem to determine the optimum location. In [40] a hybrid 138
73 of dispatchable and non-dispatchable type DG units integration in approach is used to deploy 24-h medical emergency resources by 139
74 the modern distribution system, current and future trends in this combining the HSA with the grouping encoding concept to repair 140
75 field has been addressed in [15,16]. Analytical approach based DG infeasible solutions. In [41] a multi-objective harmony search for 141
76 units sizing and its optimal operating power factor for different urban road network reconfiguration problem to offer near-optimal 142
77 types of DG unit’s for power loss reduction has been addressed in solution to improve the vehicles mobility is proposed. In [42] a 143
78 [17]. An analytical approach for simultaneous allocation and sizing quasi-oppositional harmony search algorithm is proposed to inves- 144
79 of DG units and shunt capacitor to achieve power loss reduction is tigate the optimal controller gains to enhance the performance of 145
80 proposed in [18]. Automatic Generation Control (AGC) of the power system. A new 146
81 In [19] a heuristic algorithm, mimicking the improvisation pro- hybrid PSO algorithm has been addressed in [43] to augment the 147
82 cess of musical instruments to get a pleasant harmonious melody exploration and exploitation capability by introducing the global 148
83 so called harmony search algorithm (HSA) is proposed. The perfor- dimension selection strategy using HSA and validated with the PSO 149
84 mance of the developed algorithm is demonstrated with a traveling variants, and other meta-heuristic algorithms. In [44] a new self- 150
85 salesman problem and minimum-cost pipe network design prob- adaptive HS-PSO search algorithm is proposed with an effective 151
86 lem. In [20] an improved harmony search algorithm (IHSA) is initialization scheme by utilizing the PSO algorithm to improve the 152
87 proposed which intelligently generates new solution vector for solution quality of the initial harmony memory in the HSA. A new 153
88 the best solution and convergence of HSA. HSA is a stochastic self-adaptive adjusting method for control parameters PAR and BW 154
89 search procedure which does not need any derivative information is designed to accelerate the convergence rate and solution accu- 155
90 to solve the complex combinatorial optimization problem. In [21] racy of the proposed algorithm. The poor exploitation ability of the 156
91 the discrete search strategy of HSA is utilized for structural size ABC algorithm makes an issue of slow convergence in solving non- 157
92 optimization problem with discrete design variables. In [22,23] a linear and constrained optimization nature of engineering design 158
93 hybrid heuristic search approach is proposed which makes use of problems. To overcome these insufficiencies, a modified version of 159
94 ABC algorithm and its variants to enhance the solution vector in the the ABC algorithm is suggested in [45–49] by incorporating adap- 160
G Model
M. K., J. S. / Applied Soft Computing xxx (2016) xxx–xxx 3
161 tive inertia weights and acceleration coefficient parameters of PSO 2.2. Maximum branch current capacity limit index (MBCCLI) 218
162 algorithms.
163 The combined approach of network reconfiguration problem This index is used to reflect the improvement in reserve capacity 219
164 along with the placement of distributed generators and shunt limit of the feeder lines. It is the ratio of current flow through the 220
165 capacitors is complex combinatorial problems which involves dis- feeder lines after compensation and the maximum current carrying 221
166 crete (status of sectionalizing and tie switches) and continuous limit of that feeder branch as in Eq. (3), 222
167 control variables (size of DG units and shunt capacitors). To find the
wcomp
168 optimal topology all the feasible radial topologies of the network IB(i)
MBCCLI = max( ) (3) 223
169 are to be evaluated. Tracing such optimal topology is computa- IBmax(i)
170 tionally expensive and NP- hard in nature. Solving such type of
171 hard optimization problems using meta-heuristic techniques will
2.3. Voltage deviation index (VDI) 224
172 lead to premature convergence and local optima. Improper bal-
173 ance between intensification and diversification to find the optimal
The optimal network reconfiguration along with DG units and 225
174 solution is the major issue in meta-heuristic techniques. This neces-
shunt capacitor placement provides the reduction in bus voltage 226
175 sitates the hybridization of meta-heuristics which makes use of
deviation. The bus voltage deviation index (VDI) can be defined as 227
176 complementary characteristic of different optimization strategies
in Eq. (4) 228
177 to get the improved solution.
178 In this article, hybridization of HSA with PSO embedded artificial V1 − Vi
179 bee colony algorithm (HS-PABC) has been proposed and utilized VDI = max( )∀i = 1, 2, .....nbus whereV1 229
V1
180 to provide the optimal network reconfiguration problem along
181 with the placement of DG units and shunt capacitors to realize is taken as 1.0p.u (4) 230
182 minimization of real power loss at distinct load levels. In hybrid,

231
183 HS-PABC approach, the global searching ability (exploration) of
184 HSA and the local searching (exploitation) ability of PABC are inte-
185 grated to harmoniously enhance the HS algorithm performance. In 2.4. Constraints associated with the objective function 232
186 this way, the strength of local searching ability of standard HSA is
187 improved to find out the best solution quickly and ride over the The nominal bus voltage limit, thermal limit of the feeder lines, 233
188 local optima. In this work, different test scenarios are considered maximum real and reactive power injection by the DG units and 234
189 with an aim to enumerate the benefits of RDN through graph theory shunt capacitors, radial structure of the distribution network are 235
190 based network reconfiguration along with the allocation of shunt considered as constraints in Eqs. (5)–(10). 236
191 capacitors and DG units. Simulation results point out that the pro-
spec spec
192 posed hybrid HS-PABC algorithm is proficient than standard HSA |Vmin | ≤ |Vi | ≤ |Vmax | (5) 237
193 algorithm.
194 This article is structured as: In Section 2 the formulation of the 0.9 ≤ |Vi | ≤ 1.1 i = 1, 2, ....Nbus
195 problem and graph theory based network reconfiguration is dis- (6) 238
|Ii,i+1 | ≤ |Ii,i+1max |, i = 1, 2, ....Nb
196 cussed. Outline of standard HS algorithm and the proposed hybrid
197 HS-PABC algorithm is summarized in Section 3. In Section 4, simu- nc nl
198 lation results and discussion are presented. In Section 5 conclusions Qcj ≤ 1.0 Qc L (7) 239
j=1 j=1
199 of the work are presented.
The DG units are modeled as controllable synchronous machine 240
200 2. Problem formulation based biomass plants with active and reactive power injection 241
capabilities at an operating power factor of 0.85 lagging [17]. The 242
201 A multi objective optimization model has been proposed which real power output (kW) range of DG units sizing is selected as 243
202 includes technical factors such as real power loss reduction, line 10–50% of the total active power demand of the RDN. 244
203 loading reduction and voltage profile improvement. The multi

PDG DG
min ≤ Pi ≤ PDG
max (8) 245
204 objective proposition is considered simultaneously using weight-
Q6 ing factors for obtaining maximum benefits as in Eq. (1). All the
205

n
n
performance indices are given equal weightage. where PDG PDG and PDG PDG
min = 0.1 Max = 0.5
206
246
i i
i=2 i=2
207 Min.fitness = W1 (RPLRI) + W2 (MBCCLI) + W3 (VDI) (1)
3 QDG = PDG −1
i tan(cos (p.f)) (9) 247
208 where w = 1.0 wi [0,1] .
i=1 i
i
209 A set of performance improvement indices is described mathe-

p.fDG DG
max ≥ p.fi ≥ p.fDG
min (10) 248
210 matically as stated below to enumerate the technical benefits.
249
Radiality constraints:
211 2.1. Real power loss reduction index (RPLRI)
Nb = Nbus − 1 250
212 The objective of real power loss reduction index is to quan-

213 tify the power loss reduction in the distribution network. The real where Nb is the number of branches, Nbus is the number of nodes 251
214 power loss reduction index (RPLRI) is defined as the ratio between in the RDN. 252
215 the real power loss of the network with and without DG units and
216 shunt capacitor compensation as in Eq. (2), 2.5. Spanning tree approach for network reconfiguration 253
plosswcomp A directed graph representation of the radial distribution net-

217 RPLRI = ( ) (2) 254
plosswocomp work is utilized to form the adjacency matrix (AM) with ‘1’s and
G Model
255 ‘0’s representing the presence and absence of direct edge between Step 4: Obtain the branch current [BC] by multiplying BIBC 305
256 nodes. The node isolation is identified by determining the columns matrix obtained from step 3 with bus current injection [I] 306
257 of the adjacency matrix with only zeros except the root node and in
258 degree of each node is limited to one by determining the columns
[BC] = [BIBC ∗ I] (13) 307
259 with only one nonzero element. The formation of the loop dur-
260 ing reconfiguration process is identified using MATLAB function
261 graphisdag (AM). Step 5: The distribution load flow matrix (DLF) is obtained by 308
262 In this proposed work, the graph data structure is used to con- multiplying BCBV and BIBC matrices which relates the bus current 309
263 struct the RDN as spanning tree. The vertices representing the injection and bus voltages and it is expressed as, 310
264 nodes of RDN connected with edges (branches) are used to form
265 the directed graph. The load flow algorithm developed is based
V = [BCBV] ∗ [BIBC] ∗ [I] = [DLF ∗ I] (14) 311
266 on the formation of BIBC and BCBV matrices [50]. The BIBC matrix
267 relates the bus current injection with branch current where as BCBV
268 matrix relates the branch current with bus voltage which repre- The bus current injection in i + 1th node at kth iteration can be 312
269 sents the line impedance of the RDN. In each switching operation, obtained as follows, 313
270 the adjacency matrix and BCBV matrix are changed in accordance
271 with the change in the system structure. The depth-first search
272 (DFS) discovery order is utilized for traversing the graph from P2i+1 + Q2i+1
each node. Based on the discovery order from each node, the cor- Iki+1 = (15) 314
|Vi |2
273
274 responding BIBC matrix is generated by assigning ‘1 to discover

275 paths.
276 The hypothetical 6 node RDN with sectionalizing and tie switch The change in bus voltage at i + 1th node at k + 1th iteration can be 315
277 is shown in Fig. 1(a). The adjacency matrix representation is shown obtained as follows, 316
278 in Eq. (18). Let E be a set of direct edges {(1,2),(2,3),(3,4),(4,5),(3,6)}.

279 Each element in the adjacency matrix AMij = 1 if (i,j)C -- E and AMij = 0 K+1 K
Vi+1 = [DLF ∗ Ii+1 ] (16) 317
280 if (i,j) ∈
/ E. The adjacency matrix before closing the tie switch is
281 formulated as in Eq. (11).
K+1
VK+1
I+1
= V0 + V I+1 (17) 318
0 1 0 0 0 0
0 0 1 0 0 0 In this similar fashion, the bus voltages are updated during iter- 319
0 0 0 1 0 1
282 AM = [ ] (11) ative steps. The iterative steps are terminated once the required 320
0 0 0 0 1 0 tolerance limit is reached. 321
0 0 0 0 0 0
0 0 0 0 0 0
3. Evolutionary algorithms 322

283 Fig. 1(b) represents the directed graph generated using adjacency
284 matrix represented in (11). The RDN structure modification is
3.1. Overview of harmony search algorithm 323
285 obtained by closing the tie switch between the node 5 & 6 and
286 opening the sectionalizing switch between the node 4 and 5 and
Harmony search algorithm (HSA) was proposed by Geem et al. 324
287 the corresponding modified adjacency matrix obtained is shown
[19]. The HSA algorithm requires fewer mathematical calcula- 325
288 in Eq. (12). Fig. 1(c) represents the directed graph generated using
tions and it does not require any derivative information due to its 326
289 adjacency matrix represented in Eq. (12) for the modified structure
stochastic searching nature. Various advantages of HS algorithm 327
290 of the RDN.
are (a) it does not require differential gradients, thus it can be suit- 328
able for discontinuous as well as continuous functions; (b) it can 329

0 1 0 0 0 0 have the capability to handle discrete as well as continuous deci- 330
0 0 1 0 0 0 sion variables [32]; (c) it does not require initial value setting for 331
0 0 0 1 0 1 the decision variables.
291 AM = [ ] (12) 332
0 0 0 0 0 0 HMCR, PAR and BW are considered as tuning parameters in 333
0 0 0 0 0 0 HSA. The computational procedures of HSA are a) random initial- 334
0 0 0 0 1 0 ization of decision variables in the harmony memory vector (HMV) 335
b) Improvisation of HMV values by updating the solutions by mem- 336
292 The steps involved in solving the optimal network reconfigura- ory consideration c) pitch adjustment and random re-initialization 337
293 tion problem of RDN are lined as follows: process. Repeat the improvisation process in the HMV until the 338
294 Step 1: Generate the adjacency matrix (AM) of the given RDN termination criteria. 339
295 with ‘1’ and ‘0’ representing the presence and absence of direct edge Updating of harmony memory vector is described below. 340
296 between the nodes.

297 Step2: Generate the directed graph using function, bio-graph
298 (bg) from the adjacency matrix (AM).
299 Step 3: Traverse the directed graph in depth first search (DFS)
300 order using function traverse (bg, S) where S = 1,2. . ..n to obtain dis-
301 covery vector. The power flow path matrix [i.e., BIBC] is formed by
302 filling ‘1’ in position (i,j) where ith row corresponds to each value of (18) 341
303 S and jth column corresponds to each element in discovery vector HM is pitch adjusted or modified slightly with a probability of 342
304 with respect to S. PAR as, 343
G Model
Fig. 1. (a) Example hypothetical 6 node RDN with directed graph of the hypothetical RDN (b) before closing of tie switch (c) after closing of tie switch.
344
345 (19)where “BW” represents a random distance bandwidth. In this

346 way the HMV is improved iteratively.
positions of each particle is updated using velocity and position 364
equations expressed as below, 365
Vit+1 = ω.Vit + C1 rand(1) (Pbestit − Xid

t
) + C2 rand(2) (Gbestit − Xid
t
) (20) 366
347 3.2. PSO embedded artificial bee colony algorithm (PABC
algorithm)
Xit+1 = (Xit + Vit+1 ),
348
i = 1, 2, ......., N, d=1,2,......, D (21) 367
349 Particle swarm optimization (PSO) is one of the recent stochas- where “N” is the total number of particles; D is the dimension of 368
350 tic algorithm based on the idea of swarm intelligence of social the search space. Vit symbolize the velocity of the ith particle at 369
351 models such as bird flocking, fish schooling [50]. In PSO, simul- t is the ith particle position at the tth iteration.
the tth iteration;Xid 370
352 taneous co-existence of multiple candidate solutions flies within The term “␻” indicate the inertia weight which provides a balance 371
353 the problem search space. Each particle adjusts its flying position between exploration and exploitation. The C1 and C2 are positive 372
354 based on its own experience as well as the best flying position of social weight parameters, which has control over the preeminence 373
355 the neighboring particles to reach the optimal solution point. In and neighboring particle’s position on the current one. “Pbest ” imi- 374
356 this way each particle retains its best position (Pbest —local search tates the previous best position of the particle and “Gbest ” known 375
357 through its own experience) and its neighboring particle’s best as global best position which provides the best “Pbest ” amongst all 376
358 position (Gbest —global search through neighboring particles expe- the Pbest position; rand(1) and rand(2) are the random real value in 377
359 rience) during the search process to find the optimal solution within the range of [0,1]. 378
360 the search space of the problem. Karaboga and Basturk [28] proposed artificial bee colony algo- 379
361 The PSO algorithm has three steps: The algorithm process starts rithm (ABC) from the foraging nature of honey bee swarm to solve 380
362 the initialization of randomly distributed populations within the the constrained optimization problems. The ABC algorithm con- 381
363 search solution space. Then at each iteration, the velocities and sists of the following groups; employed bee, onlooker bee and scout 382
G Model
383 bee. The role of employed bees is to search for the food source and capability of the HSA has been enhanced by crossbreeding the HSA 440
384 share the information to the onlooker bees; based on the better food with other optimization algorithms [22]. 441
385 source information shared from the employed bees, the onlooker It is evident that in the HS algorithm the harmony memory (HM) 442
386 further searches the better food source neighboring around the plays a vital part which decides the algorithm performance. In this 443
387 selected food source. If any food source is found abandoned, the work, to augment the solution accuracy of the standard HSA, the 444
388 new food source is randomly searched by the scout bees. particle ABC algorithm is integrated and a new HS-PABC has been 445
389 Let us suppose the solution space of the optimization problem proposed. In hybrid HS-PABC algorithm, the HMV in the HSA is 446
390 is D—dimensional. The algorithm is initiated with randomly gen- taken as the initial food source by the PABC algorithm to produce 447
391 erated population and it can be represented as Xj = (Xj1 , Xj2 ,. . .. a diverse range of potentially optimal solutions while at the same 448
392 XjD ) where j D {1,2,. . ..,SN). SN is the size of the food source. Then instant it intensify its search in the vicinity of optimal solution. New 449
393 the employed bee’s could modify the solution in its memory by improvised HMV values are used as initial values by the PABC algo- 450
394 randomly selecting a neighboring food source position using the rithm to search for optimal solution. If the locally optimized vector 451
395 following Eq. (22), produced by the PABC algorithm has better fitness value than those 452
in the HMV, they will be retained in the HM. The harmony vec- 453
396 Vjt+1 = (Xjt + rand(1)(Xjt − Xkt ) (22) tor (HMV) in the HSA is improved by utilizing the potency of both 454
heuristic algorithms to get the best solution. Besides this hybrid 455
397 where K D {1,2,. . .. . .,SN) is chosen randomly with j =
/ K; rand (1)
algorithm requires the tuning of the control parameters such as 456
398 is a random number within the range of [−1,1].
Harmony Memory Size (HMS), Pitch Adjustment Rate (PAR), Har- 457
399 In onlooker bee phase, the onlooker bees start searching the
mony Memory Consideration Rate (HMCR) and Maximum Cycle 458
400 solution around the best food source provided by the employed
Number (MCN). The flowchart of the proposed hybrid HS- PABC 459
401 bees using Eq. (22) based on the probability value “Pj ” as in Eq. (23)
algorithm is presented in Fig. 2. 460
fitnessj The computational procedure of the proposed hybrid HS-PABC 461

402 Pj = HMS (23) algorithm is detailed as follows, 462
j=1
fitnessj
Step 1: Initialize the parameters of HS-PABC algorithm; HMS, 463
403 Fitnessj is the fitness value of jth the solution. The fitness of each PAR, HMCR, MCN. 464
404 solution is computed by Eq. (24) as, Step 2: Initialize the Harmony Memory Vector (HMV) with ran- 465
dom population 466

405 Fitnessj = [1/(1 + min.fitness)] (24) Step 3: Cycle = Cycle + 1 467
Step 4: Employed bee phase initiation: Random populations 468

406 During this improvisation process, if any food source found
generated in the HMV is improvised by using Eq. (26) to update 469
407 abandoned after a predetermined number of cycles (called as limit)
the bee’s position. 470
408 it is replaced with new random food source by the scout bee using
409 the following Eq. (25).
Xjk new = [␻Xjk + C1 ␾jk (Xbest k − Xjk ) + C2 ␾jk (Xnk − Xjk )] (26) 471
410 Xjt+1 = (Xmax + rand(0, 1)(Xmax − Xmin ) (25)

where Xjk new is the new feasible solution on its preceding solution 472
411 where Xmax , Xmin are the maximum and minimum bound of the Xjk . The term “␻” indicate the inertia weight utilized to modify the 473
412 food source position. current position of bees with respect to its prior one. To explore 474
413 If the new candidate food source position has better quality (fit- global search at initial stage of the search, the inertia weight is 475
414 ness function) than the previous one, then such candidate food declined linearly from 0.9 to 0.4 as suggested in [51] to explore 476
415 source is retained by replacing the older one. Repeat the above steps global search initially. The C1 and C2 are positive social weight 477
416 until the optimal solution is obtained. parameter values chosen as 2.05 [51]. Xbest reflects the finest food 478
417 The standard ABC algorithm encounters the problem in explo- source amongst the population. Xnk is the random neighbor value 479
418 ration and exploitation ability when handling high dimensional of the population and ␾jk is the random value in the range of 480
419 optimization problems. The solution search space is explored and [0,1]. After the estimate of each candidate food source position, the 481
420 exploited by the employee and onlooker bees using Eq. (22) is greedy selection procedure is applied among the solution obtained 482
421 superior in terms of tracing diverged solution (well exploration through step 2 and 4. The onlooker bees utilize the nectar infor- 483
422 ability) but fails to exploit within the search space. Greedy selec- mation shared from the employed bees and select the food source 484
423 tion method on the bee’s trajectories will lead the ABC algorithm based on the probability value accompanied with that food source 485
424 to trap into local optima. In this work, to cope with this problem, Pj from Eq. (27). 486
425 the intensification and diversification capability of ABC algorithm

is balanced by integrating the PSO into the ABC algorithm. The fitnessj
Pj = HMS
426
(27) 487
427 improved ABC algorithm (PABC) updates the bee’s position by mak- fitnessj
j=1
428 ing use of Eq. (26). The inertia weight and acceleration constants are
429 incorporated into the employed bee phase of the ABC algorithm to The fitness value of jth solution is taken as fitnessj . The fitness 488
430 improve the searching capability of the bees. These capabilities alle- of each solution is computed by Eq. (28) as, 489
431 viate the premature convergence and avoidance of the local optima
432 in standard ABC algorithm. Fitnessj = [1/(1 + min.fitness)] (28) 490
433 3.3. Proposed hybrid HS-PABC algorithm “min.fitness” is obtained through the BFS based power flow solution 491
for each fitnessj using Eq. (1). 492
434 It is a common practice to combine different search procedure to Step 5: Onlooker bee phase initiation: Based on the employed 493
435 overcome the deficiencies of the individual heuristic optimization bee phase best fitness probability, the new food source position 494
436 algorithms in order to provide a robust optimization technique. Xjk new in the vicinity of food source Xj,k old is determined by using 495
437 Standard HSA is incompetent in local search ability [27] and it Eq. (29) 496
438 encounters the problem of slow convergence, traps in local optimal

439 solution over multi model fitness landscape. The local searching Xjk new = {Xj,k old + u(Xj,k old − Xn,k )} (29) 497
G Model
Fig. 2. Flowchart of proposed hybrid HS-PABC algorithm.
498 where n D {1,2,. . ..CS} and k D {1,2,. . .. D} are random indices; which put the violating boundary values to the adjoining boundary 504
505
499 D is the total number of parameters in the optimization problem limit value, i.e.,
500 which decides the dimension of the problem; n = / j; ‘u’ signifies a
501 random number produced which lies between [−1,1]. If the calcu- Xjk new = Xjmin , ifXjk new ≤ Xjmin 506
502 lated value of the candidate food position Xjk new using Eq. (25) may
503 violate the predefined boundary, resetting scheme is employed, Xjk new = Xjmax , ifXjk new ≥ Xjmax 507
G Model
508 After the estimation of each candidate food source position, the
509 greedy selection method is adopted.
510 Step 6: Scout bee phase initiation: If any abandoned solution is
511 present, it is swapped with a new arbitrary solution generated by
512 the scout bee phase using Eq. (30).
513 Xjk new = min(Xk j ) + rand(0, 1)[max(Xk j ) − min(Xk j )] (30)
514 Step 7: Retain the best solution obtained so far in HMV

515 Step 8: Improvise the populations (decision variables) in the
516 HMV by HMCR, PAR and Random selection steps in the HS algo-
517 rithm. Step 6: IF f(Xj new ) > f(Xj ), update the HM as Xj = Xi new 536
518 Step 9: Evaluate new HMV and check new HMV is better than Step 7: Retain the best solution in the HMV. 537
519 existing HMV. If so, update HMV Step 8: Next Cycle = Cycle + 1 538
520 Step 10: Cycle = Cycle + 1 Step 9: IF Cycle number < MCN, goto step 4; else stop the itera- 539
521 Step 11: Check if the stopping criterion is met. i.e., total number tive procedure 540
522 of MCN is reached.

523 Step 12: If not, go to step 4 and reiterate from step 4 until 4. Numerical results and discussions 541
524 Cycle = MCN

This article proposes an efficient hybrid heuristic search algo- 542
rithm for the enhancement of distribution network performance by 543
implementing the four test scenarios with an objective of optimiz- 544
ing the multi objective performance improvement indices such as 545
power loss reduction, bus voltage deviation and branch current of 546
525 3.4. The pseudo code of proposed hybrid HS-PABC algorithm the RDN. The backward/forward sweep based power flow method 547
(BFS) is utilized to find the load flow solution [50]. The proposed 548
526 Step 1: Initialize the parameters: Machine cycle (MCN), Har- hybrid optimization algorithm is applied on 69 and 118 node radial 549
527 mony memory size (HMS), Harmony memory consideration rate test networks on a Matlab7.7.0 platform. For both the test net- 550
528 (HMCR) and Pitch Adjustment Rate (PAR) works, substation voltage is taken as 1.0 p.u. The four test scenarios 551
529 Step 2: Initialize HMS. at seven discrete load levels are, 552
530 For i = 1 to HMS do Scenario 1: Optimal allocation (location and sizing) of shunt 553
531 For j = 1 to D do capacitors 554
532 Xij = lbj + rand ()*(ubj − lbj) Scenario 2: Optimal allocation and sizing of DG units 555
533 Step 2.1: Calculate fitness value. Scenario 3: Simultaneous allocation (location and sizing) of DG 556
units and shunt capacitors

534 F(xi ) = Evaluate(xi ) 557
535 Step 3: Start Cycle = 1
G Model
Table 1
Q8 Average load duration schedule of a day.
69-12-70-
69-12-70-
69-12-70-
69-12-70-
10-12-70-
10-12-70-
10-12-70-
switches)
topology
Optimal
Time duration % load level No of hours/day No of hours/year
61-58
62-57
61-57
61-56
61-57
61-58
61-55
(open
6 a.m.–7 a.m. 40 1 365
7 a.m.–8 a.m. 50 1 365
node (19)
Capacitor
8 a.m.–9 a.m. 60 1 365
(kVAR) @
9 a.m.–1 p.m. 80 4 1460
1p.m.–2 p.m. 70 1 365
250
150
200
100
100
200
50
2 p.m.–5 p.m. 100 3 1095
5 p.m.–6 p.m. 90 1 365
DG (kW) @ node
6 p.m.–9 p.m. 80 3 1095
9 p.m.–12 a.m. 60 3 1095
Scenario-IV
12 a.m.–6 a.m. 40 6 2190
Total annual hours 8760
433.45
527.75
612.35
706.12
819.46
897.14
343.04
(60)
558 Scenario 4: Simultaneous optimal network reconfiguration
node (19)
Capacitor
(kVAR) @
559 along with DG units and shunt capacitor compensation
The daily load schedule for the RDN is modeled for 24 h time
200
100
560
50
50
50
50
–
561 interval of the day and it is assumed to be repetitive for 365 days
562 to get the annual load schedule. The average daily load interval
DG (kW) @ node
563 schedule is depicted in Table 1. There are seven discrete load levels
varying from 40% to 100% as shown in Fig. 3. The DG unit’s injection
Scenario-III
564
565 is considered as a negative PQ load with real and reactive power
379.57
473.24
563.37
747.37
842.34
948.44
661.01
injection capability with 0.85 lagging as an operating power factor.
(60)
566
567 The status of sectionalizing and tie switches, location and size of
568 DG unit and shunt capacitors are taken as the decision variables.
DG3 (kW) @ node

569 The maximum penetration limit of DG unit is limited to 50% of the
570 total system real power demand.
181.76
241.47
241.47
122.03
100.50
205.05
235.01
4.1. Simulation results of 69 node RDN
(64)
571
572 To test the effectiveness of the proposed hybrid HS-PABC algo- DG2 (kW) @ node
573 rithm a medium scale 69 node radial distribution network is
574 considered with the real power load demand of 3802.1 kW and
575 reactive power load demand of 2694.5 kVAR. This test system has
248.51
197.45
341.88
163.89
286.48
286.48
263.02
73 line segments with five tie switches. The tie switches are num-
(61)
576
577 bered from 69 to 73 and the sectionalizing switches are numbered

578 from 1 to 68.The active and reactive power loss at nominal load level
DG1 (kW) @ node
579 is 224.97 kW and 102.15 kVAR respectively. The system operates

580 with the nominal bus voltage magnitude of 12.66 kV with 100 MVA
Scenario-II
Simulation results of 69 node RDN at different load levels obtained with hybrid HS-PABC.
581 base. Minimum bus voltage magnitude of 0.9092 p.u. is observed at

9.44
111.46
190.77
327.99
118.06
361.08
422.04
582 65th node. The test network data are obtained from [7].
(63)
583 The simulation results obtained with proposed hybrid algo-

584 rithm and standard HSA for scenarios I–IV is summarized in
node (61)
Capacitor
(kVAR) @
585 Tables 2 and 3. The real power loss reduction obtained with HSA and
hybrid HS-PABC algorithm is summarized in Table 4. From Table 4,
150
150
350
500
100
586
50
–
587 it is observed that at 40% load level the base case loss of 32.50 kW
588 is reduced to 21.65 kW, 10.54 kW, 11.09 kW and 7.07 kW with
node (23)
Capacitor
(kVAR) @
589 respect to the scenarios I–IV using hybrid approach. At medium

load level (70%), the power loss reduction is from 104.52 kW to
100
590
50
50
50
–
591 68.91 kW, 32.99 kW, 35.37 kW and 21.72 kW. At nominal load level
(100%) the power loss reduction is from 224.96 kW to 145.68 kW,
Scenario-I
node (67)
592
Capacitor
(kVAR) @
593 69.08 kW, 73.11 kW and 48.62 kW. The bus voltage profile with
200
100
594 the scenarios I–IV at 100% load level is depicted in Fig. 4. It can
50
50
50
–
595 be seen that, the proposed hybrid algorithm provides a significant

596 improvement of minimum bus voltage profile with scenario-IV as
Fixed/Switched
597 compared with other scenarios. Fig. 5 depicts the optimal reconfig-
598 uration topology and directed graph representation of 69 node RDN
Capacitor
Switched
Switched
Switched
Switched
Switched
Switched
599 along with DG units and shunt capacitor placement (Scenario-IV)

Fixed
600 at 100% load level. It is observed that scenario-IV provides bet-

601 ter loss reduction with improved bus voltage profile at all load
602 level as compared with scenario-III. Since the power flow path is
%Load level
603 optimized in scenario IV by retaining the DG and shunt capacitor

604 location obtained in scenario-III (@ node 19 for capacitor location
Table 2
and @ node 60 for DG unit location) better power loss reduction is

100
605
40
50
60
70
80
90
606 achieved. Table 5 shows the minimum bus voltage profile obtained
G Model
Fig. 3. Daily load schedule.
0.98
Bus Voltage (p.u)
0.96
0.94 Base Case

Capacitor Placement
0.92 DG Units Placement
Simultaneous DG units + Capacitor Placement
Simultaneous reconf + DG units + capacitors
0.9
1 6 11 16 21 26 31 36 41 46 51 56 61 66
Bus Number
Fig. 4. Bus voltage profile of 69 node RDN with different scenarios with hybrid HS-PABC algorithm at 100% load level.
607 with standard HSA and hybrid HS-PABC algorithm. The results pre- Simulation results obtained using HS and hybrid HS-PABC algo- 628
608 sented in Tables 4 and 5 reflect the efficiency of the proposed hybrid rithm at different load level of the network along with the candidate 629
609 algorithm over standard HSA with respect to loss reduction and nodes for the placement of DG units and shunt capacitors are pre- 630
610 voltage profile improvement in all load levels. sented in Tables 6–11. The real power loss reduction obtained 631
611 The active power loss in each branch of the RDN with the sce- with HSA and hybrid HS-PABC algorithms with four test scenar- 632
612 narios I–IV at 100% load level is depicted in Fig. 6. The reduction ios are summarized in Table 12. From Table 12, it is observed 633
613 in total real power loss of the RDN with scenario-IV at 40%, 70% that loss reduction realized using proposed hybrid algorithm at 634
614 and 100% load level using hybrid HS-PABC algorithm is depicted in light load level (40%) is from 187.17 kW (base case) to 126.13 kW, 635
615 Fig. 7. It is revealed that scenario-IV offers more loss reduction with 59.59 kW, 66.75 kW and 54.91 kW with respect to the scenarios 636
616 enhanced bus voltage profile as compared with other scenarios. I–IV. At medium load level (70%), the power loss reduction is from 637
602.18 kW to 399.78 kW, 185.32 kW, 209.21 kW and 163.91 kW 638
617 4.2. Simulation results of 118 node RDN with respect to the scenarios I–IV. At nominal load level (100%) the 639
power loss reduction is from 1298.09 kW to 847.75 kW, 384.65 kW, 640
618 To show the suitability of the proposed hybrid approach on large 435.62 kW and 340.12 kW with respect to the scenarios I–IV. It 641
619 scale RDN it is applied on 118 node balanced three phase radial dis- is observed that scenario-IV provides better loss reduction with 642
620 tribution network. The system operates under 11 kV voltage level at improved bus voltage profile at all load level as compared with 643
621 100 MVA base with active load demand of 22709.7 kW and reactive other scenarios. Fig. 8 depicts the optimal reconfiguration topol- 644
622 load demand of 17041.1 kVAR respectively. The active and reac- ogy and directed graph representation of 118 node RDN along with 645
623 tive power loss before compensation is 1298.1 kW and 978.74 kVAR DG units and shunt capacitor placement (Scenario-IV) at 100% load 646
624 respectively. The line and load data are obtained from [8]. This test level. 647
625 network consists of 118 nodes with 117 sectionalizing switches Table 13 shows the minimum bus voltage profile with differ- 648
626 (normally closed) and fifteen tie switches (normally open). The tie ent scenarios obtained with standard HSA and hybrid HS-PABC 649
627 switches are numbered from 118 to 132. algorithms. It can be inferred from the results presented in 650
G Model
Fig. 5. Directed graph representation of optimal topology of 69 Node RDN (Scenario-IV at 100% load level).
651 Tables 12 and 13, that the proposed hybrid algorithm provides bet- that the combined approach of optimal network reconfiguration 662
652 ter loss reduction with improved bus voltage profile as compared problem along with DG units and shunt capacitors compensation 663
653 with the standard HSA (with scenarios I–IV) at all load levels. The (Scenario-IV) yields more loss reduction with improved bus voltage 664
654 bus voltage profile with the scenarios I–IV at 100% load level is profile as compared with other scenarios. 665
655 depicted in Fig. 9. It can be seen that, the proposed hybrid algorithm
656 provides a significant improvement of minimum bus voltage profile 4.3. Parameter settings 666
657 in scenario-IV as compared with other scenarios. The active power
658 loss in each branch of the RDN with the scenarios I–IV at 100% load An empirical study is performed to determine the impact of 667
659 level is depicted in Fig. 10.The reduction in total real power loss of control parameters towards the solution quality of the hybrid HS- 668
660 the RDN with scenario-IV at 40%, 70% and 100% load levels using PABC algorithm. To demonstrate the effects of single parameter 669
661 hybrid HS-PABC algorithm is depicted in Fig. 11. It can be inferred change on the fitness function for 69-node RDN, 12 different cases 670
G Model
Fig. 6. Variation of power loss at each line of 69 node RDN with different scenarios with hybrid HS-PABC algorithm at 100% load level.
250 224.96
200
Power loss (kW)
150
104.52
100
48.62
50 21.72 32.5
7.07
0
100% 70% 40%
% load level
Base Case Scenario-I Scenario-II Scenario-III Scenario-IV
Fig. 7. Power loss reduction in 69 node RDN with different scenarios with hybrid HS-PABC algorithm at 100% load level.
Fig. 8. Directed graph representation of optimal reconfiguration topology of 118 node RDN (Scenario-IV at 100% load level).
G Model
1.05
Bus Voltage (p.u)

1
0.95
0.9
0.85
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116
Bus Number
Base Case Capacitor Placement
DG units Placement Simultaneous DG units + capacitor placement
Simultaneous reconf + DG units + capacitors placement
Fig. 9. Bus voltage profile of 118 node RDN with different scenarios (100% load level).
Fig. 10. Variation of real power loss at each line of 118 node RDN with different scenarios (100% load level).
1500 1298.09
Power loss ( kW)
1000
602.18
340.12
500 187.17
163.91 54.91
0
100% 70% 40%
% Load level
Base Case Scenario-I Scenario-II Scenario-III Scenario-IV
Fig. 11. Real power loss reduction in 118 node RDN with different scenarios (100% load level).
671 are tested as shown in Table 14. Each case is tested over 50 runs solution. Selection of low PAR will limit the exploration and it 680
672 in three stages. In stage 1–3, HMCR, HMS, PAR and MCN are var- can lead to slow convergence. Higher PAR selection will lead the 681
673 ied respectively, and other parameters are kept constant. If the solution to scatter around some potential optima similar to ran- 682
674 HMCR value selection is too low, the algorithm behaves like a dom search. The selection of larger and smaller values of HMS will 683
675 pure random search which leads to slow convergence. If the HMCR decrease the efficiency of the algorithm as seen in Table 14. From 684
676 value is higher, all the harmonies in the HM are utilized, the other the empirical study, the parameters chosen for 69 node network 685
677 harmonies are not well explored and this may lead to local opti- are HMS = 20, HMCR = 0.9, PAR = 0.3, MCN = 15. Similarly for 118 686
678 mal solution. As shown in Table 14, large and small HMCR values node RDN the control parameter values are selected as HMS = 50, 687
679 lead to a decrease in the solution quality. The pitch adjustment MCN = 50, HMCR = 0.9, PAR = 0.3. 688
(PAR) is used to produce new solution around the existing best
14
ASOC 3714 1–22

G Model
Table 3
Simulation results of 69 node RDN at different load levels obtained with HSA.
%Load level Fixed/Switched Scenario-I Scenario-II Scenario-III Scenario-IV

Capacitor
Capacitor Capacitor Capacitor DG1 (kW) DG2 (kW) DG3 (kW) DG (kW) @ Capacitor DG (kW) @ Capacitor Optimal
(kVAR) @ (kVAR) @ (kVAR) @ @ node (63) @ node (61) @ node (64) node (60) (kVAR) @ node (60) (kVAR) @ topology
node (67) node (23) node (61) node (19) node (19) (open
switches)
40 Fixed 200 100 450 11.13 244.83 121 380.22 200 294.15 350 4-13-70-
63-57
50 Switched 50 50 200 121.29 246 107.15 425.24 100 396 150 69-11-70-
62-52
60 Switched 50 – 100 138.57 316.32 120.03 555.12 50 500.76 100 7-12-70-
63-53
70 Switched 100 – 150 71.94 387.32 205.66 661.02 50 592.08 100 69-14-70-
ARTICLE IN PRESS
61-55
80 Switched 50 – 350 128.07 450.29 171.28 734.72 100 699.1 50 69-13-70-
M. K., J. S. / Applied Soft Computing xxx (2016) xxx–xxx

62-55
90 Switched – – – 133.97 503.13 207.74 823.75 – 791.7 100 7-11-70-
62-58
100 Switched – – 100 104.06 538.30 227.03 900.02 – 878.9 300 69-12-70-
63-57
Table 4
Reduction in real power loss with different scenarios using HSA and hybrid HS – PABC.
%Load level Real Power loss (kW)
Before Compensation Scenario-I Scenario-II Scenario-III Scenario-IV

Capacitors @ node: DG units @ node DG unit @ node(60) Reconfiguration along
(67),(23),(61) (63),(61),(64) Capacitor @ node (19) with DG @ node (60)
Capacitor @ node (19)
HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC
40 32.50 21.65 21.65 10.65 10.54 11.07 11.09 9.204 7.07

50 51.59 34.41 34.26 16.64 16.61 19.92 17.53 13.16 11.49
60 75.51 49.88 49.88 24.81 24.13 26.41 25.98 19.68 15.71
70 104.52 68.91 68.91 32.99 32.99 35.37 35.37 23.57 21.72
80 138.88 93.96 93.89 44.22 43.66 48.59 47.12 29.55 28.50
90 178.92 117.44 117.58 56.23 55.41 61.52 59.98 43.87 36.48
100 224.96 146.11 145.68 77.25 69.08 77.65 73.11 51.41 48.62
ASOC 3714 1–22
G Model
Table 5
Minimum bus voltage magnitude with different scenarios using HSA and hybrid HS-PABC.
%Load level Minimum bus Voltage (p.u) @(Node)
Before compensation Scenario-I Scenario-II Scenario-III Scenario-IV
HSA @65 HS-PABC @ (65) HSA @ (65) HS-PABC @ (65) HSA @ (65) HS-PABC @ (65) HSA @63 HS-PABC @63
40 0.9656@18 0.9732 0.9739 0.9835 0.9836 0.9816 0.9815 0.9878 0.9878

50 0.9566@18 0.9675 0.9675 0.9792 0.9792 0.9750 0.9768 0.9849 0.9866
60 0.9474@18 0.9604 0.9604 0.9752 0.9755 0.9716 0.9720 0.9826 0.9848
70 0.9382@18 0.9539 0.9539 0.9711 0.9711 0.9673 0.9673 0.9816 0.9821
80 0.9287@18 0.9500 0.9500 0.9658 0.9670 0.9617 0.9620 0.9776 0.9779
ARTICLE IN PRESS
90 0.9191@18 0.9409 0.9412 0.9616 0.9628 0.9563 0.9571 0.9738 0.9778
100 0.9091@18 0.9330 0.9326 0.9538 0.9583 0.9504 0.9526 0.9733 0.9749

Table 6
Optimal results obtained with hybrid HS-PABC algorithm—Scenario- I & II.
Load level Fixed/Switched Capacitor Scenario-I Scenario-II
Qc1 @(86) Qc2 @(82) Qc3 @(111) Qc4 @(49) Qc5 @(32) Qc6 @(107) Qc7 @(71) Qc8 @(96) DG1 @ (109) DG2 @(52) DG3 @(70)
40 Fixed 350 300 600 800 750 500 700 400 1315.43 1118.27 1195.43
50 Switched 100 100 100 200 200 150 250 150 1646.62 1402.83 1498.12
60 Switched 100 50 200 200 200 100 200 100 1978.76 1689.54 1802.42
70 Switched 100 100 100 200 200 150 200 100 2311.87 1978.46 2108.36
80 Switched 100 100 200 200 200 100 250 150 2646.42 2267.10 2415.11
90 Switched 50 100 100 300 – 150 250 100 2980.41 2563.56 2726.37
100 Switched – – 150 100 – – 200 50 3224.02 2757.83 2964.16
15
16
ASOC 3714 1–22

G Model
Table 7
Optimal results obtained with hybrid HS-PABC algorithm—Scenario-III.
%Load level Fixed/Switched Capacitor Scenario-III
Qc1 Node (32) Qc2 Node (50) Qc3 Node (15) Qc4 Node (93) Qc5 Node (103) DG (kW)@ Node (71) DG (kW) @ Node (110)
40 Fixed 800 850 250 400 550 1147.65 1166.04

50 Switched 200 250 100 50 150 1435.09 1458.18
60 Switched 200 250 50 100 100 1727.78 1747.56
70 Switched 250 300 50 100 150 2018.77 2041.41
80 Switched 250 250 100 100 150 2315.09 2334.47
90 Switched 250 250 50 100 150 2608.78 2629.85
ARTICLE IN PRESS
100 Switched 200 250 50 100 100 2904.64 2924.14

Table 8
Optimal results obtained with hybrid HS-PABC algorithm—Scenario-IV.
%Load level Type of capacitor Scenario-IV
Fixed/Switched Capacitor size in kVAR Optimal DG units sizing in kW Optimal topology (Open Switches)
@ Node (32) @ Node (50) @ Node (15) @ Node (93) @ Node (103) @ Node (71) @ Node (110)
40 Fixed 750 750 300 250 550 1366.76 1594.45 25-23-39-42-34-60-51-125-126-96-127-107-79-130-121

50 Switched 350 100 150 200 450 1527.37 1453.32 25-22-38-42-34-59-51-125-126-98-86-107-82-130-121
60 Switched 250 200 – 150 – 1750.64 2090.32 25-22-38-42-34-61-50-125-126-97-87-131-80-130-121
70 Switched 400 – 50 50 50 2252.65 2457.30 26-23-38-42-34-59-50-125-126-97-86-131-80-130-121
80 Switched 200 300 150 – 100 2404.66 2769.04 26-22-39-42-34-59-50-125-126-97-86-131-80-130-121
90 Switched – 350 – 150 250 2562.01 3146.19 25-23-39-42-34-60-51-125-126-97-86-131-80-130-121
100 Switched – – 200 150 50 1366.76 1594.45 25-23-39-42-34-60-51-125-126-97-127-131-80-130-121
ASOC 3714 1–22
G Model
Table 9
Optimal results obtained with HSA—Scenario-I & II.
Load level Fixed/Switched Capacitor Scenario-I Scenario-II
Qc1 @ (86) Qc2 @ (82) Qc3 @ (111) Qc4 @ (49) Qc5 @ (32) Qc6 @ (107) Qc7 @ (71) Qc8 @ (96) DG1 @ (109) DG2 @ (52) DG3 @ (70)
40 Fixed 300 300 650 1050 1000 450 750 350 1323.79 1276.49 1229.82
50 Switched 100 50 150 50 550 150 250 200 1843.42 1654.55 1458.50
60 Switched 150 150 100 200 50 150 150 100 2586.35 1686.08 1804.49
70 Switched – 100 150 50 50 150 200 100 2515.81 1282.55 2108.83
80 Switched – 150 150 100 50 150 250 150 2646.66 2269.03 2715.29
90 Switched – 150 100 100 50 150 250 100 2998.59 2666.05 2821.19
ARTICLE IN PRESS
100 Switched – 700 50 100 50 150 100 – 3222.79 2759.82 2962.21

Table 10
Optimal results obtained with HSA—Scenario-III.
%Load level Fixed/Switched Capacitor Scenario-III
Qc1 @Node (32) Qc2 @Node (50) Qc3 @Node (15) Qc4 @Node (93) Qc5 @Node (103) DG (kW) @ Node (71) DG (kW) @ Node (110)
40 Fixed 700 850 250 350 500 1143.54 1463.33

50 Switched 250 250 100 150 150 1132.03 1758.55
60 Switched 200 250 50 50 150 1525.89 1846.73
70 Switched 200 300 50 50 150 1909.97 2231.70
80 Switched 250 250 100 – 100 2119.18 2333.56
90 Switched 250 250 50 250 100 2127.67 2632.98
100 Switched 350 200 – 100 600 2936.47 3172.35
17
18
ASOC 3714 1–22

G Model
Table 11
Optimal results obtained with HSA—Scenario-IV.
%Load level Type of capacitor Scenario-IV
Fixed/Switched Capacitor size in kVAR Optimal DG units sizingin kW Optimal topology (Open Switches)
@ Node (32) @ Node (50) @ Node (15) @ Node (93) @ Node (103) @ Node (71) @ Node (110)
40 Fixed 50 1000 100 550 350 1322.98 1226.79 119-20-124-42-34-61-52-125-126-96-86-131-82-130-121

50 Switched 1700 550 300 150 250 1715.63 1540.30 25-22-124-42-33-58-50-125-126-98-86-131-81-130-121
60 Switched 150 50 50 50 100 1911.94 2605.31 26-20-124-43-34-59-53-125-126-128-86-106-80-130-121
70 Switched 50 50 50 50 50 1802.79 2662.57 26-23-39-44-33-59-50-125-126-97-87-104-81-130-121
80 Switched – – 50 – – 2074.84 2930.43 25-22-38-43-33-59-50-125-126-97-74-131-80-130-121
ARTICLE IN PRESS
90 Switched 450 50 – – – 1795.53 3475.60 25-21-38-45-33-58-52-57-126-98-74- 131-79-130-121
100 Switched 750 100 – – 600 2103.25 3984.31 25-21-39-44-34-59-52-125-70-128-127-131-80-130-121

Table 12
Reduction in real power loss with scenarios: I–IV using HSA and hybrid HS-PABC.
%Load level Real Power loss (kW)
Before Compensation Scenario-I Scenario-II Scenario-III Scenario-IV

Capacitors DG units at node DG at node (71),(110) Reconfiguration along
at node: (86),(82),(111), (109),(52),(70) Capacitors at node with DG at node
(49),(32),(107),(71),(96) (32),(50),(15),(93),(103) (71),(110)
Capacitors at node
(32),(50),(15),(93),(103)
HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC
40 187.17 128.07 126.13 60.25 59.59 69.33 66.75 61.57 54.91

50 297.14 200.12 199.52 96.15 93.59 110.95 105.03 99.53 86.78
60 434.97 290.52 290.44 144.48 135.46 154.03 152.27 135.16 118.51
70 602.18 400.06 399.78 197.31 185.32 210.83 209.21 179.83 163.91
80 800.46 529.08 528.82 246.15 243.31 277.10 275.19 226.10 212.11
90 1031.72 678.89 677.84 310.25 309.57 358.85 350.70 320.82 268.64
100 1298.09 853.35 847.75 385.89 384.65 438.19 435.62 380.21 340.12
G Model
Table 13
Minimum bus voltage magnitude with scenarios: I–IV using HSA and hybrid HS-PABC.
%Load level Minimum bus Voltage (p.u) @(Node)
Before Scenario-I Scenario-II Scenario-III Scenario-IV

compensation
@ (78)
HSA @ (78) HS-PABC @ (78) HSA @ (47) HS-PABC @ (47) HSA @ (55) HS-PABC HSA HS-PABC @ (51)
40 0.9513 0.9658 0.9662 0.9844 0.9844 0.9821 0.9824@55 0.9857@97 0.9884

50 0.9385 0.9581 0.9582 0.9803 0.9803 0.9784 0.9784@55 0.9846@51 0.9842
60 0.9253 0.9487 0.9486 0.9763 0.9763 0.9744 0.9744@55 0.9796@43 0.9811
70 0.9117 0.9396 0.9397 0.9713 0.9722 0.9713 0.9713@55 0.9774@51 0.9764
80 0.8978 0.9312 0.9315 0.9682 0.9682 0.9674 0.9675@55 0.9732@51 0.9728
90 0.8835 0.9226 0.9228 0.9641 0.9642 0.9635 0.9635@47 0.9629@58 0.9725
100 0.8687 0.9122 0.9127 0.9598 0.9599 0.9589 0.9593@47 0.9655@ 99 0.9684
692
Table 14 tion seeds. Both the algorithms are allowed to run over generation 693
Q9 Impact of hybrid HS-PABC algorithm parameter variations on the fitness function with the same initial population. The robustness of hybrid approach 694
(Scenario-IV) at 100% load level.
over standard HSA is depicted in Tables 15 and 16. Scenarios-I–IV 695
Stage Parameter settings Fitness at 100% load level is considered to visualize the convergence plot 696
HMS HMCR PAR MCN

of fitness function using HSA and hybrid HS-PABC algorithm as 697
shown in Figs. 12 and 14(a–d). Undoubtedly Figs. 12 and 14(a–d) 698

1 5 0.3 0.3 5 0.37640
demonstrates the effectiveness of hybrid HS-PABC algorithm over 699
5 0.3 0.3 10 0.35424
5 0.3 0.3 15 0.35324 HSA with stable and improved performance in terms of solu- 700
tion accuracy and fast convergence to reach the optimal solution 701
2 5 0.3 0.3 15 0.35324
in all the scenarios. This is because of high probability of best 702
5 0.3 0.4 15 0.36058
5 0.3 0.5 15 0.35789 solutions updated in HMV by integrating PABC algorithm with 703
HSA. 704
3 5 0.3 0.3 15 0.38324
5 0.7 0.3 15 0.35746
5 0.95 0.3 15 0.37668
4 2 0.9 0.3 15 0.37757

20 0.9 0.3 15 0.34404 Table 17
30 0.9 0.3 15 0.36862 Statistical study using Wilcoxon signed rank sum test.
Algorithms Scenario 69 node RDN 118 node RDN

689 4.4. Robustness and convergence rate of hybrid HS-PABC and HS P-Value P-Value
690 algorithm HS-PABC vs I 1.415656224849e-009 1.415656224849e-009

HSA II 1.415656224849e-009 1.390302548729e-009
III 1.795360263143e-007 1.415656224849e-009
691 The formulated optimization problem (Scenarios-I–IV with
IV 1.651603235538e-005 1.991413222361e-009
100% load level) is solved 50 times repeatedly with random popula-
Table 15
Statistical study to test the robustness of the HS and hybrid HS −PABC algorithm with 50 independent runs at 100% load level – 69 node RDN.
Scenarios Max.fitness Min.fitness Mean fitness Median Std deviation
HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC
I 0.525175272 0.515918944 0.515953 0.5158 0.517805366 0.515901409 0.5176626 0.5158997 0.001751687 4.14644E-06
II 0.37432 0.370575 0.3703 0.36998 0.371701 0.370158 0.371478 0.370096 0.000884 0.000162
III 0.381935 0.368598 0.3473269 0.347311 0.368063 0.352237 0.36821 0.352933 0.008454 0.005464
IV 0.385383465 0.384051451 0.3457147 0.344045 0.368376043 0.353342188 0.370414 0.351459 0.010975562 0.009082648
Table 16
Statistical study to test the robustness of the HS and hybrid HS-PABC algorithm with 50 independent runs at 100% load level –118 node RDN.
Scenarios Max.fitness Min.fitness Mean fitness Median Std deviation
HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC HSA HS-PABC
I 0.38792539 0.379718752 0.3811177 0.37960 0.38356404 0.379684188 0.3831836 0.3796802 0.001672306 9.68724E-06
II 0.186545315 0.180605777 0.180605 0.18060 0.183937826 0.18060575 0.1840858 0.1806057 0.001740765 7.13674E-09
III 0.202517727 0.188078869 0.188329 0.188075 0.197148549 0.18807596 0.1976716 0.18807568 0.003243615 7.62759E-07
IV 0.198168189 0.166000427 0.1630003 0.14670 0.180038661 0.153444577 0.1814853 0.1531285 0.009251209 0.004382506
G Model
(a) Convergence plot of fitness function with HSA and Hybrid HS -PABC algorithms ( Scenario-I) (a) 0.392 Convergence plot of fitness function with HS and hybrid HS-PABC algorithms ( Scenario-I)
0.523 HSA- Capacitor placement
HSA- Cap Placenment Hybrid HS-PABC- Capacitor Placement
0.522 Hybrid HS -PABC -Cap Placement 0.39
Fitness ( Objective function)

0.521
0.388
0.52
0.386
0.519
0.518 0.384
0.517 0.382
0.516
0.38
0.515
0 50 100 150 200 250 300 350 400 450 500 550 600
Iterations 0.378
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Iterations
(b) Convergence plot of fitness function with HSA and Hybrid HS -PABC algorithms ( Scenario-II)
(b)
0.376 Convergence plot of fitness function with HS and hybrid HS-PABC algorithms ( Scenario-II)
HSA- DG Placement 0.194
Hybrid HS -PABC- DG Placement HSA - DG Placement
0.375 Hybrid HS- PABC- DG Placement
0.192
Fitness (Objective functions)
Fitness ( objective function)

0.374
0.19
0.373
0.188
0.372
0.186
0.371 0.184
0.37 0.182
0 50 100 150 200 250 300 350 400 450 500 550 600
Iterations
0.18
(c) Convergence plot of fitness function with HSA and Hybrid HS -PABC algorithms ( Scenario-III) 0 500 1000 1500 2000 2500
Iterations
3000 3500 4000 4500 5000
0.36
HSA- DG + Capacitor Placement (c) Convergence plot of fitness function with HSA and hybrid HS- PABC algorithms ( Scenario-III)
Hybrid HS -PABC - DG + Capacitor Placement
0.358 Hybrid HS-PABC DG + Capacitor Placement
HSA- DG + Capacitor Placement
0.23

0.356
0.22
0.354
0.21
0.352
0.2
0.35
0.19
0.348
0.18
0.346 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0 50 100 150 200 250 300 350 400 450 500 550 600 Iterations
Iterations
(d) Convergence plot of fitness function with HS and hybrid HS-PABC algorithms (Scenario-IV)
(d) Convergence plot of fitness function with HSA and Hybrid HS -PABC algorithms ( Scenario-IV)
0.23 Hybrid HS-PABC- Recon+ DG+ Capacitors
0.39
HSA - Rec on+ DG+ Capacitors
HSA- Recon + DG+Capacitor placement
0.385 0.22
Hybrid HS -PABC- Recon+ DG+ Capacitor placement
0.38
0.21
0.375
0.2
0.37
0.365 0.19
0.36 0.18
0.355
0.17
0.35
0.16
0.345
0.34 0.15
0 50 100 150 200 250 300 350 400 450 500 550 600 0 300 600 900 1200 1500 1800 2100 2400 2700 3000 3300 3600 3900 4200 4500 4800 5000
Iterations Iterations
Fig. 12. (a–d) Convergence plot of fitness function with HS and hybrid HS-PABC Fig. 14. (a–d) Convergence plot of fitness function with HSA and Hybrid HS-PABC
algorithms (Scenario- I–IV)—100% load level-69 node RDN. algorithm (Scenario- I–IV)—100% load level-118 node RDN.
(a) Scenario-I -69 node RDN (b) Scenario-II- 69 node RDN

0.526
0.524 0.375
0.522 0.374
0.373
Fitness
0.52
Fitness
0.372
0.518
0.371
0.516 0.37
0.514 0.369
0.512 0.368
0.51 0.367
HS-PABC HSA HS-PABC HSA
Algorithm Algorithm
(c) Scenario-III- 69 node RDN (d) Scenario-IV - 69 node RDN

0.39 0.39
0.38 0.38
0.37 0.37
Fitness
Fitness
0.36 0.36
0.35 0.35
0.34 0.34
0.33 0.33
0.32 0.32
Algorithm Algorithm
Fig. 13. (a–d) Box plot for the results of HS-PABC and HS algorithms—Scenarios I–IV for 69 node RDN.
G Model
(a) Scenario-I 118 node RDN (b) Scenario-II 118 node RDN
0.39
0.388 0.188
0.386 0.186
0.384 0.184
Fitness
Fitness
0.382 0.182
0.38
0.378 0.18
0.376 0.178
0.374 0.176
Algorithm Algorithm
(c ) Scenario-III 118 node RDN (d) Scenario-IV - 118 node RDN

0.204 0.24
0.201
0.198 0.22
0.2
Fitness
0.195
Fitness
0.192 0.18
0.189
0.16
0.186
0.183 0.14
0.18 0.12
Algorithm Algorithm
Fig. 15. (a–d) Box plot for the results of HS-PABC and HS algorithms—Scenarios I–IV for 118 node RDN.
737
5. Conclusion 738
705 4.4.1. Statistical analysis
706 The box plot and Wilcoxon rank sum test is used to show the This article presents a hybrid HS-PABC algorithm based 739
707 quality of the solution obtained by hybrid HS-PABC with respect to approach for network reconfiguration problem along with optimal 740
708 HSA. location and sizing of DG units and shunt capacitors for optimal 741
planning of distribution networks. Different test scenarios are sim- 742
ulated on 69 and 118 node distribution network at varying load 743

709 4.4.1.1. Box-plot. Repeatedly 50 runs are performed to visualize the conditions. The hybrid HS-PABC algorithm offers a promising and 744
710 distribution of the samples in HS-PABC and HS algorithms with four preferable performance in terms of power loss reduction, improved 745
711 scenarios in box plot format as shown in Figs. 13 and 15(a–d). It bus voltage profile than standard HSA at all discrete load levels. 746
712 is observed that the distribution of data based on minimum, first From the simulation results, it is inferred that integrated approach 747
713 quartile, median, third quartile, and maximum are close to each of network reconfiguration problem with the simultaneous instal- 748
714 other in proposed hybrid HS-PABC algorithm when compared with lation of DG units and shunt capacitors is an efficient approach 749
715 HSA. The results in box plots reveal the robustness and superiority towards more loss reduction with the enhanced bus voltage profile 750
716 of the proposed hybrid HS-PABC over standard HS algorithm. as compared with individual loss reduction approaches. A statistical 751
analysis is performed using 50 independent trials for each approach 752
on the test system. To measure and prove the quality of the solu- 753
717 4.4.1.2. Wilcoxon rank sum test. Wilcoxon rank sum test is a non- tions obtained by proposed approach, box plot and Wilcoxon rank 754
718 parametric test utilized to show the statistical comparison between sum test are performed. The HS-PABC is a co-evolutionary algo- 755
719 the proposed hybrid HS-PABC and HS algorithm. This test assumes rithm that runs in parallel to accomplish the optimal solution in 756
720 selection of different random samples with continuous probability high-dimensional optimization problems. The simulation results 757
721 distribution. The following hypotheses are tested. confirm the capability of proposed HS-PABC algorithm to reach 758
the optimal solution with faster convergence rate than standard 759
722 H0 . The probability distributions of fitness values obtained by HS-
HS algorithm. Finally, this work provides interesting basis for 760
723 PABC and HSA are identical.
combining promising aspects of different evolutionary algorithm 761
techniques into a new hybrid meta-heuristic algorithm that shows 762

724 H1 . Distribution of the fitness values differ between HS-PABC and
significant performance. 763
725 HSA
726 Wilcoxon rank sum test results are termed p-values (observed References 764
727 significance level). When p-values less than or equal to level of sig-
728 nificance (␣ = 5%), we reject the null hypothesis. The approximation [1] Leonardo W. De Oliveira, Edimar J. de Oliveira, Flávio V. Gomes, Ivo C. Silva, 765
André L.M. Marcato, Paulo V.C. Resende, Artificial immune systems applied to Q7 766
729 method is used to calculate the p-values and the results are pre- the reconfiguration of electrical power distribution networks for energy loss 767
730 sented in Table 17 to illustrate the efficiency of proposed hybrid minimization, Int. J. Electr. Power Energy Syst. 56 (2014) 64–74. 768
731 HS-PABC algorithm over standard HS algorithm. For both the RDNs [2] Almoataz Youssef Abdelaziz, F.M. Mohammed, S.F. Mekhamer, M.A.L. Badr, 769
Distribution systems reconfiguration using a modified particle swarm 770
732 with scenario-IV, all the p-values are less than 0.05 (refer Table 17) optimization algorithm, Electr. Power Syst. Res. 79 (11) (2009) 1521–1530. 771
733 which signifies the choice of selecting alternative hypothesis. It can [3] Saeed Teimourzadeh, Kazem Zare, Application of binary group search 772
734 be concluded that these test implies a significant statistical differ- optimization to distribution network reconfiguration, Int. J. Electr. Power 773
Energy Syst. 62 (2014) 461–468. 774
735 ence between the HS-PABC and standard HS algorithm. Therefore [4] Anil Swarnkar, Nikhil Gupta, K.R. Niazi, Adapted ant colony optimization for 775
736 we can come to the conclusion that the proposed HS-PABC can efficient reconfiguration of balanced and unbalanced distribution systems for 776
maintain superior performance than the standard HSA. loss minimization, Swarm Evol. Comput. 1 (3) (2011) 129–137. 777
G Model
778 [5] A. Mohamed Imran, M. Kowsalya, A new power system reconfiguration [29] D. Karaboga, B. Basturk, On the performance of artificial bee colony algorithm, 848
779 scheme for power loss minimization and voltage profile enhancement using Appl. Soft Comput. 8 (2008) 687–697. 849
780 Fireworks Algorithm, Int. J. Electr. Power Energy Syst. 62 (2014) 312–322. [30] D. Karaboga, Akay Bahriye, A comparative study of artificial bee colony 850
781 [6] Rayapudi Srinivasa Rao, Sadhu Venkata Lakshmi Narasimham, Manyala algorithm, Appl. Math. Comput. 214 (2009) 108–132. 851
782 Ramalinga Raju, Optimal network reconfiguration of large-scale distribution [31] Z.W. Geem, Novel derivative of harmony search algorithm for discrete design 852
783 system using harmony search algorithm, IEEE Trans. Power Syst. 26 (3) (2011) variables, Appl. Math. Comput. 199 (1) (2008) 223–230. 853
784 1080–1088. [32] K.S. Lee, Z.W. Geem, A new meta-heuristic algorithm for continuous 854
785 [7] J.S. Savier, Debapriya Das, Impact of network reconfiguration on loss engineering optimization: harmony search theory and practice, Comput. 855
786 allocation of radial distribution systems, IEEE Trans. Power Deliv. 22 (4) Methods Appl. Mech. Eng. 194 (36–38) (2005) 3902–3933. 856
787 (2007) 2473–2480. [33] D. Manjarres, I. Landa-Torres, S. Gil-Lopez, J. Del Ser, M.N. Bilbao, S. 857
788 [8] D. Zhang, Z. Fu, L. Zhang, An improved TS algorithm for loss minimum Salcedo-Sanz, Z.W. Geem, A survey on applications of the harmony search 858
789 reconfiguration in large scale distribution systems, Electr. Power Syst. Res. 77 algorithm, Eng. Appl. Artif. Intell. 26 (8) (2013) 1818–1831. 859
790 (2007) 685–694. [34] Swagatam Das, Arpan Mukhopadhyay, Anwit Roy, Ajith Abraham, Bijaya K. 860
791 [9] Y. Mishima, K. Nara, T. Satoh, et al., Method for minimum-loss reconfiguration Panigrahi, Exploratory power of the harmony search algorithm: analysis and 861
792 of distribution system by Tabu search, Electr. Eng. Jpn. 152 (2005) 18–25. improvements for global numerical optimization, IEEE Trans. Syst. Man 862
793 [10] A.Y. Abedelaziz, F.M. Mohamed, S.F. Mekhamer, et al., Distribution system Cybern.—B: Cybern. 41 (1) (2011) 89–106. 863
794 reconfiguration using a modified Tabu search algorithm, Elect. Power Syst. [35] Z.W. Geem, Optimal scheduling of multiple dam system using harmony 864
795 Res. 80 (2010) 943–953. search algorithm, Lect. Notes Comput. Sci. 4507 (2007) 316–323. 865
796 [11] Hamed Ahmadi, Jose R. Matri, Mathematical representation of radiality [36] I. Ahmad, M. Gh. Mohammad, A.A. Salman, S.A. Hamdan, Broadcast 866
797 constraints in distribution system reconfiguration problem, Int. J. Electr. scheduling in packet radio networks using harmony search algorithm, Expert 867
798 Power Energy Syst. 64 (2015) 293–299. Syst. Appl. 39 (1) (2012) 1526–1535. 868
799 [12] R. Srinivasa Rao, Kumudhini Ravindra, K. Satish, S.V.L. Narasimham, Power [37] I. Landa-Torres, E.G. Ortiz-Garcia, S. Salcedo-Sanz, M.J. Segovia-Vargas, S. 869
800 loss minimization in distribution system using network reconfiguration in the Gil-L’opez, M. Miranda, J.M. Leiva-Murillo, J. Del Ser, Evaluating the 870
801 presence of distributed generation, IEEE Trans. Power Syst. 28 (1) (2013) internationalization success of companies through a hybrid grouping 871
802 317–325. harmony search—extreme learning machine approach, IEEE J. Sel. Top. Signal 872
803 [13] A. Mohamed Imran, M. Kowsalya, D.P. Kothari, A novel integration technique Process. 6 (4) (2012) 388–397. 873
804 for optimal network reconfiguration and distributed generation placement in [38] Z.W. Geem, K.S. Lee, Y. Park, Application of harmony search to vehicle routing, 874
805 power distribution networks, Int. J. Electr. Power Energy Syst. 63 (2014) Am. J. Appl. Sci. 2 (12) (2005) 1552–1557. 875
806 461–472. [39] I. Landa-Torres, S. Gil-Lopez, S. Salcedo-Sanz, J. Del Ser, J.A. Portilla-Figueras, A 876
807 [14] Attia A. El-Fergany, Almoataz Y. Abdelaziz, Artificial bee colony algorithm to novel grouping harmony search algorithm for the multiple-type access node 877
808 allocate fixed and switched static shunt capacitors in radial distribution location problem, Expert Syst. Appl. 39 (5) (2012) 5262–5270. 878
809 networks, Electr. Power Compon. Syst. 42 (5) (2014) 427–438. [40] I. Landa-Torres, D. Manjarres, S. Salcedo-Sanz, J. Del Ser, S. Gil-Lopez, A 879
810 [15] Pavlos S. Georgilakis, Nikos D. Hatziargyriou, Optimal distributed generation multi-objective grouping harmony search algorithm for the optimal 880
811 placement in power distribution networks: models, methods, and future distribution of 24-h medical emergency units, Expert Syst. Appl. 40 (6) (2013) 881
812 research, IEEE Trans. Power Syst. 28 (3) (2013) 3420–3428. 2343–2349. 882
813 [16] Pavlos S. Georgilakis, Nikos D. Hatziargyriou, A review of power distribution [41] S. Salcedo-Sanz, D. Manjarres, A. Pastor-Sanchez, J. Del Ser, J.A. 883
814 planning in the modern power systems era: models, methods and future Portilla-Figueras, S. Gil-Lopez, One-way urban traffic reconfiguration using a 884
815 research, Electr. Power Syst. Res. 121 (2015) 89–100. multi-objective harmony search approach, Expert Syst. Appl. 40 (9) (2013) 885
816 [17] Duong Quoc Hung, Nadarajah Mithulananthan, R.C. Bansal, Analytical 3341–3350. 886
817 expressions for DG allocation in primary distribution networks, IEEE Trans. [42] Chandan Kumar Shiva, V. Mukherjee, A novel quasi-oppositional harmony 887
818 Energy Convers. 25 (3) (2010) 814–820. search algorithm for automatic generation control of power system, Appl. Soft 888
819 [18] S. Gopiya Naik, D.K. Khatod, M.P. Sharma, Optimal allocation of combined DG Comput. 35 (2015) 749–765. 889
820 and capacitor for real power loss minimization in distribution networks, Int. J. [43] Hai-bin Ouyang, Li-qun Gao, Xiang-yong Kong, Steven Li, De-xuan Zou, 890
821 Electr. Power Energy Syst. 53 (2013) 967–973. Hybrid harmony search particle swarm optimization with global dimension 891
822 [19] Zong Woo Geem, Joong Hoon Kim, G.V. Loganathan, A new heuristic selection, Inf. Sci. 346–347 (2016) 318–337. 892
823 optimization algorithm: harmony search, Simulation 76 (2) (2001) 60–68. [44] Fuqing Zhao, Yang Liu, Chuck Zhang, Junbiao Wang, A self-adaptive harmony 893
824 [20] M. Mahdavi, M. Fesanghary, E. Damangir, An improved harmony search PSO search algorithm and its performance analysis, Expert Syst. Appl. 42 894
825 algorithm for solving optimization problems, Appl. Math. Comput. 188 (2007) (2015) 7436–7455. 895
826 1567–1579. [45] Bahriye Akay, Dervis Karaboga, Artificial bee colony algorithm for large-scale 896
827 [21] Kang Seok Lee, et al., The harmony search heuristic algorithm for discrete problems and engineering design optimization, J. Intell. Manuf. 23 (2012) 897
828 structural optimization, Eng. Optim. 37 (7) (2005) 663–684. 1001–1014. 898
829 [22] Bin Wu, Cunhua Qian, Weihong Ni, Shuhai Fan, Hybrid harmony search and [46] Dervis Karaboga, Bahriye Akay, A modified artificial bee colony (ABC) 899
830 artificial bee colony algorithm for global optimization problems, Comput. algorithm for constrained optimization problems, Appl. Soft Comput 11 900
831 Math. Appl. 64 (8) (2012) 2621–2634. (2011) 3021–3031. 901
832 [23] K. Muthukumar, S. Jayalalitha, Optimal placement and sizing of distributed [47] G. Li, P. Niu, X. Xiao, Development and investigation of efficient artificial bee 902
833 generators and shunt capacitors for power loss minimization in radial colony algorithm for numerical function optimization, Appl. Soft Comput. 12 903
834 distribution networks using hybrid heuristic search optimization technique, (2012) 320–332. 904
835 Int. J. Electr. Power Energy Syst. 78 (June) (2016) 299–319. [48] Ahmad Nickabadi, Mohammad Mehdi Ebadzadeh, Reza Safabakhsh, A novel 905
836 [24] Z.W. Geem, Optimal cost design of water distribution networks using particle swarm optimization algorithm with adaptive inertia weight, Appl. 906
837 harmony Search, Eng. Optim. 38 (2006) 259–277. Soft Comput. 11 (2011) 3658–3670. 907
838 [25] Z.W. Geem, Particle-swarm harmony search for water network design, Eng. [49] Zhiyong Li, Weiyou Wang, Yanyan Yan, Zheng Li, PS–ABC: a hybrid algorithm 908
839 Optim. 49 (2009) 297–311. based on particle swarm and artificial bee colony for high-dimensional 909
840 [26] Pan Quan-Ke, P.N. Suganthan, M. Fatih Tasgetiren, A self-adaptive global best optimization problems, Expert Syst. Appl. 42 (2015) 8881–8895. 910
841 harmony search algorithm for continuous optimization problems, Appl. Math. [50] Teng Jen-Hao, A direct approach for distribution system load flow solutions, 911
842 Comput. 216 (2010) 830–848. IEEE Trans. Power Deliv. 18 (3) (2003) 882–887. 912
843 [27] M. Fesanghary, M. Mahdavi, M. Minary-Jolandan, Y. Alizadeh, Hybridizing [51] Leandro Dos Santos Coelho, Cezar Augusto Sierakowski, A software tool for 913
844 sequential quadratic programming with HS algorithm for engineering teaching of particle swarm optimization fundamentals, Adv. Eng. Softw. 39 914
845 optimization, Comput. Methods Appl. Mech. Eng. 197 (2008) 3080–3091. (11) (2008) 877–887. 915
846 [28] D. Karaboga, B. Basturk, A powerful and efficient algorithm for numerical
847 function optimization: artificial bee colony algorithm, J. Global Optim. 3
(2007) 9459–9471.

Article in Press

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Article in Press

Uploaded by

Copyright:

Available Formats

G Model

ASOC 3714 1–22 ARTICLE IN PRESS

Contents lists available at ScienceDirect

Applied Soft Computing

1 Integrated approach of network reconﬁguration with distributed

24 1. Introduction must be implemented to get the beneﬁts of integrating the DG units 39

and shunt capacitors into the distribution networks. 40

such as Artiﬁcial Immune (AIS) Systems [1], Modiﬁed particle 55

swarm optimization [2], Binary group search optimization [3], 56

Adapted ant colony [4], Fireworks algorithm [5], Harmony search 57

[6] algorithm. A fuzzy multi objective network reconﬁguration 58

(IHSA), global harmony search algorithm (GHSA) and self adaptive 97

tion programming technique (HSA-SQL) to accelerate the local search 110

tion problems with continuous as well as discrete design variables. 124

In [33,34] the overview of the recent applications of HSA, which 125

182 minimization of real power loss at distinct load levels. In hybrid,

capabilities at an operating power factor of 0.85 lagging [17]. The 242

203 loading reduction and voltage proﬁle improvement. The multi

209 A set of performance improvement indices is described mathe-

212 The objective of real power loss reduction index is to quan-

plosswcomp A directed graph representation of the radial distribution net-

274 responding BIBC matrix is generated by assigning ‘1 to discover

278 in Eq. (18). Let E be a set of direct edges {(1,2),(2,3),(3,4),(4,5),(3,6)}.

3. Evolutionary algorithms 322

able for discontinuous as well as continuous functions; (b) it can 329

b) Improvisation of HMV values by updating the solutions by mem- 336

296 between the nodes.

345 (19)where “BW” represents a random distance bandwidth. In this

equations expressed as below, 365

Vit+1 = ω.Vit + C1 rand(1) (Pbestit − Xid

ﬁtnessj The computational procedure of the proposed hybrid HS-PABC 461

dom population 466

Step 4: Employed bee phase initiation: Random populations 468

410 Xjt+1 = (Xmax + rand(0, 1)(Xmax − Xmin ) (25)

425 the intensiﬁcation and diversiﬁcation capability of ABC algorithm

for each ﬁtnessj using Eq. (1). 492

438 encounters the problem of slow convergence, traps in local optimal

Fig. 2. Flowchart of proposed hybrid HS-PABC algorithm.

513 Xjk new = min(Xk j ) + rand(0, 1)[max(Xk j ) − min(Xk j )] (30)

514 Step 7: Retain the best solution obtained so far in HMV

522 of MCN is reached.

524 Cycle = MCN

rithm for the enhancement of distribution network performance by 543

implementing the four test scenarios with an objective of optimiz- 544

ing the multi objective performance improvement indices such as 545

units and shunt capacitors

565 is considered as a negative PQ load with real and reactive power

DG3 (kW) @ node

577 bered from 69 to 73 and the sectionalizing switches are numbered

579 is 224.97 kW and 102.15 kVAR respectively. The system operates

581 base. Minimum bus voltage magnitude of 0.9092 p.u. is observed at

583 The simulation results obtained with proposed hybrid algo-

589 respect to the scenarios I–IV using hybrid approach. At medium

595 be seen that, the proposed hybrid algorithm provides a signiﬁcant

599 along with DG units and shunt capacitor placement (Scenario-IV)

600 at 100% load level. It is observed that scenario-IV provides bet-

603 optimized in scenario IV by retaining the DG and shunt capacitor

and @ node 60 for DG unit location) better power loss reduction is

Fig. 3. Daily load schedule.

0.94 Base Case

602.18 kW to 399.78 kW, 185.32 kW, 209.21 kW and 163.91 kW 638

Bus Voltage (p.u)

(PAR) is used to produce new solution around the existing best

274 responding BIBC matrix is generated by assigning ‘1 to discover