SSRG International Journal of Electrical and Electronics Engineering Volume 11 Issue 8, 217-225, August 2024

ISSN: 2348-8379/ https://doi.org/10.14445/23488379/IJEEE-V11I8P119 © 2024 Seventh Sense Research Group®

Original Article

A New Approach to Automatic and Optimal Membership

Function Generation for Fuzzy System Modelling
Anup Kumar Mallick1, Dwaipayan Ghosh2, Kabita Purkait3
1,2,3
Department of Electronics & Communication Engineering, Kalyani Government Engineering College, West Bengal, India.
1
Corresponding Author : [email protected]

Received: 09 June 2024 Revised: 11 July 2024 Accepted: 10 August 2024 Published: 31 August 2024

Abstract - During the last few decades, the optimization-based data-driven approach has been widely used for generating
membership functions in fuzzy-based systems, where the shapes of membership functions are mostly considered either triangular
or trapezoidal. However, the number of parameters that are required to be estimated for a triangular membership function is
three (left vertex, center, and right vertex). For a trapezoidal membership function, it is four (left base point, left shoulder, right
base point, and right shoulder). Whereas, the number of parameters required for a Gaussian membership function is two (mean
and standard deviation). Therefore, a fuzzy system modelled using the Gaussian membership function can significantly reduce
the number of parameters when the number of subsets for the antecedent and consequent membership functions is large.
However, not much attention is given to designing fuzzy models with Gaussian-shaped membership functions; most of the
existing fuzzy modelling techniques impose many restrictions on the membership functions’ parameters. As a result, the flexibility
and scope of the optimization techniques are reduced. This study, therefore, suggests a novel optimization-based technique to
frame fuzzy membership functions in which the membership functions are Gaussian-shaped, and very few restrictions are
imposed on the parameter selection. A comparative analysis is carried out between the conventional method and the proposed
method with different optimization techniques (Differential Evolution (DE), Particle Swarm Optimization (PSO), and Genetic
Algorithm (GA)) to approximate four standard nonlinear functions.

Keywords - Differential Evolution, Fuzzy system, Genetic Algorithm, Membership Function, Particle Swarm Optimization.

1. Introduction conventional way to formulate membership functions is to

The basic principles of fuzzy modelling were formulated divide the input and output spaces equally to generate the
by Zadeh [1] with the aim of approximately but effectively antecedent and consequent membership functions,
describing the behaviours of complex or ill-defined systems. respectively.
In the last few decades, fuzzy models have been widely used
in different fields of science, engineering, social science, Another method of modelling fuzzy systems is to use
medical diagnosis and treatment, etc. [2-4]. There are two expert knowledge. However, knowledge-based membership
broad types of fuzzy models, viz., the Mamdani fuzzy model function generation has some limitations. Sometimes, the
and the Takagi-Sugeno fuzzy model. The Mamdani model is expert in the particular domain is not available, or the experts'
a linguistic model that is based on a collection of IF-THEN opinions may differ from one another. Therefore, researchers
rules, with the antecedent and the consequent both fuzzy [5]. [9-11] have developed different approaches, such as
The Sugeno models are formed by rules with fuzzy antecedent clustering-based approach [12-14], neural network-based
and functional consequent [6]. In this paper, the experiment is approach [15], density-based technique [16], statistical
conducted on the modelling of the Mamdani-type fuzzy approach [17, 18], and optimization-based method [19-23],
system. etc., to derive the fuzzy models’ membership functions.

The major tasks involved in modelling a Mamdani-type In the last few decades, optimization techniques, due to
fuzzy-based system are the generation of fuzzy Membership their ability to automatically generate the optimum value of
Functions (MFs) for the antecedent and consequent, formation parameters, have been extensively used for the generation of
of the rule base, development of a fuzzy inference engine, and MFs of fuzzy systems. In such an approach, the optimization
finally, the defuzzification for the crisp output [7, 8]. The techniques are used either to fine-tune the membership
construction of membership functions plays a pivotal role in functions after initial guesses or to generate the membership
the design process of a Mamdani-type fuzzy model. One functions automatically. Different optimization techniques,

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

viz., GA, PSO, and DE, have been used to generate the 2.1. Mamdani-Type Fuzzy System
membership functions optimally.
Knowledge
Karr et al. [19] have employed GA to model a fuzzy Crisp Base Crisp

Defuzzifier
controller for balancing a cart pole. Herrera et al. [20] have

Fuzzifier
used a genetic algorithm to tune fuzzy rules and fit fuzzy Input Output
membership functions for balancing an inverted pendulum.
The membership functions for a single input-single output
system where the output points are squares of the input points Fuzzy
are determined using a genetic algorithm in [21]. Zhang et al. Inference
[22] optimize the membership function for a general industrial Engine
Fig. 1 Block diagram of a Mamdani-type fuzzy system
process with dead time, saturation, and time delay using a
genetic algorithm. Safaee et al. [23] apply PSO and GA to
Figure 1 shows the block diagram of a Mamdani-type
generate the membership functions in the design of a quad
fuzzy system. It consists of a fuzzifier, a knowledge base, a
rotor.
fuzzy inference engine, and a defuzzifier. The knowledge base
Although several research findings claim the efficacy of comprises a rule base and a database. The rule base consists
the optimized fuzzy systems over the conventional fuzzy of a set of fuzzy If-Then rules, and the database keeps the MFs
systems, the major problem in the optimum generation of of the input and output variables. The fuzzifier, with the help
membership functions is the restrictions imposed on the of antecedent MF, converts real-world crisp input to its
parameters, which in turn reduces the search space for the corresponding fuzzy value. The fuzzy inference engine fires
optimization problem. For example, in [20], the authors appropriate rules of the rule base with varying firing strength
update a fuzzy number by four parameters, and the and provides fuzzy outputs of each fired rule. The defuzzifier
formulation of these four parameters sufficiently restricts the generates a crisp output from the aggregated FIS output for
parameters of membership functions. real-world applications using the defuzzification method.

Zhang et al. [22] put several constraints on the position of 2.2. Mamdani-type Fuzzy System
membership functions. They [22] consider triangular-shaped A membership function µA(x) is defined by the following
membership functions and restrict the position of the centers mapping:
and the two vertexes of the membership functions. As the
number of constraints increases, the flexibility and scope of  A : x  [0,1], x  X. (1)
generating new positions for the parameters for the
membership functions reduces. Moreover, it is found that Where x is a real number describing an object, X is the
most of the optimization-based MF generation techniques universe of discourse, and A is a subset of X. The membership
have considered the shape of the MFs, either triangular [21- function in fuzzy logic maps an attribute or object to a positive
23] or trapezoidal [20], whereas both shapes require more real number in the interval [0, 1]. Because of its function-like
parameters to be selected for their formation. mapping characteristics, it is called a membership function
[24].
Motivated by the above-discussed problems in
membership function generation, this paper suggests a novel The membership functions used for mapping input
technique for constructing MFs for the antecedent and variables and output variables are termed antecedent
consequent in a Mamdani-type fuzzy system. The shapes of membership function and consequence membership function,
the membership functions of both the antecedent and respectively. Theoretically, any function can serve as a
consequent are considered as Gaussian. The performances of membership function for a given fuzzy set [25]. The shape of
the proposed model with varying optimization techniques the membership function depends on the context of the
(DE, PSO, and GA) are compared with the conventional fuzzy applications. Different methods of generating membership
system modelling method in predicting four standard functions have been proposed in the literature, some of which
nonlinear functions, viz., a cube function, a square function, a are outlined in the introduction section of this article.
square root function, and an exponential function.
2.3. Optimization Techniques
2. Theoretical Preliminaries Optimization is a technique that aims to maximize or
The presented paper devises an optimization-based minimize a function in the design of a system. The function,
membership function generation technique for a Mamdani- here, termed a cost or objective function, attempts to fulfil
type fuzzy system. Hence, basic concepts of a Mamdani-type some performance specifications or targets. The optimization
fuzzy system, fuzzy membership functions, and optimization techniques return a set of parameters’ values to obtain the best
techniques are discussed briefly in this section. possible result for the design of the underlying system.

218
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

There are several optimization techniques or algorithms, middle and the Standard Deviations (SDs) of all three subsets
such as GA, PSO, DE, etc. GA relies on the possibility of for the antecedent and the consequent are determined using an
generating better children from fitter parents and the survival optimization technique. In the evolutionary algorithm, the
of the fittest. PSO is inspired by the social movement of composition of an individual is given in Figure 2.
organisms in a bird flock or fish school.

Differential evolution [28–30] is a search algorithm that SDLA CMA SDMA SDHA SDLC CMC SDMC SDHC
optimizes a problem by iteratively developing a candidate
solution through an evolutionary process with little or no Fig. 2 Individual representing the membership functions’ parameters
assumption about the underlying optimization problem and is
capable of rapidly exploring large design areas. The Here, SDLA, SDMA, SDHA denote SDs of the MFs
optimization techniques are used for various purposes; representing the input variable for its low, medium, and high
optimal generation of membership functions in a fuzzy system subsets, respectively; CMA and CMC represent the medium
is one such application. subset’s mean values for the input variable and the output
variable, respectively; and SDLC, SDMC, and SDHC are the
3. Proposed Model notions of SDs for the low, medium, and high subsets of the
The design of the proposed model broadly involves four output variable, respectively. The optimization algorithm in
major steps: generation of fuzzy MFs, formation of a fuzzy each iteration aims to reduce the objective function given in
rule base, rule firing by the inference engine, and Equation (1).
defuzzification of the fuzzy output. The novelty of the
suggested method lies in its first step, i.e., the generation of N
the fuzzy MFs, which is thus illustrated in detail. The other f   (( g (i)  d (i))2 (2)
three steps, viz., formation of fuzzy rules, the firing of rules i 1
by the fuzzy inference engine, and finally, defuzzification, are
done in the conventional methods of fuzzy system design; Here, N represents the number of data points, g(i) and d(i)
hence, these three steps are described in brief. denote, respectively, the given output and the derived output
obtained using the proposed technique for the i-th input x(i).
3.1. Generation of Fuzzy Membership Functions The best population found at the last iteration of the
The proposed model is designed for a single-input, single- optimization technique is translated as membership functions’
output system. The input variable represented by antecedent parameters.
MF and the output variable represented by consequent MF are
both comprised of three fuzzy subsets: low, medium, and high, This paper proposes and compares the generation of
and the MFs are assumed to be Gaussian-shaped. The membership functions' parameters using three different
minimum and maximum values of the variables are considered evolutionary algorithms (GA, PSO, and DE). The procedures
as the mean of low subset and high subset, respectively, for for generating the parameters of MFs using GA, PSO, and DE
both antecedent and consequent. The mean of the subset are given in Algorithms 1, 2, and 3, respectively.

Algorithm 1. Parameter optimization of member functions using GA

Input: Population size (n), maximum number of iterations (itrmax), crossover, and mutation probabilities.
Output: Membership functions’ parameters.
1: Generate initial population or parents (nx8)
2: Set iteration=0
3: Compute the fitness of the parents using objective function in Equation (1)
4: while iteration < itrmax do
5: Increment iteration by 1
6: Form the mating pool from the parents through binary tournament selection
7: Perform crossover & mutation
8: Compute the fitness of the offspring using objective function in Equation (1)
9: Update parents for the next generation by taking the best fit n chromosomes among the parents &
the children
10: end while
Return best chromosome

219
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

Algorithm 2. Parameter optimization of member functions using PSO

Input: Swarm size (n), maximum number of iterations (itrmax), w, c1, c2.
Output: Membership functions’ parameters.
1: Generate initial particles (nx8) and initial velocity(old)
2: Set iteration=0
3: Compute the fitness of each particle using objective function in Equation (1)
4: Assign the current position as the pbest of each particle
5: Assign the best position of all particles as the gbest
6: while iteration < itrmax do
7: Increment iteration by 1
8: Compute the new velocity of each particle as,
velocity(new)=w*velocity(old)+ c1*(pbest- position(old))+c2*(gbest-position(old))
9: Update the position of each particle by,
position(new)=position(old)+velocity(new)
10: Compute fitness of each particle with its updated position using objection function in Equation (1)
11: Update the pbest of each particle and the gbest of all particles
12: end while
Return best chromosome

Algorithm 3. Parameter optimization of member functions using DE

Input: Population size (n), maximum number of
Iterations (itrmax), crossover probability (CR).
Output: Membership functions’ parameters.
1: Generate initial target vectors (nx8)
2: Set iteration=0
3: Compute the fitness of the target vectors using objective function in Equation (1) and find the best
target vector pbest
4: while iteration < itrmax do
5: Increment iteration by 1
6: Generate F by the Quantile function of the Cauchy distribution
7: for i=1:n
8: Generate donor vector by,
donor vector(i)=target vector(i)+ F(pbest-target vector(i)+target vector(r1)-target vector(r2)),
such that r1≠r2≠i
9: Perform recombination between target vector(i) and donor vector(i) with CR to generate trial
vector(i)
10: Compute the fitness of the trial vector(i) using objective function in Equation (1)
11: Replace the target vector(i) by trial vector(i), if trial vector is fitter
end for
12: Update target vectors and pbest for the next iteration
13: end while
Return best chromosome

220
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

3.2. Fuzzy Rule Base Where x and y are the notions of input and output,
To illustrate the proposed method, a single input-single respectively, to quantitatively measure and compare the
output system is considered with a set of rules (R1-R3) as outputs of the suggested fuzzy model and the conventional
given below. fuzzy model, two error metrics, viz., average sum squared
error ((here denoted by ASSE) and average error (here denoted
R1: If input is low, then output is low. by AE) are used. The expressions of these two error indices
R2: If input is medium, then output is medium. are given by Equations (7) and (8).
R3: If input is high, then output is high.
1
𝐴𝑆𝑆𝐸 = 𝑟 ∑𝑟𝑘=1(𝑒(𝑘) − 𝑝(𝑘))2 (7)
3.3. Fuzzy Inference Engine
The Fuzzy Inference Engine (FIS) infers the three rules in 1
𝐴𝐸 = ∑𝑟𝑘=1 |𝑒(𝑘) − 𝑝(𝑘)| (8)
the rule base with different firing strengths of [0, 1]. The final 𝑟
fuzzy output of the FIS is obtained by aggregating the outputs
of all three rules using the fuzzy MAX operator. Here, e(k) and p(k) represent the exact output and
predicted output of the k-th test point, and r is the size of the
3.4. Defuzzification test points. For unbiased comparison, the shape of the
There are different defuzzification techniques [31], each membership functions for the conventional fuzzy system is
with some advantages and disadvantages. The proposed also considered Gaussian, with the low subset and high subset
model uses one of the most popular defuzzification as the right-sided Gaussian and left-sided Gaussian,
techniques, Mean of Maxima (MOM) [32], to find the final respectively. The mean of the subset middle of the
crisp output. conventional fuzzy system is set at the midpoint of the input
variable for the antecedent and the midpoint of the output
variable for the consequent. The standard deviations of the
4. Simulation Results membership functions for the conventional fuzzy system are
The performance of the proposed fuzzy system is
all calculated as 1/6th of the respective variable.
validated on four nonlinear functions, viz., a cube function, a
square function, a square root function, and an exponential
In this paper, the conventional fuzzy model is denoted by
function, as given in Equations (3) to (6) respectively.
CFM and the Proposed Fuzzy Models (PFMs), whose
membership functions are generated by GA, PSO, and DE, are
y  x3 for 0 x2 (3) denoted by PFM-GA, PFM-PSO, and PFM-DE, respectively.
The performances of the fuzzy models (CFM, PFM-GA,
y  x2 for 0 x3 (4) PFM-PSO, and PFM-DE) are tested on 11 equally spaced data
points of the four nonlinear functions considered here. The
actual and predicted outputs for the test data points of the cube
y x for 0 x9 (5)
function, square function, square root function, and
exponential function are depicted in Figures 3, 4, 5, and 6,
y  ex for 0 x3 (6) respectively.
8 8
Actual Actual
6 CFM 6 PFM-GA
Output

Output

4 4

2 2
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Input Input
(a) (b)
8 8
Actual Actual
6 PFM-PSO 6 PFM-DE
Output

Output

4 4

2 2

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Input Input
(c) (d)
Fig. 3 Actual (o) and predicted (*) output for the cube function as estimated by (a) CFM (b) PFM-GA (c) PFM-PSO (d) PFM-DE

221
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

10 10
8 Actual 8 Actual
CFM PFM-GA

Output

Output
6 6
4 4
2 2
0 0
0 1 2 3 0 1 2 3
Input Input
(a) (b)
10 10
Actual Actual
8 8
PFM-PSO PFM-DE
Output

Output
6 6
4 4
2 2
0 0
0 1 2 3 0 1 2 3
Input Input
(c) (d)
Fig. 4 Actual (o) and predicted (*) output for the square function as estimated by (a) CFM, (b) PFM-GA, (c) PFM-PSO, and (d) PFM-DE.

3 3

2 2
Output

1 Output 1
Actual Actual
CFM PFM-GA
0 0
0 5 10 0 5 10
Input Input
(a) (b)
3 3

2 2
Output

Output

1 Actual 1 Actual
PFM-PSO PFM-DE
0 0
0 5 10 0 5 10
Input Input
(c) (d)
Fig. 5 Actual (o) and predicted (*) output for the square root function as estimated by (a) CFM, (b) PFM-GA, (c) PFM-PSO, and (d) PFM-DE.

25 25
Actual Actual
20 CFM 20 PFM-GA
Output
Output

15 15
10 10
5 5
0 0
0 1 2 3 0 1 2 3
Input Input
(a) (b)
25 25
Actual Actual
20 PFM-PSO 20 PFM-DE
Output

Output

15 15
10 10
5 5
0 0
0 1 2 3 0 1 2 3
Input Input
(c) (d)
Fig. 6 Actual (o) and predicted (*) output for the exponential function as estimated by (a) CFM (b) PFM-GA (c) PFM-PSO (d) PFM-DE

222
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

Table 1. Comparative performances of the fuzzy models with respect to ASSE

Fuzzy Models
Test Functions
CFM PFM-GA PFM-PSO PFM-DE

Cube 5.061847 0.005818 0.109382 0.054487

Square 3.440291 0.099593 0.098489 0.006995

Square root 0.362491 0.002987 0.038995 0.003597

Exponential 18.006377 0.173327 0.080090 0.082414

Table 2. Comparative performances of the fuzzy models with respect to AE

Fuzzy Models
Test Functions
CFM PFM-GA PFM-PSO PFM-DE

Cube 1.800000 0.052364 0.219636 0.178182

Square 1.366364 0.233182 0.225000 0.065455

Square root 0.444238 0.044234 0.134281 0.044415

Exponential 3.289064 0.343060 0.218723 0.236512

The errors in predicting the four non-linear functions respectively. Similar to the cube function, the square function
were noted, and the values of the error indices are presented has not been predicted well by the conventional fuzzy model,
in Tables 1 and 2 to compare the performances of the as seen in Figure 4(a), and it encompasses very large errors, as
suggested fuzzy model and the conventional fuzzy model found in the result tables (Tables 1 and 2). From Figures 4(b)
quantitatively. and 4(c), it is seen that the performance of PFM-GA and PFM-
PSO are very similar, and they predict the square function
5. Result Discussion accurately at some points and deviate from the actual outputs
From Figure 3(a), it is seen that the Conventional Fuzzy at some other points.
Model (CFM) has predicted the correct output of the cube
function at very few points, rising to the very high values of The square function has been quite satisfactorily
ASSE (approximately 5.062) and AE (approximately 1.80), as predicted by PFM-DE, as seen in Figure 4(d). It is further
are found from Tables 1 & 2. From Figures 3(b) - 3(d), it is supported by the values of the error indices, as shown in
seen that all the proposed models (PFM-GA, PFM-PSO, and Tables 1 and 2. PFM-DE, with values of 0.007 and 0.0655 for
PFM-DE) are outperforming the conventional fuzzy model. ASSE and AE, respectively, outperforms the other three
Among the proposed models, PFM-PSO & PFM-DE are methods in predicting the square function. Figure 5(a)
predicting the cube function at some points accurately, indicates that for the square root function, the performance of
whereas, at some other points, there exist minor deviations the conventional fuzzy model is not appreciable, and it is
from the actual outputs. imperative from Tables 1 and 2, too.

From Tables 1 and 2, it is seen that the average sum From Figures 5(b)-5(d), it is noticed that PFA-GA and
squared error for PFM-PSO & PFM-DE are 0.109 and 0.054, PFM-DE are making sufficiently good predictions, whereas
respectively, and the average error for PFM-PSO & PFM-DE PFM-PSO, although performing better than the conventional
are 0.220 and 0.178, respectively. The best performance in the fuzzy model, at some points, its predicted outputs are much
prediction of the cube function is obtained by PFM-GA. From deviated from the actual output values. From the tables of
Figure 3(b), it is noticed that except at a very few points, PFM- results, it is seen that among the four models, the square root
GA has predicted the output accurately. From Tables 1 and 2, function has been best predicted by the PFM-GA. The
it appears that the values of ASSE & AE in predicting the cube exponential function, as seen from Figures 6(b)–6(d), at the
functions at the test points by PFM-GA are 0.006 and 0.052, test points is predicted very well by all three proposed models.

223
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

In terms of errors, this function has been best predicted by from data. The suggested technique has been applied to model
PFM-PSO, as found in Tables 1 and 2. a Mamdani-type fuzzy system. Efforts may be put into
designing a Sugeno-type fuzzy system utilizing the proposed
6. Conclusion method. Further research may also be conducted to achieve
This paper proposes a novel optimization-based better results using multi-objective optimization algorithms
technique for designing a fuzzy model with Gaussian-shaped instead of single-objective optimization algorithms in the
membership functions. The simulation results demonstrate design process.
that the proposed model outperforms the traditional fuzzy
model in approximating the four nonlinear functions Acknowledgments
considered here. It is expected that the suggested fuzzy The authors would like to thank their institution for
modeling technique will be adopted in various domains of providing a good research ambience that has helped them to
fuzzy-based system design to generate membership functions conduct work related to this article.

References
[1] L.A. Zadeh et al., “Fuzzy Sets,” Information and Control, vol. 8, no. 3, pp. 338-353, 1965. [CrossRef] [Google Scholar] [Publisher Link]
[2] Lotfi A Zadeh, Rafik A. Aliev, Fuzzy Logic Theory and Applications: Part I and Part II, World Scientific, Singapore, 2018. [CrossRef]
[Google Scholar] [Publisher Link]
[3] Elena Vlamou, and Basil Papadopoulos, “Fuzzy Logic Systems and Medical Applications,” AIMS Neuroscience, vol. 6, no. 4, pp. 266-
272, 2019. [CrossRef] [Google Scholar] [Publisher Link]
[4] Hangyao Wu, and Zeshui XU, “Fuzzy Logic in Decision Support: Methods, Applications and Future Trends,” International Journal of
Computers Communications & Control, vol. 16, no. 1, pp. 1-27, 2021. [CrossRef] [Google Scholar] [Publisher Link]
[5] E.H. Mamdani, and S. Assilian, “An Experiment in Linguistic Synthesis with a Fuzzy Logic Controller,” International Journal of Man-
Machine Studies, vol. 7, no.1, pp. 1-13, 1975. [CrossRef] [Google Scholar] [Publisher Link]
[6] Tomohiro Takagi, and Michio Sugeno, “Fuzzy Identification of Systems and Its Applications to Modeling and Control,” IEEE
Transactions on Systems, Man, and Cybernetics, vol. 15, no.1, pp. 116-132, 1985. [CrossRef] [Google Scholar] [Publisher Link]
[7] Dimitar P. Filev, and Ronald R. Yager, Essential of Fuzzy Modeling and Control, John Wiley & Sons Inc., Singapore, 1994. [Google
Scholar]
[8] Ajit K. Mandal, Introduction to Control Engineering Modeling, Analysis and Design, New Age International Publisher, India, 2006.
[Google Scholar]
[9] Swarup Medasani, Jaeseok Kim, and Raghu Krishnapuram, “An Overview of Membership Function Generation Techniques for Pattern
Recognition,” International Journal of Approximate Reasoning, vol. 19, no. 3-4, pp. 391-417, 1998. [CrossRef] [Google Scholar]
[Publisher Link]
[10] I. Burhan Türkşen, “Review of Fuzzy System Models with an Emphasis on Fuzzy Functions,” Transactions of the Institute of Measurement
and Control, vol. 31, no. 1, pp. 7-31, 2009. [CrossRef] [Google Scholar] [Publisher Link]
[11] Mehdi Zangeneh, Ebrahim Aghajari, and Mehdi Forouzanfar, “A Survey: Fuzzify Parameters and Membership Function in Electrical
Applications,” International Journal of Dynamics & Control, vol. 8, pp. 1040-1051, 2020. [CrossRef] [Google Scholar] [Publisher Link]
[12] T.W. Liao, Aivars K. Celmins, and Robert J. Hammell II, “A Fuzzy C-Means Variant for the Generation of Fuzzy Term Sets,” Fuzzy Sets
and Systems, vol. 135, no. 2, pp. 241-257, 2003. [CrossRef] [Google Scholar] [Publisher Link]
[13] Muhammad Hamza Azam et al., “Fuzzy Type-1 Triangular Membership Function Approximation Using Fuzzy C-Means,” 2020
International Conference on Computational Intelligence (ICCI), Bandar Seri Iskandar, Malaysia, pp. 115-120, 2020. [CrossRef] [Google
Scholar] [Publisher Link]
[14] Chun-Hao Chen et al., “Cluster-Based Membership Function Acquisition Approaches for Mining Fuzzy Temporal Association Rules,”
IEEE Access, vol. 8, pp. 123996-124006, 2020. [CrossRef] [Google Scholar] [Publisher Link]
[15] Chih-Chung Yang, and N.K. Bose, “Generating Fuzzy Membership Function with Self-Organizing Feature Map,” Pattern Recognition
Letters, vol. 27, no. 5, pp. 356-365, 2006. [CrossRef] [Google Scholar] [Publisher Link]
[16] Imen Derbel, Narjes Hachani, and Habib Ounelli, “Membership Functions Generation Based on Density Function,” 2008 International
Conference on Computational Intelligence and Security, Suzhou, China, pp. 96-101, 2008. [CrossRef] [Google Scholar] [Publisher Link]
[17] Dumidu Wijayasekara, and Milos Manic, “Data-Driven Fuzzy Membership Function Generation for Increased Understandability,” 2014
IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), Beijing, China, pp. 133-140, 2014. [CrossRef] [Google Scholar]
[Publisher Link]
[18] Hossein Pazhoumand-Dar, Chiou-Peng Lam, and Martin Masek, “Automatic Generation of Fuzzy Membership Functions Using Adaptive
Mean-Shift and Robust Statistics,” Proceedings of 8th International Conference on Agents and Artificial Intelligence, pp. 160-171, 2016.
[CrossRef] [Google Scholar] [Publisher Link]

224
 Anup Kumar Mallick et al. / IJEEE, 10(8), 217-225, 2024

[19] Charles L. Karr, “Design of a Cart-Pole Balancing Fuzzy Logic Controller Using a Genetic Algorithm,” Proceedings of the SPIE, vol.
1468, 1991. [CrossRef] [Google Scholar] [Publisher Link]
[20] F. Herrera, M. Lozano, and J.L. Verdegay, “Tuning Fuzzy Logic Controllers by Genetic Algorithms,” International Journal of
Approximate Reasoning, vol. 12, no. 3-4, pp. 299-315, 1995. [CrossRef] [Google Scholar] [Publisher Link]
[21] Ahmet Arslan, and Mehmet Kaya, “Determination of Fuzzy Logic Membership Functions Using Genetic Algorithms,” Fuzzy Sets and
Systems, vol. 118, no. 2, pp. 297-306, 2001. [CrossRef] [Google Scholar] [Publisher Link]
[22] Huai-Xiang Zhang, Feng Wang, and Bo Zhang, “Genetic Optimization of Fuzzy Membership Functions,” 2009 International Conference
on Wavelet Analysis and Pattern Recognition, Baoding, pp. 465-470, 2009. [CrossRef] [Google Scholar] [Publisher Link]
[23] B. Safaee, and S.K. Kamaleddin Mousavi Mashhadi, “Optimization of Fuzzy Membership Functions via PSO and GA with Application
to Quadrotor,” Journal of AI and Data Mining, vol. 5, no. 1, pp. 1-10, 2017. [CrossRef] [Google Scholar] [Publisher Link]
[24] Amit Konar, Computational Intelligence: Principles, Techniques and Applications, 1st ed., Springer, Berlin Heidelberg, 2005. [CrossRef]
[Google Scholar] [Publisher Link]
[25] S. Roy, and Udit Chakraborty, Introduction to Soft Computing, Neuro-Fuzzy and Genetic Algorithms, Pearson, India, 2013. [Google
Scholar]
[26] Sourabh Katoch, Sumit Singh Chauhan, and Vijay Kumar, “A Review on Genetic Algorithm: Past, Present, and Future,” Multimedia
Tools and Applications, vol. 80, pp. 8091-8126, 2020. [CrossRef] [Google Scholar] [Publisher Link]
[27] Dongshu Wang, Dapei Tan, and Lei Liu, “Particle Swarm Optimization Algorithm: An Overview,” Soft Computing, vol. 22, pp. 387-408,
2018. [CrossRef] [Google Scholar] [Publisher Link]
[28] Rainer Storn, and Kenneth Price, “Differential Evolution - A Simple and Efficient Heuristic for Global Optimization over Continuous
Spaces,” Journal of Global Optimization, vol. 11, no. 4, pp. 341-359, 1997. [CrossRef] [Google Scholar] [Publisher Link]
[29] Swagatam Das, and Ponnuthurai Nagaratnam Suganthan, “Differential Evolution: A Survey of the State-of-the-Art,” IEEE Transactions
on Evolutionary Computation, vol. 15, no. 1, pp. 4-31, 2011. [CrossRef] [Google Scholar] [Publisher Link]
[30] Manolis Georgioudakis, and Vagelis Plevris, “A Comparative Study of Differential Evolution Variants in Constrained Structural
Optimization,” Frontiers in Built Environment, vol. 6, 2020. [CrossRef] [Google Scholar] [Publisher Link]
[31] Shounak Roychowdhury, and Witold Pedrycz, “A Survey of Defuzzification Strategies,” International Journal of intelligent systems, vol.
16, no. 6, pp. 679-695, 2001. [CrossRef] [Google Scholar] [Publisher Link]
[32] Aurélie Talon, and Corinne Curt, “Selection of Appropriate Defuzzification Methods: Application to the Assessment of Dam
Performance,” Expert Systems with Applications, vol. 70, pp. 160-174, 2017. [CrossRef] [Google Scholar] [Publisher Link]

225