INFORMATICA, 2024, Vol. 35, No.

1, 155–178 155
© 2024 Vilnius University
DOI: https://doi.org/10.15388/23-INFOR538

A Fuzzy MARCOS-Based Analysis of Dragonfly

Algorithm Variants in Industrial Optimization
Problems

Kanak KALITA1,2 , Narayanan GANESH3 , Rajendran SHANKAR4 ,

Shankar CHAKRABORTY5,∗
1 Department of Mechanical Engineering,
Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Avadi, India
2 University Centre for Research & Development, Chandigarh University, Mohali, 140413, India
3 School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India
4 Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, India
5 Department of Production Engineering, Jadavpur University, Kolkata, India.
e-mail: [email protected]

Received: December 2022; accepted: November 2023

Abstract. Metaheuristics are commonly employed as a means of solving many distinct kinds of
optimization problems. Several natural-process-inspired metaheuristic optimizers have been intro-
duced in the recent years. The convergence, computational burden and statistical relevance of meta-
heuristics should be studied and compared for their potential use in future algorithm design and
implementation. In this paper, eight different variants of dragonfly algorithm, i.e. classical dragon-
fly algorithm (DA), hybrid memory-based dragonfly algorithm with differential evolution (DADE),
quantum-behaved and Gaussian mutational dragonfly algorithm (QGDA), memory-based hybrid
dragonfly algorithm (MHDA), chaotic dragonfly algorithm (CDA), biogeography-based Mexican
hat wavelet dragonfly algorithm (BMDA), hybrid Nelder-Mead algorithm and dragonfly algorithm
(INMDA), and hybridization of dragonfly algorithm and artificial bee colony (HDA) are applied
to solve four industrial chemical process optimization problems. A fuzzy multi-criteria decision
making tool in the form of fuzzy-measurement alternatives and ranking according to compromise
solution (MARCOS) is adopted to ascertain the relative rankings of the DA variants with respect
to computational time, Friedman’s rank based on optimal solutions and convergence rate. Based on
the comprehensive testing of the algorithms, it is revealed that DADE, QGDA and classical DA are
the top three DA variants in solving the industrial chemical process optimization problems under
consideration.
Key words: Dragonfly algorithm, Process optimization, MCDM, MARCOS.

∗ Corresponding author.
156 K. Kalita et al.

1. Introduction

The increasing population coupled with rapid urbanization has led to unprecedented de-
mands for natural and man-made resources. The chemical industry which acts as the raw
material provider to several critical sectors, like pharmaceuticals, construction, etc., must
keep up with the pace of these ever-increasing demands. Setting up new production fa-
cilities to serve the increasing demand requires significant resources and is also highly
capital-intensive. Improving yield and efficiency of the existing plants on the other hand
just needs deployment of better managerial practices, and sound knowledge of the pro-
cesses and their optimization.
Due to large number of input parameters involved in any of the typical industrial
chemical processes, optimizing them using classical approaches, like one-factor-at-a-time
(OFAT), Taguchi methodology, etc., may not be always feasible. Of late, metaheuristics,
which are essentially general-purpose heuristic approaches, have become quite popular
among the researchers working in the area of process optimization. Perhaps this popular-
ity is mainly due to high-level problem-independent algorithmic framework of the meta-
heuristics (Sörensen and Glover, 2013). Metaheuristics are stochastic algorithms and of-
ten draw their inspiration from nature. Any metaheuristic algorithm is an amalgamation of
two basic functions, i.e. exploration and exploitation (Blum and Roli, 2003). Exploration
and exploitation are also sometimes referred to as diversification and intensification. The
objective of diversification or exploration is to navigate through the search space to find
out ‘potential regions or zones’ with ‘good solutions’. On the other hand, intensification
or exploitation is related to thoroughly searching out the ‘potential region or zone’ to lo-
cate the best solution. All metaheuristic algorithms attempt to strike an optimal balance
between diversification and intensification. This balance has a direct bearing on the con-
vergence rate of the considered algorithm as well as its ability to find out diverse solutions.
Many metaheuristic algorithms have been proposed so far in quest of the optimal balance
between diversification and intensification. Yet, many more algorithms continue to be de-
veloped. Nevertheless, in recent times, researchers have established the suitability, appli-
cability and often superiority of metaheuristics over the traditional approaches, which are
deterministic and exact. For large-scale and complex problems, like chemical process op-
timization, structural optimization, etc., metaheuristics provide a good trade-off between
solution quality and computational time. Some of the most popular metaheuristics are
genetic algorithm (GA) (Holland, 1992), simulated annealing (Kirkpatrick et al., 1983),
particle swarm optimization (PSO) (Kennedy and Eberhart, 1995), grey wolf optimizer
(Mirjalili et al., 2014) etc.
In this decade, research on metaphor-based metaheuristics has received a tremendous
impetus. A plethora of nature-inspired metaheuristics, like flower pollination algorithm
(Yang, 2012), swallow swarm optimization algorithm (Neshat et al., 2013), grey wolf
optimizer (Mirjalili et al., 2014), moth-flame optimization algorithm (Mirjalili, 2015),
dragonfly algorithm (Mirjalili, 2016), grasshopper optimization algorithm (Saremi et al.,
2017), artificial flora optimization algorithm (Cheng et al., 2018), seagull optimization al-
gorithm (Dhiman and Kumar, 2019), marine predators algorithm (Faramarzi et al., 2020),
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 157

arithmetic optimization algorithm (Abualigah et al., 2021), rat swarm optimization (Mzili
et al., 2022), etc., has been developed in the last ten years. Besides proposing new meta-
heuristics, tons of work have also been carried out in the area of hybridization of meta-
heuristics, wherein existing algorithms are either merged with other algorithms or new
features are introduced in the existing algorithms.
The DA, a population-based nature-inspired metaheuristic, was propounded by Mir-
jalili (2016), Mafarja et al. (2018). Just like PSO, DA is also guided by swarm intelli-
gence. It mimics the static and dynamic swarming behaviours of dragonflies. Since its
inception in 2015, DA has been applied to solve various classes of optimization prob-
lems ranging from continuous (Abedi and Gharehchopogh, 2020) to discrete (Jawad et
al., 2021) and from unconstrained (Can and Alatas, 2017) to constrained (Khalilpourazari
and Khalilpourazary, 2020). It has been successfully employed in both single-objective
(Reddy, 2016) and multi-objective roles (Joshi et al., 2021). The hybridization of DA has
also received a lot of attention lately. Debnath et al. (2021) developed a hybrid memory-
based dragonfly algorithm with differential evolution (DADE), whereas, Shirani and Safi-
Esfahani (2020) proposed a biogeography-based Mexican hat wavelet dragonfly algorithm
(BMDA). Xu and Yan (2019) fused the classical DA with the Nelder-Mead algorithm to
develop a hybrid Nelder-Mead algorithm and dragonfly algorithm (INMDA) to improve
the local capacity for exploration. Ghanem and Jantan (2018) proposed a hybridization of
dragonfly algorithm and artificial bee colony (HDA) to improve the convergence rate. Sree
Ranjini and Murugan (2017) combined the exploration capability of DA with the exploita-
tion capacity of PSO to develop a memory-based hybrid dragonfly algorithm (MHDA).
Yu et al. (2020) proposed the quantum-behaved and Gaussian mutational dragonfly algo-
rithm (QGDA) and (Sayed et al., 2019) developed the chaotic dragonfly algorithm (CDA)
by seamlessly integrating chaos theory with classical DA.
In this paper, the performance of eight popular DA variants is compared based on four
industrial chemical process problems (i.e. heat exchanger network design (Floudas and
Ciric, 1989), optimal operation of alkylation unit (Sauer et al., 1964), reactor network de-
sign (Ryoo and Sahinidis, 1995) and Haverly’s pooling problem (Floudas and Pardalos,
1990). Hence, the algorithms considered in this paper are classical DA (Mirjalili, 2016),
DADE (Debnath et al., 2021), QGDA (Yu et al., 2020), MHDA (Sree Ranjini and Muru-
gan, 2017), CDA (Sayed et al., 2019), BMDA (Shirani and Safi-Esfahani, 2020), INMDA
(Xu and Yan, 2019) and HAD (Ghanem and Jantan, 2018). The algorithms are compre-
hensively tested based on the optimal solution obtained, computational time and conver-
gence rate. The derived optimal solutions are further validated from the viewpoint of the
best solution, mean best solution and dispersion (standard deviation) of the solutions on
repeated trials. The Friedman’s test rank is computed for each algorithm based on three
criteria (best, mean and standard deviation) used for optimal solution analysis. Further,
the opinions of five experts are aggregated using a fuzzy scale and a multi-criteria deci-
sion making (MCDM) tool in the form of fuzzy-measurement alternatives and ranking
according to compromise solution (MARCOS) is adopted to identify the best algorithm
based on the comprehensive analysis of the optimal solution, computational burden and
convergence rate. The basic methodology followed in this paper can be represented in the
form of a flowchart, as shown in Fig. 1.
158 K. Kalita et al.

Fig. 1. Flowchart of the adopted methodology.

2. Methods

2.1. Dragonfly Algorithm

The classical DA (Mirjalili, 2016) is a simple yet powerful metaheuristic algorithm mim-
icking the swarm behaviour of dragonflies. The social behaviour of dragonflies exhibited
during searching and gathering of food as well as during foe avoidance forms the basis of
the equation-based rules used to simulate and actuate DA. The DA is realized using the
five parameters, e.g. separation, alignment, cohesion, attraction and distraction. Collision
avoidance with neighbouring dragonflies is governed by separation. Velocity matching to
the neighbouring individuals is carried out by alignment. Attractions towards the centre
of mass of the neighbourhood and towards a food source are respectively governed by
cohesion and attraction. Movement away from the enemy is controlled by distraction.

2.2. Hybrid Dragonfly Algorithm with Differential Evolution

Differential evolution (DE), in general, has high computational ability and a fast conver-
gence rate. Akin to GA, DE explores the search space based on crossover and mutation.
At the end of each cycle, DADE (Debnath et al., 2021) stores the best solution in its mem-
ory and continues the search with DE which promotes population diversity by employing
mutation.

2.3. Quantum-Behaved and Gaussian Mutational Dragonfly Algorithm

By implementing the concept of a quantum rotation gate, Yu et al. (2020) endeavoured

to strike a better balance between exploration and exploitation traits of DA. The Gaussian
mutation is also incorporated into this algorithm to help generate diverse solutions.
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 159

2.4. Memory-Based Hybrid Dragonfly Algorithm

The lack of internal memory in the classical DA can cause premature convergence to local
optima. To overcome this problem, Sree Ranjini and Murugan (2017) introduced certain
features of PSO into DA and called the hybrid algorithm as MHDA. By endowing DA with
internal memory, MHDA allows each dragonfly to keep track of its DA-pbest solution, i.e.
coordinates of the best solution obtained by it so far. The MHDA has also access to DA-
gbest, i.e. coordinates of the overall best solution obtained by the algorithm so far. After
initial exploration of the search space by DA, exploitation of the promising search space
zones is initialized by PSO considering DA-pbest and DA-gbest solutions.

2.5. Chaotic Dragonfly Algorithm

Sayed et al. (2019) employed ten chaotic maps to fine-tune the weights involved in the
separation, alignment, cohesion, attraction and distraction parameters of the classical DA.
The authors argued that as compared to DA, CDA would have an improved convergence
rate, with the algorithmic complexity being at par with DA. The overall complexity of
CDA is O(dM + MC), where d, M and C are the dimensions of the problem, number of
dragonflies and objective function complexity respectively.

2.6. Biogeography-Based Mexican Hat Wavelet Dragonfly Algorithm

To address the issue of premature convergence under heavy loads, Shirani and Safi-
Esfahani (2020) proposed a variant of DA called BMDA (biogeography-based algorithm,
Mexican hat wavelet and dragonfly algorithm) that combines the migration process of
the biogeography-based optimization (BBO) technique with the transformation process
of DA’s Mexican hat wavelet.

2.7. Hybrid Nelder-Mead Algorithm and Dragonfly Algorithm

Xu and Yan (2019) argued that too many social interactions in DA would be responsible
for reduced solution accuracy and premature convergence to local optima. These may
be caused due to improper balance between diversification and intensification. Xu and
Yan (2019) thus suggested hybridizing DA with an improved Nelder-Mead algorithm to
improve its local search capacity.

2.8. Hybridization of Dragonfly Algorithm and Artificial Bee Colony

Ghanem and Jantan (2018) highlighted that the presence of Levy flight in the position
update phase of DA would make it unable to effectively carry out a local search. To rectify
this problem, Ghanem and Jantan (2018) suggested hybridization of DA and artificial bee
colony (ABC) algorithm to make use of the exploitation and exploration abilities of DA
along with the exploration ability of ABC.
160 K. Kalita et al.

2.9. Fuzzy MARCOS

The MARCOS is an innovative MCDM approach that can be employed in many contexts
(Chakraborty et al., 2020; Stanković et al., 2020; Stević et al., 2020; Deveci et al., 2021;
Biswal et al., 2023). Its computational strategy has been developed taking into account
both the ideal and anti-ideal solutions (Bakır and Atalık, 2021). The utility degrees of
the candidate alternatives are quantified, which are subsequently considered to evaluate
the relative performance and rank each of the alternatives (Bakır et al., 2021; Badi et
al., 2022). In this paper, MARCOS is integrated with fuzzy set theory to deal with the
individual opinions of five experts with respect to computational time, Friedman’s rank
based on optimal solutions and convergence rate leading to the relative ranking of the
eight DA variants.
The application steps of MARCOS in fuzzy environment are summarized as shown
below:

Step 1: Formulate the initial decision matrix (X), consisting of m possible choices (alter-
natives) and n evaluation criteria.

⎡ ⎤
x11 x12 ... x1j ... x1n
⎢x x2n ⎥
⎢ 21 x22 ... x2j ... ⎥
⎢ ⎥
⎢ ... ... ... ... ... ... ⎥
X=⎢ ⎥, (1)
⎢ xi1 xi2 ... xij ... xin ⎥
⎢ ⎥
⎣ ... ... ... ... ... ... ⎦
xm1 xm2 ... xmj ... xmn

where xij is the performance of ith alternative against j th criterion.

Step 2: Develop the corresponding extended decision matrix (X  ) while considering the
anti-ideal (AI) and ideal (ID) solutions.

AI = min xij if j ∈ B and max xij if j ∈ C, (2)

i i
AI = max xij if j ∈ B and min xij if j ∈ C, (3)
i i

where B is the set of beneficial criteria and C is the set of non-beneficial criteria.
⎡ ⎤
xai1 xai2 ... xaij ... xain
⎢x x1n ⎥
⎢ 11 x12 ... x1j ... ⎥
⎢ ⎥
 ⎢ ... ... ... ... ... ... ⎥
X =⎢ ⎥. (4)
⎢ xi1 xi2 ... xij ... xin ⎥
⎢ ⎥
⎣ xm1 xm2 ... xmj ... xmn ⎦
xid1 xid2 ... xidj ... xidn
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 161

Step 3: Normalize the extended decision matrix using Eqs. (5) and (6) depending on the
type of the criterion under consideration.

xid
nij = , if j ∈ C, (5)
xij
xij
nij = , if j ∈ B. (6)
xid

Step 4: Develop the weighted normalized fuzzy decision matrix.

 
ṽij = vijl , vijm , viju = nij ⊗ w̃j = nij × wjl , nij × wjm , nij × wju , (7)

where w̃j is the fuzzy weight assigned to j th criterion.

Step 5: Determine the utility degrees of each alternative using the following expressions:

S̃i sil sim siu

K̃i− = = u , sm l , (8)
S̃ai sai ai sai
S̃i sil sim siu
K̃i+ = = u , sm l , (9)
s̃id sid id sid

where S̃i (sil , sim , siu ) is the sum of elements of the weighted normalized fuzzy decision
matrix and can be estimated using Eq. (10):

n
S̃i = ṽij . (10)
j =1

Step 6: Formulate the fuzzy matrix T̃i applying the following expression:
 
T̃i = t˜i = til , tim , tiu = K̃i− ⊕ K̃i+ = ki−l + ki+l , ki−m + ki+m , ki−u + ki+u . (11)

Step 7: Evaluate the utility functions for both the ideal and anti-ideal solutions.

 K̃i− ki−l k −m k −u
f K̃i+ = = , i , i , (12)
dfcrisp dfcrisp dfcrisp dfcrisp
 K̃i+ ki+l k +m k +u
f K̃i− = = , i , i , (13)
dfcrisp dfcrisp dfcrisp dfcrisp

where

l + 4m + u
dfcrisp = . (14)
6
162 K. Kalita et al.

The value of dfcrisp is obtained from a fuzzy number D̃, which can be estimated using
Eq. (15):


D̃ = d l , d m , d u = max t˜ij . (15)
i

Step 8: Determine the utility functions of all the alternatives.

From the defuzzified values of utility degrees and utility functions, the corresponding
utility function of each of the alternatives with respect to anti-ideal and ideal solutions is
computed using the following equation:

Ki+ + Ki−
f (Ki ) = . (16)
1−f (Ki+ ) 1−f (Ki− )
1+ + f (Ki− )
f (Ki+ )

Step 9: Rank the alternatives.

Based on the descending values of the utility function, the alternatives are finally sorted
from the best to the worst, the best alternative having the maximum utility function value.

3. Problem Description

To assess and compare the relative performance of eight different DA variants, four differ-
ent industrial chemical process optimization problems are considered in this paper as the
test problems. All these four problems are constrained optimization problems. The numer-
ical experiments are carried out on a Dell Inspiron 15-3567 series Windows System with
Intel(R) CoreTM i7-7500U CPU @2.70 GHz, Clock Speed 2.9 Ghz, L2 Cache Size 512
and 8 GB RAM. To avoid any bias in the results, 30 independent trials are conducted for
each of the DA algorithms on each test problem. The initial population size and maximum
number of cycles for each DA variant are kept as 60 and 500 respectively. Thus, during
each trial, 30000 function evaluations are carried out. The weight parameter in DA vari-
ants is assumed to be linearly decreasing from 0.9 to 0.4 as the number of cycles increases
from 0 to 500. Similarly, the separation/alignment/cohesion weights in the considered DA
variants are randomly varied between 0–0.1 for cycles less than 250. At 250 or more than
250 cycles, the separation/alignment/cohesion weight becomes 0.
The DA variants are subsequently ranked by comparing the algorithm’s mean (fmean ),
standard deviation (fstd ), CPU (run time) (in sec) and Friedman ranking. While comparing
the performance of optimizers, the one having the lowest fmean value is always preferable
(for minimization problem). If the fmean values of two optimizers become equal, their fstd
values can then be compared. In such cases, the optimizer having smaller fstd value is
more stable.
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 163

4. Numerical Results on Chemical Process Optimization

4.1. Case Study 1: Heat Exchanger Network Design (HEND)

The main objective of the HEND problem (Floudas and Ciric, 1989) is to minimize the
comprehensive area of HEND. This problem contains nine control variables and eight
equality constraints. Mathematically, this minimization type HEND problem is defined as
follows:

f (x̄) = 35x10.6 + 35x20.6 ,

h1 (x̄) = 200x1 x4 − x3 = 0,
h2 (x̄) = 200x2 x6 − x5 = 0,
h3 (x̄) = x3 − 10000(x7 − 100) = 0,
h4 (x̄) = x5 − 10000(300 − x7 ) = 0, (17)
h5 (x̄) = x3 − 10000(600 − x8 ) = 0,
h6 (x̄) = x5 − 10000(900 − x9 ) = 0,
h7 (x̄) = x4 ln(x8 − 100) − x4 ln(600 − x7 ) − x8 + x7 + 500 = 0,
h8 (x̄) = x6 ln(x9 − x7 ) − x6 ln(600) − x9 + x7 + 600 = 0,
0  x1  10, 0  x2  200, 0  x3  100, 0  x4  200,
1000  x5  2000000, 0  x6  600, 100  x7  600, 100  x8  600,
100  x9  900.

The optimal values of the control variables obtained, and objective function values
(i.e. minimum (fmin ), mean (fmean ) and standard deviation (fstd ) of the eight DA variants
(DADE, QGDA, MHDA, BMDA, INMDA, HDA, CDA and DA) are provided in Table 1.
Based on the simulation results for this problem, the fmin values of DADE, QGDA,
MHDA, BMDA, INMDA, HDA, CDA and DA are respectively observed as 190.5047,
189.305, 189.3934, 189.3181, 189.3138, 189.3116, 192.599 and 189.3521. In other

Table 1
Simulation results of the HEND problem.

DADE QGDA MHDA BMDA INMDA HDA CDA DA

x1 0.052351 1.44E−06 2.92E−05 4.58E−07 8.65E−08 4.61E−13 0.011093 4.58E−06
x2 15.97275 16.66409 16.66889 16.66681 16.66669 16.66667 16.80441 16.66939
x3 87.17488 66.77105 0.83742 0.01812 0.003442 1.58E−05 57.77835 47.67126
x4 33.51656 99.85326 143.3811 197.9965 198.9125 123.661 124.0067 23.45766
x5 1971712 1999763 1999999 2000000 2000000 2000000 1958385 1999885
x6 595.4896 599.993 599.9197 599.9949 599.999 600 585.1924 599.8195
x7 101.3036 100.0403 100.0001 100 100 100 102.4928 100.0042
x8 599.2642 599.9864 599.9999 600 600 600 599.251 599.9945
x9 701.4332 700.0313 700.0001 700 700 700 704.8981 699.9442
fmin 190.5047 189.305 189.3934 189.3181 189.3138 189.3116 192.599 189.3521
fmean 191.4536 190.2539 190.3423 190.267 190.2627 190.2605 193.5479 190.301
fstd 0.789 0.082 0.915 0.864 0.525 0.727 0.940 0.836
Run time 3.2125 4.60625 7.570313 7.254688 7.303125 5.41875 7.290625 5.2125
FNRT TRank 4.7 3.2 4.7 4.8 4.8 5.1 4.7 4
164 K. Kalita et al.

Fig. 2. Convergence curve for the HEND problem.

words, the QGDA result is respectively 0.63%, 0.047%, 0.007%, 0.005%, 0.003%,
1.710% and 0.025% lower (better) than the simulation-based results obtained using
DADE, MHDA, BMDA, INMDA, HDA, CDA and DA. Table 1 also shows that DADE,
QGDA, MHDA, BMDA, INMDA, HDA, CDA and DA provide the corresponding fmean
values as 191.4536, 190.2539, 190.3423, 190.267, 190.2627, 190.2605, 193.5479 and
190.301 respectively without violating any of constraints. Thus, the resulting benefit for
QGDA is 0.627%, 0.046%, 0.007%, 0.005%, 0.003%, 1.702% and 0.025% as compared
to that obtained from DADE, MHDA, BMDA, INMDA, HDA, CDA and DA respec-
tively. On the other hand, the fstd value for QGDA is noticed to be 0.082, which is lower
by 89.607%, 91.038%, 90.509%, 84.381%, 88.721%, 91.277% and 90.191% than the
simulation-based results derived from DADE, MHDA, BMDA, INMDA, HDA, CDA and
DA respectively. Table 1 also shows the Friedman’s ranks for DADE, QGDA, MHDA,
BMDA, INMDA, HDA, CDA and DA as 4.7, 3.2, 4.7, 4.8, 4.8, 5.1, 4.7 and 4 respec-
tively. Thus, based on the Friedman’s rank test (FNRT) at 95% significance level, the
ranking of the eight DA variants can be derived as QGDA > DA > DADE > MHDA
> CDA > BMDA > INMDA > HDA. It is also interesting to note that according to the
average run time, DADE is 30.258%, 57.565%, 55.718%, 56.012%, 40.715%, 55.937%
and 38.369% faster than QGDA, MHDA, BMDA, INMDA, HDA, CDA and DA respec-
tively. Although DADE is superior to QGDA, MHDA, BMDA, INMDA, HDA, CDA,
and DA with respect to computational burden, QGDA is the second best in compu-
tational time. Figure 2 depicts the convergence curves of all the DA variants for this
problem. Based on Fig. 2, it can be unveiled that DADE has a convergence advantage,
which can find out a better solution with faster speed as compared to other DA vari-
ants. With respect to convergence rate, the considered DA variants can be ranked as
DADE > HDA > BMDA > DA > CDA > MHDA > QGDA > INMDA.
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 165

4.2. Case Study 2: Optimal Operation of Alkylation Unit (OOAU)

The basic objective of the OOAU problem (Sauer et al., 1964) (containing seven variables
and 14 inequality constraints) is to maximize the octane number of olefin feed in the
presence of acid. The minimization type of the OOAU problem can be mathematically
stated as shown below:

f (x̄) = 0.035x1 x6 + 1.715x1 + 10.0x2 + 4.0565x3 − 0.063x3 x5 ,

g1 (x̄) = 0.0059553571x62 x1 + 0.88392857x3 − 0.1175625x6 x1 − x1  0,
g2 (x̄) = 1.1088x1 + 0.1303533x1 x6 − 0.0066033x1 x62 − x3  0,
g3 (x̄) = 6.66173269x62 − 56.596669x4 + 172.39878x5 − 10000 − 191.20592x6  0,
g4 (x̄) = 1.08702x6 − 0.3762x62 + 0.32175x4 + 56.85075 − x5  0,
g5 (x̄) = 0.006198x7 x4 x3 + 2462.3121x2 − 25.125634x2 x4 − x3 x4  0,
g6 (x̄) = 161.18996x3 x4 + 5000.0x2 x4 − 489510.0x2 − x3 x4 x7  0,
g7 (x̄) = 0.33x7 x4 + 44.333333  0,
g8 (x̄) = 0.022556x5 − 1.0x2 − 0.007595x7  0, (18)
g9 (x̄) = 0.00061x3 − 1.0 − 0.0005x1  0,
g10 (x̄) = 0.819672x1 − x3 + 0.819672  0,
g11 (x̄) = 24500.0x2 − 250.0.0x2 x4 − x3 x4  0,
g12 (x̄) = 1020.4082x2 x4 + 1.2244898x3 x4 − 100000x2  0,
g13 (x̄) = 6.25x1 x6 + 6.25x1 − 7.625x3 − 100000  0,
g14 (x̄) = 1.22x3 − x1 x6 − x1  0,
10000  x1  2000, 0  x2  100, 2000  x3  4000, 0  x4  100,
0  x5  100, 0  x6  20, 0  x7  200.

When this OOAU problem is solved using the eight DA variants, the correspond-
ing values of the optimal control variables, and objective functions with respect to fmin ,
fmean and fstd are derived in Table 2. Using the simulation-based results, it is noticed

Table 2
Simulation results of the OOAU problem.
DADE QGDA MHDA BMDA INMDA HDA CDA DA
x1 1362.7004 1364.9895 1365.0069 1365.0087 1364.4943 1364.8813 1365.009 1365.0091
x2 99.957173 99.99925 99.999969 99.999997 99.997169 99.994546 100 99.999999
x3 2000.1839 2000.0086 2000.0046 2000.0001 2000.328 2000.0167 2000.0009 2000
x4 90.745206 90.740691 90.740725 90.740741 90.741674 90.740325 90.740738 90.740741
x5 91.03223 91.015261 91.015162 91.015122 91.018349 91.015422 91.015123 91.01512
x6 3.307297 3.2787429 3.2786118 3.2785546 3.2857938 3.280122 3.2785563 3.2785504
x7 141.48571 141.46021 141.46006 141.45996 141.46966 141.46005 141.45996 141.45996
fmin −136.97331 −142.65288 −142.70488 −142.71839 −141.13781 −142.43987 −142.71733 −142.71923
fmean −136.18861 −141.86818 −141.92018 −141.93369 −140.35311 −141.65517 −141.93263 −141.93453
fstd 0.005292 0.0008229 0.049782 0.0451094 0.0024717 0.0104315 0.002072 0.0328824
Run time 4.0140625 6.9 9.18125 9.4265625 9.39375 5.9734375 9.515625 5.596875
FNRT Rank 3.4 3.2 5.1 5.2 5.8 5.2 3.9 3.4
166 K. Kalita et al.

Fig. 3. Convergence diagram for the OOAU problem.

that the fmin values for DADE, QGDA, MHDA, BMDA, INMDA, HDA, CDA and
DA are −136.97331, −142.65288, −142.70488, −142.71839, −141.13781, 142.43987,
−142.71733 and −142.71923 respectively. It is revealed that the DA result is −4.195%,
−0.047%, −0.010%, −0.001%, −1.120%, −0.196% and −0.001% lower (better) than
that derived using DADE, QGDA, MHDA, BMDA, INMDA, HDA and CDA respec-
tively. It can also be observed from Table 2 that the applications of DADE, QGDA,
MHDA, BMDA, INMDA, HDA, CDA and DA result in the corresponding fmean val-
ues as −136.18861, −141.86818, −141.92018, −141.93369, −140.35311, 141.65517,
−141.93263 and −141.93453 respectively, without violating any of the constraints. Thus,
the benefit achieved for DA is −4.219%, −0.047%, −0.010%, −0.001%, −1.127%,
−0.197% and −0.001% as compared to that obtained from DADE, QGDA, MHDA,
BMDA, INMDA, HDA and CDA respectively. On the other hand, the fstd value for
QGDA is estimated as 0.000823, which is lower by 84.45%, 98.347%, 98.176%, 66.707%,
92.111%, 60.285% and 97.497% than that derived for DADE, QGDA, MHDA, BMDA,
INMDA, CDA and DA respectively. For this example, the Friedman’s ranks for DADE,
QGDA, MHDA, BMDA, INMDA, HDA, CDA and DA are noticed to be 3.4, 3.2, 5.1, 5.2,
5.8, 5.2, 3.9 and 3.4 respectively. Using the Friedman’s rank test at 95% level of signif-
icance, the DA variants under consideration can be sorted as QGDA > DA > DADE >
CDA > MHDA > BMDA > HDA > INMDA. It can be revealed that the average run time
for DADE is 41.825%, 56.280%, 57.418%, 57.269%, 32.801%, 57.816% and 28.280%
faster than QGDA, MHDA, BMDA, INMDA, HDA, CDA and DA respectively. Thus, with
respect to computational burden, the eight DA variants can be ranked as DADE > DA >
HDA > QGDA > MHDA > INMDA > BMDA > CDA. In Fig. 3, the corresponding con-
vergence curves of the considered DA variants are depicted, which reveal that DA ranks
first, followed by DADE, QGDA, HDA, MMDA, CDA, BMDA and INMDA, with respect
to rate of convergence.
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 167

4.3. Case Study 3: Reactor Network Design (RND)

The RND problem (Ryoo and Sahinidis, 1995) deals with maximization of the concen-
tration of a certain product. It consists of six variables, one inequality and four equality
constraints. Mathematically, this minimization type RND problem can be stated as shown
below:

f (x̄) = x4 ,
h1 (x̄) = k1 x2 x5 + x1 − 1 = 0,
h2 (x̄) = k3 x3 x5 + x1 + x3 − 1 = 0,
h3 (x̄) = k2 x2 x6 − x1 − x2 = 0, (19)
h4 (x̄) = k4 x4 x6 + x2 − x1 + x4 − x3 = 0,
g1 (x̄) = x50.5 + x60.5  4,
0  x1 , x2 , x3 , x4  1, 0.00001  x5 , x6  16,

where k3 = 0.0391908, k4 = 0.9k3 , k1 = 0.09755988 and k2 = 0.99k1 .

The simulation-based results for this problem determine the optimal values of the
control variables, and objective functions (i.e. fmin , fmean and fstd ) of the DA variants
under consideration, as provided in Table 3. It can be unveiled from this table that the
fmin value of HDA is respectively 1.231%, −3.072%, −3.500%, −3.489%, −0.967%,
0.000%, −3.498% and −1.408% lower (better) than that obtained using DADE, QGDA,
MHDA, BMDA, INMDA, CDA and DA respectively. Similarly, with respect to fmean
values, the resulting benefit for DA is 0.0767, 0.0952, 0.1083, 0.2252, 0.2892, 0.1374
and 0.1772 as compared to that derived from DADE, QGDA, MHDA, BMDA, INMDA,
HDA and CDA respectively. On the other hand, the fstd value for INMDA is estimated
as 0.011053, which is 93.439%, 93.602%, 93.967%, 90.739%, 93.78%, 93.603% and
92.1252% lower than that obtained from DADE, QGDA, MHDA, BMDA, HDA, CDA
and DA respectively. The corresponding Friedman’s ranks for DADE, QGDA, MHDA,
BMDA, INMDA, HDA, CDA and DA are 4, 3.7, 4.3, 4.5, 7.4, 4.8, 4.8 and 2.5 respec-
tively. Thus, based on the Friedman’s rank test at 95% significance level, the considered
DA variants can be sorted as DA > QGDA > DADE > MHDA > BMDA > HDA >

Table 3
Simulation results of the RND problem.

DADE QGDA MHDA BMDA INMDA HDA CDA DA

x1 0.9999763 0.9999369 0.3944072 0.3919993 0.9945312 0.9999841 0.3940459 0.9976073
x2 0.4354706 0.4665822 0.3943078 0.3918984 0.4360605 0.398256 0.3939394 0.4420572
x3 2.495E−09 8.385E−09 0.3746106 0.3746492 0.0054414 0.0001139 0.374615 0.0024281
x4 0.3832139 0.3763675 0.3748098 0.3748497 0.384213 0.3879295 0.3748192 0.3825421
x5 0.0025721 0.0006354 15.739914 15.899662 0.1285526 0.002037 15.764036 0.0540139
x6 13.419795 11.83318 1.037E−05 2.24E−05 13.26011 15.640824 0.0001726 13.009488
fmin −0.383214 −0.376367 −0.374810 −0.374850 −0.384213 −0.387929 −0.374819 −0.382542
fmean −0.2162199 −0.1976946 −0.1846461 −0.0677247 −0.0037783 −0.1555713 −0.1157014 −0.292938
fstd 0.1684727 0.1727503 0.1832099 0.1193487 0.0110534 0.1777097 0.1727922 0.1403655
Run time 2.68125 4.9203125 8.4046875 8.440625 4.9046875 6.065625 8.0796875 8.4
FNRT Rank 4 3.7 4.3 4.5 7.4 4.8 4.8 2.5
168 K. Kalita et al.

Fig. 4. Convergence curve for the RND problem.

CDA > INMDA. However, in terms of computational time, these eight algorithms can be
ranked as DADE > INMDA > QGDA > HDA > CDA > DA > MHDA > BMDA. Fig-
ure 4 exhibits the convergence curves of the DA variants for this problem. Thus, based on
this figure, it can be concluded that HDA has a superior convergence advantage, helping in
searching out a better solution with a faster speed, followed by DADE, MHDA, INMDA,
CDA, QGDA, BMDA and DA.

4.4. Case Study 4: Haverly’s Pooling Problem (HPP)

This HPP problem (Floudas and Pardalos, 1990) is of maximization type, containing nine
variables, two inequality and four equality constraints. Mathematically, the HPP problem
can be defined as below:

f (x̄) = 9x1 + 15x2 − 6x3 − 16x4 − 10(x5 + x6 ),

h1 (x̄) = x7 + x8 − x4 − x3 = 0,
h2 (x̄) = x1 − x5 − x7 = 0,
h3 (x̄) = x2 − x6 − x8 = 0,
h4 (x̄) = x7 x9 + x8 x9 − 3x3 − x4 = 0, (20)
g1 (x̄) = x7 x9 + 2x5 − 2.5x1  0,
g2 (x̄) = x8 x9 + 2x6 − 1.5x2  0,
0  x1 , x3 , x4 , x5 , x6 , x8  100, 0  x2 , x7 , x9  200.

This maximization problem (converted first to minimization type) is now solved using
the eight DA variants along with determination of the optimal values of the control vari-
ables, and objective functions (with respect to fmin , fmean and fstd ), as shown in Table 4.
Based on the derived results, it can be noticed that with respect to fmin value, the MHDA
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 169

Table 4
Simulation results of the HPP problem.

DADE QGDA MHDA BMDA INMDA HDA CDA DA

x1 0.0001505 9.983E−05 0.0001034 1.127E−05 0.9561836 0.0181292 0.0001372 1.9123671
x2 199.99996 199.99997 199.99999 199.99927 199.19222 199.93562 199.99961 198.38445
x3 9.63E−05 0.0001411 0.0001513 9.324E−06 0 0 2.284E−06 0
x4 99.999692 99.999637 99.999648 99.999939 99.609244 99.999977 99.999819 99.218489
x5 5.243E−05 1.227E−06 3.414E−06 2.773E−06 0.9438356 0.0176439 0.000137 1.8876711
x6 99.999991 99.999994 99.99999 99.999345 99.595327 99.936127 99.999653 99.190654
x7 8.313E−06 1.504E−06 5.335E−07 1.794E−05 0.012348 0.0004854 5.582E−07 0.0246959
x8 99.999875 99.999876 99.999899 99.999928 99.596896 99.999491 99.999921 99.193792
x9 1.0000004 1.0000009 1.000001 1.0000002 0.9999999 1 0.9999992 0.9999999
fmin −400.00475 −400.00546 −400.00555 −399.99661 −397.34948 −399.66011 −400.00041 −394.69895
fmean −399.04995 −399.05066 −399.05075 −399.04181 −396.39468 −398.70531 −399.04561 −393.74415
fstd 0.3171725 0.3181686 0.2615975 0.2643702 0.3889547 0.552458 0.3975508 0.2672359
Run time 2.0260417 4.109375 7.1916667 4.1302083 7.1614583 4.98125 7.1458333 7.1791667
FNRT Rank 4.10 3.77 5.00 4.77 4.93 5.27 4.33 3.83

Fig. 5. Convergence curve for the HPP problem.

result is −0.0002%, −0.00002%, −0.0022%, −0.6684%, −0.0864%, −0.0013% and

−1.3445% lower than that obtained with DADE, QGDA, BMDA, INMDA, HDA, CDA
and DA respectively. Similarly, with respect to fmean value, the resulting benefit for MHDA
is −0.00020%, −0.00002%, −0.00224%, −0.67006%, −0.08664%, −0.00129% and
−1.34773% as compared to that obtained using DADE, QGDA, BMDA, INMDA, HDA,
CDA and DA respectively. Based on the simulation-based results, the fstd value for MHDA
is estimated as 0.261598, which is 17.52%, 17.78%, 1.05%, 32.74%, 52.65%, 34.2% and
2.11% lower than that derived using DADE, QGDA, BMDA, INMDA, HDA, CDA and
DA respectively. Table 4 also shows the results of the Friedman’s rank test at 95% signif-
icance level, which lead to the ranking of the considered DA variants as QGDA > DA >
DADE > CDA > BMDA > INMDA > MHDA > HDA. However, with respect to compu-
tational burden, these algorithms can be sorted as DADE > QGDA > BMDA > HDA >
CDA > INMDA > DA > MHDA. Figure 5 shows the corresponding convergence dia-
gram of all the eight DA variants for this problem. Based on this figure, the ranking of the
170 K. Kalita et al.

algorithms is noted as DADE > MHDA > DA > QGDA > HDA > CDA > INMDA >
BMDA.

5. Fuzzy MARCOS-Based Ranking of the DA Variants

A summarized version of the rankings of the eight DA variants with respect to computa-
tional time (T), Friedman’s rank based on the derived optimal solutions (F) and conver-
gence rate (C) for the four case studies under consideration is provided in Table 5. It can
be interestingly noticed that for the four industrial chemical process problems, there are
some discrepancies in the optimization performance of the eight DA variants in respect
of computational time, Friedman’s rank based on the optimal solutions and convergence
rate. For example, in case of the HEND problem, DADE ranks best in terms of computa-
tional time and convergence rate, but is inferior to QGDA and classical DA with respect
to the optimal solution obtained. Thus, there is an ardent need to holistically analyse these
algorithms and their performance on multiple case studies based on different evaluation
criteria. Moreover, if the obtained rankings are directly aggregated, it would mean that
equal importance is assigned to each of the criteria which would be an oversimplification
of the problem. Thus, a fuzzy scale for assigning relative importance to each criterion is
considered, as provided in Table 6.
Five experts (decision makers) are subsequently asked to provide their opinions on
the importance on the three evaluation criteria using the fuzzy linguistic scale. Table 7
shows the assigned importance for each criterion by each expert and the corresponding
triangular fuzzy number. It can be observed that all the experts deem information related
to the optimal solution (FNRT criterion) as relatively the most important one. Based on
the aggregation of the triangular fuzzy numbers for each of the criteria for all the experts,
the corresponding fuzzy criteria weights are obtained as (3.8, 4.6, 5.8), (5.8, 7, 7.8) and
(3.8, 5, 5.8) for T, F and C respectively.
It should be noted that only the aggregated values of T, F and C in Table 5 constitute
the decision matrix. Thus, the initial decision matrix has an 8 × 3 format (Table 8). It is
normalized using equations (5) and (6). The normalized decision matrix is presented in

Table 5
Summary of performance of the DA variants on the four case studies.

Problem HEND OOAU RND HPP Aggregated

Criteria T F C T F C T F C T F C T F C
DADE 1 3 1 1 2 2 1 3 2 1 3 1 1.00 2.75 1.50
QGDA 2 1 7 4 1 3 3 2 6 2 1 4 2.75 1.25 5.00
MHDA 8 3 6 5 5 5 7 4 3 8 7 2 7.00 4.75 4.00
BMDA 5 6 3 7 6 7 8 5 7 3 5 8 5.75 5.50 6.25
INMDA 7 6 8 6 8 8 2 8 4 6 6 7 5.25 7.00 6.75
HDA 4 8 2 3 6 4 4 6 1 4 8 5 3.75 7.00 3.00
CDA 6 3 5 8 4 6 5 6 5 5 4 6 6.00 4.25 5.50
DA 3 2 4 2 2 1 6 1 8 7 2 3 4.50 1.75 4.00
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 171

Table 6
Fuzzy scale considered in this paper.

Linguistic term for Symbol Triangular fuzzy

criteria importance number
Extremely Poor EP (1, 1, 1)
Very Poor VP (1, 1, 3)
Poor P (1, 3, 3)
Medium Poor MP (3, 3, 5)
Medium M (3, 5, 5)
Medium Good MG (5, 5, 7)
Good G (5, 7, 7)
Very Good VG (7, 7, 9)
Extremely Good EG (7, 9, 9)

Table 7
Importance assigned to each criterion by the experts.

Decision Linguistic term Triangular fuzzy number

maker T F C T F C
Expert 1 VG EG G (7, 7, 9) (7, 9, 9) (5, 7, 7)
Expert 2 MG G VP (5, 5, 7) (5, 5, 7) (1, 1, 3)
Expert 3 P MG M (1, 3, 3) (5, 5, 7) (3, 5, 5)
Expert 4 MP G MG (3, 3, 5) (5, 7, 7) (5, 7, 7)
Expert 5 M VG G (3, 5, 5) (7, 7, 9) (5, 7, 7)

Table 8
Decision matrix and its normalization.

Problem Decision matrix Normalized decision matrix

Criteria T F C T F C
DADE 1.00 2.75 1.50 1.0000 0.4545 1.0000
QGDA 2.75 1.25 5.00 0.3636 1.0000 0.3000
MHDA 7.00 4.75 4.00 0.1429 0.2632 0.3750
BMDA 5.75 5.50 6.25 0.1739 0.2273 0.2400
INMDA 5.25 7.00 6.75 0.1905 0.1786 0.2222
HDA 3.75 7.00 3.00 0.2667 0.1786 0.5000
CDA 6.00 4.25 5.50 0.1667 0.2941 0.2727
DA 4.50 1.75 4.00 0.2222 0.7143 0.3750
Anti-ideal (AI) solution 1.00 1.25 1.50 1.00 1.00 1.00
Ideal (ID) solutions 7.00 7.00 6.75 0.1429 0.1786 0.2222

Table 8. Sample calculations of normalization are shown below:

xid 1 1.25 1.5

nDADE,T = = = 1; nDADE,F = = 0.45; nDADE,C = = 1;
xij 1 2.75 1.5
1 1.25 1.5
nQGDA,T = = 0.36; nQGDA,F = = 1; nQGDA,C = = 0.3.
2.75 1.25 5.0
The fuzzy weighted normalized decision matrix is presented in Table 9 along with
the computed values of S̃i parameter. Sample calculations for fuzzy weighted normalized
172 K. Kalita et al.

Table 9
Fuzzy-weighted normalized decision matrix.

Alternative T F C S̃i
AI (0.543, 0.657, 0.827) (1.036, 1.25, 1.393) (0.844, 1.111, 1.289) (2.423, 3.018, 3.510)
DADE (3.800, 4.600, 5.800) (2.636, 3.182, 3.545) (3.800, 5.000, 5.800) (10.236, 12.782, 15.145)
QGDA (1.382, 1.673, 2.109) (5.800, 7.000, 7.800) (1.140, 1.500, 1.740) (8.322, 10.173, 11.649)
MHDA (0.543, 0.657, 0.829) (1.526, 1.842, 2.053) (1.425, 1.875, 2.175) (3.494, 4.374, 5.056)
BMDA (0.661, 0.800, 1.009) (1.318, 1.591, 1.773) (0.912, 1.200, 1.392) (2.891, 3.591, 4.173)
INMDA (0.724, 0.876, 1.105) (1.036, 1.250, 1.393) (0.844, 1.111, 1.289) (2.604, 3.237, 3.786)
HDA (1.013, 1.227, 1.547) (1.036, 1.250, 1.393) (1.900, 2.500, 2.900) (3.949, 4.977, 5.839)
CDA (0.633, 0.767, 0.967) (1.706, 2.059, 2.294) (1.036, 1.364, 1.582) (3.376, 4.189, 4.843)
DA (0.844, 1.022, 1.289) (4.143, 5.000, 5.571) (1.425, 1.875, 2.175) (6.412, 7.897, 9.035)
ID (3.800, 4.600, 5.800) (5.800, 7.000, 7.800) (3.800, 5.000, 5.800) (13.400, 16.600, 19.400)

decision matrix using equation (7) are shown below:


ṽDADE,T = nij × wjl , nij × wjm , nij × wju = (1 × 3.8, 1 × 4.6, 1 × 5.8)
= (3.8, 4.6, 5.8),
ṽDADE,F = (0.4545 × 5.8, 0.4545 × 7, 0.4545 × 7.8) = (2.636, 3.182, 3.545),
ṽDADE,C = (1 × 3.8, 1 × 5, 1 × 5.8) = (3.8, 5, 5.8),
ṽQGDA,T = (0.3636 × 3.8, 0.3636 × 4.6, 0.3636 × 5.8) = (1.382, 1.673, 2.109),
ṽQGDA,F = (1 × 5.8, 1 × 7, 1 × 7.8) = (5.8, 7, 7.8),
ṽQGDA,C = (0.3 × 3.8, 0.3 × 5, 0.3 × 5.8) = (1.14, 1.5, 1.74).

Similarly, sample calculations for S̃i parameter using equation (10) are shown below:

n
S̃DADE = ṽij = (3.8 + 2.636 + 3.8, 4.6 + 3.182 + 5, 5.8 + 3.545 + 5.8)
j =1

= (10.236, 12.782, 15.145),

S̃QGDA = (1.382 + 5.8 + 1.14, 1.673 + 7 + 1.5, 2.109 + 7.8 + 1.74)
= (8.322, 10.173, 11.649).

The corresponding fuzzy values of utility degree and utility function are subsequently
calculated for each of the alternatives, as exhibited in Table 10. Sample calculations for
K̃i− using equation (8) are shown below:

− S̃i sil sim siu 10.236 12.782 15.145

K̃DADE = = u , sm l = , ,
S̃ai sai ai sai 3.510 3.018 2.423
= (2.916, 4.235, 6.251),
− 8.322 10.173 11.649
K̃QGDA = , , = (2.371, 3.370, 4.808).
3.510 3.018 2.423
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 173

Table 10
Utility degree and utility function of the DA variants.

Alternative Utility degree Utility function

K̃i− K̃i+ f (K̃i− ) f (K̃i+ )
DADE (2.916, 4.235, 6.251) (0.528, 0.770, 1.130) (0.103, 0.150, 0.220) (0.567, 0.824, 1.216)
QGDA (2.371, 3.370, 4.808) (0.429, 0.613, 0.869) (0.083, 0.119, 0.169) (0.461, 0.656, 0.935)
MHDA (0.995, 1.449, 2.087) (0.180, 0.263, 0.377) (0.035, 0.051, 0.073) (0.194, 0.282, 0.406)
BMDA (0.824, 1.190, 1.722) (0.149, 0.216, 0.311) (0.029, 0.042, 0.067) (0.160, 0.231, 0.335)
INMDA (0.742, 1.073, 1.563) (0.134, 0.195, 0.283) (0.026, 0.038, 0.055) (0.144, 0.209, 0.304)
HDA (1.125, 1.649, 2.410) (0.204, 0.300, 0.436) (0.040, 0.058, 0.085) (0.219, 0.321, 0.469)
CDA (0.962, 1.388, 1.999) (0.174, 0.252, 0.361) (0.034, 0.049, 0.070) (0.187, 0.270, 0.389)
DA (1.827, 2.616, 3.729) (0.330, 0.476, 0.674) (0.064, 0.092, 0.131) (0.355, 0.509, 0.725)

Similarly, sample calculations for K̃i+ using equation (9) are shown below:

+ S̃i sil sim siu 10.236 12.782 15.145

K̃DADE = = u , sm l = , ,
S̃id sid id sid 19.4 16.6 13.4
= (0.528, 0.770, 1.130),
+ 8.322 10.173 11.649
K̃QGDA = , , = (0.429, 0.613, 0.869).
19.4 16.6 13.4

The T̃i and dfcrisp are calculated using equations (11) and (14) respectively:
 
t˜DADE = til , tim , tiu = K̃i− ⊕ K̃i+ = ki−l + ki+l , ki−m + ki+m , ki−u + ki+u
= (2.916 + 0.528, 4.235 + 0.77, 6.251 + 1.13)
= (3.444, 5.005, 7.381),
l m u 
t˜QGDA = ti , ti , ti = K̃i− ⊕ K̃i+ = ki−l + ki+l , ki−m + ki+m , ki−u + ki+u
= (2.371 + 0.429, 3.37 + 0.613, 4.808 + 0.869)
= (2.799, 3.983, 5.677),
l + 4m + u 3.444 + (4 × 5.005) + 7.381
dfcrisp = = = 5.14.
6 6

The f (K̃i+ ) and f (K̃i− ) are computed using equation (12) and (13), as shown below:

 + K̃i− ki−l k −m k −u 2.916 4.235 6.251

f K̃DADE = = , i , i = , ,
dfcrisp dfcrisp dfcrisp dfcrisp 5.14 5.14 5.14

= (0.567, 0.824, 1.216),

 + 2.371 3.37 4.808

f K̃QGDA = , , = (0.461, 0.656, 0.935),
5.14 5.14 5.14
174 K. Kalita et al.

Table 11
Defuzzified utility degree, utility function and ranks of the DA variants.

(1−f (Ki− )) (1−f (Ki+ ))

Alternative Ki− Ki+ f (Ki− ) f (Ki+ ) f (Ki ) Rank
f (Ki− ) f (Ki+ )
DADE 4.3510 0.7896 0.1536 0.8464 5.5101 0.1815 0.7682 1
QGDA 3.4433 0.6249 0.1216 0.6698 7.2260 0.4929 0.4666 2
MHDA 1.4799 0.2686 0.0522 0.2879 18.1402 2.4737 0.0809 5
BMDA 1.2175 0.2210 0.0430 0.2368 22.2653 3.2224 0.0543 7
INMDA 1.0991 0.1995 0.0388 0.2138 24.7705 3.6770 0.0441 8
HDA 1.6884 0.3064 0.0596 0.3284 15.7763 2.0447 0.1060 4
CDA 1.4187 0.2575 0.0501 0.2760 18.9661 2.6236 0.0742 6
DA 2.6703 0.4846 0.0943 0.5194 9.6075 0.9251 0.2736 3

 − K̃i+ ki+l k +m k +u 0.528 0.77 1.13

f K̃DADE = = , i , i = , ,
dfcrisp dfcrisp dfcrisp dfcrisp 5.14 5.14 5.14

= (0.103, 0.150, 0.220),

 − 0.429 0.613 0.869

f K̃QGDA = , , = (0.083, 0.119, 0.169).
5.14 5.14 5.14

After deriving the utility degree and utility function for each alternative, their values
are finally defuzzified. These defuzzified values of utility degree and utility function along
with the final rankings of the alternatives are provided in Table 11. The Ki− and Ki+ are
computed as follows:

− 2.916 + (4 × 4.235) + 6.251

KDADE = = 4.351,
6
− 2.371 + (4 × 3.37) + 4.808
KQGDA = = 3.443,
6
+ 0.528 + (4 × 0.77) + 1.13
KDADE = = 0.7896,
6
+ 0.429 + (4 × 0.613) + 0.869
KQGDA = = 0.6249.
6

The f (Ki− ) and f (Ki+ ) are computed as follows:

 − 0.103 + (4 × 0.15) + 0.22

f KDADE = = 0.1536,
6
 − 0.083 + (4 × 0.119) + 0.169
f KQGDA = = 0.1216,
6
 + 0.567 + (4 × 0.824) + 1.216
f KDADE = = 0.8464,
6
 + 0.461 + (4 × 0.656) + 0.935
f KDADE = = 0.6698.
6
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 175

Similarly, the f (Ki ) is computed using equation (16):

Ki+ + Ki− 0.7896 + 4.3510

f (KDADE ) = = = 0.7682,
1−f (Ki+ ) 1−f (Ki− ) 1+ (1−0.8464)
+ (1−0.1536)
1+ f (Ki+ )
+ f (Ki− )
0.8464 0.1536

0.6249 + 3.4433
f (KQGDA ) = (1−0.6698) (1−0.1216)
= 0.4666.
1+ 0.6698 + 0.1216

Thus, based on the considered industrial chemical process optimization problems, the
application of fuzzy MARCOS method leads to relative ranking of the eight DA variants
as DADE > QGDA > DA > HDA > MHDA > CDA > BMDA > INMDA.

6. Conclusions

In the last decade, a plethora of optimization algorithms has been developed by the re-
searchers to solve a variety of complex problems. Among various application fields, in-
dustrial process optimization is a realistic application area where an optimized solution
can directly lead to real-world benefits. In this paper, the performance of eight different
variants of DA is comprehensively studied based on four complex industrial chemical
process optimization case studies. Evaluation of the considered DA variants is carried out
from the standpoint of convergence criterion, time intensiveness and quality of the solu-
tion obtained. The quality of the solution is assessed while measuring the best solution
derived, mean best solution obtained and dispersion of the derived solutions on 30 re-
peated trials. To amalgamate all this information on the solution quality, Friedman’s test
ranks are also computed. Finally, employing a group decision-making approach under
fuzzy environment, the information derived from convergence criterion, time intensive-
ness and solution quality is translated into a relative ranking of the eight DA variants as
DADE > QGDA > DA > HDA > MHDA > CDA > BMDA > INMDA. It can be in-
terestingly noted that despite its simplicity, DA outperforms many of its better endowed
variants. The derived observations would thus help the future researchers in identifying
the most promising DA variants. Moreover, the comprehensive methodology followed to
evaluate the optimization techniques can also be replicated by the researchers for analysis
of other algorithms as well.
However, despite the comprehensiveness of the study, these findings also come with
caveats. The scalability of the tested algorithms and the computational resources required
are potential limitations, as is the transferability of the current results to other, perhaps
larger-scale industrial contexts. These factors may influence the broader applicability of
the conclusions and are critical considerations for future research endeavours.
In terms of future scope, an expansion of this research to include a broader array of
DA subtypes, such as those enhanced through hybridization with grey wolf optimization,
genetic algorithms, and binary dragonfly improved particle swarm optimization can be un-
dertaken. Furthermore, the potential of multi-objective DA variations remains an enticing
176 K. Kalita et al.

prospect for further investigations. In light of this study’s scope and its constraints, partic-
ularly the length of this paper, a comprehensive discussion on every existing DA variant
was not feasible. Yet, this constraint opens the door for future work that can explore these
additional variants, ideally leading to the development of more refined, context-specific
optimization tools. It is hoped the methodological rigour and the analytical framework
presented herein will not only inform but also inspire subsequent research in this domain.

References

Abedi, M., Gharehchopogh, F.S. (2020). An improved opposition based learning firefly algorithm with dragonfly
algorithm for solving continuous optimization problems. Intelligent Data Analysis, 24(2), 309–338.
Abualigah, L., Diabat, A., Mirjalili, S., Abd Elaziz, M., Gandomi, A.H. (2021). The arithmetic optimization
algorithm. Computer Methods in Applied Mechanics and Engineering, 376, 113609.
Badi, I., Muhammad, L., Abubakar, M., Bakır, M. (2022). Measuring sustainability performance indicators using
FUCOM-MARCOS methods. Operational Research in Engineering Sciences: Theory and Applications, 5(2),
99–116.
Bakır, M., Atalık, Ö. (2021). Application of fuzzy AHP and fuzzy MARCOS approach for the evaluation of
e-service quality in the airline industry. Decision Making: Applications in Management and Engineering,
4(1), 127–152.
Bakır, M., Akan, Ş., Özdemir, E. (2021). Regional aircraft selection with fuzzy PIPRECIA and fuzzy MARCOS:
A case study of the Turkish airline industry. Facta Universitatis, Series: Mechanical Engineering, 19(3),
423–445.
Biswal, S., Sahoo, B.B., Jeet, S., Barua, A., Kumari, K., Naik, B., Pradhan, S. (2023). Experimental investiga-
tion based on MCDM optimization of electrical discharge machined Al-WC–B4 C hybrid composite using
Taguchi-MARCOS method. Materials Today: Proceedings, 74, 587–594.
Blum, C., Roli, A. (2003). Metaheuristics in combinatorial optimization: Overview and conceptual comparison.
ACM Computing Surveys (CSUR), 35(3), 268–308.
Can, U., Alatas, B. (2017). Performance comparisons of current metaheuristic algorithms on unconstrained
optimization problems. Periodicals of Engineering and Natural Sciences, 5(3), 328–340.
Chakraborty, S., Chattopadhyay, R., Chakraborty, S. (2020). An integrated D-MARCOS method for supplier
selection in an iron and steel industry. Decision Making: Applications in Management and Engineering,
3(2), 49–69.
Cheng, L., Wu, X.-h., Wang, Y. (2018). Artificial flora (AF) optimization algorithm. Applied Sciences, 8(3),
329.
Debnath, S., Baishya, S., Sen, D., Arif, W. (2021). A hybrid memory-based dragonfly algorithm with differential
evolution for engineering application. Engineering with Computers, 37, 2775–2802.
Deveci, M., Özcan, E., John, R., Pamucar, D., Karaman, H. (2021). Offshore wind farm site selection using
interval rough numbers based Best-Worst Method and MARCOS. Applied Soft Computing, 109, 107532.
Dhiman, G., Kumar, V. (2019). Seagull optimization algorithm: theory and its applications for large-scale in-
dustrial engineering problems. Knowledge-Based Systems, 165, 169–196.
Faramarzi, A., Heidarinejad, M., Mirjalili, S., Gandomi, A.H. (2020). Marine predators algorithm: a nature-
inspired metaheuristic. Expert Systems with Applications, 152, 113377.
Floudas, C., Ciric, A. (1989). Strategies for overcoming uncertainties in heat exchanger network synthesis. Com-
puters & Chemical Engineering, 13(10), 1133–1152.
Floudas, C.A., Pardalos, P.M. (1990). A Collection of Test Problems for Constrained Global Optimization Al-
gorithms. Springer.
Ghanem, W.A., Jantan, A. (2018). A cognitively inspired hybridization of artificial bee colony and dragonfly
algorithms for training multi-layer perceptrons. Cognitive Computation, 10, 1096–1134.
Holland, J.H. (1992). Genetic algorithms. Scientific American, 267(1), 66–73.
Jawad, F.K., Mahmood, M., Wang, D., Al-Azzawi, O., Al-Jamely, A. (2021). Heuristic dragonfly algorithm for
optimal design of truss structures with discrete variables. Structures, 29, pp. 843–862.
Joshi, M., Ghadai, R.K., Madhu, S., Kalita, K., Gao, X.-Z. (2021). Comparison of NSGA-II, MOALO and
MODA for multi-objective optimization of micro-machining processes. Materials, 14(17), 5109.
 A Fuzzy MARCOS-Based Analysis of Dragonfly Algorithm Variants in Industrial OP 177

Kennedy, J., Eberhart, R. (1995). Particle swarm optimization. In: Proceedings of ICNN’95 – International
Conference on Neural Networks, Vol. 4, pp. 1942–1948.
Khalilpourazari, S., Khalilpourazary, S. (2020). Optimization of time, cost and surface roughness in grind-
ing process using a robust multi-objective dragonfly algorithm. Neural Computing and Applications, 32,
3987–3998.
Kirkpatrick, S., Gelatt Jr., C., Vecchi, M.P. (1983). Optimization by simulated annealing. Science, 200, 671–680.
Mafarja, M., Aljarah, I., Heidari, A.A., Faris, H., Fournier-Viger, P., Li, X., Mirjalili, S. (2018). Binary dragon-
fly optimization for feature selection using time-varying transfer functions. Knowledge-Based Systems, 161,
185–204.
Mirjalili, S. (2015). Moth-flame optimization algorithm: a novel nature-inspired heuristic paradigm. Knowledge-
Based Systems, 89, 228–249.
Mirjalili, S. (2016). Dragonfly algorithm: a new meta-heuristic optimization technique for solving single-
objective, discrete, and multi-objective problems. Neural computing and applications, 27, 1053–1073.
Mirjalili, S., Mirjalili, S.M., Lewis, A. (2014). Grey wolf optimizer. Advances in Engineering Software, 69,
46–61.
Mzili, T., Riffi, M.E., Mzili, I., Dhiman, G. (2022). A novel discrete Rat swarm optimization (DRSO) algorithm
for solving the traveling salesman problem. Decision Making: Applications in Management and Engineering,
5(2), 287–299.
Neshat, M., Sepidnam, G., Sargolzaei, M. (2013). Swallow swarm optimization algorithm: a new method to
optimization. Neural Computing and Applications, 23(2), 429–454.
Reddy, A.S. (2016). Optimization of distribution network reconfiguration using dragonfly algorithm. Journal of
Electrical Engineering, 16(4), 10.
Ryoo, H.S., Sahinidis, N.V. (1995). Global optimization of nonconvex NLPs and MINLPs with applications in
process design. Computers & Chemical Engineering, 19(5), 551–566.
Saremi, S., Mirjalili, S., Lewis, A. (2017). Grasshopper optimisation algorithm: theory and application. Ad-
vances in Engineering Software, 105, 30–47.
Sauer, R., Colville, A., Burwick, C. (1964). Computer points way to more profits. Hydrocarbon Processing,
84(2).
Sayed, G.I., Tharwat, A., Hassanien, A.E. (2019). Chaotic dragonfly algorithm: an improved metaheuristic al-
gorithm for feature selection. Applied Intelligence, 49, 188–205.
Shirani, M.R., Safi-Esfahani, F. (2020). BMDA: applying biogeography-based optimization algorithm and Mex-
ican hat wavelet to improve dragonfly algorithm. Soft Computing, 24(21), 15979–16004.
Sörensen, K., Glover, F. (2013). Metaheuristics. Encyclopedia of Operations Research and Management Science,
62, 960–970.
Sree Ranjini, K.S., Murugan, S. (2017). Memory based hybrid dragonfly algorithm for numerical optimization
problems. Expert Systems with Applications, 83, 63–78.
Stanković, M., Stević, Ž., Das, D.K., Subotić, M., Pamučar, D. (2020). A new fuzzy MARCOS method for road
traffic risk analysis. Mathematics, 8(3), 457.
Stević, Ž., Pamučar, D., Puška, A., Chatterjee, P. (2020). Sustainable supplier selection in healthcare industries
using a new MCDM method: measurement of alternatives and ranking according to Compromise solution
(MARCOS). Computers & Industrial Engineering, 140, 106231.
Xu, J., Yan, F. (2019). Hybrid Nelder–Mead algorithm and dragonfly algorithm for function optimization and
the training of a multilayer perceptron. Arabian Journal for Science and Engineering, 44, 3473–3487.
Yang, X.-S. (2012). Flower pollination algorithm for global optimization. In: International Conference on Un-
conventional Computing and Natural Computation, pp. 240–249, Springer.
Yu, C., Cai, Z., Ye, X., Wang, M., Zhao, X., Liang, G., Chen, H., Li, C. (2020). Quantum-like mutation-induced
dragonfly-inspired optimization approach. Mathematics and Computers in Simulation, 178, 259–289.
178 K. Kalita et al.

K. Kalita is an associate professor at the Mechanical Engineering Department at Vel

Tech University, India. With a dedicated focus on research, K. Kalita specializes in opti-
mizing composite laminated structures. This specialization is complemented by a strong
background in computational mechanics and soft computing techniques, contributing sig-
nificantly to the field through innovative research and academic excellence.
N. Ganesh brings a wealth of experience to his role as a senior associate professor at the
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai
Campus. With a career spanning nearly two decades in teaching, training and research, he
has established himself as an authority in this field. His research interests are diverse and
forward-thinking, encompassing software engineering, agile software development, pre-
diction and optimization techniques, deep learning, image processing and data analytics.
R. Shankar serves as an associate professor in the Department of Computer Science and
Engineering at the Koneru Lakshmaiah Education Foundation in Vaddeswaram, India.
His research interests are software development optimization, deep learning and data an-
alytics.
S. Chakraborty is a distinguished member of the faculty at Jadavpur University’s De-
partment of Production Engineering in India. Renowned for contributions to academia
and research, professor Chakraborty is a regular reviewer for several journals of interna-
tional repute. The research interests of professor Chakraborty encompass a broad range
of topics including operations research, multi-criteria decision making, statistical quality
control and soft computing.