Expert Systems With Applications 199 (2022) 116858

Contents lists available at ScienceDirect

Expert Systems With Applications

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

An efficient optimizer for optimal overcurrent relay coordination in power

distribution system
Debasis Acharya, Dushmanta Kumar Das ∗
Department of Electrical and Electronics Engineering National Institute of Technology Nagaland, Chumukedima 797103, India

ARTICLE INFO ABSTRACT

Keywords: In this paper, a student-learning strategy based efficient optimizer has been developed to solve optimal over
Optimization algorithm current relay coordination problem in the power system. The power system is a massive, interrelated network
Time dial setting (TDS) with inherent nonlinearity. The conventional analytical methods to solve an overcurrent relay coordination
Plug setting (PS)
issue is difficult for such systems. Therefore, the metaheuristic optimization technique can be a good choice for
Coordination time interval (CTI)
relay coordination problem in the power network. Class topper optimization (CTO) is a stochastic technique
inspired by the student learning behavior to improve his/her performance to be the best in his/her class
which is successfully applied for different optimization problems. Like the other metaheuristic algorithm,
CTO faces the problem of local optima with premature convergence to deal with complex problems. In this
paper, a sigmoid acceleration coefficient based chaotic search class topper optimization (SCCTO) is proposed
to develop an efficient optimizer and applied to solve optimal relay coordination problem. The logistic map
based chaotic search is used at initialization stage of SCCTO algorithm. In the update stage of the proposed
algorithm, a slowly varying sigmoid function is used along with normal varying function. It will keep diversity
in local solution to explore the optimum global solution in the search space. To validate the effectiveness
of the proposed algorithm to solve an optimization problem, the optimal overcurrent relay coordination in
the distribution network is tested. The optimal setting of time dial of the relay (over current) with total
operating time minimization is the aim of such optimization problem considering relay coordination time
gap as constraints. Three types of power distribution networks are examined optimal relay coordination
problems. The simulation result shows the efficiency of the developed optimizer in relay coordination with an
improvement of 26 to 35% in total relay operating time than existing results.

1. Introduction used as backup protection. They operate if the primary relays are not
operated within specified time. Thus, a coordination interval of time
Electricity supply and distribution network is a complex intercon- must be there among primary and backup relay scheme (Korashy,
nected system. With the increasing demand of electricity, the complex- Kamel, Nasrat, & Jurado, 2020). The optimal coordination between
ity in the electricity supply system becomes high. To supply continuous primary and backup relay is a prior requirement in power system
quality power to consume, the protective relay system in distribution protection scheme to maintain reliability in electricity supply. The relay
network should be efficient. The purpose of protection scheme in power coordination in electricity system must be accurate and efficient to clear
network is to sense and detect the fault and separate the faulted part of fault as fast as possible. Total operational time of relay system should
the healthy part of the system. The protection system of power supply
be maintained minimum for optimal coordination between primary and
also relates to the revenue in powerful industries. As the relay systems
backup relay. Thus, the primary objective of optimal relay coordination
in power networks work coordinately, improper relay coordination can
is to obtain minimal operating relay time under different operational
lead to financial loss. Directional over current relays are the primary
condition.
choice for their selectivity, sensitivity and accuracy (Elsadd, Kawady,
There are two different settings in relay in terms of plug setting
Taalab, & Elkalashy, 2021). Generally, the primary protective relays
operate when any fault occurred in the area of the ruling relay system. and setting of time decided by expert’s knowledge. As the distribution
It leads the circuit breakers (CBs) open to isolate the faulty sections network of electricity is complex, the conventional analytical methods
of the network. To secure the protection system, some relays are also are time consuming and not exactly efficient. There are many research

∗ Corresponding author.
E-mail addresses: deba.konnagar@gmail.com (D. Acharya), dushmantakumardas29@gmail.com (D.K. Das).

https://doi.org/10.1016/j.eswa.2022.116858
Received 30 April 2021; Received in revised form 25 December 2021; Accepted 7 March 2022
Available online 19 March 2022
0957-4174/© 2022 Elsevier Ltd. All rights reserved.
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

works available for studying relay coordination problem such as Alka- Table 1
𝛽 and 𝛼 of overcurrent relays (Saha, Datta, & Das,
ran, Vatani, Sanjari, Gharehpetian, and Naderi (2016), Elsadd et al.
2016).
(2021) and Korashy, Kamel, Jurado, and Youssef (2019). In Swathika
Type of relay 𝛽 𝛼
and Hemamalini (2020), graph theory based methods have been ap-
IDMT 0.140 0.020
plied to relay coordination solution. Trial and error technique is used
Very inverse 13.50 1.0
in Korashy et al. (2019) for coordination of relay where a larger Extremely inverse 80.0 2.0
iterative process is applied which leads the convergence process slow.
Some classical methods like curve fitting (Bedekar & Bhide, 2010),
analytical method (Sarwagya, Nayak, & Ranjan, 2020) also apply to the
problem. In most of work, the nonlinear relay coordinating problem The structure of the paper is as follows: After introduction in
has been solved by representing the problem as a problem of linear Section 1, the problem formulation is presented in Section 2. Simulation
optimization at a fixed plug setting (Elsadd et al., 2021). and result analysis is discussed in Section 3. Conclusion of the present
Presently, swarm based evolutionary methods such as: genetic work has been drawn in Section 4.
algorithm (GA) (Alkaran et al., 2016), particle swarm optimization
(PSO) (Khurshaid, Wadood, Farkoush, Kim, Cho et al., 2019), ant 2. Problem formulation
colony optimization (ACO) (Shih, Salazar, & Enríquez, 2015), differ-
ential evolution (Shih, Conde, & Ángeles-Camacho, 2019) are used 2.1. Overcurrent relay coordination in distribution power network
for optimal relay coordination. Some other methods such as harmonic
search algorithm (HAS) (Rajput & Pandya, 2017), chaotic firefly algo- In this part, the problem formulation for relay coordination in
rithm (CFA) (Gokhale & Kale, 2016), hybrid gray wolf optimization power distribution network is presented. For this purpose, a radial
(GWO) (Korashy et al., 2019) is also verified for the problem. In Be- based two bus distribution network is considered as illustrated in Fig. 1.
dekar and Bhide (2010), hybrid nonlinear programming based GA was Suppose, there is a fault occurred at location 𝐾 near 𝐵 bus. From Fig. 1,
used for relay-coordination. For the issue of over-current relays, a it is clear that relay 𝑅2 is responsible to sense the fault and operate the
Firefly-based algorithm is utilized in Khurshaid, Wadood, Farkoush, corresponding circuit breaker (CB) to separate the faulty section from
Kim, Yu et al. (2019). In Kudkelwar and Sarkar (2019), The same the healthy part of the network. If 𝑅2 does not operate at the set time,
issue is addressed using the water cycle algorithm (WCA). For relay the relay 𝑅1 must work as a secondary relay at a specified time gap.
coordination, the majority of the known optimization strategies use Therefore, the time setting for relay 𝑅1 is set high compared to the relay
irregular start where the TDSs of relays are generated randomly keeping 𝑅2 . The time setting for the relay 𝑅1 depends on the operating time of
plug setting multiplier (PSM) constant (Alkaran et al., 2016; Kudkelwar relay (𝑅1 ), circuit breaker’s (CB). Assume an eight-relay standard ring-
& Sarkar, 2019; Rajput & Pandya, 2017). main distribution scheme. 𝑅1 and 𝑅8 are bilateral relays, whereas the
The intellectual learning strategy of a student to become best among rest is unilateral overcurrent relays. The relays are set to operate under
other students in a class is the key concept of well known class topper a safety zone. Thus, relays 2, 4, 6, and 8 will make one set and relays 1,
optimization (CTO) algorithm introduced in Das, Das, and Dey (2018). 3, 5, and 7 will make another set of relay. The co-ordination between
Due to few parameters, clustering nature, easy implementation and relays in set 1 is as follows: 𝑇𝑅8 > 𝑇𝑅6 > 𝑇𝑅4 > 𝑇𝑅2 . The coordination
lower computational complexity, CTO is applied in different engineer- relays in set 2 is as follows: 𝑇𝑅7 > 𝑇𝑅5 > 𝑇𝑅3 > 𝑇𝑅1 .
ing applications (Acharya & Das, 2021b; Balakrishnan & Nisi, 2020;
Choudhary & Das, 2021). In Srivastava and Das (2020), an aggrandized 2.1.1. Characteristic of relay
CTO is suggested for optimal load dispatch problem in power system. There are different types of overcurrent relays available for power
In Acharya and Das (2021b), a modified constricted CTO is exam- system application. An inverse definite minimum time (IDMT) relay is
ined for the artificial respiratory support system. In Mohanta and Das generally applied as an overcurrent relay for their best features (Elsadd
(2021), the author has successfully applied CTO for wireless sensor net- et al., 2021). The responding time of IDMT is expressed as follows:
work. In Pradhan, Acharya, and Das (2022), the authors have applied 𝛽 × 𝑇 𝐷𝑆 0.14 × 𝑇 𝐷𝑆
CTO for core nuclear power control. Similar to the other metaheuristic 𝑇𝑜𝑝 = = , (1)
(𝑃 𝑆𝑀)𝛼 − 1 (𝑃 𝑆𝑀)0.02 − 1
algorithm, CTO also suffers with premature convergence problem with
local solutions. The performance of CTO depends on initial generation, where 𝑇𝑜𝑝 is relay operational time, 𝛽 and 𝛼 are constants mentioned
inertia weight, acceleration coefficients. As the performance of any in Table 1. The plug multiplier setting (PMS) expressed as follows:
evolutionary algorithm depends on initial generation, the application of 𝐼𝑓
𝑃 𝑆𝑀 = , (2)
chaotic variable in initialization stage can provide a diversity in search- 𝐼𝑝𝑟
ing optimum solution by the algorithm. Moreover, the factor of the
where 𝐼𝑓 is the fault-current and 𝐼𝑝𝑟 is the pick-up current of the relay.
acceleration coefficient in most of algorithm is assumed as constant. It
(1) can be reformulated for a specific PSM with a fixed pick-up current
has an important effect on global searching solution. Most of evolution-
as follows:
ary algorithm suffers from premature convergence problem for wrong
selection of such factors. Therefore, a special care must be considered 𝑇𝑜𝑝 = 𝑎𝑖 × 𝑇 𝐷𝑆, (3)
in selecting accelerating coefficient. From different literatures, one can
0.14
clearly identify that the proper acceleration coefficients are able to where 𝑎𝑖 = .
(𝑃 𝑆𝑀)0.02 −1
improve the performance of an evolutionary algorithm to search global
solution till the end of the search process (Tian, Zhao, & Shi, 2019). 2.2. Objective function
To maintain a balance between exploration and exploitation abilities
in CTO, this paper proposed a sigmoid acceleration coefficient based The objective function is formed with the aim of minimizing the
chaotic search class topper optimization algorithm. Chaotic variable is total relay operation time as
adapted in initial stage to generate properly distributed initial particles ∑
𝑛
(students in CTO). The sigmoid coefficient will balance the global min 𝑧 = 𝜔𝑗 × 𝑇𝑖𝑘 , (4)
search in starting stage and exponentially converge on exploitation 𝑖=0

stage. To verify the SCCTO algorithm in power system application, the where 𝑛 is quantity of relay, 𝑇𝑖𝑘 is operational time of 𝑖th relay for fault
optimization problem is formulated for relay coordination in different at location 𝑘, 𝑤𝑗 is the weighting scale of system. 𝑤 is set to 1 for a
types of distribution networks and compared with some existing results. distribution network.

2
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Fig. 1. Radial power distribution network.

2.2.1. Constraints overall relay performance time is long enough, this could affect that
In this section, different constraints for the relay system in power much delays in the preventative entire system and harm devices in the
network are discussed. electrical network. It may diminish the device’s efficiency (Sarkar &
Kudkelwar, 0000). As a result, the relay mechanism must defend the
1. Operating time constraints: A Relay should work for a fault ini-
power network as quickly as feasible while adhering to the various
tiated in its protective area or zone. For variable load associated
limits mentioned before. A high sensitivity and selectivity relay system
with a distribution network, the problem of over current may
helps keep the electrical network’s reliability and sustainability (Shih
arise in load connection or disconnection process in the network.
et al., 2019). The overall working time over all relays involved in the
Such overcurrent does not exist for a long time. Therefore, the
mechanism is minimized in the optimal relay coordinating issue, so that
relay must not work for such overcurrent. The relay operating
each relay should function in the shortest time possible. As a result,
time must be maintained within an operating range as follows:
optimizing overall relay running time has the benefit of maintaining
𝑇𝑖𝑘 min ≤ 𝑇𝑖𝑘 ≤ 𝑇𝑖𝑘 max , (5) power network resilience, attentiveness, and stable. Commercially man-
ufactured overcurrent relays such as the MiCOM P111, P115, and P116
where 𝑇𝑖𝑘 min is minimum operational time and 𝑇𝑖𝑘 max is max-
(overcurrent relays) are defined, including a working time with less
imum operational time. Generally, 𝑇min = 0.1 s and 𝑇max =
than 10 ms and the P127, P521 (overcurrent relays) with a working
1.1 s (Khurshaid, Wadood, Farkoush, Kim, Yu et al., 2019).
time of the less than 0.7 milliseconds (Hemmati & Mehrjerdi, 2019;
2. Coordination time interval (CTI): The secondary relays are set
Schneider, 2015).
to operate if the primary relay does not work at specified times
in its protective area during a fault in the network. The operating
time of the secondary relay should maintain the CTI as: 2.3. Sigmoid acceleration coefficient based chaotic search class topper op-
timization (SCCTO)
𝑇𝑏𝑘 − 𝑇𝑝𝑘 ≥ 𝐶𝑇 𝐼, (6)
2.3.1. Basic class topper optimization (Das et al., 2018)
where 𝑇𝑏𝑘 is secondary relay operational time at 𝑘 and 𝑇𝑝𝑘 is
In standard CTO (Das et al., 2018), the optimal solution of an
time of primary relay operation.
optimization problem is searched by dividing the complete search space
3. Time dial setting (TDS): TDS of overcurrent relay must be in a
with some sections. In each section, all students (searching object or
range to get a fast response, ignoring false response as follows:
particle) are verified for an objective function to check their fitness.
𝑇 𝐷𝑆 min ≤ 𝑇 𝐷𝑆 ≤ 𝑇 𝐷𝑆 max , (7) The best, fittest student in each section represents section topper (local
optimum solution). Among all section toppers (local optimum solution)
where min stand for minimum and max stand for maximum.
in the class (searching space) are compared and the best among them
In this paper, the fitness function of the relay coordination problem is chosen as a class topper (global optimum solution). The main idea
is presented in Eq. (4) considering constraints presented in Eqs. (5)–(7). of finding a class topper (global solution) is taken from the human in-
In (4), the operating time of each relay in a network is calculated by telligence of learning to improve self efficiency by a student (searching
using Eqs. (1)–(3). For simulation purpose, three types of distribution object). To update the performance of section toppers (local optimum
networks are considered. For each of distribution network, the total solutions) and others students (local solutions), the following equations
relay time is optimized maintaining the required constraints which are are suggested in CTO.
discussed in simulation section. If section is defined as 𝑆𝐸𝐶 in a class with (𝐿𝑠) number of pupils in
each sections, (𝑗) times the test is held, as per the CTO, the participant
Remark 1. Most engineering optimization based problems in reality ability 𝑃 𝑟𝑓 is upgraded as follows:
are so complex and nonlinear in nature, such as optimal relay co-
ordination problems in power system. For optimal relay coordination The section toppers in a class are updated as follows:
problem in a distributed power network, some authors have suggested
swarm based algorithm (Acharya & Das, 2021a; Sarkar & Kudkelwar,
0000). The aim of relay system in power distribution network is to
provide a safe electricity to the utilizer in an efficient way. Due to the 𝐸𝑥𝑆𝑇 (𝐿𝑠,𝑗+1) = 𝑊 ℎ𝑓𝑡1 ∗ 𝐸𝑥𝑆𝑇 (𝐿𝑠,𝑗) + 𝐶1 ∗ 𝑟𝑑1 ∗ (𝐶𝑇𝑗 − 𝑆𝑇𝑗 ), (8)
interconnection between distribution networks, the electricity supply
system is complex. Therefore, the protection system in distribution net- 𝑃 𝑟𝑓 𝑆𝑇 (𝐿𝑠,𝑗+1) = 𝑃 𝑟𝑓 𝑆𝑇 (𝐿𝑠,𝑗) + 𝐸𝑥𝑆𝑇 (𝐿𝑠,𝑗+1) , (9)
work should be safe and reliable. To prevent a blackout, the protective
relay system’s active period must be short. In the safety zone, which where 𝐸𝑥𝑆𝑇 (𝐿𝑠,𝑗+1) is the marks obtained by 𝐿𝑠th student or learner
is also secured by the alternative relay mechanism, the principal relay at (𝑗 + 1)th examination, 𝑊 ℎ𝑓𝑡1 is weighting factor, 𝐶1 is coefficient
system must operate promptly. The secondary relaying unit must wait of acceleration, 𝑟𝑑1 is random value, 𝐸𝑥𝑆𝑇 (𝐿𝑠,𝑗) is the marks of 𝐿𝑠th
for a predetermined amount of time. As a result, the relay working time student or learner at 𝑗th examination, 𝐶𝑇𝑗 is the best in the class at
is dependent on a number of variables, including the principal relay’s 𝑗th test, 𝑆𝑇𝑗 is the best at 𝑗th test in a section, 𝑃 𝑟𝑓 𝑆𝑇 (𝐿𝑠,𝑗+1) is the
operational time, the delay in terms of backup relay cooperation time, capability of 𝐿𝑠th pupils in (𝑗 + 1)th test, 𝑃 𝑟𝑓 𝑆𝑇 (𝐿𝑠,𝑗) is the capability
circuit-breaker processing times, and traveling time of signal. If the of 𝐿𝑠 pupils at 𝑗th test.

3
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

All the students in a section except topper of that section is updated as optimization approaches (Tian et al., 2019). There are different equa-
follows: tions like ten map, Gauss map, Lozi map, sinusoidal iterator, Chua’s
oscillator, Mackey Glass system, Lorenz system, Ikeda map, and others
are used, and extremely interesting results were obtained (Caponetto,
Fortuna, Fazzino, & Xibilia, 2003; dos Santos Coelho & Mariani, 2009;
𝐸𝑥𝑆 (𝐿𝑠′ ,𝑗+1) = 𝑊 ℎ𝑓𝑡2 ∗ 𝐸𝑥𝑆 (𝐿𝑠′ ,𝑗) + 𝐶2 ∗ 𝑟𝑑2 ∗ (𝑆𝑇𝑗 − 𝐿𝑠′ 𝑗 ), (10)
Singh & Sinha, 2010).
The searching efficiency of CTO can be accelerated by introducing
𝑃 𝑟𝑓 𝑆 (𝐿𝑠′ ,𝑗+1) = 𝑃 𝑟𝑓 𝑆 (𝐿𝑠′ ,𝑗) + 𝐸𝑥𝑆 (𝐿𝑠′ ,𝑗+1) , (11) chaotic variable based global optimal neighbor in each iteration. Thus,
the student’s knowledge (particles fitness) in CTO can be improved to
where 𝐸𝑥𝑆(𝐿𝑠′ ,𝑗+1) is marks of 𝐿𝑠′ th
learner in the (𝑗 + 1)th exami-
reach topper position through the complete search space. The method
nation, 𝑊 ℎ𝑓𝑡2 is weight-factor, 𝐶2 is cofactor of acceleration, 𝑟𝑑2 is
of chaotic based local search is used in each examination or iteration
random value, 𝐸𝑥𝑆 (𝐿𝑠′ ,𝑗) is marks of 𝐿𝑠′ learner in the 𝑗th examination.
of CTO to improve the exploitation ability and maintain a balance
𝑆𝑇𝑗 is the section-topper at 𝑗th examination, 𝐿𝑠′ 𝑗 is the result in the
between different searching stages. It can be defined as follows:
𝑗th examination (𝐿𝑠′ 𝑗 ≠ 𝐶𝑇𝑗 ≠ 𝑆𝑇𝑗 ), 𝑃 𝑟𝑓 𝑆 (𝐿𝑠′ ,𝑗+1) is the performance
measured for 𝐿𝑠′ learner at (𝑗 + 1)th examination, 𝑃 𝑟𝑓 𝑆 (𝐿𝑠′ ,𝑗) is the 𝜚𝑖+1 𝑗 = 𝜇 × 𝜚𝑖 𝑗 × (1 − 𝜚𝑖+1 𝑗 ), (13)
performance of 𝐿𝑠′ learner at 𝑗th examination.
The weight-factor 𝑊 ℎ𝑓𝑡𝑖 𝑖=1,2 is adapted to limit the particle (stu- where 𝜚𝑖 is in 0 to 1, 𝜚0 ∈ [0, 1] and 𝜚0 ∉ [0, 0.25, 0.5, 0.75, 1] 𝜇 is coeffi-
dent) performance in search space. It should not cross beyond the limit cient of bifurcation (set to 4), 𝑖 is the iteration count, 𝑗 is the chaotic
of searching area. Therefore, this factor is adapted with the increasing variable. The logistic mapping of chaotic variables with search variable
iteration must be reduced as defined: is carried out as follows:
( )
𝑡𝑖max − 𝑡𝑖min 𝑃𝑖𝑗 = 𝑃min,𝑗 + 𝜚𝑗 𝑖 (𝑃max,𝑗 − 𝑃min,𝑗 ), (14)
𝑊 ℎ𝑓𝑡𝑖 = 𝑊 ℎ𝑓𝑡𝑖 max − × 𝐼𝑡𝑟, (12)
𝐼𝑡𝑟max
The variable for chaotic is generated by (13) and using (14) is used to
where 𝐼𝑡𝑟 is running examination (iteration) and 𝐼𝑡𝑟max is the maximum map the chaotic variables 𝜚 into vector of chaotic 𝑃𝑗 . The vector 𝑃𝑗 is
number of examination or iteration. combined with target student performance to be best in class.

Issues in CTO 2.3.4. Sigmoid acceleration coefficients

The issues in basic CTO are as follows: Generally, an evolutionary algorithm must start with exploration
• The initial performances of students (searching objects) are ran- and increase iteration, it should exploit the optimum solution. Thus,
domly generated within the range. The initialization, in an opti- a balance between exploration and exploitation must be established
mization algorithm is important to get the best solution from the by using a proper acceleration coefficient. In this work, a sigmoid
algorithm. coefficient is considered to balance the local and global search (Tian
• The acceleration coefficients in CTO are chosen constant. Im- et al., 2019). Sigmoid function is used for its ability to change state
proper selection of such parameters may divert the searching from linear to nonlinear. It is defined as follows:
( )2
process towards local solution. 1 ( ) 𝑖𝑡𝑒𝑟
𝐶1 = ( ( )) + 2 𝐶1𝑓 − 𝐶1𝑖 − 1 , (15)
• It suffers with local optimum convergence problems due to lack of 1 + exp −𝜆 𝑖𝑡𝑒𝑟∕𝑖𝑡𝑒𝑟max 𝑖𝑡𝑒𝑟max
diversity in global search solution in each iteration (examination). ( )2
1 ( ) 𝑖𝑡𝑒𝑟
𝐶2 = ( ( )) + 𝐶2𝑓 − 𝐶2𝑖 , (16)
1 + exp −𝜆 𝑖𝑡𝑒𝑟∕𝑖𝑡𝑒𝑟max 𝑖𝑡𝑒𝑟max
2.3.2. Proposed SCCTO
In this paper, the classical class topper optimization algorithm is where 𝜆 is adjustable parameter to change the sigmoid function value.
modified to balance between core exploration and exploitation by 𝐶1 and 𝐶2 are positive value, initial value 𝐶1𝑖 and final value 𝐶1𝑓 , 𝑖𝑡𝑒𝑟
applying sigmoid based acceleration coefficient and the facility of local is present iteration and 𝑖𝑡𝑒𝑟max is maximum iteration. This 𝐶1 and 𝐶2
search by chaotic method. will be used in (8) and (10).

2.3.3. Chaotic based local search Algorithm: Pseudo-code for proposed SCCTO
Most of swarm algorithm starts with randomly generated initial
particle to check the initial fitness of particles by using an objective 1. Input: Iteration (𝑗max ), section number (𝑠𝑒𝑐 max ) in the search
function for an optimization problem. The initialization of particle in space, population (student) in each sections (𝐿𝑠max ), course num-
any swarm optimization algorithm is a crucial stage as it affects the ber for each student 𝑞 and define other constants.
convergence behavior of the algorithm. Therefore, to achieve faster 2. ∕ ∗ Chaos based initialization ∗ ∕
convergence and quality of the global solution, the initial particles 3. Generate variable of initial chaos for (𝐿𝑠max ) students in each
must be assigned correctly. The use of chaotic initialization instead of section ((𝑠𝑒𝑐 max )) using (13) and map chaos variable with search
random initialization is found in different literature (Chen, Zhou, & Liu, variable for the specified range by using (14).
2018; Dong et al., 2012; Tian, 2017). Chaos theory based optimization 4. ∕ ∗Start examination ∗ ∕
algorithms are search approaches that are distinct from any existing 5. 𝑓 𝑜𝑟 𝑗 ≤ 𝑗max
standard probabilistic optimization methods (Tian et al., 2019). For 6. ∕ ∗For each sections of a class Evaluate the
a small change in initial values have a vast change in system output fitness function∗ ∕
under chaos which is random, unpredictable in nature (Chen et al., 7. 𝑓 𝑜𝑟 𝑠𝑒𝑐 = 1 … 𝑠𝑒𝑐 max and 𝐿𝑠 = 1 … 𝐿𝑠max
2018). It can determine the nonlinear stochastic system’s unpredictable 8. Evaluate the initial fitness function or performance for each
condition (Tian et al., 2019). In CTO, using chaotic sequences rather students in each sections using (4).
than random sequences is a powerful method for diversifying the 9. 𝑖𝑓 (𝐿𝑠 > 1)
particle population and improving the CTO’s performance in preventing 10. ∕ ∗ For whole class use ranking system∗ ∕
premature convergence to local minima. Thus, the notions of opti- 11. Select (𝑆𝑇 ) of each sections with highest fitness in a section
mization based on chaotic sequences can be a useful alternative in and select 𝐶𝑇 with best 𝑆𝑇 .
CTO design to create diversity in CTO approach populations. In the 12. ∕ ∗In class∗ ∕
literature, various forms of equations have been proposed for use in 13. Update 𝑆𝑇 using (8) and (9).

4
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 2 applied clusters in multi-dimensional space and performed better than

Parameters of proposed SCCTO algorithm.
PSO, GSA, ACO etc. For this reason, CTO is chosen in the current
Parameter Value work. Like other methods such as PSO, ACO, GSA, the CTO algorithm
Section number 3 also starts searching optimal solution with randomly generated initial
Students or particle in each sections 500
particles (student initial performance). To enhance the performance
Iteration number 500
𝐶1 2–0.5 of CTO algorithm for complex power system application (such as
𝐶2 0.5–2 optimal relay coordination problem), the chaotic search initialization
𝑤 0.9–0.1 is applied in current work. Moreover, to accelerate the convergence
rate, instead of using constant acceleration coefficient used in the CTO,
an exponentially decaying sigmoid function is used which helps fast
14. ∕ ∗In student level∗ ∕ convergence. The reason of avoiding constant acceleration coefficient
15. Update other students using (10) and (11). is that improper selection of such value can lead local solution stacking
16. 𝑒𝑛𝑑 𝑖𝑓 problem. If the acceleration coefficient related with local solution gives
17. 𝑒𝑛𝑑 𝑓 𝑜𝑟 more priority than exploiting towards global solutions get diverted
18. ∕ ∗Identify class topper∗ ∕ into local solution. If the acceleration coefficient related with global
19. Assign best 𝑆𝑇 as 𝐶𝑇 in class level. solution is more than exploring best solution will be affected which also
results of stickiness in local solution. Thus, the chaotic initialization
20. 𝑒𝑛𝑑 𝑓 𝑜𝑟
21. Output: Class Topper or global optimum solution and sigmoid acceleration coefficient is combined with classical CTO to
form SCCTO. The unpredictability of chaotic sequences, as well as their
stochastic qualities and scatter features, justifies their use as a suitable
adjunct to CTO.
2.3.5. SCCTO for relay coordination problem
The proposed SCCTO is used to find optimum value of TDS such that 2.4.1. Benchmark function-parametric and non-parametric statistical test
total operational timing of relay become minimum satisfying necessary of SCCTO
constraints. The steps of SCCTO for rely coordinating problem are as The convergence behavior of the proposed SCCTO algorithm in
follows: terms of exploration and exploitation is tested with 13 types of bench-
mark test functions (Wang, Guo, Gandomi, Hao, & Wang, 2014). The
1. Initialize iteration numbers (𝑖max ), local search space number in
details of benchmark test functions are found in Acharya and Das
total search space (sections) (secmax ), size of populations (𝑆max )
(2021b) and Wang et al. (2014). The test functions are presented in Ta-
in each local search space or section, dimension or course (𝐷),
ble 3. The termination condition is a predetermined period of iterations.
assign the range of TDS and other constraints according to
The algorithm is repeated multiple times for every test function. The
distribution protection system scheme.
mean and standard deviation (SD) of fitness value for each reference
2. Generate chaos variables for (𝑆max ) populations of TDS in each
functions are calculated after several rounds. The SCCTO algorithm
section ((𝑠𝑒𝑐 max )) using (13) and map chaos variable with search
is used to examine the convergence test for each function. The fittest
variable (TDS) for the specified range by using (14). For a dis-
scores of every loop for several chosen variables are evaluated and visu-
tribution network, if there is the 𝑁 number of TDS, the chaotic
ally depicted to demonstrate the suggested method’s convergence rate.
variable will also be 𝑁 number.
SCCTO findings are checked with GWO (Mirjalili, Mirjalili, & Lewis,
3. Start the test.
2014), PSO (Kennedy & Eberhart, 1995), GSA (Rashedi, Nezamabadi-
4. Using (4) obtain fitness or total operating time for the relay
Pour, & Saryazdi, 2009), DE (Storn & Price, 1997) and C–CTO (Acharya
system for (𝑆max ) size for each section.
& Das, 2021b) data for every criterion. In the SCCTO algorithm, the
5. Determine the regional top or optimal (section toppers) for each
degree of students (seeking objects) at every section (local search
local region with the fittest obtained by TDSs (pupils) at section
area) is assumed 500 for each standard function, and the maximum
stage (local search space). Obtain the finest global remedy (class
examination or iteration is 500, while all other parameters are kept as
topper) from all the regional best options.
they are in the SCCTO method. The search space is assumed to be 30
6. According to SCCTO, the native optimum TDSs (ST) in each
(𝑛 = 30) in size. For every function, the SCCTO algorithm is tested 30
native area (sections) are upgraded based on the outcome of
times for 500 iterations and repeated 30 times for each function. For
optimum TDS (CT) using (8). Other TDSs in a specific native
analysis needs, the mean and standard deviation (SD) of best fitness
area (section level) are upgraded in native lookup space using
are calculated for each function. Tables 4 to 5 show the results of
(10)
test functions. By observing the Tables 4 and 5, the performance of
7. Steps 4 to 8 should be run as many times as necessary to pick
the SCCTO algorithm for functions F2, F3, F4, F5, F10, and F11 are
the ideal TDSs for the distribution network.
found better than that of GSA, GWO, PSO. For functions F1 and F6,
SCCTO performed better than GWO and PSO. For function F7, SCCTO
2.4. Advantage in SCCTO performed better than all of GWO, PSO, GSA, DE and C–CTO. For the
functions F8 and F9, SCCTO is better than GWO, PSO, GSA and DE. For
• Chaotic based initial particle generation at the initial stage to the functions F12 and F13, SCCTO is better than GWO, PSO, GSA and
calculate initial fitness of an optimization problem will keep C–CTO. To show the convergence performance of SCCTO algorithm,
diversity in searching method. some randomly selected functions such as F1, F6, F9, F11 and F13 are
• Exponentially decaying sigmoid acceleration coefficient ensures illustrated graphically as shown in Figs. 2–6 respectively.
fast convergence. A t-test analysis is carried out for SCCTO with the results obtained
from 13 numbers of benchmark function as presented in Table 3. The
Remark 2. The class topper optimization (CTO) (Das et al., 2018) is test criterion is same as stated above. The results of 500 iterations with
clustering based algorithm where it divides the searching space into 500 particles each in 30 sections are analyzed for T-test hypothetical
small sections and start exploring better on each section. Making group mean value zero and significant level of 𝛼 = 0.05. The results (Mean,
among all section bests (locally optimal), it starts exploiting the class Standard Deviation and Two-tailed P value) for the T-test is presented
topper (global solution) in the group (Das et al., 2018). Thus, CTO in Table 6. The performance of SCCTO for 12 numbers of benchmark

5
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 3
Benchmark functions.
Function D R 𝐹min
∑
𝑛
𝐹1 (𝑥) = 𝑥2𝑖 30 [−100, 100] 0
𝑖=1
∑ 𝑛
∏ 𝑛
𝐹2 (𝑥) = ||𝑥𝑖 || + ||𝑥𝑖 || 30 [−10, 10] 0
𝑖=1
( )2
𝑖=1

∑𝑛
∑𝑖
𝐹3 (𝑥) = 𝑥𝑗 30 [−100, 100] 0
𝑖=1 𝑗−1
{ }
𝐹4 (𝑥) = max𝑖 ||𝑥𝑖 || , 1 ≤ 𝑖 ≤ 𝑛 30 [−100, 100] 0
∑
𝑛−1 ( )2 ( )2
𝐹5 (𝑥) = [𝑋 + 𝑌 ] where 𝑋 = 100 𝑥𝑖+1 − 𝑥2𝑖 and 𝑌 = 𝑥𝑖 − 1 30 [−30, 30] 0
𝑖=1
𝑛−1 ([ ])2
∑
𝐹6 (𝑥) = 𝑥𝑖 + 0.5 30 [−100, 100] 0
𝑖=1
∑𝑛
𝐹7 (𝑥) = 𝑖𝑥4𝑖
+ 𝑟𝑎𝑛𝑑𝑜𝑚 [0, 1) 30 [−1.28, 1.28] 0
𝑖=1 (√ )
∑𝑛
| |
𝐹8 (𝑥) = −𝑥𝑖 sin |𝑥𝑖 | 30 [−500, 500] −418.98*n
𝑖=1
𝑛 [ ]
∑ 2
𝐹9 (𝑥) = 𝑥𝑖 − 10 cos(2𝜋𝑥𝑖 ) + 10 30 [−5.12, 5.12] 0
𝑖=1 ( √ ) [ √ ]
∑𝑛
∑
𝑛 ( )
𝐹10 (𝑥) = −20 exp −0.2 1𝑛 𝑥2𝑖 − exp 1𝑛 cos 2𝜋𝑥𝑖 + 20 + 𝑒 30 [−32, 32] 0
𝑖=1 𝑖=1
∑
𝑛
∏
𝑛 ( )
1 𝑥
𝐹11 (𝑥) = 4000
𝑥2𝑖 − cos √𝑖 +1 30 [−600, 600] 0
𝑖=1 𝑖=1 𝑖
( ) 𝜋
𝑛−1 (
∑ )2 { ( )} 𝜋 ( )2 ∑
𝑛 ( )
𝐹12 (𝑥) = 10 sin 𝜋𝑦1 𝑛
+ 𝑦𝑖 − 1 1 + 10sin2 𝜋𝑦𝑖+1 𝑛
+ 𝑦𝑛 − 1 𝜋𝑛 + 𝑢 𝑥𝑖 , 10, 100, 4
𝑖=1 𝑖=1
⎧𝑘(𝑥 − 𝑎)𝑚 𝑥 > 𝑎
(𝑥𝑖 +1 ) ⎪ 𝑖 𝑖
where 𝑦𝑖 = 1 + 𝑢 𝑥𝑖 , 𝑎, 𝑘, 𝑚 = ⎨ 0
4
𝑥𝑖 > 𝑎 30 [−50, 50] 0
⎪
⎩
∑𝑛
∑
𝑛 ( )
𝐹13 (𝑥) = 0.1[sin2 (3𝜋𝑥1 ) + (𝑥𝑖 − 1)2 {1 + sin2 (3𝜋𝑥1 + 1)} + (𝑥𝑛 − 1)2 {1 + sin2 (2𝜋𝑥𝑛 )}] + 𝑢 𝑥𝑖 , 5, 100, 4 30 [−50, 50] 0
𝑖=1 𝑖=1

Table 4
Comparison of results obtained of benchmark functions for SCCTO algorithm.
GWO (Mirjalili et al., 2014) PSO (Kennedy & Eberhart, 1995) GSA (Rashedi et al., 2009)
Function
Avg SD Avg SD Avg SD
F1 6.590E−28 6.340E−05 0.0001360 0.0002020 2.530E−16 9.670E−17
F2 7.180E−17 0.0290 0.042140 0.0454210 0.05560 0.194070
F3 3.290E−17 79.14950 70.12560 22.11920 896.5350 318.9560
F4 5.610E−07 1.315080 1.086480 0.317030 7.35480 1.74140
F5 26.813 69.90490 96.71830 60.11560 67.5431 62.225340
F6 0.8165 0.000120 0.000100 8.28E−050 2.50E−16 1.740E−16
F7 0.0022 0.100280 0.122850 0.044950 0.089440 0.043390
F8 −6123.10 −4087.440 −4841.290 1152.8410 −2821.070 493.03750
F9 0.31052 47.356120 46.704230 11.629380 25.96840 7.47010
F10 1.061E−13 0.0778350 0.2760150 0.509010 0.06210 0.23630
F11 0.004480 0.0066590 0.0092150 0.007720 27.70150 5.040340
F12 0.053440 0.0207340 0.0069170 0.0263010 1.79960 0.95110
F13 0.654460 0.0044740 0.0066750 0.008910 8.899080 7.12620

Table 5
Comparison of results obtained of benchmark functions for SCCTO algorithm.
DE (Storn & Price, 1997) C-CTO (Acharya & Das, 2021b) SCCTO
Function
Avg SD Avg SD Avg SD
F1 8.20E−14 5.90E−14 7.650E−169 0 1.40068E−06 1.3021E−05
F2 1.50E−09 9.90E−10 1.610E−94 1.0E−93 4.23424E−05 0.00016311
F3 6.80E−11 7.40E−11 2.850E−171 0 1.56467E−06 1.63801E−05
F4 0 0 1.870E−73 9.840E−73 0.000314717 0.002704793
F5 0 0 0 0 0.77111 0.188744598
F6 0 0 0 0 3.37243E−06 3.34788E−05
F7 0.0046 0.00120 0.000130 0.000264 5.79445E−05 0.000139726
F8 −11080.10 574.70 12569.50 0 −418.983 6.25904E−13
F9 69.20 38.80 0 0 4.32E−09 3.78548E−08
F10 9.70E−08 4.20E−08 0 0 1.68439312 6.97653E−05
F11 0 0 0 0 8.14792E−07 1.523E−05
F12 7.90E−15 80E−15 −0.643550 0.109 −0.916817966 7.60263E−07
F13 5.10E−14 4.80E−14 1.350E−32 0.00014 00.10000069 9.96537E−06

functions are found significant except function F11. For some bench- Woźniak, 2021). The comparative results in terms of Optimal solutions
mark functions from Table 3 such as F1 (Sphere), F3 (Hyper-Ellipsoids), (OS) and Standard deviation (SD) are presented in Table 7. From
F5 (Rosenbrock), F6 (Weierstrass), F8 (Schwefel), F9 (Rastrigin) and Table 7, it is clear that SCCTO is performing better to find optimal
F11 (Griewank), SCCTO is compared with Polar Bear Optimization solutions when compared with PBO and RFO for test functions F1
(PBO) (Połap et al., 2017) and Red fox optimization (RFO) (Połap & (Sphere), F3 (Hyper-Ellipsoids), F6 (Weierstrass), F8 (Schwefel), F9

6
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Fig. 5. Convergence graph of benchmark function F11 with SCCTO.

Fig. 2. Convergence graph of benchmark function F1 with SCCTO.

Fig. 6. Convergence graph of benchmark function F13 with SCCTO.

Fig. 3. Convergence graph of benchmark function F4 with SCCTO.
Table 6
T-test result of SCCTO algorithm for benchmark functions (significant level of 𝛼 = 0.05).
SCCTO
Function
𝑝-value Mean SD
F1 0.0165 1.40068E−06 1.3021E−05
F2 0.0001 4.23424E−05 0.00016311
F3 0.0001 1.56467E−06 1.63801E−05
F4 0.0095 0.000314717 0.002704793
F5 0.0010 0.77111 0.188744598
F6 0.0247 3.37243E−06 3.34788E−05
F7 0.0001 5.79445E−05 0.000139726
F8 0.0001 −418.983 6.25904E−13
F9 0.0110 4.32E−09 3.78548E−08
F10 0.0001 1.68439312 6.97653E−05
F11 0.2320 8.14792E−07 1.523E−05
F12 0.0001 −0.916817966 7.60263E−07
F13 0.0001 00.10000069 9.96537E−06

(Rastrigin) and F11 (Griewank). For F5 (Rosenbrock), both of PBO and

RFO are performing better than SCCTO to find the optimal solution. In
terms of small deviation for all chosen functions, SCCTO is better than
others. The implementation of non-parametric testing for statistical
Fig. 4. Convergence graph of benchmark function F9 with SCCTO.
features are also demonstrated for the SCCTO method. Numerous uses

7
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 7
Comparison of SCCTO algorithm with existing results for benchmark functions.
PBO (Połap et al., 2017) RFO (Połap & Woźniak, 2021) SCCTO
Function
OS SD OS SD OS SD
F1 (Sphere) 0.000064 0.001936 0.01299 0.00322 8.73775E−37 1.3021E−05
F3 (Hyper-Ellipsoids) 0.000039 0.003240 0.01507 0.00298 3.02094E−52 1.63801E−05
F5 (Rosenbrock) 0.027936 0.311527 0.00371 0.00287 0.7711 0.188745
F6 (Weierstrass) −0.50901 0.322731 0.00891 0.36547 0 3.34788E−05
F8 (Schwefel) 0.04352 0.295528 0.009102 0.00116 −418.983 6.25904E−13
F9 (Rastrigin) 0.004483 0.240397 0.01374 0.00372 0 3.78548E−08
F11 (Griewank) 0.000052 0.000382 0.011 0.00351 0 1.523E−05

Table 8
Friedman test statistical data for benchmark function from twelve simulations.
GWO (Mirjalili et al., 2014) PSO (Kennedy & Eberhart, 1995) GSA (Rashedi et al., 2009) SCCTO
Function
SD Rank SD Rank SD Rank SD Rank
F1 6.34E−05 3 0.000202 4 9.67E−17 1 1.3021E−05 2
F2 0.02901 2 0.045421 3 0.19407 4 0.00016 1
F3 79.1495 3 22.1192 2 318.956 4 1.638E−05 1
F4 1.31508 3 0.31703 2 1.7414 4 0.0027 1
F5 69.9049 4 60.1156 2 62.22534 3 0.1887 1
F6 0.00012 4 8.28E−05 3 1.74E−16 1 3.3478E−05 2
F7 0.10028 4 0.04495 3 0.04339 2 0.00014 1
F8 −4087.44 1 1152.841 4 493.0375 3 6.259E−13 2
F9 47.35612 4 11.62938 3 7.4701 2 3.7855E−08 1
F10 0.077835 2 0.50901 4 0.2363 3 6.9765E−05 1
F11 0.006659 2 0.00772 3 5.04034 4 1.523E−05 1
F12 0.020734 2 0.026301 3 0.9511 4 7.6026E−07 1
Total rank 34 36 35 15
Average rank 2.8 3 2.9 1.3
Statistic 13.968
𝑝-value 0.0017

Table 9 As, the power distribution network is a complex network, the relay
Wilcoxon signed ranks test for benchmark function at significant level 𝛼 = 0.05 with protective scheme is also complex. Therefore, to establish optimal
2-tailed hypothesis.
coordination among relays, the proposed method will be efficient.
Comparison 𝑅+ 𝑅− 𝑝-value
SCCTO is also better than some other algorithm such as Monarch
SCCTO versus GWO (Mirjalili et al., 2014) 12 66 0.034 butterfly optimization (MBO) (Wang, Deb, & Cui, 2019) where the
SCCTO versus PSO (Kennedy & Eberhart, 1995) 0 78 0.00222
SCCTO versus GSA (Rashedi et al., 2009) 3 75 0.0048
old fitness (parent) is replaced by new fitness (child) to keep the
SCCTO versus DE (Storn & Price, 1997) 47 31 0.5287 population unchanged. In SCCTO, all searching objects are allowed to
update according to their fitness. All local optimal solutions in each
section are updated according to the class topper or global solution
at a particular examination or iteration. Therefore, The old fitness
of such tests can be discovered in Wang et al. (2014) and Zeng, Xie, gets a chance to improve itself which may be very near to global
Chen, and Weng (2019). The Friedman test as well as the Wilcoxon solution. Most of optimization methods such as an earthworm optimiza-
signed rank test are explored in this research for benchmark functions, tion algorithm (EWA) (Wang, Deb, & Coelho, 2018), elephant herding
as seen in Table 3. One can look up the protocols for such non- optimization (EHO) (Wang, Deb, & Coelho, 2015), moth search (MS)
parametric tests in detail in Derrac, García, Molina, and Herrera (2011) algorithm (Wang, 2018), exploration is done in the whole search region
and Wang et al. (2014). The Friedman statistical test result for SCCTO globally and exploitation of the optimal solution is done locally. In
is presented in Table 8 where the p value and statistic indicates the SCCTO, the exploration is carried out in each sections rather than
significance of SCCTO as compared with other methods. 𝑅+ is the searching globally.
total of all rankings where the first algorithm performs better than the
second, and 𝑅− denotes the opposite situation at a significant level 3. Simulation and result analysis
of 0.05 in the Wilcoxon signed test. The Wilcoxon signed test result
for SCCTO is presented in Table 9 where the p value indicates the The proposed SCCTO algorithm is applied to the problem of optimal
significance of SCCTO as compared with PSO, GSA and GWO methods coordination of relays in power system. For this purpose, the single
except DE. The Friedman statistical test for SCCTO is also examined and multi mesh distribution systems are considered. LabVIEW©2015
with PBO (Połap et al., 2017) and RFO (Połap & Woźniak, 2021) for platform is used for simulation purpose of optimization problem. The
7 numbers of benchmark functions such as F1 (Sphere), F3 (Hyper- main purpose is to search optimal TDS gain of relays for proper
Ellipsoids), F6 (Weierstrass), F8 (Schwefel), F9 (Rastrigin) and F11 coordination in a distribution network. The performance of the SCCTO
(Griewank) as presented in Table 10. The 𝑝-value in Table 10 indicates is checked for three (3) types of distribution networks. The associated
the significance of SCCTO compared with PBO (Mirjalili et al., 2014) system parameters of the networks are chosen from Kudkelwar and
and RFO (Połap & Woźniak, 2021). Sarkar (2019).

Remark 3. The proposed SCCTO algorithm uses nonlinear chaotic ini- 3.1. Case study 1: Relay coordination in single end multi parallel feeder
tial search and exponentially decaying acceleration coefficient, which distribution network
ensure fast convergence by making balance exploration and exploita-
tion stage. Moreover, the sub-division of complete search space will The coordinating of relay in single-end multiparallel feeder distribu-
allow diversity in the exploration process of the proposed method. tion network is chosen in which six over current relays are applied for

8
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 10
Friedman statistical test data for benchmark functions.
PBO (Połap et al., 2017) RFO (Połap & Woźniak, 2021) SCCTO
Function
SD Rank SD Rank SD Rank
F1 (Sphere) 0.001936 2 0.00322 3 1.3021E−05 1
F3 (Hyper-Ellipsoids) 0.003240 3 0.00298 2 1.63801E−05 1
F5 (Rosenbrock) 0.311527 3 0.00287 1 0.188745 2
F6 (Weierstrass) 0.322731 2 0.36547 3 3.34788E−05 1
F8 (Schwefel) 0.295528 3 0.00116 2 6.25904E−13 1
F9 (Rastrigin) 0.240397 3 0.00372 2 3.78548E−08 1
F11 (Griewank) 0.000382 2 0.00351 3 1.523E−05 1
Total rank 18 16 8
Average rank 2.6 2.3 1.1
Statistic 8.82
𝑝-value 0.0183

Fig. 7. Single end multi parallel feeder distribution system.

protection purpose as shown in Fig. 7. The admittance of distribution Table 11

lines is neglected due to the short distance. The line data for the CT ratios and plug setting of relays for case
study-1 (Gokhale & Kale, 2016).
distribution network is presented in Table 12. The rating of current
Relay CT ratio Plug setting
transformer and plug setting of relays are mentioned in Table 11.
The relationship among primary and backup relays for fault at R2 300.0 ∶ 1 1
R4 600.0 ∶ 1 1
separate points is given in Table 13. The various fault locations as
R3 600.0 ∶ 1 1
assumed in the system, 𝑎𝑖 constants, and current setting of relays R6 600.0 ∶ 1 1
are placed in Table 14. The performance index for this optimization R1 1000.0 ∶ 1 1
problem is chosen as follows: R5 1000.0 ∶ 1 1

min 𝑧 = 102.40𝑥1 + 6.060𝑥2 + 98.760𝑥3 + 24.4𝑥4 (17)

Table 12
+ 35.32𝑥5 + 11.54𝑥6
Line data for case study-1 (Gokhale
Subject to the constraints as follows: & Kale, 2016).
Line Impedance
• 3.6460𝑥1 ≥ 0.10, 1–2 0.080 + 𝑗1.0
• 6.0550𝑥2 ≥ 0.10, 1–3 0.160 + 𝑗2.0
• 8.8440𝑥3 ≥ 0.10, 2–3 0.080 + 𝑗1.0
• 8.8440𝑥4 ≥ 0.10,
• 4.0440𝑥5 ≥ 0.10, Table 13
• 11.5390𝑥6 ≥ 0.10. Main and secondary relay combination for different
fault point for case study-1 (Kudkelwar & Sarkar,
The range of TDSs are taken between 0.025 to 1.2. Therefore, the 2019).
corrections in the constraints of 𝑥2 , 𝑥3 , 𝑥4 and 𝑥5 are considered as Fault point Main relay Back-up relay
greater than or equal to 0.025. The relations of primary and backup 𝐾 5, 6 3
relay in terms of coordinating time interval are as follows: 𝐿 3, 5 1
𝑀 3, 4 1, 5
• 15.550𝑥4 − 6.0650𝑥2 ≥ 0.30, 𝑁 1, 2 4
• 8.8440𝑥1 − 8.8440𝑥3 ≥ 0.30,
• 4.8440𝑥5 − 8.8440𝑥4 ≥ 0.30,
• 75.910𝑥3 − 11.530𝑥6 ≥ 0.30,
with SCCTO are presented in Table 2. The optimum TDSs and total re-
• 13.9980𝑥1 − 13.9980𝑥3 ≥ 0.30.
lay operating time of the system under test is compared with some exist-
ing results such as Firefly algorithm (FA) (Gokhale & Kale, 2014), Con-
3.1.1. Result analysis for case study-1 tinuous genetic algorithm (CGA) (Moirangthem, Krishnanand, Dash,
The simulation for single-end loop system with six overcurrent & Ramaswami, 2013), Chaotic firefly algorithm (CFA) (Gokhale &
relays are done in LabVIEW©2015 platform. The parameters associated Kale, 2016), Root tree algorithm (RTA) (Wadood et al., 2018), Water

9
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 14 R2 will operate for fault at 𝑀 location. Thus,

𝑎𝑖 constants at different fault locations for case study-1 (Gokhale & Kale, 2016).
Fault location Relay 1 Relay 2 Relay 3 Relay 4 Relay 5 Relay 6 𝑎2 = 43.776, (19)
𝐾 75.91 – 75.91 – – 4.044
𝐿 13.99 – 13.99 – 6.872 –
R3 can sense fault at 𝐾, 𝐿 and 𝑁 locations. Thus,
𝑀 8.844 – 8.844 8.844 8.844 –
𝑁 3.646 6.065 – 15.55 15.55 – 𝑎3 = 6.872 + 5.809 + 4.044 = 16.725, (20)

R4 detects fault for 𝑀, 𝑁 locations. It is written as follows:

𝑎4 = 8.159 + 5.811 = 13.97, (21)

R5 can sense fault at 𝐾, 𝐿 and 𝑁 places. Thus,

𝑎5 = 4.811 + 5.809 + 14.1 = 24.63, (22)

R6 works for fault at 𝐾 location. Thus,

𝑎6 = 22.165, (23)

R7 is set for fault at 𝐿 location and can be defined as follows:

𝑎7 = 4.145, (24)

Thus, the objective function for this case study is defined as follows:

min 𝑧 = 10.24𝑥1 + 43.77𝑥2 + 16.725𝑥3 + 13.97𝑥4

Fig. 8. Convergence curve for case study 1.
+ 24.63𝑥5 + 22.16𝑥6 + 4.145𝑥7 , (25)

where 𝑥𝑖 is 𝑇 𝐷𝑆𝑖 of 𝑅𝑒𝑙𝑎𝑦𝑖 and 𝑖 = 1, 2 … , 7.

cycle algorithm (Kudkelwar & Sarkar, 2019), crow search algorithm The constrains for this problem are as below:
(CSA) (Sarkar, Kudkelwar, & Saha, 2019) and Opposition based learn-
ing fractional order CTO (OBL-FOCTO) (Choudhary & Das, 2021). The • 3.255𝑥1 ≥ 0.1,
optimum TDSs for all 6 number of relays and total operating time for • 43.77𝑥2 ≥ 0.1,
the system found with SCCTO is presented in Table 15 and graphically • 4.044𝑥3 ≥ 0.1,
shown in Fig. 8. • 5.811𝑥4 ≥ 0.1,
• 4.881𝑥5 ≥ 0.1,
3.1.2. Discussion
• 22.165𝑥6 ≥ 0.1,
The optimal coordination of relay for single-end distribution net-
• 4.14𝑥7 ≥ 0.1.
work is performed with SCCTO in LabVIEW©2015 platform. The opti-
mum value of TDSs and total operating time is tabulated in Table 15. The CTI criteria for this case study are as follows:
The total minimum relay operational time is achieved 11.87. It is
compared to FA (Gokhale & Kale, 2014), CGA (Moirangthem et al., • 43.77𝑥2 − 8.159𝑥4 ≥ 0.2,
2013), CFA (Gokhale & Kale, 2016) and RTA (Wadood et al., 2018) • 6.872𝑥1 − 4.044𝑥3 ≥ 0.2,
and found 4.38 s, 3.71 s, 2.52 s and 0.06 s improvement in total op- • 14.01𝑥5 − 5.811𝑥4 ≥ 0.2,
erating time respectively. The iteration numbers to reach convergence • 6.872𝑥3 − 22.165𝑥6 ≥ 0.2,
in RTA (Wadood et al., 2018) and CSA (Sarkar et al., 2019) are 200 • 5.809𝑥3 − 13.998𝑥7 ≥ 0.2.
and 150 respectively. From the Fig. 8, it is clear that in SCCTO takes
less than 100 iterations which is superior than RTA (Wadood et al.,
2018) and CSA (Sarkar et al., 2019). Only for OBL-FOCTO (Choudhary 3.2.1. Result analysis for case study 2
& Das, 2021), it takes 23 iterations to achieve the optimal solution. The The relay optimum coordination problem for single loop distri-
statistical results of the proposed method in terms of mean, standard bution network with 7 numbers of relays is implemented in Lab-
deviation, 𝑝-value showed the significance of the proposed method VIEW©2015 platform. For this purpose, the TDS range is taken as
for relay coordination problem. Thus, fast protection is gained by the 0.025 to 1.2. The SCCTO is also developed in the LabVIEW©2015
developed algorithm. platform for the same parameters as in case study 1 (Table 2). The
proposed optimization method is run for 500 initial populations for 500
3.2. Case study 2: Single loop distribution network iterations. The optimum TDSs for all relays and total operating time
achieved with SCCTO are tabulated in Table 19. The convergence curve
In this case study, the problem of relay co-ordination in distribution
of total relay operating time is shown in Fig. 10.
networks with single loop is chosen as shown in Fig. 9. The system
is supplied from one side and contains seven numbers of overcurrent
relay. Relay 1 and relay 5 are chosen as non-directional keeping rest of 3.2.2. Discussion
the relays as directional. The specification for current transformer and The coordination of relay system for the single loop distribution
plug setting criterion are placed in Table 16. The relationship among network is simulated with SCCTO in LabVIEW©2015. The comparative
primary and secondary relays during fault at different points is given result with existing results are given in Table 19. From the Table 19,
in Table 17. In Table 18, the 𝑎𝑖 constant are given. the minimum total operation time is achieved 5.51 s which is 10.19 s
In this case studies, the minimum TDS for all relays is chosen as
less than Simplex method (SM) (Bedekar & Bhide, 2010). Though, the
0.025 s and CTI is defined for 0.2 s. The relay 𝑅1 is assigned for fault
total operating time of relays are better in WCA (Kudkelwar & Sarkar,
at 𝑀 and 𝑁 location as shown in Fig. 9. It is expressed as follows:
2019) and CSA (Sarkar et al., 2019), the convergence performance in
𝑎1 = 3.252 + 6.872 = 10.124, (18) SCCTO is superior than WCA and CSA.

10
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 15
Optimal TDSs and total operating time for case study 1.
Approaches FA (Gokhale & CGA CFA (Gokhale & RTA (Wadood WCA (Kudkelwar CSA (Sarkar OBL-FOCTO Proposed
Kale, 2014) (Moirangthem Kale, 2016) et al., 2018) & Sarkar, 2019) et al., 2019) (Choudhary & Das, SCCTO
et al., 2013) 2021)
TDS 1 0.027 0.0765 0.027 0.059 0.589 0.0589 0.0589 0.0589
TDS 2 0.13 0.034 0.221 0.025 0.025 0.025 0.025 0.025
TDS 3 0.025 0.0339 0.025 0.025 0.025 0.025 0.025 0.025
TDS 4 0.025 0.036 0.025 0.029 0.029 0.029 0.02904 0.029
TDS 5 0.025 0.0711 0.029 0.065 0.0626 0.0626 0.06297 0.063
TDS 6 0.489 0.0294 0.363 0.025 0.025 0.025 0.025 0.025
Total operating 16.25 15.58 14.39 11.93 11.86 11.86 11.785 11.87
time in second

Fig. 9. Single loop distribution system.

Table 16
CT ratio and plug-setting of relay for case study-
2 (Kudkelwar & Sarkar, 2019).
Relay CT rating Plug setting
R1 1000.0 ∶ 1 0.8
R2 1000.0 ∶ 1 0.8
R3 500.0 ∶ 1 0.8
R4 500.0 ∶ 1 0.8
R5 1000.0 ∶ 1 0.8
R6 1000.0 ∶ 1 0.8
R7 500.0 ∶ 1 0.5

Table 17
Fault location and respective primary, backup relay and fault current for case
study-2 (Kudkelwar & Sarkar, 2019).
Fault location Primary relay Backup relay Fault current (A)
K 5, 6 –, 3 3289, 1096.5
L 7 3, 5 1315.8 Fig. 10. Convergence curve for case study 2.
M 3, 4 1, 5 2193, 1315.5
N 1, 2 –, 4 6759, 939

• 2.97𝑥1 ≥ 0.1,
• 2.97𝑥2 ≥ 0.1,
3.3. Case study 3: Multi loop single end fed distribution network • 5.584𝑥3 ≥ 0.1,
• 4.980𝑥4 ≥ 0.1,
The coordination of optimal relay in distribution networks with • 3.297𝑥5 ≥ 0.1,
multiloop, single end fed is chosen. It belongs to eight numbers of • 3.297𝑥6 ≥ 0.1,
over current relays as illustrated in Fig. 11. The line admittance is • 4.980𝑥7 ≥ 0.1,
ignored for the distribution network. The specification for CT and PS
• 10.035𝑥8 ≥ 0.1.
is placed in Table 20. The details of main and secondary relays of
various fault location is given in Table 21. In Table 22, the 𝑎𝑖 constant within the TDS range of:
are mentioned. In this case, the minimum relay operation time is
considered as 0.10 s and CTI is chosen as 0.60 s. The TDSs are assigned • 𝑥3 ≥ 0.025,
as 𝑥𝑖 where 𝑖 = 1, 2, … , 8. • 𝑥4 ≥ 0.025,
The fitness function for this case study is defined as follows: • 𝑥7 ≥ 0.025,
• 𝑥8 ≥ 0.025.
min 𝑧 = 28.975𝑥1 + 28.967𝑥2 + 37.73𝑥3 + 11.502𝑥4
+ 3.297𝑥5 + 3.297𝑥6 + 4.9804𝑥7 + 30.799𝑥8 (26) The CTI criterion are as follows:

Subject to : • 5.749𝑥3 − 5.749𝑥8 ≥ 0.6,

11
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 18
𝑎𝑖 constant at different fault locations for case study-2 (Kudkelwar & Sarkar, 2019).
Relay 𝐾 (Fault location) 𝐿 (Fault location) 𝑀 (Fault location) 𝑁 (Fault location)
R1 – – 3.252 6.872
R2 – – 43.776 –
R3 6.872 5.809 – 4.044
R4 – – 8.159 5.811
R5 4.881 −5.809 – 14.01
R6 22.165 – – –
R7 – 4.145 – –

Table 19
Simulation result for case study-2: TDS and total operating time.
Time dial setting SM (Bedekar & OLGSA (Acharya WCA (Kudkelwar CSA (Sarkar Proposed SCCTO
Bhide, 2010) & Das, 2021a) & Sarkar, 2019) et al., 2019)
TDS 1 0.238 0.026 0.059 0.059 0.0936821
TDS 2 0.12 0.025 0.025 0.025 0.025
TDS 3 0.36 0.055 0.052 0.052 0.1097
TDS 4 0.031 0.087 0.025 0.025 0.025
TDS 5 0.027 0.153 0.025 0.025 0.025
TDS 6 0.025 0.025 0.025 0.026 0.025
TDS 7 0.08 0.036 0.025 0.025 0.025
Total operating 15.70 6.05 4.2 4.23 5.51156
time in second

Fig. 11. Multi loop single end fed distribution system.

Table 20 Table 21
CT ratio and plug settings of relays (Kudkelwar & Fault location and respective primary, backup relay and fault current for case
Sarkar, 2019). study-3 (Kudkelwar & Sarkar, 2019).
Relay CT rating Plug setting Fault location Primary relay Backup relay Fault current (A)
R1 1000.0 ∶ 1 0.8 𝐾 5, 6 –, 3 3289, 1096.5
R2 1000 ∶ 1 0.8 𝐿 7 3, 5 1315.8
R3 500 ∶ 1 0.8 𝑀 3, 4 1, 5 2193, 1315.5
R4 500 ∶ 1 0.8 𝑁 1, 2 –, 4 6759, 939
R5 1000 ∶ 1 0.8
R6 1000 ∶ 1 0.8
R7 500 ∶ 1 0.5
R8 500 ∶ 1 0.5 SCCTO is also developed in LabVIEW©2015 platform and run for the
system with 500 initial populations for 500 iterations. The solution of
optimum TDSs and total operating relay time is presented in Table 23.
• 5.584𝑥1 − 3.551𝑥4 ≥ 0.6,
• 5.584𝑥2 − 3.551𝑥4 ≥ 0.6,
3.3.2. Discussion
• 4.980𝑥4 − 4.980𝑥7 ≥ 0.6,
The optimal relay coordination of the multiloop singled distribution
• 4.281𝑥1 − 2.971𝑥4 ≥ 0.6, network is examined with SCCTO in LabVIEW©2015. The outcomes
• 4.980𝑥3 − 4.980𝑥8 ≥ 0.6, of the optimal relay setting (TDS and total operating time of relays
• 10.035𝑥8 − 3.297𝑥5 ≥ 0.6, in protecting area) are mentioned in Table 23 and the convergence
• 10.035𝑥8 − 3.297𝑥6 ≥ 0.6. of total operating time is shown in Fig. 12. From the Table 23, it
is observed that the minimum operational time for the network is
achieved 23.9 s which is lesser compared to existing results. It is gained
3.3.1. Result analysis for case study-3 by 7.93 s, 2.731 s, 0.191 s, 0.566 s and 0.176 s in comparison with the
In this section, the optimum relay coordinating problem for eight genetic algorithm (GA) (Bedekar & Bhide, 2011), RTA (Wadood et al.,
relays based multiloop single fed distribution network is simulated in 2018), WCA (Kudkelwar & Sarkar, 2019), CSA (Sarkar et al., 2019) and
LabVIEW©2015 platform. The TDS range is taken as 0.025 to 1.2 s The OBL-FOCTO (Choudhary & Das, 2021) respectively.

12
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Table 22
𝑎𝑖 constant at different fault location for case study-3 (Kudkelwar & Sarkar, 2019).
Fault location Relay 1 Relay 2 Relay 3 Relay 4 Relay 5 Relay 6 Relay 7
𝐾 10.035 10.035 2.971 4.9804 – – 4.9804
𝐿 5.584 5.584 4.227 3.551 – – –
𝑀 4.281 4.281 4.9804 2.971 – – –
𝑁 2.267 3.837 10.035 – 3.297 – –
𝑂 3.837 2.267 10.035 – – 3.297 –
𝑃 2.971 2.971 5.749 – – – –

Table 23
Simulation result for case study-3: TDS and total operating time.
Time dial setting GA (Bedekar & RTA (Wadood WCA (Kudkelwar CSA (Sarkar OBL-FOCTO (Choudhary & Proposed
Bhide, 2011) et al., 2018) & Sarkar, 2019) et al., 2019) Das, 2021) SCTO
TDS 1 0.297 0.252 0.194 0.25 0.24142 0.24112
TDS 2 0.297 0.252 0.145 0.19 0.200006 0.1999
TDS 3 0.227 0.20 0.030 0.191 0.190364 0.18849
TDS 4 0.173 0.151 0.145 0.145 0.14555 0.14548
TDS 5 0.06 0.03 0.030 0.034 0.03033 0.025
TDS 6 0.06 0.03 0.030 0.034 0.030713 0.025
TDS 7 0.04 0.025 0.025 0.025 0.025021 0.025
TDS 8 0.113 0.08 0.069 0.071 0.06988 0.068
Total operating 31.883 26.681 24.141 24.516 24.126 23.95
time in seconds

4. Conclusion and future scope

A sigmoid acceleration coefficient based chaotic search class topper

optimization (SCCTO) algorithm is developed to solve optimum relay
coordination problem in power distribution network. The aim of the
optimization problem is to make a suitable coordination among pri-
mary and secondary or backup relays in protection system along with
the minimum total operating time that can provide reliable protection
in power distribution system. The optimum setting of time dial setting
(TDS) is found in a constant plug setting, maintaining coordination time
gap with the developed algorithm. The proposed SCCTO is verified for
single-end multi-parallel feeder distribution system, multi-loop single-
fed distribution and single loop distribution network. A comparative
analysis with existing results for all 3 cases have been presented. The
Fig. 12. Convergence curve for case study 3. resulting analysis show that the total minimum operational time with
optimum TDS is achieved in all cases with an improvement of 26 to
35% in total operating time. Thus, the developed SCCTO optimizer
Remark 4. In the present paper, a single objective optimizer is pre- is performed efficiently for optimal over-current relay coordination
sented and the efficiency of the proposed algorithm is validated with problem in power system. The proposed modified SCCTO algorithm
13 numbers of benchmark functions. The non-parametric test and is tested for three different simple distribution network with relay
time complexity test of the SCCTO is also carried out. To check the protection scheme.
performance of the algorithm for engineering optimization problem, A comparative studies on SCCTO with other newly developed al-
the optimal relay coordination in power distribution network is se- gorithms such as monarch butterfly optimization (MBO) (Wang et al.,
2019), earthworm optimization algorithm (EWA) (Wang et al., 2018),
lected. The aim of such optimization problem is to minimize total relay
elephant herding optimization (EHO) (Wang et al., 2015), moth search
operational time by finding an optimal time dial setting (TDS) of relays
(MS) algorithm (Wang, 2018), Slime mould algorithm (SMA) (Li, Chen,
applied in power system protection purpose. Therefore, three types
Wang, Heidari, & Mirjalili, 2020), and Harris hawks optimization
of power distribution networks well verified in recent literatures are
(HHO) (Heidari et al., 2019) can be extended as future scope.
considered in the present paper. The real power network is intercon-
One of the drawback of such heuristic methods is that most of all
nected and complex. The protection scheme in real scenario is also
such methods are off-line based method. For, optimal relay coordi-
complex. There are some complex real world power networks available
nation problem, the suitable optimal relay setting for power network
in literatures such as 8 bus IEEE 150 kV transmission system, 30 bus protection is required. Therefore, offline mode methods for optimal
IEEE 33 kV distribution system, etc. For such system, the optimal relay setting of relays are suggested by many authors. In a real scenario, as
coordination problem must be done for both plug setting multiplier a complex network, the relay setting may be required to change due to
(PSM) and TDS as the plug setting will not be constant due to the uncertainties in the system. Presently, numerical relay such as MiCOM
complexity in the system. So, multi-objective based SCCTO can be overcurrent relay are used in the power network. The extension of the
verified for such real world system as a future scope of the present present work may also be carried out to develop fuzzy based online
work. optimization technique.

13
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

CRediT authorship contribution statement Hemmati, R., & Mehrjerdi, H. (2019). Non-standard characteristic of overcurrent relay
for minimum operating time and maximum protection level. Simulation Modelling
Practice and Theory, 97, Article 101953.
Debasis Acharya: Conceptualization, Methodology, Writing – orig-
Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization, PSO. In Proc. IEEE
inal draft. Dushmanta Kumar Das: Conceptualization, Methodology, international conference on neural networks. Perth, Australia (pp. 1942–1948).
Supervision, Review and editing. Khurshaid, T., Wadood, A., Farkoush, S. G., Kim, C. -H., Cho, N., & Rhee, S. -B.
(2019). Modified particle swarm optimizer as optimization of time dial settings
Declaration of competing interest for coordination of directional overcurrent relay. Journal of Electrical Engineering &
Technology, 14(1), 55–68.
Khurshaid, T., Wadood, A., Farkoush, S. G., Kim, C. -H., Yu, J., & Rhee, S. -B. (2019).
The authors declare that they have no known competing finan- Improved firefly algorithm for the optimal coordination of directional overcurrent
cial interests or personal relationships that could have appeared to relays. IEEE Access, 7, 78503–78514.
influence the work reported in this paper. Korashy, A., Kamel, S., Jurado, F., & Youssef, A. -R. (2019). Hybrid whale optimization
algorithm and grey wolf optimizer algorithm for optimal coordination of direction
overcurrent relays. Electric Power Components and Systems, 47(6–7), 644–658.
Acknowledgments
Korashy, A., Kamel, S., Nasrat, L., & Jurado, F. (2020). Developed multi-objective
grey wolf optimizer with fuzzy logic decision-making tool for direction overcurrent
Funding relays coordination. Soft Computing, 1–13.
Kudkelwar, S., & Sarkar, D. (2019). Online implementation of time augmentation of
This research did not receive any specific grant from funding agen- over current relay coordination using water cycle algorithm. SN Applied Sciences,
1(12), 1628.
cies in the public, commercial, or not-for-profit sectors. Li, S., Chen, H., Wang, M., Heidari, A. A., & Mirjalili, S. (2020). Slime mould algorithm:
A new method for stochastic optimization. Future Generation Computer Systems, 111,
Statements of ethical approval 300–323.
Mirjalili, S., Mirjalili, S. M., & Lewis, A. (2014). Grey wolf optimizer. Advances in
Engineering Software, 69, 46–61.
For this research article, the authors did not undertake work that
Mohanta, T. K., & Das, D. K. (2021). Class topper optimization based improved
involved human participants or animals. localization algorithm in wireless sensor network. Wireless Personal Communications,
1–20.
References Moirangthem, J., Krishnanand, K., Dash, S. S., & Ramaswami, R. (2013). Adaptive
differential evolution algorithm for solving non-linear coordination problem of
Acharya, D., & Das, D. K. (2021a). Optimal coordination of over current relay directional overcurrent relays. IET Generation, Transmission & Distribution, 7(4),
using opposition learning-based gravitational search algorithm. The Journal of 329–336.
Supercomputing, 1–21. Połap, D., & Woźniak, M. (2021). Red fox optimization algorithm. Expert Systems with
Acharya, D., & Das, D. K. (2021b). Swarm optimization approach to design PID Applications, 166, Article 114107.
controller for artificially ventilated human respiratory system. Computer Methods Połap, D., et al. (2017). Polar bear optimization algorithm: Meta-heuristic with fast
and Programs in Biomedicine, 198, Article 105776. population movement and dynamic birth and death mechanism. Symmetry, 9(10),
Alkaran, D. S., Vatani, M., Sanjari, M., Gharehpetian, G. B., & Naderi, M. S. (2016). 203.
Optimal overcurrent relay coordination in interconnected networks by using Pradhan, S. K., Acharya, D., & Das, D. K. (2022). Internal model control based
fuzzy-based GA method. IEEE Transactions on Smart Grid, 9(4), 3091–3101. proportional-integral controller with class topper optimization for power control
Balakrishnan, N., & Nisi, K. (2020). A deep analysis on optimization techniques for of molten salt breeder reactor core. Annals of Nuclear Energy, 165, Article 108675.
appropriate PID tuning to incline efficient artificial pancreas. Neural Computing and Rajput, V. N., & Pandya, K. S. (2017). Coordination of directional overcurrent relays
Applications, 32(12), 7587–7596. in the interconnected power systems using effective tuning of harmony search
Bedekar, P. P., & Bhide, S. R. (2010). Optimum coordination of directional overcurrent algorithm. Sustainable Computing: Informatics and Systems, 15, 1–15.
relays using the hybrid GA-NLP approach. IEEE Transactions on Power Delivery, Rashedi, E., Nezamabadi-Pour, H., & Saryazdi, S. (2009). GSA: A gravitational search
26(1), 109–119. algorithm. Information Sciences, 179(13), 2232–2248.
Bedekar, P. P., & Bhide, S. R. (2011). Optimum coordination of overcurrent relay Saha, D., Datta, A., & Das, P. (2016). Optimal coordination of directional overcurrent
timing using continuous genetic algorithm. Expert Systems with Applications, 38(9), relays in power systems using symbiotic organism search optimisation technique.
11286–11292. IET Generation, Transmission & Distribution, 10(11), 2681–2688.
Caponetto, R., Fortuna, L., Fazzino, S., & Xibilia, M. G. (2003). Chaotic sequences dos Santos Coelho, L., & Mariani, V. C. (2009). A novel chaotic particle swarm
to improve the performance of evolutionary algorithms. IEEE Transactions on optimization approach using Hénon map and implicit filtering local search for
Evolutionary Computation, 7(3), 289–304. economic load dispatch. Chaos, Solitons & Fractals, 39(2), 510–518.
Chen, K., Zhou, F., & Liu, A. (2018). Chaotic dynamic weight particle swarm Sarkar, D., & Kudkelwar, S. (0000). An over current relay coordination: A comparative
optimization for numerical function optimization. Knowledge-Based Systems, 139, analysis of metaheuristic and linear program approach. International Transactions
23–40. on Electrical Energy Systems, Article e13242.
Choudhary, P. K., & Das, D. K. (2021). Optimal coordination of over-current relay in a Sarkar, D., Kudkelwar, S., & Saha, D. (2019). Optimal coordination of overcurrent relay
power distribution network using opposition based learning fractional order class using crow search algorithm. Smart Science, 7(4), 282–297.
topper optimization (OBL-FOCTO) algorithm. Applied Soft Computing, 113, Article Sarwagya, K., Nayak, P. K., & Ranjan, S. (2020). Optimal coordination of directional
107916. overcurrent relays in complex distribution networks using sine cosine algorithm.
Das, P., Das, D. K., & Dey, S. (2018). A new class topper optimization algorithm with Electric Power Systems Research, 187, Article 106435.
an application to data clustering. IEEE Transactions on Emerging Topics in Computing, Schneider, E. (2015). Micom overcurrent relay-technical guide. https://www.se.com/
8(4), 948–959. in/en/product-range-download/60758-micom-p12x/#/documents-tab.
Derrac, J., García, S., Molina, D., & Herrera, F. (2011). A practical tutorial on the use Shih, M. Y., Conde, A., & Ángeles-Camacho, C. (2019). Enhanced self-adaptive
of nonparametric statistical tests as a methodology for comparing evolutionary and differential evolution multi-objective algorithm for coordination of directional over-
swarm intelligence algorithms. Swarm and Evolutionary Computation, 1(1), 3–18. current relays contemplating maximum and minimum fault points. IET Generation,
Dong, N., Wu, C. -H., Ip, W. -H., Chen, Z. -Q., Chan, C. -Y., & Yung, K. -L. (2012). Transmission & Distribution, 13(21), 4842–4852.
An opposition-based chaotic GA/PSO hybrid algorithm and its application in circle Shih, M. Y., Salazar, C. A. C., & Enríquez, A. C. (2015). Adaptive directional overcurrent
detection. Computers & Mathematics with Applications, 64(6), 1886–1902. relay coordination using ant colony optimisation. IET Generation, Transmission &
Elsadd, M. A., Kawady, T. A., Taalab, A. -M. I., & Elkalashy, N. I. (2021). Adaptive Distribution, 9(14), 2040–2049.
optimum coordination of overcurrent relays for deregulated distribution system Singh, N., & Sinha, A. (2010). Chaos based multiple image encryption using multiple
considering parallel feeders. Electrical Engineering, 1–19. canonical transforms. Optics and Laser Technology, 42(5), 724–731.
Gokhale, S. S., & Kale, V. S. (2014). Application of the firefly algorithm to optimal Srivastava, A., & Das, D. K. (2020). A new aggrandized class topper optimization
over-current relay coordination. In 2014 international conference on optimization of algorithm to solve economic load dispatch problem in a power system. IEEE
electrical and electronic equipment (pp. 150–154). IEEE. Transactions on Cybernetics.
Gokhale, S. S., & Kale, V. S. (2016). An application of a tent map initiated chaotic Storn, R., & Price, K. (1997). Differential evolution–A simple and efficient heuristic for
firefly algorithm for optimal overcurrent relay coordination. International Journal global optimization over continuous spaces. Journal of Global Optimization, 11(4),
of Electrical Power & Energy Systems, 78, 336–342. 341–359.
Heidari, A. A., Mirjalili, S., Faris, H., Aljarah, I., Mafarja, M., & Chen, H. (2019). Swathika, O. V. G., & Hemamalini, S. (2020). Graph theory and optimization algo-
Harris Hawks optimization: Algorithm and applications. Future Generation Computer rithms aided adaptive protection in reconfigurable microgrid. Journal of Electrical
Systems, 97, 849–872. Engineering & Technology, 15(1), 421–431.

14
D. Acharya and D.K. Das Expert Systems With Applications 199 (2022) 116858

Tian, D. (2017). Particle swarm optimization with chaos-based initialization for Wang, G. -G., Deb, S., & Coelho, L. d. S. (2015). Elephant herding optimization. In 2015
numerical optimization. Intelligent Automation & Soft Computing, 1–12. 3rd international symposium on computational and business intelligence (pp. 1–5). IEEE.
Tian, D., Zhao, X., & Shi, Z. (2019). Chaotic particle swarm optimization with sigmoid- Wang, G. -G., Deb, S., & Coelho, L. D. S. (2018). Earthworm optimisation algorithm: A
based acceleration coefficients for numerical function optimization. Swarm and bio-inspired metaheuristic algorithm for global optimisation problems. International
Evolutionary Computation, 51, Article 100573. Journal of Bio-Inspired Computation, 12(1), 1–22.
Wadood, A., Gholami Farkoush, S., Khurshaid, T., Kim, C. -H., Yu, J., Geem, Z. W., et al. Wang, G. -G., Deb, S., & Cui, Z. (2019). Monarch butterfly optimization. Neural
(2018). An optimized protection coordination scheme for the optimal coordination Computing and Applications, 31(7), 1995–2014.
of overcurrent relays using a nature-inspired root tree algorithm. Applied Sciences, Wang, G. G., Guo, L., Gandomi, A. H., Hao, G. -S., & Wang, H. (2014). Chaotic krill
8(9), 1664. herd algorithm. Information Sciences, 274, 17–34.
Wang, G. -G. (2018). Moth search algorithm: A bio-inspired metaheuristic algorithm Zeng, G. -Q., Xie, X. -Q., Chen, M. -R., & Weng, J. (2019). Adaptive population extremal
for global optimization problems. Memetic Computing, 10(2), 151–164. optimization-based PID neural network for multivariable nonlinear control systems.
Swarm and Evolutionary Computation, 44, 320–334.

15