Professional Documents
Culture Documents
To cite this article: Sajjan Kumar, Kamal K. Mandal & Niladri Chakraborty (2020) A novel
opposition-based tuned-chaotic differential evolution technique for techno-economic analysis
by optimal placement of distributed generation, Engineering Optimization, 52:2, 303-324, DOI:
10.1080/0305215X.2019.1585832
1. Introduction
Because of the rapid depletion of fossil fuels, appalling environmental impacts, and higher trans-
mission and distribution losses in conventional power systems, small localized non-conventional
source-based power plants are being developed rapidly around the world. Owing to their smaller
size, low operational and maintenance costs and lower environmental impacts, non-conventional
source-based generating units in distributed generation (DG) systems are becoming more pop-
ular. The installation of different types of DG units (intermittent/non-intermittent) of appropri-
ate size and in the correct locations is a challenging task. It is necessary to install DG units of
optimal sizes at optimal locations in the power system network to achieve the maximum ben-
efits in terms of minimal cost and loss, with improvements in voltage profile, reliability, and
so on.
In recent years, a lot of research has been reported in the literature using different classical
and heuristic techniques to solve the optimal distributed generation placement (ODGP) problem.
Singh, Mukherjee, and Tiwari (2016) used the genetic algorithm (GA) to study the impact assess-
ment of different types of optimally placed DG in IEEE 37-bus radial distribution system (RDS).
The combination of analytical and GA methods was used by Vatani et al. (2016) for optimal allo-
cation of multiple DGs in different RDSs to minimize the system losses. An improved particle
swarm optimization (IPSO) algorithm was presented by Kanwar et al. (2017) for simultaneous allo-
cation of distributed energy resources. A modified teaching–learning-based optimization (TLBO)
algorithm was proposed by García and Mena (2013) to solve the ODGP problem. Yammani, Mah-
eswarapu, and Matam (2016) proposed a multi-objective shuffled bat algorithm to evaluate the
impact of DG placement and sizing considering different load models. Several other state-of-the-
art algorithms, including stochastic fractal search algorithm (SFSA) (Nguyen and Vo 2018), chaotic
stochastic fractal search with Gauss/mouse map (CSFS3) algorithm (Nguyen, Tran, and Vo 2018),
swine influenza model-based optimization with quarantine (SIMBO-Q) and quasi-oppositional
SIMBO-Q (QOSIMBO-Q) (Sharma, Bhattacharjee, and Bhattacharya 2016), quasi-oppositional
teaching–learning-based optimization (QOTLBO) technique (Sultana and Roy 2014), clonal differ-
ential evolution (Madihah, Junichi, and Hirotaka 2017), krill herd algorithm (KHA) (Sultana and
Roy 2016), harmony search algorithm with particle swarm optimization embedded artificial bee
colony (HSA-PABC) algorithm (Muthukumar and Jayalalitha 2016), cuckoo search (CS) algorithm
(Yuvaraj, Ravi, and Devabalaji 2017), augmented Lagrangian genetic algorithm (ALGA) (Hassan et
al. 2017) and stud krill herd algorithm (SKHA) (ChithraDevi, Lakshminarasimman, and Balamu-
rugan 2017), have been proposed to solve the ODGP problem in different aspects and scenarios.
Most of these authors investigated either DGs with unity power factor (upf) or some other power
factors for the enhancement of technical aspects only. For more feasible implementation in practi-
cal scenarios, economic aspects are as important as technical aspects, and therefore these should be
investigated.
In this work, the capacities and locations of DG sources in the network system operating at differ-
ent power factors are chosen optimally in such a way that the maximum techno-economic benefits
can be achieved. The overall cost of the system is directly linked with the investment in DG sources
considering the lifespan of the system and the costs incurred owing to losses during electricity dis-
tribution. The proposed approach has been implemented in the different test systems considering
economic as well as technical aspects simultaneously, which have been neglected in most of the
literature.
Previous work by this group (Kumar, Mandal, and Chakraborty 2017) showed that the opposition-
based chaotic differential evolution (OCDE) technique performs well for the single-objective case
and for smaller and medium-sized instances. But, for larger instance and for better performance in
both single- and multi-objective cases, it requires some tuning. Therefore, for better performance,
a new improved opposition-based tuned-chaotic differential evolution (OTCDE) algorithm is pro-
posed. In this modified technique, the mutation factor is tuned and made self-adaptive using chaotic
sequence by logistic mapping, while the crossover ratio is linearly varied from 0.95 to 0.75 to avoid
premature convergence. The mutation scheme ‘DE/tournament-best/1/bin’ is used for better per-
turbation of the base vector towards the best vector. The mutation process is executed within the
crossover operation to reduce the computational time. The effectiveness of the technique is tested
on the standard IEEE 33-bus, 69-bus and 118-bus RDSs. The proposed technique is found to be
capable of producing superior results to other metaheuristic techniques in terms of techno-economic
benefits.
2. Problem formulation
The main objective of optimal sizing and placement of DG units in RDSs is to optimize the over-
all system cost with optimum technical benefits. The overall system cost depends mainly on losses
occurring in the network system, as well as the cost of penetration. Thus, a cumulative cost index
(CI) is considered as one of the objectives of the proposed multi-objective optimization. Two other
important objectives are the voltage deviation index (VDI) and line flow capacity index (LFCI). These
three indices are described in the following subsections.
ENGINEERING OPTIMIZATION 305
where Pi , Pj and Qi , Qj are the total active and reactive power injections at buses i and j; Rij is the
line resistance between bus i and j; Vi and Vj are the voltage magnitudes, and δi and δj are the voltage
angles at bus i and j, respectively.
After installing one or more DGs in the system, the total active power loss (PLwithDG ) is reduced
compared to the total active power loss occurring in the system without any DGs (PLwithoutDG ).YELDG
reflects the yearly economic loss due to the loss occurring in the power distribution in the presence of
DG, as well as the yearly extra burden due to DG integration. Thus, YEL and YELDG can be calculated
as:
YEL = PLwithoutDG × Ce × 8760 (2)
NDG
CDG PDGi
YELDG = PLwithDG × Ce × 8760 + i=1
LDG (3)
where Ce is the energy loss cost per kWh ($); NDG is the number of installed DGs; CDG is the cost
of DG-generated power per kW, which includes capital investment in the DG, with installation,
operation and maintenance costs; and LDG is the total DG life (years). Therefore,
limit is also taken as one of the constraints on power flow. However, power flow near to the line limit
is also not a healthy situation with regard to the stability and reliability of the system. Therefore, the
minimization of LFCI is taken as another objective. LFCI can be considered as a measure of relative
line loading and can be represented as (Singh, Singh, and Verma 2009):
nb |Si |
LFCI = max (7)
i=1 Si0
where nb is the total number of branches in the network system; Si is the power flow (MVA) in the
ith branch after penetration; and Si0 is the power flow capacity limit (MVA) of the ith branch. Since
|Si | ≤ Si0 ∀i, LFCI ∈ [0,1].
where
3
ωi = 1.0; ∧ωi ∈ (0, 1) (9)
i=1
Several constraints, such as power conservation limit, line power flow limits and min–max bus voltage
limits, are considered and their limits strictly maintained to minimize MOI.
3.1. Initialization
The random initialization and mutation in DE of type ‘DE/rand/1/bin’ has an exploratory effect but
simultaneously slows down the convergence of DE. Therefore, in the OTCDE technique, the initial-
ization process is modified based on the concept of opposition-based learning (OBL) for generating
parent vectors (xjk ). This concept was first proposed by Tizhoosh (2005). According to this concept,
the random estimated data and corresponding opposite estimated data are considered simultaneously
to obtain a better approximation of the current candidate solution. The random estimated data can
be generated as:
xij = xmin,i + σi,j .(xmax,i − xmin,i ) (10)
where i = 1, 2, . . . D; j = 1, 2, . . . NP. NP and D are the population number and the dimension of the
problem or number of decisive variable parameters, respectively. xmin,i and xmax,i are the minimum
and maximum (min–max) limits of the ith decisive variable parameter. σi,j are random numbers, such
that σi,j ∈ [0, 1).
ENGINEERING OPTIMIZATION 307
The opposite estimated data (x̃ij ) can be generated by taking the component of xij . Mathematically,
it can be calculated as:
By calculating the fitness values for both points Xj and X̃j , where Xj = (x1j , x2j , . . . xij ) and X̃j =
(x̃1j , x̃2j , . . . x̃ij ), only superior points among them are selected as the parent vectors. Mathematically,
this can be expressed as:
where k indicates the generation or iteration number. For the initialization process, k = 1.
and the remaining two vectors are used to find the differential vector (Kaelo and Ali 2006). So, the
region around each (xtbk ) explores for each mutated point. This exploration maintains its exploratory
feature and, at the same time, expedites the convergence. This scheme of mutation is indicated as
‘DE/tournament-best/1/bin’; it is carried out to increase the diversity of the population and it can be
expressed as:
vjk if (μj ≤ CR k )
ukj = (13)
xjk else
where
vjk = xtb
k
+ F k .(xr2
k k
− xr3 ) (14)
μj denotes a uniformly distributed random number within the range [0, 1], generated anew for each
value of j. Equation (13) is known as a crossover process whereas Equation (14) is known as a mutation
process. The crossover ratio CR and mutation factor Fk can take any value from within the range (0,
1). A higher value of CR increases the probability of mutation, which helps the solution to increase
exploration. Similarly, a lower value of CR helps the solutions to converge rapidly. Hence, the value of
CR decreases linearly from 0.95 to 0.75 with generation advancement to achieve the proper balance
between explorations in the beginning and exploitations in the later stage. Fk plays a significant role in
the generation for generating perturbed mutant vectors. For better control of diversity and to prevent
stagnation of solutions to local minima, a factor Zk is varied on the basis of a chaotic sequence by
logistic mapping (Thangraj et al. 2012), as described below:
where k indicates the generation or iteration number and η is the chaos attractor. Equation (15) is
deterministic, displaying chaotic dynamics when η = 4 and Z0 ∈ (0, 1), where Z0 ∈ / {0, 14 , 12 , 34 , 1}.
k
By the above chaotic sequence generator, Z varies within the range (0,1). Since smaller values of the
308 S. KUMAR ET AL.
mutation factor cannot perturb the parent vectors effectively, and this can lead to premature con-
vergence or stagnate the solution to local minima. Therefore, the factor Zk can be tuned within the
specified range for generating Fk , and this can be formulated as:
where F min and F max are the min–max allowed range of Fk . After hundreds of trial runs, it is found
that the best choices for F min and F max are 0.7 and 0.95, respectively, for most problems.
3.3. Selection
The selection process is carried out to determine which one, between the trial vector and the parent
vector, will survive in the next generation. Survival of the vector (trial or parent) depends on its fitness
values. Thus, for the generation of a new parent vector, the selection process can be expressed as:
where f (.) indicates the fitness function or objective function (OF). This newly generated parent
vector becomes the parent vector of the (k + 1)th generation. Using this updated parent vector, repeat
the mutation, crossover and selection processes until the termination criterion is met.
Step 1: Randomly initialize the parent vectors using (10) within their corresponding min–max limits.
For the present work, DG sizes and their corresponding locations in the network system are
taken as the elements of the parent vectors. The parent matrix (X) is formed in such a way that
the upper half of the matrix is associated with the DG sizes, whereas the lower half is associated
with their corresponding locations in the network system. Mathematically, this can be modelled
as:
⎡ ⎤
⎡ ⎤ PDG1,1 PDG1,2 · · · PDG1,NP
x11 · · · x1j · · · x1,NP ⎢ . .. .. ⎥
⎢ .. .. .. .. ⎥ ⎢ .. . ··· . ⎥
⎢ . . . . ⎥
··· ⎥ ⎢ ⎢ ⎥
⎢ P P · · · P ⎥
⎢
X = ⎢ xi1 · · · xij · · · xi,NP ⎥ = ⎢⎥ ⎢ DGN,1 DGN,2 DGN,NP ⎥
(18)
L L · · · L ⎥
⎢ . . .. . .. ⎥ ⎢ DG1,1 DG1,2 DG1,NP ⎥
⎣ . . . . . . . . ⎦ ⎢ ⎣ ..
. .. .. ⎥
⎦
. ··· .
xD,1 · · · xD,j · · · xD,NP
LDGN,1 LDGN,2 · · · LDGN,NP
where subscript (DG1, . . . ,DGN) indicates the DG number and subscript (1,2, . . . ,NP) indi-
cates the population number. PDGN,NP and LDGN,NP indicate the size and corresponding location
of the Nth DG for the NPth population, respectively. So, the dimensions of the parent matrix
would be (2DGN × NP). Set the generation count = 0.
Step 2: Calculate the corresponding opposite estimated data (X̃) using (11).
Step 3: Calculate MOI for all the populations of both X and the X̃ matrix using (8), and then select
the fittest populations among them as per (12) to form the fittest parent matrix.
Step 4: Perform the mutation operation using (14) only for populations which satisfy μj ≤ CR , and
then perform the crossover operation using (13) to form trial vectors.
Step 5: Check the variable limits, such as DG size and location. If a limit is violated, then reinitialize
that population using (10).
ENGINEERING OPTIMIZATION 309
Step 6: Calculate MOI for all the trial vectors using (8).
Step 7: Check the system constraints limits, such as bus voltage limits and line flow limit. If a limit is
violated, then multiply the penalty factor by MOI to omit that population in the next generation
automatically.
Step 8: Perform the selection operation as per (17) to form a new updated parent matrix for the next
generation. Increase the generation count by 1.
Step 9: Check the termination criteria. Here, the termination criterion is set as the maximum number
of generations. If it is met, go to Step 10; otherwise, go to Step 4.
Step 10: Display the optimal result corresponding to minimum MOI and Stop.
DG location DG size Minimum bus Weakest Real power loss Real power from
No. of DGs Method (bus no.) (kW/kVAr) voltage (p.u.) bus (kW) substation (kW)
Scenario 1: Penetration with DGs operating at upf
3 OTCDE 13 801.80/0.0 0.9687 33 72.785 841.075
24 1091.31/0.0
30 1053.60/0.0
3 SFSAa 13 802.0/0.0 – – 72.785 –
24 1092.0/0.0
30 1053.7/0.0
3 KHAb 13 810.7/0.0 0.9610 18 75.412 –
25 836.8/0.0
30 841.0/0.0
4 OTCDE 6 926.27/0.0 0.9703 18 67.631 556.031
14 646.77/0.0
24 967.24/0.0
31 686.32/0.0
Scenario 2: Penetration with DGs operating at 0.95 lagging pf
3 OTCDE 13 830.23/272.88 0.9880 33 28.533 549.093
24 1124.65/369.66
30 1239.56/407.42
3 SFSAa 13 830.6/273.0 – – 28.533 –
24 1125.6/370.0
30 1239.6/407.4
3 QOSIMBO-Qc 13 830.3/272.9 – – 28.5 –
24 1123.9/369.4
30 1239.8/407.5
4 OTCDE 7 787.85/258.96 0.9879 33 23.599 294.189
14 657.92/216.25
24 1012.22/332.70
30 986.42/324.22
Note: DG = distributed generation; upf = unity power factor; pf = power factor; OTCDE = opposition-based tuned-chaotic dif-
ferential evolution; SFSA = stochastic fractal search algorithm; KHA = krill herd algorithm; QOSIMBO-Q = quasi-oppositional
swine influenza model-based optimization with quarantine.
a Nguyen and Vo (2018); b Sultana and Roy (2016); c Sharma, Bhattacharjee, and Bhattacharya (2016).
5.1.1.1. Statistical analysis. The same problem is also tackled by other variants of DE, namely
OCDE, chaotic differential evolution (CDE) and DE, to study their comparative performance and
check their robustness. For the performance study, all the results of 50 independent trial runs by each
technique for all scenarios are plotted in Figure 1 as box-and-whisker plots, or simply box plots. Here,
variations in results are indicated by small dots, the small square box indicates the mean value and the
min–max points are indicated by cross marks. The standard deviations (σ ) are considered as the box
range and the min–max is taken as the whiskers range. So, the interquartile range (IQR) will be 2σ .
Detailed results of all the above-mentioned variants of DE for both scenarios are shown in Table 2.
The smaller value of σ in all scenarios for the OTCDE technique can be interpreted as a smaller
deviation in the results. For clarification of its deviation, the histogram plots of OTCDE under dif-
ferent scenarios are shown in Figure 2. This figure shows that the highest frequency or count is close
to the mean of the respective scenarios, demonstrating that the central tendency rule is followed.
Normality tests, using the Shapiro–Wilk and Kolmogorov–Smirnov tests, have also been performed
to determine whether the results are derived from the normal distribution curve. The p values for
different scenarios obtained by these normality tests are presented in Table 3. For these tests, a confi-
dence level ‘α’ is set to the level of 0.05. From Table 3, it can be seen that the p values for all scenarios
are smaller than α, which signifies that the test rejects normality with a 95% level of confidence.
Therefore, it can be concluded that all the results have not been derived from normal distribution
curves.
For better understanding of the comparative performance studies, non-parametric tests, i.e. Fried-
man–ANOVA and Mann–Whitney tests, have also been performed and their results are reported
ENGINEERING OPTIMIZATION 311
Figure 1. Box-and-whisker plot of different techniques under different scenarios: (a) distributed generation (DG) at unity
power factor (upf); (b) DG at 0.95 lagging power factor (pf). OTCDE = opposition-based tuned-chaotic differential evolution;
OCDE = opposition-based chaotic differential evolution; CDE = chaotic differential evolution; DE = differential evolution.
in Table 4. By the Friedman–analysis of variance (ANOVA) test, a comparative ranking among all
the variants of DE has been generated for both scenarios. The OTCDE technique has the low-
est ranking for every scenario, showing the superiority of the algorithm. From this analysis, it
can be concluded that the performance of OTCDE is better than that of OCDE, followed by DE
and CDE. Similarly, the Mann–Whitney test (also called the Mann–Whitney U test or Wilcoxon
rank sum test) is performed for post hoc analysis to compare the performance of OTCDE with
the other variants of DE. The U value of all other techniques is much lower than the expected U
value (1250) compared with OTCDE. Therefore, the performance of OTCDE is much better than
the other variants of DE, and it can produce optimum results in every situation. The convergence
characteristics of the proposed OTCDE technique for different scenarios are shown in Figure 3.
From this figure, it can be observed that the proposed technique is capable of avoiding premature
convergence.
312 S. KUMAR ET AL.
Table 2. Performance study of different variants of differential evolution (DE) techniques: case study 1.
Figure 2. Histogram plots of the proposed opposition-based tuned-chaotic differential evolution (OTCDE) technique under differ-
ent scenarios. DG = distributed generation; upf = unity power factor; pf = power factor
Table 3. Normality tests of opposition-based tuned-chaotic differential evolution (OTCDE) technique (α = 0.05): case study 1.
p Value obtained by
Table 4. Non-parametric tests of different variants of differential evolution (DE): case study 1.
3 DGs at 0.95 4 DGs at 0.95
Case 3s DGs at upf 4 DGs at upf lagging pf lagging pf
Ranking of different variants of DE by Friedman–ANOVA test
OTCDE 1.38 1.79 1.54 1.74
OCDE 2.28 2.13 2.34 2.40
CDE 3.56 3.42 3.12 3.06
DE 2.78 2.66 3.00 2.80
Post hoc analysis of OTCDE through Mann–Whitney test
U value OCDE 655 1050.5 949 743
CDE 60 246 173 492
DE 263.5 646 300 535
Note: DG = distributed generation; upf = unity power factor; pf = power factor; OTCDE = opposition-
based tuned-chaotic differential evolution; OCDE = opposition-based chaotic differential evolution;
CDE = chaotic differential evolution.
160
3 DG @ upf
140 4 DG @ upf
3 DG @ 0.95 pf
4 DG @ 0.95 pf
120
Real Power Loss (kW)
100
80
60
40
20
0 50 100 150 200 250 300
Iteration No.
Figure 3. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for power loss minimiza-
tion in different scenarios: case study 1. DG = distributed generation; upf = unity power factor; pf = power factor.
in Table 5. For scenario 1, it can be observed that after the placement of three and four DGs of opti-
mal size at their optimal locations, the loss reduces to 73.712 and 68.293 kW, respectively, which is
a 65.1% and 67.6% reduction in comparison with the original system without DGs. These losses are
slightly greater compared to the single-objective case (the case with only loss minimization) for the
respective scenarios, but this compromise leads to more cost savings. After ODGP, the YELDG reduces
to $40,230.196 for three DGs and to $38,561.987 for four DGs at upf. So, the annual net savings
314 S. KUMAR ET AL.
Table 5. Multi-objective case considering cost index (CI), voltage deviation index (VDI) and line flow capacity index (LFCI) simulta-
neously: case study 1.
Total yearly
DG location DG size Minimum bus Weakest bus Real power loss economic loss Total yearly
No. of DGs (bus no.) (kW/kVAr) voltage (p.u.) no. (kW) ($) saving ($)
Scenario 1: Penetration with DGs operating at upf
3 13 814.47/0.0 0.9681 33 73.712 40,230.196 52,182.306
25 764.30/0.0
30 1069.35/0.0
4 6 790.99/0.0 0.9704 33 68.293 38,561.987 53,850.515
14 693.81/0.0
25 681.07/0.0
31 717.31/0.0
Scenario 2: Penetration with DGs operating at 0.95 lagging pf
3 13 848.57/278.91 0.9883 33 28.613 21,974.328 70,438.174
24 1051.91/345.75
30 1246.72/409.78
4 8 640.37/210.48 0.9882 33 24.587 20,002.679 72,409.823
15 532.06/174.88
25 769.54/252.94
30 1135.89/373.35
Note: DG = distributed generation; upf = unity power factor; pf = power factor.
0.4
0.35
Multiobjective Index (MOI)
0.3 3 DG @ upf
4 DG @ upf
3 DG @ 0.95 pf
0.25 4 DG @ 0.95 pf
0.2
Figure 4. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for multi-objective index
(MOI) minimization in different scenarios: case study 1. DG = distributed generation; upf = unity power factor; pf = power factor.
for the respective cases are $52,182.306 and $53,850.515, i.e. yearly economic benefits of 56.467%
and 58.272%, respectively, can be achieved. Similarly, for scenario 2, the losses reduce to 28.613 and
24.587 kW, while the YELDG reduces to $21,974.328 and $20,002.679 after placement of three and
four DGs, respectively. So, the corresponding total yearly net savings increase to $70,438.174 and
$72,409.823, i.e. yearly economic benefits of 76.221% and 78.355% can be achieved by placement of
three and four DGs, respectively.
Figure 4 shows the convergence characteristics of MOI with respect to iteration number for three-
and four-DG cases in both scenarios. From the figure, it can be seen that as the number of DGs
increases, more iterations are required to find the optimal values. This is due to the fact that with the
increase in number of DGs, the number of decision variable also increases.
ENGINEERING OPTIMIZATION 315
Figure 5. Bus voltage profile in p.u. for the IEEE 33-bus system. DG = distributed generation; upf = unity power factor;
pf = power factor.
60
No DG
3 DG @ upf
50 4 DG @ upf
3 DG @ 0.95 pf
Branch Power Loss (kW)
4 DG @ 0.95 pf
40
30
20
10
0
0 5 10 15 20 25 30 35
Branch Number
Figure 6. Line power loss profile for the IEEE 33-bus system. DG = distributed generation; upf = unity power factor; pf = power
factor.
Figure 5 depicts the bus voltage profile in p.u. in the form of a radar chart. This shows that without
penetration, the minimum voltage level occurs at bus number 18 because it is an end bus of the longest
branch. After installing three and four DGs, the end bus of the second longest branch, i.e. bus number
33, becomes the weakest bus. Since reactive power management mainly depends on the operating
power factor of DG, it plays a vital role in the improvement of voltage deviations, which can also be
seen in Figure 5.
Figure 6 shows the variation in line power loss with respect to branch number. Power loss in branch
numbers 1–5 is very high without DG; it decreases drastically after installing DGs because DG shares
316 S. KUMAR ET AL.
90
80
70
Real Power Loss (kW)
1 DG @ upf
60 2 DG @ upf
3 DG @ upf
1 DG @ 0.95 pf
50
2 DG @ 0.95 pf
3 DG @ 0.95 pf
40
30
20
0 50 100 150 200 250 300
Iteration No.
Figure 7. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for power loss minimiza-
tion in different scenarios: case study 2. DG = distributed generation; upf = unity power factor; pf = power factor.
some of the load and hence the current flow from these branches reduces. Current flow from branch
number 1 is the maximum but impedance of that branch is very low, which is why power loss in this
branch is less than that in branch number 2.
ENGINEERING OPTIMIZATION 317
Table 7. Multi-objective case considering cost index (CI), voltage deviation index (VDI) and line flow capacity index (LFCI) simulta-
neously: case study 2.
Total yearly
DG location DG size Minimum bus Weakest bus Real power loss economic loss Total yearly
No. of DGs (bus no.) (kW/kVAr) voltage (p.u.) no. (kW) ($) saving ($)
Scenario 1: Penetration with DGs operating at upf
1 61 1850.12/0.0 0.9682 27 83.242 42,010.537 56,540.591
2 18 507.56/0.0 0.9790 65 71.701 38,288.495 60,262.633
61 1787.00/0.0
3 12 379.26/0.0 0.9788 65 69.761 37,917.819 60,633.309
22 328.39/0.0
61 1746.46/0.0
Scenario 2: Penetration with DGs operating at 0.95 lagging pf
1 61 2069.10/680.08 0.9719 27 38.426 23,037.755 75,513.374
2 18 604.89/198.82 0.9939 69 23.510 17,932.221 80,618.908
61 1940.11/637.68
3 11 489.36/160.84 0.9942 50 20.786 17,451.057 81,100.072
17 429.37/141.13
61 1863.49/612.50
Note: DG = distributed generation; upf = unity power factor; pf = power factor.
0.4
0.35
Multiobjective Index (MOI)
1 DG @ upf
0.3
2 DG @ upf
3 DG @ upf
1 DG @ 0.95 pf
0.25 2 DG @ 0.95 pf
3 DG @ 0.95 pf
0.2
Figure 8. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for multi-objective index
(MOI) minimization in different scenarios: case study 2. DG = distributed generation; upf = unity power factor; pf = power factor.
Figure 9. Bus voltage profile in p.u. for the IEEE 69-bus system. DG = distributed generation; upf = unity power factor;
pf = power factor.
50
No DG
45 1 DG @ upf
2 DG @ upf
40
3 DG @ upf
35 1 DG @ 0.95 pf
Branch Power Loss (kW)
2 DG @ 0.95 pf
30 3 DG @ 0.95 pf
25
20
15
10
0
0 10 20 30 40 50 60 70
Branch Number
Figure 10. Line power loss profile for the IEEE 69-bus system. DG = distributed generation; upf = unity power factor; pf = power
factor.
the proposed technique are also compared with other techniques from the literature, namely TLBO,
SFSA and KHA. A detailed comparative study with some more techniques can be found in supple-
mentary Table S2. From this table, it can be found that the proposed technique outperforms the other
methods and can produce superior results for both scenarios.
DG location DG size Minimum bus Weakest Real power loss Real power from
No. of DG s Method (bus no.) (kW/kVAr) voltage (p.u.) bus (kW) substation (kW)
Scenario 1: Penetration with DGs operating at upf
5 OTCDE 30 3998.65/0.0 0.9546 43 569.044 9022.334
39 2371.32/0.0
72 2576.37/0.0
80 2513.72/0.0
118 2796.65/0.0
5 SFSAa 50 3181.3/0.0 – – 578.741 –
73 2324.1/0.0
80 2042.8/0.0
91 2015.6/0.0
110 2868.6/0.0
7 OTCDE 20 1789.69/0.0 0.9543 43 524.924 7650.474
39 2734.58/0.0
47 1821.19/0.0
72 2525.51/0.0
80 2326.88/0.0
90 1661.39/0.0
110 2725.21/0.0
7 SFSAa 21 1375.7/0.0 – – 525.277 –
42 1199.7/0.0
50 2741.8/0.0
71 2891.5/0.0
81 1702.5/0.0
97 1332.1/0.0
110 2667.4/0.0
Scenario 2: Penetration with DGs operating at 0.866 lagging pf
5 OTCDE 39 3145.46/1816.24 0.9668 96 212.907 9632.967
47 3100.12/1790.07
74 1960.94/1132.28
80 2406.27/1389.42
110 2677.15/1545.84
5 SFSAa 50 3078.7/1777.7 – – 236.528 –
74 2384.6/1376.9
79 2508.8/1448.7
91 2099.5/1212.3
110 3100.4/1790.2
7 OTCDE 30 3865.04/2231.74 0.9732 27 149.627 4579.477
39 2747.45/1586.43
47 1669.11/963.78
74 2423.12/1399.15
80 2569.10/1483.45
90 1902.42/1098.49
110 3103.91/1792.25
7 SFSAa 21 1935.1/1117.4 – – 155.159 –
40 2081.0/1201.6
50 3130.1/1807.4
71 2892.0/1669.9
80 2054.1/1186.1
96 1385.9/800.3
110 3230.6/1865.4
Note: DG = distributed generation; upf = unity power factor; pf = power factor; OTCDE = opposition-based tuned-chaotic
differential evolution; SFSA = stochastic fractal search algorithm.
a Nguyen and Vo (2018).
reduces drastically, to 38.426, 23.510 and 20.786 kW after the placement of one, two and three DGs,
respectively. Hence, YELDG also reduces, to $42,010.537, $38,288.495 and $37,917.819, respective to
penetration level. Therefore, total yearly cost savings of $56,540.591, $60,262.633 and $60,633.309
can be achieved by the placement of one, two and three DGs, respectively, i.e. cost savings per annum
equivalent to 57.37%, 61.15% and 61.52% can be achieved. Similarly, in scenario 2, the cost savings
have increased, to 76.62%, 81.80% and 82.29%, for the respective penetration levels.
320 S. KUMAR ET AL.
900
5 DG @ upf
7 DG @ upf
800
5 DG @ 0.866 pf
7 DG @ 0.866 pf
700
500
400
300
200
100
0 100 200 300 400 500 600 700 800 900 1000
Iteration No.
Figure 11. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for power loss minimiza-
tion in different scenarios: case study 3. DG = distributed generation; upf = unity power factor; pf = power factor.
Table 9. Multi-objective case considering cost index (CI), voltage deviation index (VDI) and line flow capacity index (LFCI) simulta-
neously: case study 3.
Total yearly
DG location DG size Minimum bus Weakest bus Real power loss economic loss Total yearly
No. of DGs (bus no.) (kW/kVAr) voltage (p.u.) no. (kW) ($) saving ($)
Scenario 1: Penetration with DGs operating at upf
5 39 3038.42/0.0 0.9586 43 574.428 291,550.645 276,050.393
46 2402.97/0.0
72 2617.20/0.0
80 2387.66/0.0
109 2870.74/0.0
7 30 2894.71/0.0 0.9572 43 529.599 279,094.337 288,506.701
39 2595.00/0.0
47 1584.20/0.0
74 2323.38/0.0
80 2301.68/0.0
96 1051.99/0.0
108 2959.05/0.0
Scenario 2: Penetration with DGs operating at 0.866 lagging pf
5 39 3273.55/1890.21 0.9670 96 226.407 144,959.950 422,641.088
44 3712.68/2143.77
74 2356.75/1360.83
80 2814.80/1625.32
110 3106.80/1793.92
7 39 2867.78/1655.91 0.9675 96 175.389 126,949.366 440,651.672
48 1744.41/1007.25
56 1090.59/629.73
61 2270.15/1310.83
72 2643.43/1526.36
80 2744.77/1584.88
109 3348.47/1933.47
Note: DG = distributed generation; upf = unity power factor; pf = power factor.
The convergence characteristics of MOI for all the scenarios are depicted in Figure 8. Similarly
to the IEEE 33-bus system, as the penetration level increases it takes more iterations to converge.
However, the proposed OTCDE technique is still capable of producing the optimum results and does
not stagnate to its local minima. Figure 9 shows a radar plot of the bus voltage profile in p.u. of the
ENGINEERING OPTIMIZATION 321
standard system as well as the system with one, two and three DGs for both scenarios simultaneously.
From this figure, it can be seen that the voltage profile improves with penetration level. The operating
power factor of DGs also affects the bus voltage level owing to better reactive power management.
From Figure 10, it can be seen that line power loss reduces significantly in branch numbers 5, 6 and
52–60 after installing one or more DGs for both scenarios.
0.9
5 DG @ upf
0.8 7 DG @ upf
5 DG @ 0.866 pf
7 DG @ 0.866 pf
0.7
Multiobjective Index (MOI)
0.6
0.5
0.4
0.3
0.2
0 100 200 300 400 500 600 700 800 900 1000
Iteration No.
Figure 12. Convergence characteristics of opposition-based tuned-chaotic differential evolution (OTCDE) for multi-objective index
(MOI) minimization in different scenarios: case study 3. DG = distributed generation; upf = unity power factor; pf = power factor.
1.04
1.02
0.98
Bus Voltage (pu)
0.96
0.94
0.92
No DG
5 DG @ upf
0.9
7 DG @ upf
0.88 5 DG @ 0.866 pf
7 DG @ 0.866 pf
0.86
0 20 40 60 80 100 120
Bus Number
Figure 13. Bus voltage profile in p.u. for the IEEE 118-bus system. DG = distributed generation; upf = unity power factor;
pf = power factor.
322 S. KUMAR ET AL.
120
No DG
5 DG @ upf
100 7 DG @ upf
5 DG @ 0.866 pf
7 DG @ 0.866 pf
60
40
20
0
0 20 40 60 80 100 120
Branch Number
Figure 14. Line power loss profile for the IEEE 118-bus system. DG = distributed generation; upf = unity power factor;
pf = power factor.
11 kV and 100 MVA, respectively. For this system, the total standard losses are 1295.893 kW and
978.971 kVAr, and hence, consumed power from substation is 24,005.893 kW. The minimum bus
voltage drops to 0.866 p.u. at bus number 77.
6. Conclusion
This article presents a novel OTCDE technique for the optimal sizing and placement of multiple
DG units in IEEE 33-bus, 69-bus and 118-bus RDSs. The MOI has been formulated to obtain maxi-
mum techno-economic benefits, including cost, voltage deviation and line flow. Optimal placement
of multiple DGs with different operating power factors by the OTCDE method accomplished the
above-stated benefits. The results show that OTCDE can also handle more decision variables with
good convergence characteristics for a larger system such as the IEEE 118-bus system. The results
obtained by OTCDE have also been compared with some other newly proposed techniques, and the
proposed technique produced superior results in every scenario for all cases. The performance of
OTCDE and its robustness compared with OCDE, CDE and DE have also been tested statistically
and found to be better. It may be concluded that the proposed OTCDE technique performs better
than other techniques and can successfully avoid premature convergence.
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
The authors acknowledge the support from the Green Energy Project of UGC UPE-II Program [grant number
D.O.No.S.14-2/2008(NS/PE) dated 1 December 2011] and the Departmental Research Scheme of the Power Engineer-
ing Department, Jadavpur University, India, which is funded by UGC-DRS Phase-I Program, Government of India
[grant number F.3-31/2012 (SAP-II)].
ORCID
Sajjan Kumar http://orcid.org/0000-0002-7321-278X
Niladri Chakraborty http://orcid.org/0000-0002-7506-6624
References
Chakravorty, M., and D. Das. 2001. “Voltage Stability Analysis of Radial Distribution Networks.” International Journal
of Electrical Power and Energy Systems 23 (2): 129–135.
ChithraDevi, S. A., L. Lakshminarasimman, and R. Balamurugan. 2017. “Stud Krill Herd Algorithm for Multiple DG
Placement and Sizing in a Radial Distribution System.” Engineering Science and Technology, An International Journal
20 (2): 748–759.
García, J. A. M., and A. J. G. Mena. 2013. “Optimal Distributed Generation Location and Size Using a Modified Teach-
ing–Learning Based Optimization Algorithm.” International Journal of Electrical Power and Energy Systems 50:
65–75.
Hassan, A. A., F. H. Fahmy, A. E. A. Nafeh, and M. A. Abu-elmagd. 2017. “Genetic Single Objective Optimisation for
Sizing and Allocation of Renewable DG Systems.” International Journal of Sustainable Energy 36 (6): 545–562.
Kaelo, P., and M. M. Ali. 2006. “A Numerical Study of Some Modified Differential Evolution Algorithms.” European
Journal of Operational Research 169 (3): 1176–1184.
Kanwar, N., N. Gupta, K. R. Niazi, A. Swarnkar, and R. C. Bansal. 2017. “Simultaneous Allocation of Distributed Energy
Resource Using Improved Particle Swarm Optimization.” Applied Energy 185 (2): 1684–1693.
Kumar, K. S., and T. Jayabarathi. 2012. “Power System Reconfiguration and Loss Minimization for an Distribution
Systems Using Bacterial Foraging Optimization Algorithm.” International Journal of Electrical Power and Energy
Systems 36 (1): 13–17.
Kumar, S., K. K. Mandal, and N. Chakraborty. 2017. “Multiple DG Allocation in Radial Distribution System Using
Opposition Based Chaotic Differential Evolution Technique.” Paper presented at i-PACT, India, April 21–22.
Madihah, M. R., M. Junichi, and T. Hirotaka. 2017. “Fossil Fuel Cost Saving Maximization: Optimal Allocation and Siz-
ing of Renewable-Energy Distributed Generation Units Considering Uncertainty via Clonal Differential Evolution.”
Applied Thermal Engineering 114: 1424–1432.
Muthukumar, K., and S. Jayalalitha. 2016. “Optimal Placement and Sizing of Distributed Generators and Shunt Capac-
itors for Power Loss Minimization in Radial Distribution Networks Using Hybrid Heuristic Search Optimization
Technique.” International Journal of Electrical Power and Energy Systems 78: 299–319.
324 S. KUMAR ET AL.
Nguyen, T. P., T. T. Tran, and D. N. Vo. 2018. “Improved Stochastic Fractal Search Algorithm with Chaos for Opti-
mal Determination of Location, Size, and Quantity of Distributed Generators in Distribution Systems.” Neural
Computing and Applications. doi:10.1007/s00521-018-3603-1.
Nguyen, T. P., and D. N. Vo. 2018. “A Novel Stochastic Fractal Search Algorithm for Optimal Allocation of Distributed
Generators in Radial Distribution Systems.” Applied Soft Computing 70: 773–796.
Rao, B. H., and S. Sivanagaraju. 2012. “Optimum Allocation and Sizing of Distributed Generations Based on Clonal
Selection Algorithm for Loss Reduction and Technical Benefit of Energy Savings.” Paper presented at APCET, India,
August 2–4.
Sharma, S., S. Bhattacharjee, and A. Bhattacharya. 2016. “Quasi-oppositional Swine Influenza Model Based Opti-
mization with Quarantine for Optimal Allocation of DG in Radial Distribution Network.” International Journal
of Electrical Power and Energy Systems 74: 348–373.
Singh, B., V. Mukherjee, and P. Tiwari. 2016. “Genetic Algorithm for Impact Assessment of Optimally Placed Dis-
tributed Generations with Different Load Models from Minimum Total MVA Intake Viewpoint of Main Substation.”
Renewable and Sustainable Energy Reviews 57: 1611–1636.
Singh, D., D. Singh, and K. S. Verma. 2009. “Multiobjective Optimization for DG Planning with Load Models.” IEEE
Transactions on Power Systems 24 (1): 427–436.
Sultana, S., and P. K. Roy. 2014. “Multi-objective Quasi-oppositional Teaching Learning Based Optimization for Opti-
mal Location of Distributed Generator in Radial Distribution Systems.” International Journal of Electrical Power and
Energy Systems 63: 534–545.
Sultana, S., and P. K. Roy. 2016. “Krill Herd Algorithm for Optimal Location of Distributed Generator in Radial
Distribution System.” Applied Soft Computing 40: 391–404.
Thangraj, R., M. Pant, T. R. Chelliah, and A. Abraham. 2012. “Opposition Based Chaotic Differential Evolution
Algorithm for Solving Global Optimization Problem.” Paper presented at Fourth World Congress on Nature and
Biologically Inspired Computing (NaBIC), November 5.
Tizhoosh, H. R. 2005. “Opposition-Based Learning: A New Scheme for Machine Intelligence.” Proceedings of Compu-
tational Intelligence for Modeling Control and Automation Conference 1: 695–701.
Vatani, M., D. S. Alkaran, M. J. Sanjari, and G. B. Gharehpetian. 2016. “Multiple Distributed Generation Units Allo-
cation in Distribution Network for Loss Reduction Based on a Combination of Analytical and Genetic Algorithm
Methods.” IET Generation, Transmission & Distribution 10 (1): 66–72.
Yammani, C., S. Maheswarapu, and S. K. Matam. 2016. “A Multi-objective Shuffled Bat Algorithm for Optimal Place-
ment and Sizing of Multi Distributed Generations with Different Load Models.” International Journal of Electrical
Power and Energy Systems 79: 120–131.
Yuvaraj, T., K. Ravi, and K. R. Devabalaji. 2017. “Optimal Allocation of DG and DSTATCOM in Radial Distribution
System Using Cuckoo Search Optimization Algorithm.” Modelling and Simulation in Engineering 2017: 1–11.