You are on page 1of 14

Journal of Computational Science 63 (2022) 101782

Contents lists available at ScienceDirect

Journal of Computational Science


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

Fuzzy harmonic mean technique for solving fully fuzzy multi-objective


transportation problem
Yadvendra Kacher, Pitam Singh ∗
Department of Mathematics, Motilal Nehru National Institute of Technology Allahabad, Prayagraj, 211004 (U.P.), India

ARTICLE INFO ABSTRACT

Keywords: In this research article, an algorithm is developed using a fuzzy harmonic mean technique to solve a fully fuzzy
Fully fuzzy transportation problem multi-objective transportation problem (FFMOTP). First, the FFMOTP will be formalized mathematically, and
Multi-objective transportation problem then by using fuzzy arithmetic, it will be decomposed into three-level (lower, middle, & upper) multi-objective
Harmonic mean
linear programming problems. After then, by using the fuzzy harmonic mean as a tool, the three-level multi-
Linear programming
objective linear programming problems are converted into three-level single objective linear programming
Fuzzy numbers
problems. Further, these single objective linear programming problems for all the three levels are solved to
give the Combine fuzzy optimal solution. In the end, two numerical examples are solved to elaborate our
proposed algorithm completely and compare the result obtained by our proposed algorithm with the results
obtained using the other methods.

1. Introduction necessary information of the simplex table, such as, which variable will
enter the basis and which variable will leave the basis and optimality
The transportation system is defined as the displacement of goods condition. Later on, various specific methods such as the North-West
or passengers from one place to another. It has always played a sig- Corner Rule Method (NWC), Row Minima Method (RMM), Column
nificant role in the country’s economic and social development. In Minima Method (CMM), Least Cost Method (LCM), Vogel’s Approxi-
today’s highly competitive market scenario, the major challenge for any mation Methods (VAM), etc. developed by different researchers to find
organization is to sustain their customers in the market and provide the initial basic feasible solution to the transportation problem. For
them with cost-effective service quality. When it comes to providing optimality condition, we generally use the Modi Method and Stepping
a quality service, it is essential to maintain the supply and demand Stone Method (Charnes et al. 1954) [6]. Besides these methods, there
requirements in the market. Hence, the transportation model comes are many other commercial and academic packages like linear interac-
into the picture, it provides a powerful framework that ensures the tive and discrete optimization (LINDO), general interactive optimizer
cost and time-effective movement of products, raw materials, labor, (GINO), and TORA, which are available to solve the transportation
machinery, etc. in their respective locations.
problem efficiently.
The transportation problem was first formalized by the French
In general, transportation problems are solved with the assumption
mathematician Gaspard Monge in 1781 [1]. Later on, in 1920, A.N.
that the parameters involved in the problem are known precisely,
Tolstoi [2] was the first who has studied it mathematically. However,
i.e., they are crisp in nature. However, in real-world transportation
the major advancement was made in this field during world war second
problems, it is very difficult to construct a model with crisp parameters
by the Soviet mathematician and economist Leonid Kantrowitch [3].
due to several uncontrollable and uncertain factors. To deal with these
The linear programming formulation of the transportation problem was
uncertain data, the concept of fuzzy set theory was introduced by Zadeh
given by F.L. Hitchcock in 1941 [4]. Further, Koopmans (1951) [5],
Charnes and Cooper (1954) [6], Dentzig (1951) [7] and many other in 1965. Since then, many other researchers have taken their interest
researchers have contributed in the development of the transportation in this direction and developed various techniques and methods based
problem. Since the transportation problem can be expressed as a special on fuzzy concepts to solve transportation problems. The transportation
type of LPP. It can also be solved by the simplex method. However, problem with fuzzy parameters is known as the fuzzy transportation
because of its special mathematical structure, it has been observed problem. Zimmermann [8] showed that an efficient solution could
that the simplex method could be made more effective in getting the be obtained by using fuzzy linear programming techniques, and it

∗ Corresponding authors.
E-mail addresses: yadvendra.2019rma03@mnnit.ac.in (Y. Kacher), pitams@mnnit.ac.in (P. Singh).

https://doi.org/10.1016/j.jocs.2022.101782
Received 9 November 2021; Received in revised form 7 June 2022; Accepted 4 July 2022
Available online 21 July 2022
1877-7503/© 2022 Elsevier B.V. All rights reserved.
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

has further developed into several fuzzy optimization methods for with three objective functions under a fuzzy intuitionistic environment.
solving fuzzy transportation problems. If in transportation problems, all The expected value operator is used to convert the fuzzy model into the
the involved parameters such as transportation cost, available supply, deterministic model, and the weighted Tchebycheff programming and
required demand, and the decision variables are all fuzzy, then the min–max goal programming are used to solve the modeled problem.
formed transportation problem is called the fully fuzzy transportation Ghosh and Roy (2020) [30] considered an important aspect of multi-
problem. Gupta et al. (2011) [9] proposed a new method, namely objective transportation problems as product blending constraints for
Mehar’s method, to find the exact fuzzy optimal solution of a fully transporting raw materials with different purity levels for customer sat-
fuzzy multi-objective transportation problem. Kundu et al. (2013) [10] isfaction. Mondal and Roy (2021) [31], in this study, a multi-objective
proposed two different methods for solving multi-objective multi-item multi-product multi-period two-stage sustainable open and closed-loop
solid transportation problems with fuzzy parameters. The first method supply chain is considered to optimize the economic and social aspects
is the defuzzification method based on fuzzy linear programming, and during the Covid-19 pandemic. A hybrid algorithm is developed for
the second method is based on the concept of minimum of fuzzy cost-saving and pollution-reducing caused by sharing materials, equip-
numbers. Giri et al. (2015) [11] proposed a method to solve a fully ment, and infrastructural development. Kacher and Singh (2021) [32]
fuzzy fixed charged multi-item solid transportation problem. Mozaffari discussed various types of transportation problems and their develop-
(2015) [12] presented an algorithm for solving a fully fuzzy trans- ments in their respective field. This article help researchers by giving
portation problem (FFTP) which transforms the problem into the triple collective information on different transportation model on a single
objective problem and then scalarizes these objectives by using the platform. Mardanya et al. (2021) [33] have focussed more on delivery
weighted method and finally solves this using the simplex method. within the scheduled time. A multi-objective, multi-item just-in-time
Ebrahimnejad and Ali (2015) [13] developed a two-step method for transportation problem is modeled, and a priority-based algorithm is
solving fuzzy transportation problems (FTP). In the first step, FTP is developed to find the Pareto optimal solution. Ghosh et al. (2021)
converted into a linear programming problem (LPP) with fuzzy cost and [34] considered multi-objective fixed charged stochastic transportation
crisp constraints using fuzzy arithmetic. In the second step, obtained problems in an intuitionistic fuzzy environment with membership and
LPP is transformed into three crisp bounded transportation problems non-membership functions. Midya et al. (2021) [35], in this study, a
using decomposition techniques. Dhanasekar et al. (2016) [14] de- multi-objective fractional fixed charged transportation problem is con-
veloped a hybrid fuzzy Hungarian MODI method to solve a fully sidered under a rough environment. Also, a fuzzy chance-constrained
fuzzy multi-objective transportation problem. Singh et al. (2016) [15] rough approximation(FCRA) technique is used to find the more accept-
developed an efficient method for solving MOTP using fuzzy program- able solution to the problem. Fathy and Hassanien (2022) [36] in this
ming, goal programming, and interactive programming techniques all article, a fully fuzzy multi-level multi-objective LPP is considered, and a
together to get a fuzzy compromise solution to the modeled prob- fuzzy harmonic mean technique is used to aggregate multiple objectives
lem. Bharati et al. (2017) [16] defines a new distance function that into a single objective at a different level. At last, all the single objective
measures a degree of deviation between two trapezoidal numbers, LPP is solved to get the fuzzy compromise solution.
which is being used to solve an FFMOTP. Maity and Roy (2017) [17] The rest of this research article is organized as follows; In Section 2,
analyze a situation where the optimum goal function is combined some basic terminologies related to fuzzy set theory and fuzzy harmonic
with the optimal product allocation for distinct routes. A multi-choice mean are introduced. Also, the FFMOTP is formulated at the end of
goal programming is used to solve the formulated problem. Vidhya this section. In Section 3, the proposed method is explained in detail,
and Ganesan (2018) [18] developed an algorithm for the solution of followed by Section 4, where two numerical examples are solved to
FMOTP with time and cost coefficients are trapezoidal fuzzy numbers. support our proposed method. Further, Section 5 contains the results
Mahmoodirad et al. (2018) [19] proposed a method to overcome the and discussion, followed by Section 6, in which the advantages and
shortcomings that have been found in the previously existing meth- disadvantages of this proposed method are discussed. Section 7 includes
ods. Maheswari and Ganesan (2018) [20] developed a new method the main conclusion of this research article.
for solving FFTP in which the involved parameters are presented by
pentagonal fuzzy numbers and used robust ranking techniques to deal
2. Preliminaries
with these fuzzy numbers. Baidya and Bera (2019) [21] formulated a
fully fuzzy solid transportation problem with an additional constraint
In this section, some basic concepts of fuzzy set theory and other
on the total budget of each destination and used a generalized reduced
terminologies are to be defined which will help to understand the
gradient technique (GRT) to solve this problem. Anukokila and Rad-
proposed method.
hakrishnan (2019) [22] used a goal programming approach to solve the
fully fuzzy fractional multi-objective transportation problem. Bagheri
et al. (2020) [23] developed a new fuzzy DEA-based approach to solve 2.1. Some basic definitions of fuzzy sets and fuzzy numbers
FFMOTP. Krishnaveni and Ganesan (2020) [24] discuss a solution for
a fully fuzzy multi-objective fractional transportation problem with Definition 2.1.1. Let 𝑋 be a non-empty universal set, finite or infinite.
the use of triangular fuzzy number without transforming it to a crisp Then a fuzzy set 𝐴̃ on 𝑋 is a set of ordered pair 𝐴̃ = {(𝑥, 𝜇𝐴̃(𝑥))|𝑥 ∈ 𝑋},
equivalent to the existing model. Mishra and Kumar (2020) [25] pro- where 𝜇𝐴̃ ∶ 𝑋 → [0, 1] is a membership function on 𝑋 defined as the
degree of membership of any element 𝑥 of 𝑋 in 𝐴. ̃
posed a new method for solving fully fuzzy transportation problems
and also resolved the drawback of Ebrahimnejad’s method given in
2015 [13]. Mishra and Kumar (2020) [26] also developed a JMD Definition 2.1.2. A fuzzy set 𝐴̃ is said to be convex if for any two
approach for solving unbalanced fully trapezoidal intuitionistic fuzzy points 𝑥 and 𝑦 in 𝑋 and for any 𝜆 ∈ [0, 1], the membership function of
transportation problems. Nishad and Abhishekh (2020) [27] developed 𝐴̃ holds the inequality 𝜇𝐴̃ (𝜆𝑥 + (1 − 𝜆)𝑦) ≥ 𝑚𝑖𝑛{𝜇𝐴̃(𝑥), 𝜇𝐴̃(𝑦)} , ∀ 𝑥, 𝑦 ∈
a new approach for ranking generalized intuitionistic fuzzy numbers 𝑋 & 𝜆 ∈ [0, 1].
and normal intuitionistic fuzzy numbers that meets Wang and Kerre’s
(2001) standard axiom [28]. Also, Sankar Kumar Roy and his team Definition 2.1.3. A fuzzy set 𝐴̃ is said to be normalized if there exit
made a significant contribution to the development of different solution at least one element 𝑥 in 𝑋 such that 𝜇𝐴̃(𝑥) = 1.
algorithms for fuzzy transportation problems (FTP). Various problem
situations are discussed that can occurs in modeling real-life problems Definition 2.1.4. A fuzzy set 𝐴̃ on R with membership function
as a transportation problem. Midya et al. (2020) [29] have considered a 𝜇𝐴̃ ∶ R → [0, 1] is said to be a fuzzy number if it satisfies the following
multi-stage multi-objective fixed charged solid transportation problem criteria:

2
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

1. 𝐴̃ is a convex fuzzy set, i.e. 𝜇𝐴̃ (𝜆𝑥 + (1 − 𝜆)𝑦) ≥ 𝑚𝑖𝑛{𝜇𝐴̃(𝑥), 𝜇𝐴̃(𝑦)}, 2.2.3. Fuzzy Harmonic mean :
∀ 𝑥, 𝑦 ∈ R & 𝜆 ∈ [0, 1], One needs to analyze various information or data collected from
2. 𝐴̃ is a normalized fuzzy set, i.e. there is an 𝑥 ∈ R such that different sources to make the optimum decision in any decision-making
𝜇𝐴̃(𝑥) = 1, problem. Information aggregation is a critical tool or technique to get
3. Its membership function 𝜇𝐴̃(𝑥) is a piecewise continuous. the relevant information about the solution procedure of any decision-
making problem. However, as we know, in real-life decision-making
Definition 2.1.5. A fuzzy number 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) is called a triangular problems, the input data or information is not precisely known. To
fuzzy number (TFN) if its membership function is given by: overcome this situation, there are various ways, among which using
if 𝑥 ≤ 𝑎𝑙 , fuzzy data as input is the most common approach. So, when the input
⎧ 0
⎪ 𝑥−𝑎𝑙 data is a fuzzy quantity,an operator that is used to deal with them is
⎪ 𝑎 −𝑎 if 𝑎𝑙 ≤ 𝑥 ≤ 𝑎𝑚 ,
known as a fuzzy harmonic mean operator. Let 𝑎̃𝑗 = (𝑎𝑙𝑗 , 𝑎𝑚 𝑢
𝑗 , 𝑎𝑗 ) for
𝜇𝐴̃(𝑥) = ⎨ 𝑎𝑚 −𝑥𝑙
𝑢
⎪ 𝑎𝑢 −𝑎𝑚 if 𝑎𝑚 ≤ 𝑥 ≤ 𝑎𝑢 , 𝑗 = 1, 2, … 𝑛 are 𝑛 triangular fuzzy numbers, then the fuzzy harmonic
⎪ mean of 𝑎̃1 , 𝑎̃2 , … , 𝑎̃𝑛 is defined as:
⎩ 0 if 𝑥 ≥ 𝑎𝑢 .
𝑛
𝐹 𝐻𝑀(𝑎̃1 , 𝑎̃2 , … , 𝑎̃𝑛 ) = ∑𝑛
1
Definition 2.1.6. A triangular fuzzy number 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) is said 𝑗=1 𝑎̃𝑗
to be a non-negative (positive) triangular fuzzy number if and only if 𝑛 𝑛 𝑛 𝑛
𝑎𝑙 ≥ 0 (𝑎𝑙 > 0). = ∑𝑛 = ( ∑𝑛 , ∑𝑛 , ∑𝑛 )
1 1 1 1
𝑗=1 (𝑎𝑙𝑗 ,𝑎𝑚 𝑢 𝑗=1 𝑎𝑙 𝑗=1 𝑎𝑚 𝑗=1 𝑎𝑢
𝑗 ,𝑎𝑗 ) 𝑗 𝑗 𝑗
Definition 2.1.7. Two triangular fuzzy numbers 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) and (1)
̃ = (𝑏𝑙 , 𝑏𝑚 , 𝑏𝑢 ) are said to be equal if and only if 𝑎𝑙 = 𝑏𝑙 , 𝑎𝑚 = 𝑏𝑚 , 𝑎𝑢 =
𝐵
𝑏𝑢 .
Remark 1. Since, in this paper, triangular fuzzy numbers are taken
as fuzzy quantities. Thus, the fuzzy harmonic mean is defined for
Definition 2.1.8. Let 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) and 𝐵
̃ = (𝑏𝑙 , 𝑏𝑚 , 𝑏𝑢 ) are two TFNs
triangular fuzzy numbers only. However, one can also find the FHM
and 𝑘 ∈ R. Then, some fundamental arithmetic operations are defined
for other types of fuzzy numbers.
as follows:

• Addition : 𝐴̃ ⊕ 𝐵
̃ = (𝑎𝑙 + 𝑏𝑙 , 𝑎𝑚 + 𝑏𝑚 , 𝑎𝑢 + 𝑏𝑢 ).
2.3. Fully fuzzy multi-objective transportation problem (FFMOTP )
• Subtraction : 𝐴̃ ⊖ 𝐵
̃ = (𝑎𝑙 − 𝑏𝑢 , 𝑎𝑚 − 𝑏𝑚 , 𝑎𝑢 − 𝑏𝑙 ).
• Multiplication : Let 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) be an arbitrary triangular A transportation problem can be of a single objective or multiple
fuzzy number and 𝐵 ̃ = (𝑏𝑙 , 𝑏𝑚 , 𝑏𝑢 ) be a non-negative triangular
objectives depending upon the number of objectives chosen to be
fuzzy number then, optimized by the decision-maker. As it is known, the mathematical
⎧(𝑎𝑙 𝑏𝑙 , 𝑎𝑚 𝑏𝑚 , 𝑎𝑢 𝑏𝑢 ) if 𝑎𝑙 ≥ 0, formulation of this problem consists of objective functions, some set
𝐴̃ ⊙ 𝐵 ̃=⎪
̃ = 𝐴̃𝐵 ⎨(𝑎𝑙 𝑏𝑢 , 𝑎𝑚 𝑏𝑚 , 𝑎𝑢 𝑏𝑢 ) if 𝑎𝑙 < 0 , 𝑎𝑢 ≥ 0,
of constraints, and non-negative restrictions. These are nothing but the
⎪ combination of constant and variable quantities or parameters. If any
⎩(𝑎𝑙 𝑏𝑢 , 𝑎𝑚 𝑏𝑚 , 𝑎𝑢 𝑏𝑙 ) if 𝑎𝑢 < 0. of the constant or variable quantities are taken as fuzzy numbers, that
• Division : Let 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) be an arbitrary triangular fuzzy transportation problem is known as the fuzzy transportation problem
number and 𝐵 ̃ = (𝑏𝑙 , 𝑏𝑚 , 𝑏𝑢 ) ≠ ̃
0 be a non-zero triangular fuzzy (FTP). Suppose all the parameters of TP, such as transportation cost,
number then, available supply, required demand, and decision variables are taken
̃ 𝐵̃ = (𝑎𝑙 ∕𝑏𝑢 , 𝑎𝑚 ∕𝑏𝑚 , 𝑎𝑢 ∕𝑏𝑙 ). as a fuzzy quantity. In that case, the formed transportation problem
𝐴∕
is called the fully fuzzy transportation problem (FFTP). Therefore,
• Multiplication by a constant number : a fully fuzzy multi-objective transportation problem is a problem in
{ which multiple objectives are taken for optimization and also all the
(𝑘𝑎𝑙 , 𝑘𝑎𝑚 , 𝑘𝑎𝑢 ) 𝑖𝑓 𝑘 ≥ 0,
𝑘𝐴̃ = parameters are taken as fuzzy numbers.
(𝑘𝑎𝑢 , 𝑘𝑎𝑚 , 𝑘𝑎𝑙 ) 𝑖𝑓 𝑘 < 0. Mathematical formulation of a fully fuzzy multi-objective trans-
• Order relation [17] : 𝐴̃ ⪯ 𝐵
̃ iff 𝑎𝑙 ≤ 𝑏𝑙 , 𝑎𝑚 ≤ 𝑏𝑚 , 𝑎𝑢 ≤ 𝑏𝑢 . portation problem (FFMOTP):

𝑚 ∑
𝑛
2.2. Ranking of a fuzzy number and fuzzy harmonic mean operator ̃𝑘 (̃
min 𝑍 𝑥) = 𝑐̃𝑖𝑗𝑘 ⊗ 𝑥
̃𝑖𝑗 (𝑘 = 1, 2, … , 𝑝) (2)
𝑖=1 𝑗=1
2.2.1. Ranking of a fuzzy number : subject to
In many fuzzy decision-making problems, the final solution comes ∑𝑛

out to be a fuzzy number. To give preference ordering to these fuzzy ̃𝑖𝑗 ≈ ̃


𝑥 𝑠𝑖 (𝑖 = 1, 2, … , 𝑚), (3)
𝑗=1
solutions, we need a method or technique which can provide a crisp
preference ordering for these fuzzy solutions. This technique or method ∑𝑚
̃𝑖𝑗 ≈ 𝑑̃𝑗
𝑥 (𝑗 = 1, 2, … , 𝑛), (4)
is known as a ranking of a fuzzy number. Here, in this paper, we
𝑖=1
consider the ranking of a triangular fuzzy number as an arithmetic
mean of its components, i.e. if 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) be a fuzzy number, then ̃𝑖𝑗 ⪰ 0 (𝑖 = 1, 2, … , 𝑚 ; 𝑗 = 1, 2, … , 𝑛).
𝑥 (5)
𝑎 + 𝑎𝑚 + 𝑎𝑢
ranking of 𝐴̃ is denoted by (𝐴)
̃ and is defined as (𝐴) ̃ = 𝑙 . Where,
3
𝑚 = total number of sources.
2.2.2. Harmonic mean : 𝑛 = total number of destinations.
The harmonic mean is generally used as a tool to aggregate central 𝑥̃𝑖𝑗 = fuzzy quantity of products that needs to be transported from
tendency data. It is generally defined as the reciprocal of the arith-
source 𝑖 to destination 𝑗.
metic mean of reciprocal of observations, i.e. if 𝑥1 , 𝑥2 , … , 𝑥𝑛 are 𝑛
𝑠𝑖 = fuzzy availability of products at 𝑖𝑡ℎ source.
̃
observations, then harmonic mean of these 𝑛 observations is defined
as: 𝑑̃𝑗 = fuzzy demand of products at 𝑗𝑡ℎ destination.
𝑛 𝑐̃𝑖𝑗𝑘 = fuzzy penalty per unit product transported from source 𝑖 to
𝐻.𝑀.(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = destination 𝑗 associated with the 𝑘𝑡ℎ objective function.
1∕𝑥1 + 1∕𝑥2 + … + 1∕𝑥𝑛

3
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

The above problem (2)–(5) can also be written in the following ∑


𝑚 ∑
𝑛
OR min (𝑣𝑙𝑘 , 𝑣𝑚 𝑢
𝑘 , 𝑣𝑘 ) = min (𝑐𝑖𝑗𝑘𝑙 𝑥𝑙𝑖𝑗 , 𝑐𝑖𝑗𝑘𝑚 𝑥𝑚 𝑘𝑢 𝑢
𝑖𝑗 , 𝑐𝑖𝑗 𝑥𝑖𝑗 )
manner; 𝑥
̃
𝑖=1 𝑗=1

̃𝑘 (̃
min 𝑍 𝑥) = min (𝑓̃11 (̃
𝑥), 𝑓̃12 (̃
𝑥), . . . , 𝑓̃1𝑝 (̃
𝑥)) (6) (𝑘 = 1, 2, … , 𝑝) (16)
𝑥
̃
subject to
subject to the constraints (3)–(5). (7)
∑𝑛

Where, 𝑥𝑙𝑖𝑗 = 𝑠𝑙𝑖 (𝑖 = 1, 2, … , 𝑚), (17)


∑ ∑𝑛 𝑗=1
𝑥) = 𝑚
𝑓̃11 (̃ 𝑖=1 𝑗=1 𝑐̃𝑖𝑗1 ⊗ 𝑥
̃𝑖𝑗 ⎫ ∑𝑛
∑ ∑ ⎪ 𝑥𝑚 𝑚
(18)
̃
𝑓12 (̃ 𝑚 𝑛 2
𝑥) = 𝑖=1 𝑗=1 𝑐̃𝑖𝑗 ⊗ 𝑥 ̃𝑖𝑗 ⎪ 𝑖𝑗 = 𝑠𝑖 (𝑖 = 1, 2, ..., 𝑚),
(8) 𝑗=1

⋮ ⎪ ∑𝑛

̃ ∑𝑚 ∑ 𝑛 𝑝 ⎪ 𝑥𝑢𝑖𝑗 = 𝑠𝑢𝑖 (𝑖 = 1, 2, … , 𝑚), (19)


𝑓1𝑝 (̃𝑥) = 𝑖=1 𝑐̃ ⊗ 𝑥 ̃𝑖𝑗 ⎭
𝑗=1 𝑖𝑗 𝑗=1
̃𝑘 ∈ ℜ𝑝 , 𝑓̃𝑟𝑘 ∈ ℜ𝑝 , 𝑐̃𝑘 ∈ ℜ𝑚∗𝑛 , ̃
Here, 𝑍 𝑠𝑖 ∈ ℜ𝑚 , 𝑑̃𝑗 ∈ ℜ𝑛 , 𝑥
̃𝑖𝑗 ∈ ℜ𝑚∗𝑛 ∑𝑚
𝑖𝑗 𝑥𝑙𝑖𝑗 = 𝑑𝑗𝑙 (𝑗 = 1, 2, ..., 𝑛), (20)
𝑖=1
∑𝑚 ̃ ∑𝑛
Remark 2. If 𝑖=1 ̃ 𝑠𝑖 = 𝑗=1 𝑑𝑗 , then the FFMOTP (2)–(5) is called ∑𝑚

balanced. Otherwise it is called unbalanced. 𝑥𝑚 𝑚


𝑖𝑗 = 𝑑𝑗 (𝑗 = 1, 2, … , 𝑛), (21)
𝑖=1

Remark 3. Without loss of generality, we assume that FFMOTP (2)–(5) ∑𝑚


𝑥𝑢𝑖𝑗 = 𝑑𝑗𝑢 (𝑗 = 1, 2, … , 𝑛), (22)
is a balanced transportation problem. 𝑖=1

𝑥𝑙𝑖𝑗 ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛), (23)


Remark 4. Because there is no physical significance to the negative
quantity of products and the negative transportation cost. As a result, 𝑥𝑚
𝑖𝑗 − 𝑥𝑙𝑖𝑗 ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, ..., 𝑛), (24)
all of the parameters in problems (6)–(8) are considered to be non-
𝑥𝑢𝑖𝑗 − 𝑥𝑚
𝑖𝑗 ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, ..., 𝑛), (25)
negative triangular fuzzy numbers with component-wise order relation,
i.e. if 𝐴̃ = (𝑎𝑙 , 𝑎𝑚 , 𝑎𝑢 ) then 𝑎𝑙 ≤ 𝑎𝑚 ≤ 𝑎𝑢 . Now, in order to show the equivalency of the model (2)–(5) and
̃𝑘 , 𝑓̃𝑟𝑘 , 𝑐̃𝑘 , ̃ (16)–(25), the following theorem is introduced.
Regarding to Remark 4, assume that 𝑍 𝑖𝑗 𝑠𝑖 , 𝑑̃𝑗 𝑎𝑛𝑑 𝑥 ̃𝑖𝑗
̃ 𝑚 𝑢 ̃
are all represented by 𝑍𝑘 = (𝑣𝑘 , 𝑣𝑘 , 𝑣𝑘 ) , 𝑓𝑟𝑘 = (𝑢𝑟𝑘 , 𝑢𝑟𝑘 , 𝑢𝑢𝑟𝑘 ) , 𝑐̃𝑖𝑗𝑘 =
𝑙 𝑙 𝑚
Theorem 2.3.1. Formulation (2)–(5) is equivalent to the formulation
𝑠𝑖 = (𝑠𝑙𝑖 , 𝑠𝑚
((𝑐𝑖𝑗𝑘 )𝑙 , (𝑐𝑖𝑗𝑘 )𝑚 , (𝑐𝑖𝑗𝑘 )𝑚 ) = (𝑐𝑖𝑗𝑘𝑙 , 𝑐𝑖𝑗𝑘𝑙 , 𝑐𝑖𝑗𝑘𝑙 ) , ̃ 𝑢 ̃ 𝑙 𝑚 𝑢
𝑖 , 𝑠𝑖 ) , 𝑑𝑗 = (𝑑𝑗 , 𝑑𝑗 , 𝑑𝑗 ) , 𝑥
̃𝑖𝑗 (16)–(25).
= (𝑥𝑙𝑖𝑗 , 𝑥𝑚 , 𝑥
𝑖𝑗 𝑖𝑗
𝑢 ) respectively. Thus by using fuzzy arithmetic (Defini-

tion 2.1.8), FFMOTP (6)–(8) can be reformulated as follows: Proof. The equivalency of the model (2)–(5) and (16)–(25) is proved in
terms of feasibility and optimality of solution sets. Let 𝐷 and 𝐷∗ are fea-
min (𝑣𝑙𝑘 , 𝑣𝑚 𝑢
𝑘 , 𝑣𝑘 ) = min ((𝑢𝑙11 , 𝑢𝑚 , 𝑢𝑢 ), (𝑢𝑙12 , 𝑢𝑚
11 11
, 𝑢𝑢 ), .
12 12
.. , (𝑢𝑙1𝑝 , 𝑢𝑚 , 𝑢𝑢 ))
1𝑝 1𝑝
(9)
𝑥
̃ sible sets of models (formulations) (2)–(5) and (16)–(25) respectively.
subject to Let 𝑋 = {̃ 𝑥𝑖𝑗 } be the solution of the formulation (2)–(5), then 𝑋 ∈ 𝐷
∑𝑛 iff it satisfies all the constraints from Eq. (3)–(5) and consequently all
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢 𝑙 𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ≈ (𝑠𝑖 , 𝑠𝑖 , 𝑠𝑖 ) (𝑖 = 1, 2, … , 𝑚), (10) the constraints from Eq. (10)–(12). Now by using the definition 2.1.7
𝑗=1
& 2.1.8 on the set of constraints (10)–(12), the following new set of
∑𝑚
constraints are obtained:
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢 𝑙 𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ≈ (𝑑𝑗 , 𝑑𝑗 , 𝑑𝑗 ) (𝑗 = 1, 2, … , 𝑛), (11)
𝑖=1 ∑ 𝑛 ∑𝑛 ∑
𝑛
( 𝑥𝑙𝑖𝑗 , 𝑥𝑚
𝑖𝑗 , 𝑥𝑢𝑖𝑗 ) = (𝑠𝑙𝑖 , 𝑠𝑚 𝑢
𝑖 , 𝑠𝑖 ) (𝑖 = 1, 2, … , 𝑚), (26)
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ⪰ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (12) 𝑗=1 𝑗=1 𝑗=1

Where, ∑
𝑛 ∑
𝑛 ∑
𝑛
( 𝑥𝑙𝑖𝑗 , 𝑥𝑚
𝑖𝑗 , 𝑥𝑢𝑖𝑗 ) = (𝑑𝑗𝑙 , 𝑑𝑗𝑚 , 𝑑𝑗𝑢 ) (𝑗 = 1, 2, … , 𝑛), (27)
∑ ∑𝑛 𝑗=1 𝑗=1 𝑗=1
𝑓̃11 (̃
𝑥) = (𝑢𝑙11 , 𝑢𝑚 , 𝑢𝑢 ) = 𝑚
11 11 𝑖=1
1𝑙 1𝑚 1𝑢 𝑙 𝑚 𝑢
𝑗=1 (𝑐𝑖𝑗 , 𝑐𝑖𝑗 , 𝑐𝑖𝑗 ) ⊗ (𝑥𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 ), ⎫
∑ ∑ ⎪ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛), (28)
𝑓̃12 (̃
𝑥) = (𝑢𝑙12 , 𝑢𝑚 , 𝑢𝑢 ) = 𝑚
12 12 𝑖=1
𝑛 2𝑙 2𝑚 2𝑢 𝑙 𝑚 𝑢
𝑗=1 (𝑐𝑖𝑗 , 𝑐𝑖𝑗 , 𝑐𝑖𝑗 ) ⊗ (𝑥𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 ), ⎪
⎬ According to definition Definition 2.1.7 and non-negative constraints
⋮ ⎪
∑ ∑𝑛 ⎪ (28)), the equations ((26)–(28) is converted into (17)–(25). Since the
𝑥) = (𝑢𝑙 , 𝑢𝑚 , 𝑢𝑢 ) = 𝑚
𝑓̃1𝑝 (̃ 1𝑝 1𝑝 1𝑝
(𝑐 𝑝𝑙 , 𝑐 𝑝𝑚 , 𝑐 𝑝𝑢 ) ⊗ (𝑥𝑙 , 𝑥𝑚 , 𝑥𝑢 ). ⎭
𝑖=1 𝑗=1 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 constraints set (3)–(5) converted into (10)–(10) and (12)–(12) con-
verted into (17)–(25). This means that any solution 𝑋 = {̃ 𝑥𝑖𝑗 } that
(13)
satisfies constraints set (3)–(5) will satisfy the constraints set (17)–(25),
Now, using the fuzzy arithmetic defined in Definitions 2.1.7 and 2.1.8, meaning that 𝑋 ∈ 𝐷∗ . Therefore it concludes that 𝑋 = {̃ 𝑥𝑖𝑗 } ∈ 𝐷
the above formulation can be rewritten as follows: iff 𝑋 ∈ 𝐷∗ . This implies 𝐷 = 𝐷∗ . Hence, the feasible set of both the
(𝑚 𝑛 formulation (2)–(5) and (16)–(25) are the same.
∑∑
𝑙 𝑚 𝑢
min (𝑣𝑘 , 𝑣𝑘 , 𝑣𝑘 ) = min (𝑐𝑖𝑗1𝑙 , 𝑐𝑖𝑗1𝑚 , 𝑐𝑖𝑗1𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ,
Next we show that, the optimal solution of problems (2)–(5) & (16)–
𝑥
̃
𝑖=1 𝑗=1 (25) are also same. For this let 𝑋 ∗ = {̃ 𝑥∗𝑖𝑗 } be the optimal solution

𝑚 ∑
𝑛 of the problem (2)–(5), meaning 𝑋 ∈ 𝐷. Since 𝐷 = 𝐷∗ ⇒ 𝑋 ∗ ∈

(𝑐𝑖𝑗2𝑙 , 𝑐𝑖𝑗2𝑚 , 𝑐𝑖𝑗2𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ), 𝐷∗ . Next, we have to show that 𝑋 ∗ is an optimal solution in 𝐷∗ . On
𝑖=1 𝑗=1
) contrary, let us assume that 𝑋 ∗ is not an optimal solution in 𝐷∗ . Then
∑𝑚 ∑ 𝑛 ∑ ∑𝑛 ̃𝑘 ∑𝑛 ̃𝑘
𝑝𝑙 𝑝𝑚 𝑝𝑢 𝑙 𝑚 𝑢 ∃ another solution 𝑋 ∈ 𝐷∗ such that 𝑚 𝑖=1 𝑗=1 𝐶𝑖𝑗 ⊗ 𝑋 ≤ 𝑗=1 𝐶𝑖𝑗 ⊗
..., (𝑐𝑖𝑗 , 𝑐𝑖𝑗 , 𝑐𝑖𝑗 ) ⊗ (𝑥𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 ) (14) ∗
𝑋 for every (𝑘 = 1, 2, … , 𝑝)
𝑖=1 𝑗=1

𝑚 ∑
𝑛


𝑚 ∑
𝑛 ⇔ (𝐶𝑖𝑗𝑘𝑙 , 𝐶𝑖𝑗𝑘𝑚 , 𝐶𝑖𝑗𝑘𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 )
OR min (𝑣𝑙𝑘 , 𝑣𝑚 ,
𝑘 𝑘 𝑣𝑢
) = min (𝑐𝑖𝑗𝑘𝑙 , 𝑐𝑖𝑗𝑘𝑚 , 𝑐𝑖𝑗𝑘𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 )
𝑖=1 𝑗=1
𝑥
̃
𝑖=1 𝑗=1 ∑𝑚 ∑
𝑛
≤ (𝐶𝑖𝑗𝑘𝑙 , 𝐶𝑖𝑗𝑘𝑚 , 𝐶𝑖𝑗𝑘𝑢 ) ⊗ (𝑥𝑙∗ 𝑚∗ 𝑢∗
𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 )
(𝑘 = 1, 2, … , 𝑝) (15) 𝑖=1 𝑗=1

4
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782


𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛
𝑥𝑚 𝑙
⇔ ( 𝐶𝑖𝑗𝑘𝑙 𝑥𝑙𝑖𝑗 , 𝐶𝑖𝑗𝑘𝑚 𝑥𝑚 𝐶𝑖𝑗𝑘𝑢 𝑥𝑢𝑖𝑗 ) ≤ 𝑖𝑗 ≥ 𝑥𝑖𝑗 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (36)
𝑖𝑗 ,
𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1
The Eqs. (33)–(36) are a crisp multi-objective linear programming
∑𝑚 ∑ 𝑛 ∑𝑚 ∑ 𝑛 ∑𝑚 ∑ 𝑛
problem with 𝑝 objective functions for middle-level. The important
( 𝐶𝑖𝑗𝑘𝑙 𝑥𝑙∗
𝑖𝑗 , 𝐶𝑖𝑗𝑘𝑚 𝑥𝑚∗
𝑖𝑗 , 𝐶𝑖𝑗𝑘𝑢 𝑥𝑢∗
𝑖𝑗 )
𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 point that needs to be noticed here is that Eq. (36) which guarantees

𝑚 ∑ 𝑛 ∑
𝑚 ∑ 𝑛 ∑
𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛 that the solutions obtained for this problem will always be greater than
⇔ 𝐶𝑖𝑗𝑘𝑙 𝑥𝑙𝑖𝑗 ≤ 𝐶𝑖𝑗𝑘𝑙 𝑥𝑙∗
𝑖𝑗 , 𝐶𝑖𝑗𝑘𝑙 𝑥𝑚
𝑖𝑗 ≤ 𝐶𝑖𝑗𝑘𝑙 𝑥𝑚∗
𝑖𝑗 ,
or equal to the solutions obtained in lower-level problem.
𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 U - upper level (U-MOLPP)
∑𝑚 ∑ 𝑛 ∑𝑚 ∑ 𝑛
𝐶𝑖𝑗𝑘𝑙 𝑥𝑢𝑖𝑗 ≤ 𝐶𝑖𝑗𝑘𝑙 𝑥𝑢∗
𝑖𝑗
min (𝑢𝑢11 , 𝑢𝑢12 , … , 𝑢𝑢1𝑝 )
𝑥
̃
𝑖=1 𝑗=1 𝑖=1 𝑗=1 (𝑚 𝑛 )
∑∑ ∑
𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛
This gives a contradiction that 𝑋 ∗ is an optimal solution in 𝐷. Since = min 𝑐𝑖𝑗1𝑢 ⊗ 𝑥𝑢𝑖𝑗 , 𝑐𝑖𝑗2𝑢 ⊗ 𝑥𝑢𝑖𝑗 , … , 𝑐𝑖𝑗𝑝𝑢 ⊗ 𝑥𝑢𝑖𝑗 (37)
𝑥
̃
𝑋 ∈ 𝐷∗ = 𝐷 (From 𝐼𝑠𝑡 part) ⇒ 𝑋 ∈ 𝐷. Thus our initial assumption 𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1
was wrong and therefore 𝑋 ∗ is an optimal solution in 𝐷∗ . Hence the subject to
theorem is proved. ∑𝑛
𝑥𝑢𝑖𝑗 = 𝑠𝑢𝑖 (𝑖 = 1, 2, … , 𝑚), (38)
In the next section a new approach is proposed to solve FFMOTP 𝑗=1
using harmonic mean technique. ∑𝑚
𝑥𝑢𝑖𝑗 = 𝑑𝑗𝑢 (𝑗 = 1, 2, … , 𝑛), (39)
3. Proposed method for solving fully fuzzy multi-objective trans- 𝑖=1
portation problem (FFMOTP) 𝑥𝑢𝑖𝑗 ≥ 𝑥𝑚 (40)
𝑖𝑗 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛).

While going through the existing literature, it has found that, till The Eqs. (37)–(40) are a crisp multi-objective linear programming
now, very few attempts [9,11,12,14,16] have been made in the litera- problem with 𝑝 objective functions for upper-level, where Eq. (40)
ture to solve FFMOTP. This section proposes a new approach/algorithm guarantees that the solutions obtained for this problem will always be
based on the fuzzy harmonic mean technique as a tool to solve FF- greater than or equal to the solution obtained in middle-level problem.
MOTP. Consider an FFMOTP as a model in Eq. (2)–(5) with p objective
Step-4. In this step, find the individual optimal solution of each objec-
functions that need to be minimized. The proposed algorithm can be
tive function 𝑢𝑠𝑟𝑘 , (for 𝑟 = 1, 𝑘 = 1, 2, … , 𝑝 and 𝑠 = 𝑙, 𝑚, 𝑢) involved
explain mathematically as follows: 𝑠 (for 𝑟 = 1 , 𝑘 =
in respective three levels and name them 𝜓𝑟𝑘
Step-1. Write all the objectives of a given problem in minimization 1, 2, … , 𝑝 & 𝑠 = 𝑙, 𝑚, 𝑢). If any linear programming problem comes out to
form and reduce them in the form of the problem given in Eqs. (6)–(8). be infeasible then by using dual-simple method and sensitivity analysis,
we will remove the infeasibility of that problem. Also, if optimal values
Step-2. Put all the values of fuzzy parameters given as a non-negative
of all the objective functions in all three levels are zero, re-assign the
triangular fuzzy numbers in the problem (6)–(8) and write it in the form
of Eq. (9)–(13). fuzzy parameters in step-2 and repeat the process. Table 1 shows the
decision variable values and optimal values for respective objective
Step-3. In this step, divide the entire problem (9)–(13) in three level functions involves in all the three levels.
crisp multi-objective linear programming problem (TL-CMOLPP).
L - lower level (L-MOLPP) Remark 5. In Table 1, ’*’ denotes the value of 𝑥𝑠𝑖𝑗 (𝑖 = 1, 2, … , 𝑚, 𝑗 =
min (𝑢𝑙11 , 𝑢𝑙12 , … , 𝑢𝑙1𝑝 ) 1, 2, … , 𝑛 and 𝑠 = 𝑙, 𝑚, 𝑢).
𝑥
̃
(𝑚 𝑛 ) Step-5. This step finds the fuzzy harmonic mean of individual
∑∑ 𝑚 ∑
∑ 𝑛 𝑚 ∑
∑ 𝑛
= min 𝑐𝑖𝑗1𝑙 ⊗ 𝑥𝑙𝑖𝑗 , 𝑐𝑖𝑗2𝑙 ⊗ 𝑥𝑙𝑖𝑗 , … , 𝑐𝑖𝑗𝑝𝑙 ⊗ 𝑥𝑙𝑖𝑗 (29) optimal solutions obtained in Step-4 for each of these three-levels
𝑥
̃
𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 respectively. The fuzzy harmonic mean can be calculated as follows:
subject to ( )
𝑠
∑𝑛 𝐹 𝐻𝑀1 (𝜓𝑟𝑘 ) = 𝐻𝑀1𝑙 , 𝐻𝑀1𝑚 , 𝐻𝑀1𝑢
𝑥𝑙𝑖𝑗 = 𝑠𝑙𝑖 (𝑖 = 1, 2, … , 𝑚), (30) 𝑝 𝑝 𝑝
= ( ∑𝑝 , ∑𝑝 , ∑𝑝 ) (41)
𝑗=1 1 1 1
𝑘=1 𝜓 𝑙 𝑘=1 𝜓 𝑚 𝑘=1 𝜓 𝑢
∑𝑚 1𝑘 1𝑘 1𝑘
𝑥𝑙𝑖𝑗 = 𝑑𝑗𝑙 (𝑗 = 1, 2, ..., 𝑛), (31) Here,
𝑖=1 𝑝
𝐻𝑀1𝑙 = ∑𝑝 1 = Harmonic mean of 𝑝 optimal solutions obtained
𝑥𝑙𝑖𝑗 ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (32) 𝑘=1 𝜓 𝑙
1𝑘
in lower level,
As we have noticed, the problem (29)–(32) is a crisp multi-objective 𝑝
𝐻𝑀1𝑚 = ∑𝑝 1 = Harmonic mean of 𝑝 optimal solutions obtained
linear programming problem of lower-level with 𝑝 objective functions 𝑘=1 𝜓 𝑚
1𝑘
that can be easily solved using the optimization tool TORA. Also, in middle level,
Eq. (32) guarantees that the solutions obtained for this problem are 𝐻𝑀1𝑢 = ∑𝑝 𝑝 1 = Harmonic mean of 𝑝 optimal solutions obtained
always greater than or equal to zero. 𝑘=1 𝜓 𝑢
1𝑘
M - middle level (M-MOLPP) in upper level.

min (𝑢𝑚 , 𝑢𝑚 , … , 𝑢𝑚 ) Step-6. In this step, three-level crisp multi-objective linear program-
𝑥
̃ 11 12 1𝑝
( ) ming problem (TL-CMOLPP) transformed into three-level crisp single-

𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛 ∑
𝑚 ∑
𝑛
objective linear programming problem (TL-CSOLPP).
= min 𝑐𝑖𝑗1𝑚 ⊗ 𝑥𝑚
𝑖𝑗 , 𝑐𝑖𝑗2𝑚 ⊗ 𝑥𝑚
𝑖𝑗 , … , 𝑐𝑖𝑗𝑝𝑚 ⊗ 𝑥𝑚
𝑖𝑗 (33)
𝑥
̃
𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 L - lower level (L-SOLPP)
subject to min 𝑢𝑙11 + min 𝑢𝑙12 + … + min 𝑢𝑙1𝑝
∑𝑛 min 𝜆𝑙1 = (42)
𝑥𝑚 𝑚
(34) 𝑥
̃ 𝐻𝑀1𝑢
𝑖𝑗 = 𝑠𝑖 (𝑖 = 1, 2, … , 𝑚),
𝑗=1 subject to

𝑚 ∑𝑛
𝑥𝑚 𝑚
𝑖𝑗 = 𝑑𝑗 (𝑗 = 1, 2, ..., 𝑛), (35) 𝑥𝑙𝑖𝑗 = 𝑠𝑙𝑖 (𝑖 = 1, 2, … , 𝑚), (43)
𝑖=1 𝑗=1

5
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 1
Optimal values table.
Lower level Middle level Upper level
𝑢𝑙11 𝑢𝑙12 … 𝑢𝑙1𝑝 𝑢𝑚11 𝑢𝑚12 … 𝑢𝑚1𝑝 𝑢𝑢11 𝑢𝑢12 … 𝑢𝑢1𝑝
𝑥𝑙11 ∗ ∗ … ∗ 𝑥𝑚11 ∗ ∗ … ∗ 𝑥𝑢11 ∗ ∗ … ∗

𝑥𝑙12 ∗ ∗ … ∗ 𝑥𝑚12 ∗ ∗ … ∗ 𝑥𝑢12 ∗ ∗ … ∗

⋮ ⋮ ⋮ ⋮ ⋮ ⋮
𝑥𝑙𝑚𝑛 ∗ ∗ … ∗ 𝑥𝑚𝑚𝑛 ∗ ∗ … ∗ 𝑥𝑢𝑚𝑛 ∗ ∗ … ∗
𝑙 𝑙 𝑙 𝑚 𝑚 𝑚 𝑢 𝑢 𝑢
𝜓11 𝜓12 … 𝜓1𝑝 𝜓11 𝜓12 … 𝜓1𝑝 𝜓11 𝜓12 … 𝜓1𝑝


𝑚
the fuzzy optimal solution of FFMOTP (2)–(5) is bigger or equal to its
𝑥𝑙𝑖𝑗 = 𝑑𝑗𝑙 (𝑗 = 1, 2, … , 𝑛), (44)
𝑖=1 mid-point.
𝑥𝑙𝑖𝑗 ≥ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (45) Step-7. By solving these different level crisp single-objective linear
programming problems, we will get the values of 𝑥 ̃𝑖𝑗 (𝑖 = 1, 2, … , 𝑚; 𝑗 =
Here we can see that the problem (42)–(45) is a crisp single-objective 1, 2, … , 𝑛) for all the three-levels, which eventually gives a combine
linear programming problem that can be easily solved by using the fuzzy optimal solution for the required problem. Here the optimal
optimization tool TORA. The optimal solution of this problem gives solution of problems (42)–(45), (46)–(49) and (50)–(53) respectively,
the left point of fuzzy optimal solution of FFMOTP (2)–(5). Also the guarantees that the fuzzy optimal solution of FFMOTP (2)–(5) is as
constraints in Eq. (45) guarantees that the obtained fuzzy optimal non-negative triangular fuzzy number. In sum, if 𝑥𝑙∗ 𝑚∗ 𝑢∗
𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 are the
solution of FFMOTP is as a non-negative triangular fuzzy number. optimal solutions of crisp problems (42)–(45), (46)–(49) and (50)–(53)
Now, assuming that 𝑥𝑙∗ = (𝑥𝑙∗
𝑖𝑗 )𝑚𝑛×1 is the optimal solution of prob-
respectively, then 𝑥 ̃𝑖𝑗 = (𝑥𝑙∗ 𝑚∗ 𝑢∗
𝑖𝑗 , 𝑥𝑖𝑗 , 𝑥𝑖𝑗 ) will be the fuzzy optimal solution
lem (42)–(45). We now solve the following middle-level problem in of FFMOTP (2)–(5). Finally, the optimal values of FFMOTP (2)–(5) is
order to obtain the mid-point of the fuzzy optimal solution of FFMOTP obtained by putting 𝑥 ̃𝑖𝑗 in equation number (2).
(2)–(5). Step-8. In this step, the ranking function will be used to defuzzify the
M - middle level (M-SOLPP) fuzzy compromise solution obtained in step-7 to get the crisp values.
min 𝑢𝑚
11
+ min 𝑢𝑚
12
+ … + min 𝑢𝑚
1𝑝
This will help us in comparing the results with the other existing results.
min 𝜆𝑚 = (46)
𝑥
̃ 1 𝐻𝑀1𝑚
3.1. Summary of the proposed algorithm
subject to
∑𝑛
This algorithm is to obtain the fuzzy optimal solution for the fully
𝑥𝑚 𝑚
𝑖𝑗 = 𝑠𝑖 (𝑖 = 1, 2, … , 𝑚), (47)
𝑗=1 fuzzy multi-objective transportation problem (FFMOTP) defined in Sec-

𝑚 tion 2.3 can be summarized as follows:
𝑥𝑚 𝑚
𝑖𝑗 = 𝑑𝑗 (𝑗 = 1, 2, ..., 𝑛), (48)
𝑖=1
Step-1. Write all the objectives of the given problem into minimization
type.
𝑥𝑚 𝑙
𝑖𝑗 ≥ 𝑥𝑖𝑗 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (49)
Step-2. Assign values to the fuzzy parameters of the problem and
It can be seen that, problem (46)–(49) is a crisp single-objective linear simplify it.
programming problem with 𝑥𝑙∗ 𝑚
𝑖𝑗 as a lower bound for variable 𝑥𝑖𝑗 . Also,
Step-3. Divide the entire problem in three-level crisp multi-objective
constraints (49) guarantees that the mid-point of the fuzzy optimal
linear programming problem, namely: L-MOLPP, M-MOLPP, U-MOLPP.
solution of FFMOTP (2)–(5) is not less than the left point of it.
Finally assuming that 𝑥𝑚∗ = (𝑥𝑚∗ 𝑖𝑗 )𝑚𝑛×1 is the optimal solution of
Step-4. In this step, find the individual optimal solution for each
problem (46)–(49). Now, we solve the following upper-level problem in objective function 𝑢𝑠𝑟𝑘 and name them as 𝜓𝑟𝑘 𝑠 (for 𝑟 = 1, 𝑘 = 1, 2,
order to obtain the right-point of the fuzzy optimal solution of FFMOTP … , 𝑝 and 𝑠 = 𝑙, 𝑚, 𝑢) involved in respective three levels. If any linear
(2)–(5). programming problem comes out to be infeasible then by using dual
U - upper level (U-SOLPP) simple method and sensitivity analysis we will remove the infeasibility
of that LPP. Also, if optimal values of all the objective functions in all
min 𝑢𝑢11 + min 𝑢𝑢12 + … + min 𝑢𝑢1𝑝 three levels are zero, re-assign the fuzzy parameters in step-2 and repeat
min 𝜆𝑢1 = (50)
𝑥
̃ 𝐻𝑀1𝑙 the process.
subject to 𝑠 (𝑟 = 1, 𝑘 = 1, 2,
Step-5. Determine fuzzy harmonic mean (FHM) of 𝜓𝑟𝑘
∑𝑛
… , 𝑝 and 𝑠 = 𝑙, 𝑚, 𝑢).
𝑥𝑢𝑖𝑗 = 𝑠𝑢𝑖 (𝑖 = 1, 2, … , 𝑚), (51)
𝑗=1 Step-6. Using FHM as a tool, convert the TL-CMOLPP into TL-CSOLPP.
∑𝑚
𝑥𝑢𝑖𝑗 = 𝑑𝑗𝑢 (𝑗 = 1, 2, … , 𝑛), (52) Step-7. Repeat step-4, and combine the obtained result 𝑥𝑖𝑗 (𝑖 = 1, 2,
𝑖=1 … , 𝑚, 𝑗 = 1, 2, … , 𝑛) for all the three-levels to get the fuzzy optimal
solution.
𝑥𝑢𝑖𝑗 ≥ 𝑥𝑚
𝑖𝑗 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (53)
Step-8. Defuzzify the fuzzy compromise solution obtained in step-7 by
Again, as we notice that the problem (50)–(53) is a crisp single-
using ranking function defined in Section 2.2.1 to get the required crisp
objective linear programming problem with 𝑥𝑚∗ 𝑖𝑗 as a lower bound for solution of the problem.
variable 𝑥𝑢𝑖𝑗 . In addition, constraint (53) confirm that the right point of

6
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

3.2. Flowchart of the proposed algorithm

7
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

be reduced in the loss matrix by subtracting all the elements of the


matrix by the highest element of that matrix. Table 3 shows the reduced
4. Numerical experiments loss matrix of problem 1.
Now, after converting all the objectives into minimization type,
In this section, two numerical examples will be solve to illustrate write the entire problem in form of Eq. (9)–(13). For this, let us assume
the efficiency of our proposed method. that 𝑍 ̃𝑘 = (𝑣𝑙 , 𝑣𝑚 , 𝑣𝑢 ) 𝑥 ̃𝑖𝑗 = (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢 𝑓̃𝑟𝑘 = (𝑢𝑙𝑟𝑘 , 𝑢𝑚 , 𝑢𝑢𝑟𝑘 ),
𝑘 𝑘 𝑘 𝑖𝑗 , 𝑥𝑖𝑗 ) 𝑟𝑘
𝑠𝑖 = (𝑠𝑙𝑖 , 𝑠𝑚
̃ , 𝑠𝑢 ), 𝑑̃ = (𝑑 𝑙 , 𝑑 𝑚 , 𝑑 𝑢 ) (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, ..., 𝑛; 𝑟 =
𝑖 𝑖 𝑗 𝑗 𝑗 𝑗
Example 1. Consider an online delivery problem of dairy products 1 & 𝑘 = 1, 2, … , 𝑝) are all triangular fuzzy numbers, then reconstruct
from three different towns to eight different cities which has already
the problem by using the data given in Table 3 as follows:
been solved by M.Bagheri et al. (2020) [23] in their paper (Solution of
FFMOTP by fuzzy DEA based approach). Table 2 gives all the required min(𝑣𝑙𝑘 , 𝑣𝑚 𝑢 𝑙 𝑚 𝑢 𝑙 𝑚 𝑢 𝑙 𝑚 𝑢
𝑘 , 𝑣𝑘 ) = min ((𝑢11 , 𝑢11 , 𝑢11 ), (𝑢12 , 𝑢12 , 𝑢12 ), . . . , (𝑢1𝑝 , 𝑢1𝑝 , 𝑢1𝑝 )) (54)
𝑥
̃
data of this problem. All the parameters of this problem, such as
subject to
delivery time, loss during transportation, total profit per unit product,
available supply, and required demand, are considered triangular fuzzy ∑𝑛
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢 𝑙 𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ≈ (𝑠𝑖 , 𝑠𝑖 , 𝑠𝑖 ) (𝑖 = 1, 2, … , 𝑚), (55)
numbers. 𝑗=1

To solve this problem, the proposed approach first converts all the ∑𝑚
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢 𝑙 𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ≈ (𝑑𝑗 , 𝑑𝑗 , 𝑑𝑗 ) (𝑗 = 1, 2, … , 𝑛), (56)
objective functions as minimization types. thereby the profit matrix will
𝑖=1

8
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 2
Data of the given problem.
𝐶1 𝐶2 𝐶3 𝐶4 𝐶5 𝐶6 𝐶7 𝐶8 ̃
𝑠𝑖
𝑇1 Delivery t. (7.5, 8, 9) (4.5, 5, 6) (5.5, 6, 7) (7, 9, 9.5) (8.5, 10, 11) (3.5, 4, 5) (7.6, 8, 9.5) (6.5, 7, 8)
Loss d.t. (2.5, 3, 3.5) (1, 1, 2) (1.5, 2, 2.5) (2, 3, 3.5) (1, 2.5, 3) (1, 1.5, 3) (2, 2.6, 3) (1, 1.5, 2) (105, 120, 140)
Profit (115, 125, 130) (82, 85, 98) (71, 75, 83) (92, 100, 108) (60, 65, 69) (90, 95, 96) (40, 45, 60) (86, 90, 110)
𝑇2 Delivery t. (6.8, 7, 7.5) (3.8, 4, 5) (9, 10, 10.5) (5, 6, 7.5) (7.8, 8, 8.5) (6, 6, 8) (4.5, 5, 7) (9, 10, 11.5)
Loss d.t. (1.5, 2, 2.8) (1, 1.5, 2.5) (0.5, 1, 2.5) (5, 6, 8) (2.5, 3, 3.5) (2, 2.6, 3.5) (1, 1.7, 3.5) (2.5, 2.5, 4) (72, 85, 108)
Profit (60, 65, 72) (42, 50, 51) (135, 145, 150) (180, 205, 210) (50, 59, 63) (59, 60, 70) (62, 65, 75) (95, 105, 120)
𝑇3 Delivery t. (9, 10, 11) (8, 9, 9.5) (8.5, 9, 9.5) (10.5, 11, 11.5) (10, 12, 12) (7, 8, 10) (8.6, 9, 10.5) (10.4, 11, 12)
Loss d.t. (3.5, 4, 5) (1.5, 2, 3) (2.5, 3.5, 4.5) (1, 2, 4) (4, 4.2, 6) (2.8, 3, 3.5) (2, 2.8, 3) (3.5, 4, 5) (73, 75, 83)
Profit (132, 135, 150) (65, 70, 72) (129, 130, 140) (130, 165, 170) (79, 80, 83) (70, 79, 90) (80, 82, 85) (70, 93, 110)
𝑑̃𝑗 (26, 30, 33) (22, 25, 35) (32, 35, 40) (37, 40, 49) (50, 55, 60) (22, 25, 29) (24, 30, 40) (37, 40, 45)

Table 3
Minimized data/loss matrix of the given problem.
𝐶1 𝐶2 𝐶3 𝐶4 𝐶5 𝐶6 𝐶7 𝐶8 ̃
𝑠𝑖
𝑇1 Delivery t. (7.5, 8, 9) (4.5, 5, 6) (5.5, 6, 7) (7, 9, 9.5) (8.5, 10, 11) (3.5, 4, 5) (7.6, 8, 9.5) (6.5, 7, 8)
Loss d.t. (2.5, 3, 3.5) (1, 1, 2) (1.5, 2, 2.5) (2, 3, 3.5) (1, 2.5, 3) (1, 1.5, 3) (2, 2.6, 3) (1, 1.5, 2) (105, 120, 140)
Profit (50, 80, 95) (82, 120, 128) (97, 130, 139) (72, 105, 118) (111, 140, 150) (84, 110, 120) (120, 160, 170) (70, 115, 124)
𝑇2 Delivery t. (6.8, 7, 7.5) (3.8, 4, 5) (9, 10, 10.5) (5, 6, 7.5) (7.8, 8, 8.5) (6, 6, 8) (4.5, 5, 7) (9, 10, 11.5)
Loss d.t. (1.5, 2, 2.8) (1, 1.5, 2.5) (0.5, 1, 2.5) (5, 6, 8) (2.5, 3, 3.5) (2, 2.6, 3.5) (1, 1.7, 3.5) (2.5, 2.5, 4) (72, 85, 108)
Profit (108, 140, 150) (129, 155, 168) (30, 60, 75) (−30, 0, 30) (117, 146, 160) (110, 145, 151) (105, 140, 148) (60, 100, 115)
𝑇3 Delivery t. (9, 10, 11) (8, 9, 9.5) (8.5, 9, 9.5) (10.5, 11, 11.5) (10, 12, 12) (7, 8, 10) (8.6, 9, 10.5) (10.4, 11, 12)
Loss d.t. (3.5, 4, 5) (1.5, 2, 3) (2.5, 3.5, 4.5) (1, 2, 4) (4, 4.2, 6) (2.8, 3, 3.5) (2, 2.8, 3) (3.5, 4, 5) (73, 75, 83)
Profit (30, 70, 78) (108, 135, 145) (40, 75, 81) (10, 40, 80) (97, 125, 131) (90, 126, 140) (95, 123, 130) (70, 112, 140)
𝑑̃𝑗 (26, 30, 33) (22, 25, 35) (32, 35, 40) (37, 40, 49) (50, 55, 60) (22, 25, 29) (24, 30, 40) (37, 40, 45)

)
(𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 ) ⪰ 0 (𝑖 = 1, 2, … , 𝑚; 𝑗 = 1, 2, … , 𝑛). (57) +3𝑥𝑢37 + 5𝑥𝑢38 (59)

Where, ∑
𝑚 ∑
𝑛


𝑚 ∑
𝑛 𝑥) = (𝑢𝑙13 , 𝑢𝑚
𝑓̃13 (̃ 13
, 𝑢𝑢13 ) = (𝑐𝑖𝑗3𝑙 , 𝑐𝑖𝑗3𝑚 , 𝑐𝑖𝑗3𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 )
𝑥) = (𝑢𝑙11 , 𝑢𝑚
𝑓̃11 (̃ 11
, 𝑢𝑢11 ) = (𝑐𝑖𝑗1𝑙 , 𝑐𝑖𝑗1𝑚 , 𝑐𝑖𝑗1𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 )
𝑖=1 𝑗=1
𝑖=1 𝑗=1 = (50𝑥𝑙11 + 82𝑥𝑙12 + 97𝑥𝑙13 + 72𝑥𝑙14 + 111
(
= 7.5𝑥𝑙11 + 4.5𝑥𝑙12 + 5.5𝑥𝑙13 + 7𝑥𝑙14 + 8.5𝑥𝑙15 𝑥𝑙15 + 84𝑥𝑙16 + 120𝑥𝑙17 + 70𝑥𝑙18 + 108𝑥𝑙21 + 129𝑥𝑙22 + 30𝑥𝑙23
+ 3.5𝑥𝑙16 + 7.6𝑥𝑙17 + 6.5𝑥𝑙18 + 6.8𝑥𝑙21 + 3.8𝑥𝑙22 + 9𝑥𝑙23 + 5𝑥𝑙24 + 7.8𝑥𝑙25 + 6𝑥𝑙26 −30𝑥𝑙24 + 117𝑥𝑙25 + 110𝑥𝑙26 + 105𝑥𝑙27 + 60𝑥𝑙28
+ 4.5𝑥𝑙27 + 9𝑥𝑙28 + 9𝑥𝑙31 + 8𝑥𝑙32 +30𝑥𝑙31 + 108𝑥𝑙32 + 40𝑥𝑙33 + 10𝑥𝑙34 + 97𝑥𝑙35 + 90𝑥𝑙36 + 95𝑥𝑙37
+ 8.5𝑥𝑙33 + 10.5𝑥𝑙34 + 10𝑥𝑙35 + 7𝑥𝑙36 + 8.6𝑥𝑙37 + 10.4𝑥𝑙38 , 8𝑥𝑚 11
+ 5𝑥𝑚 12 +70𝑥𝑙38 , 80𝑥𝑚 + 120𝑥𝑚 + 130𝑥𝑚
11 12 13
𝑚 𝑚 𝑚 𝑚 𝑚
+ 6𝑥13 + 9𝑥14 + 10𝑥15 + 4𝑥16 + 8𝑥17 +105𝑥𝑚 + 140𝑥𝑚 + 110𝑥𝑚 + 160𝑥𝑚 + 115𝑥𝑚
14 15 16 17 18
+ 7𝑥𝑚 18
+ 7𝑥𝑚21
+ 4𝑥𝑚
22
+ 10𝑥𝑚23
+ 6𝑥𝑚24
+ 8𝑥𝑚
25
+ 6𝑥𝑚26
+ 5𝑥𝑚27
+ 10𝑥𝑚28 +140𝑥𝑚 + 155𝑥𝑚 + 60𝑥𝑚 + 0𝑥𝑚 + 146𝑥𝑚 + 145𝑥𝑚
21 22 23 24 25 26
+ 10𝑥𝑚 31
+ 9𝑥𝑚 32
+ 9𝑥𝑚
33
+ 11𝑥𝑚34
+ 12𝑥𝑚 35 +140𝑥𝑚 + 100𝑥𝑚 + 70𝑥𝑚 + 135𝑥𝑚 + 75𝑥𝑚 + 40𝑥𝑚
27 28 31 32 33 34
+ 8𝑥𝑚 36
+ 9𝑥𝑚37
+ 11𝑥𝑚
38
, 9𝑥𝑢11 + 6𝑥𝑢12 + 7𝑥𝑢13 + 9.5𝑥𝑢14 + 11𝑥𝑢15 + 5𝑥𝑢16 +125𝑥𝑚 + 126𝑥𝑚 + 123𝑥𝑚 + 112𝑥𝑚 , 95𝑥𝑢11
35 36 37 38
𝑢 𝑢 𝑢
+ 9.5𝑥17 + 8𝑥18 + 7.5𝑥21 + 5𝑥𝑢22 + 10.5 +128𝑥𝑢12 + 139𝑥𝑢13 + 118𝑥𝑢14 + 150𝑥𝑢15 + 120𝑥𝑢16 + 170𝑥𝑢17
𝑥𝑢23 + 7.5𝑥𝑢24 + 8.5𝑥𝑢25 + 8𝑥𝑢26 + 7𝑥𝑢27 + 11.5𝑥𝑢28 + 11𝑥𝑢31 + 9.5𝑥𝑢32 + 9.5𝑥𝑢33 +124𝑥𝑢18 + 150𝑥𝑢21 + 168𝑥𝑢22 + 75𝑥𝑢23 + 30𝑥𝑢24
+ 11.5𝑥𝑢34 + 12𝑥𝑢35 + 10𝑥𝑢36 + 10.5 +160𝑥𝑢25 + 151𝑥𝑢26 + 148𝑥𝑢27 + 115𝑥𝑢28 + 78𝑥𝑢31 + 145𝑥𝑢32
)
𝑥𝑢37 + 12𝑥𝑢38 (58) +81𝑥𝑢33 + 80𝑥𝑢34 + 131𝑥𝑢35 + 140𝑥𝑢36 + 130𝑥𝑢37


𝑚 ∑
𝑛 +140𝑥𝑢38 ) (60)
𝑥) = (𝑢𝑙12 , 𝑢𝑚
𝑓̃12 (̃ 12
, 𝑢𝑢12 ) = (𝑐𝑖𝑗2𝑙 , 𝑐𝑖𝑗2𝑚 , 𝑐𝑖𝑗2𝑢 ) ⊗ (𝑥𝑙𝑖𝑗 , 𝑥𝑚 𝑢
𝑖𝑗 , 𝑥𝑖𝑗 )
𝑖=1 𝑗=1 Now, we will divide the entire problem into three-level crisp multi-
( objective linear programming problem (TL-CMOLPP).
= 2.5𝑥𝑙11 + 1𝑥𝑙12 + 1.5𝑥𝑙13 + 2𝑥𝑙14 + 1𝑥𝑙15
L - lower level
+ 1𝑥𝑙16 + 2𝑥𝑙17 + 1𝑥𝑙18 + 1.5𝑥𝑙21 + 1𝑥𝑙22 + 0.5𝑥𝑙23 + 5𝑥𝑙24 + 2.5𝑥𝑙25 + 2𝑥𝑙26
+ 1𝑥𝑙27 + 2.5𝑥𝑙28 + 3.5𝑥𝑙31 + 1.5𝑥𝑙32 min (𝑣𝑙 ) = min (𝑢𝑙11 , 𝑢𝑙12 , 𝑢𝑙13 )
𝑥
̃ 𝑥
̃
+ 2.5𝑥𝑙33 + 1𝑥𝑙34 + 4𝑥𝑙35 + 2.8𝑥𝑙36 + 2𝑥𝑙37 + 3.5𝑥𝑙38 , = (7.5𝑥𝑙11 + 4.5𝑥𝑙12 + 5.5𝑥𝑙13 + 7𝑥𝑙14
3𝑥𝑚
11
+ 1𝑥𝑚
12
+ 2𝑥𝑚
13
+ 3𝑥𝑚
14
+ 2.5𝑥𝑚
15
+ 1.5𝑥𝑚
16
+ 2.6𝑥𝑚
17
+8.5𝑥𝑙15 + 3.5𝑥𝑙16 + 7.6𝑥𝑙17 + 6.5𝑥𝑙18 + 6.8𝑥𝑙21
+ 1.5𝑥𝑚
18
+ 2𝑥𝑚
21
+ 1.5𝑥𝑚
22
+ 1𝑥𝑚
23
+ 6𝑥𝑚
24
+ 3𝑥𝑚
25
+ 2.6𝑥𝑚
26
+3.8𝑥𝑙22 + 9𝑥𝑙23 + 5𝑥𝑙24 + 7.8𝑥𝑙25 + 6𝑥𝑙26 + 4.5𝑥𝑙27 + 9𝑥𝑙28 9𝑥𝑙31
+ 1.7𝑥𝑚 27
+ 2.5𝑥𝑚 28
+ 4𝑥𝑚
31
+ 2𝑥𝑚32
+ 3.5𝑥𝑚33
+ 2𝑥𝑚34
+8𝑥𝑙32 + 8.5𝑥𝑙33 + 10.5𝑥𝑙34 + 10𝑥𝑙35
+ 4.2𝑥𝑚 35
+ 3𝑥𝑚36
+ 2.8𝑥𝑚
37
+ 4𝑥𝑚38
, +7𝑥𝑙36 + 8.6𝑥𝑙37 + 10.4𝑥𝑙38 , 2.5𝑥𝑙11 + 1𝑥𝑙12 + 1.5𝑥𝑙13 + 2𝑥𝑙14
3.5𝑥𝑢11 + 2𝑥𝑢12 + 2.5𝑥𝑢13 + 3.5𝑥𝑢14 + 3𝑥𝑢15 + 3𝑥𝑢16 + 3𝑥𝑢17 + 2𝑥𝑢18 + 2.8𝑥𝑢21 +1𝑥15 + 1𝑥16 + 2𝑥17 + 1𝑥18 + 1.5𝑥𝑙21
𝑙 𝑙 𝑙 𝑙

+ 2.5𝑥𝑢22 + 2.5𝑥𝑢23 + 8𝑥𝑢24 + 3.5𝑥𝑢25 + 3.5𝑥𝑢26 + 3.5𝑥𝑢27 + 4𝑥𝑢28 + 5𝑥𝑢31 + 3𝑥𝑢32 +1𝑥𝑙22 + 0.5𝑥𝑙23 + 5𝑥𝑙24 + 2.5𝑥𝑙25 + 2𝑥𝑙26 + 1𝑥𝑙27 + 2.5𝑥𝑙28 + 3.5𝑥𝑙31
+ 4.5𝑥𝑢33 + 4𝑥𝑢34 + 6𝑥𝑢35 + 3.5𝑥𝑢36 +1.5𝑥𝑙32 + 2.5𝑥𝑙33 + 1𝑥𝑙34 + 4𝑥𝑙35

9
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 4
Optimal values table.
Lower level Middle level Upper level
𝑢𝑙11 𝑢𝑙12 𝑢𝑙13 𝑢𝑚11 𝑢𝑚12 𝑢𝑚13 𝑢𝑢11 𝑢𝑢12 𝑢𝑢13
𝑥𝑙11 0 0 0 𝑥𝑚11 2 0 4 𝑥𝑢11 2 0 7
𝑥𝑙12 14 0 22 𝑥𝑚12 17 3 25 𝑥𝑢12 23 9 35
𝑥𝑙13 32 0 0 𝑥𝑚13 35 0 0 𝑥𝑢13 40 0 0
𝑥𝑙14 0 0 0 𝑥𝑚14 0 1 0 𝑥𝑢14 0 10 0
𝑥𝑙15 0 50 27 𝑥𝑚15 1 55 32 𝑥𝑢15 1 55 35
𝑥𝑙16 22 18 22 𝑥𝑚16 25 21 25 𝑥𝑢16 29 21 29
𝑥𝑙17 0 0 0 𝑥𝑚17 0 0 0 𝑥𝑢17 0 0 0
𝑥𝑙18 37 37 34 𝑥𝑚18 40 40 34 𝑥𝑢18 45 45 34
𝑥𝑙21 3 26 0 𝑥𝑚21 3 30 0 𝑥𝑢21 3 33 0
𝑥𝑙22 8 0 0 𝑥𝑚22 8 0 0 𝑥𝑢22 12 4 0
𝑥𝑙23 0 32 32 𝑥𝑚23 0 35 35 𝑥𝑢23 0 40 40
𝑥𝑙24 37 0 37 𝑥𝑚24 40 0 40 𝑥𝑢24 49 0 49
𝑥𝑙25 0 0 0 𝑥𝑚25 4 0 0 𝑥𝑢25 4 5 0
𝑥𝑙26 0 0 0 𝑥𝑚26 0 0 0 𝑥𝑢26 0 4 0
𝑥𝑙27 24 14 0 𝑥𝑚27 30 20 4 𝑥𝑢27 40 22 8
𝑥𝑙28 0 0 3 𝑥𝑚28 0 0 6 𝑥𝑢28 0 0 11
𝑥𝑙31 23 0 26 𝑥𝑚31 25 0 26 𝑥𝑢31 28 0 26
𝑥𝑙32 0 22 0 𝑥𝑚32 0 22 0 𝑥𝑢32 0 22 0
𝑥𝑙33 0 0 0 𝑥𝑚33 0 0 0 𝑥𝑢33 0 0 0
𝑥𝑙34 0 37 0 𝑥𝑚34 0 39 0 𝑥𝑢34 0 39 0
𝑥𝑙35 50 0 23 𝑥𝑚35 50 0 23 𝑥𝑢35 55 0 25
𝑥𝑙36 0 4 0 𝑥𝑚36 0 4 0 𝑥𝑢36 0 4 0
𝑥𝑙37 0 10 24 𝑥𝑚37 0 10 26 𝑥𝑢37 0 18 32
𝑥𝑙38 0 0 0 𝑥𝑚38 0 0 0 𝑥𝑢38 0 0 0
𝑣𝑙 1607.3 275.2 14350 𝑣𝑚 2026 526 25613 𝑣𝑢 2684 971.9 34473

+2.8𝑥𝑙36 + 2𝑥𝑙37 + 3.5𝑥𝑙38 , 50𝑥𝑙11 + 82𝑥𝑙12 + 97𝑥𝑙13 + 72𝑥𝑙14 +160𝑥𝑚


17
+ 115𝑥𝑚
18
+ 140𝑥𝑚
21
+ 155𝑥𝑚
22
+ 60𝑥𝑚
23
+111𝑥𝑙15 + 84𝑥𝑙16 + 120𝑥𝑙17 + 70𝑥𝑙18 +0𝑥𝑚
24
+ 146𝑥𝑚
25
+ 145𝑥𝑚
26
+ 140𝑥𝑚
27
+ 100𝑥𝑚
28
+108𝑥𝑙21 + 129𝑥𝑙22 + 30𝑥𝑙23 − 30𝑥𝑙24 + 117𝑥𝑙25 + 110𝑥𝑙26 + 105𝑥𝑙27 +70𝑥𝑚
31
+ 135𝑥𝑚
32
+ 75𝑥𝑚
33
+ 40𝑥𝑚
34
+ 125𝑥𝑚
35
+60𝑥𝑙28 + 30𝑥𝑙31 + 108𝑥𝑙32 + 40𝑥𝑙33 +126𝑥𝑚 + 123𝑥𝑚 + 112𝑥𝑚 ) (63)
36 37 38
+10𝑥𝑙34 + 97𝑥𝑙35 + 90𝑥𝑙36 + 95𝑥𝑙37 + 70𝑥𝑙38 ) (61)
subject to 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 120 ⎫
11 12 13 14 15 16 17 18 ⎪
subject to 𝑥𝑙11 + 𝑥𝑙12 + 𝑥𝑙13 + 𝑥𝑙14 + 𝑥𝑙15 + 𝑥𝑙16 + 𝑥𝑙17 + 𝑥𝑙18 = 105 ⎫ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 85 ⎪
⎪ 21 22 23 24 25 26 27 28 ⎪
𝑥𝑙21 + 𝑥𝑙22 + 𝑥𝑙23 + 𝑥𝑙24 + 𝑥𝑙25 + 𝑥𝑙26 + 𝑥𝑙27 + 𝑥𝑙28 = 72 ⎪ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 75 ⎪
⎪ 31 32 33 34 35 36 37 38

𝑥𝑙31 + 𝑥𝑙32 + 𝑥𝑙33 + 𝑥𝑙34 + 𝑥𝑙35 + 𝑥𝑙36 + 𝑥𝑙37 + 𝑥𝑙38 = 73 ⎪ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 30 ⎪
⎪ 11 21 31

𝑥𝑙11 + 𝑥𝑙21 + 𝑥𝑙31 = 26 ⎪ 𝑥𝑚
12
+ 𝑥𝑚22
+ 𝑥𝑚
32
= 25 ⎪
⎪ ⎪
𝑥𝑙12 + 𝑥𝑙22 + 𝑥𝑙32 = 22 ⎪ 𝑥𝑚
13
+ 𝑥𝑚23
+ 𝑥𝑚
33
= 35 ⎪
⎪ ⎬
𝑥𝑙13 + 𝑥𝑙23 + 𝑥𝑙33 = 32 ⎪ 𝑥𝑚
14
+ 𝑥𝑚24
+ 𝑥𝑚
34
= 40 ⎪
⎬ ⎪
𝑥𝑙14 + 𝑥𝑙24 + 𝑥𝑙34 = 37 ⎪ 𝑥𝑚
15
+ 𝑥𝑚25
+ 𝑥𝑚
35
= 55 ⎪
⎪ ⎪
𝑥𝑙15 + 𝑥𝑙25 + 𝑥𝑙35 = 50 ⎪ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 25
16 26 36 ⎪
𝑥𝑙16 + 𝑥𝑙26 + 𝑥𝑙36 = 22 ⎪ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 30 ⎪
⎪ 17 27 37 ⎪
𝑥𝑙17 + 𝑥𝑙27 + 𝑥𝑙37 = 24 ⎪ 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 40 ⎪
⎪ 18 28 38 ⎪
𝑥𝑙18 + 𝑥𝑙28 + 𝑥𝑙38 = 37 ⎪ 𝑥𝑚 ≥ 𝑥𝑙𝑖𝑗 ∀ 𝑖, 𝑗 ⎪
⎪ 𝑖𝑗 ⎭
𝑥𝑙𝑖𝑗 ≥ 0 ∀ 𝑖, 𝑗 ⎪
⎭ (64)
(62) U - Upper level
M - Middle level min (𝑣𝑢 ) = min (𝑢𝑚 , 𝑢𝑚 , 𝑢𝑚 ) = (9𝑥𝑢11 + 6𝑥𝑢12 + 7𝑥𝑢13 + 9.5𝑥𝑢14 + 11𝑥𝑢15
𝑥
̃ 𝑥
̃11 12 13
𝑚
min (𝑣 ) = min (𝑢𝑚
11
, 𝑢𝑚
12
, 𝑢𝑚
13
) = (8𝑥𝑚
11
+ 5𝑥𝑚
12
+ 6𝑥𝑚
13
+ 9𝑥𝑚
14 +5𝑥𝑢16 + 9.5𝑥𝑢17 + 8𝑥𝑢18 + 7.5𝑥𝑢21 + 5𝑥𝑢22
𝑥
̃ 𝑥
̃
+10𝑥𝑚
15
+ 4𝑥𝑚
16
+ 8𝑥𝑚
17
+ 7𝑥𝑚
18
+ 7𝑥𝑚
21
+ 4𝑥𝑚
22
+10.5𝑥𝑢23 + 7.5𝑥𝑢24 + 8.5𝑥𝑢25 + 8𝑥𝑢26 + 7𝑥𝑢27
+10𝑥𝑚23
+ 6𝑥𝑚24
+ 8𝑥𝑚25
+ 6𝑥𝑚 26
+ 5𝑥𝑚27
+ 10𝑥𝑚 28
+11.5𝑥𝑢28 + 11𝑥𝑢31 + 9.5𝑥𝑢32 + 9.5𝑥𝑢33
+10𝑥𝑚31
+ 9𝑥𝑚32
+ 9𝑥𝑚33
+ 11𝑥𝑚 34
+ 12𝑥𝑚35
+ 8𝑥𝑚36
+11.5𝑥𝑢34 + 12𝑥𝑢35 + 10𝑥𝑢36
+9𝑥𝑚37
+ 11𝑥 𝑚
38
, 3𝑥 𝑚
11
+ 1𝑥 𝑚
12
+ 2𝑥 𝑚
13
+ 3𝑥 𝑚
14
+10.5𝑥𝑢37 + 12𝑥𝑢38 , 3.5𝑥𝑢11 + 2𝑥𝑢12 + 2.5𝑥𝑢13 + 3.5𝑥𝑢14
+2.5𝑥𝑚 15
+ 1.5𝑥𝑚16
+ 2.6𝑥𝑚 17
+ 1.5𝑥𝑚 18
+ 2𝑥𝑚21
+3𝑥𝑢15 + 3𝑥𝑢16 + 3𝑥𝑢17 + 2𝑥𝑢18 + 2.8𝑥𝑢21 + 2.5𝑥𝑢22
+1.5𝑥𝑚 22
+ 1𝑥𝑚23
+ 6𝑥𝑚 24
+ 3𝑥𝑚 25
+ 2.6𝑥𝑚26
+ 1.7𝑥𝑚 27
+2.5𝑥𝑢23 + 8𝑥𝑢24 + 3.5𝑥𝑢25 + 3.5𝑥𝑢26 + 3.5𝑥𝑢27
𝑚 𝑚 𝑚 𝑚 𝑚
+2.5𝑥28 + 4𝑥31 + 2𝑥32 + 3.5𝑥33 + 2𝑥34 +4𝑥𝑢28 + 5𝑥𝑢31 + 3𝑥𝑢32 + 4.5𝑥𝑢33 + 4𝑥𝑢34 + 6𝑥𝑢35 + 3.5𝑥𝑢36 +
+4.2𝑥𝑚 35
+ 3𝑥𝑚36
+ 2.8𝑥𝑚 37
+ 4𝑥𝑚 38
, 80𝑥𝑚 11
3𝑥𝑢37 + 5𝑥𝑢38 , 95𝑥𝑢11 + 128𝑥𝑢12 + 139𝑥𝑢13 + 118𝑥𝑢14
+120𝑥𝑚 12
+ 130𝑥𝑚 13
+ 105𝑥𝑚 14
+ 140𝑥𝑚 15
+ 110𝑥𝑚 16
+150𝑥𝑢15 + 120𝑥𝑢16 + 170𝑥𝑢17 + 124𝑥𝑢18 + 150𝑥𝑢21

10
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 5 Now, the three-level crisp multi-objective linear programming prob-


Final optimal values table. lem (TL-CMOLPP) transform into three-level crisp single objective lin-
Lower level Middle level Upper level Composite fuzzy ear programming problem (TL-CSOLPP) as follows:
optimal solution
L - CSOLPP :
𝑥𝑙11 = 0 𝑥𝑚11 = 4 𝑥𝑢11 = 7 𝑥
̃11 = (0 4 7)
𝑥𝑙12 = 22 𝑥𝑚12 = 25 𝑥𝑢12 = 35 𝑥
̃12 = (22 25 35)
min 𝑢𝑙11 + min 𝑢𝑙12 + min 𝑢𝑙13
min 𝜆𝑙1 =
𝑥𝑙13 = 0 𝑥𝑚13 = 0 𝑥𝑢13 = 0 𝑥
̃13 = (0 0 0) 𝑥
̃ 𝐻𝑀1𝑢
𝑥𝑙14 = 0 𝑥𝑚14 = 0 𝑥𝑢14 = 0 𝑥
̃14 = (0 0 0)
𝑥𝑙15 = 27 𝑥𝑚15 = 32 𝑥𝑢15 = 35 𝑥
̃15 = (27 32 35) = 0.0286𝑥𝑙11 + 0.0417𝑥𝑙12 + 0.0495𝑥𝑙13 + 0.0386𝑥𝑙14
𝑥𝑙16 = 22 𝑥𝑚16 = 25 𝑥𝑢16 = 29 𝑥
̃16 = (22 25 29)
𝑥𝑙17 = 0 𝑥𝑚17 = 0 𝑥𝑢17 = 0 𝑥
̃17 = (0 0 0) +0.0574𝑥𝑙15 + 0.0421𝑥𝑙16 + 0.0617𝑥𝑙17 + 0.0369𝑥𝑙18
𝑥𝑙18 = 34 𝑥𝑚18 = 34 𝑥𝑢18 = 35 𝑥
̃18 = (34 34 34)
𝑥𝑙21 = 0 𝑥𝑚21 = 0 𝑥𝑢21 = 0 𝑥
̃21 = (0 0 0) +0.0554𝑥𝑙21 + 0.0638𝑥𝑙22 + 0.0188𝑥𝑙23 + 0.0095𝑥𝑙24
𝑥𝑙22 = 0 𝑥𝑚22 = 0 𝑥𝑢22 = 0 𝑥
̃22 = (0 0 0)
𝑥𝑙23 = 32 𝑥𝑚23 = 35 𝑥𝑢23 = 40 𝑥
̃23 = (32 35 40) +0.0607𝑥𝑙25 + 0.0562𝑥𝑙26 + 0.0526𝑥𝑙27 + 0.0340𝑥𝑙28
𝑥𝑙24 = 37 𝑥𝑚24 = 40 𝑥𝑢24 = 49 𝑥
̃24 = (37 40 49)
𝑥𝑙25 = 0 𝑥𝑚25 = 0 𝑥𝑢25 = 0 𝑥
̃25 = (0 0 0)
+0.0202𝑥𝑙31 + 0.0560𝑥𝑙32 + 0.0243𝑥𝑙33 + 0.0102𝑥𝑙34
𝑥𝑙26 = 0 𝑥𝑚26 = 0 𝑥𝑢26 = 0 𝑥
̃26 = (0 0 0)
𝑥𝑙27 = 0 𝑥𝑚27 = 4 𝑥𝑢27 = 8 𝑥
̃27 = (0 4 8) +0.0529𝑥𝑙35 + 0.0475𝑥𝑙36 + 0.0503𝑥𝑙37 + 0.0400𝑥𝑙38 (67)
𝑥𝑙28 = 3 𝑥𝑚28 = 6 𝑥𝑢28 = 11 𝑥
̃28 = (3 6 11)
𝑥𝑙31 = 26 𝑥𝑚31 = 26 𝑥𝑢31 = 26 𝑥
̃31 = (26 26 26)
subject to 𝑥𝑙11 + 𝑥𝑙12 + 𝑥𝑙13 + 𝑥𝑙14 + 𝑥𝑙15 + 𝑥𝑙16 + 𝑥𝑙17 + 𝑥𝑙18 = 105 ⎫
𝑥𝑙32 = 0 𝑥𝑚32 = 0 𝑥𝑢32 = 0 𝑥
̃32 = (0 0 0)

𝑥𝑙33 𝑥𝑚33 𝑥𝑢33
= 0 = 0 = 0 𝑥
̃33 = (0 0 0)
𝑥𝑙21 + 𝑥𝑙22 + 𝑥𝑙23 + 𝑥𝑙24 + 𝑥𝑙25 + 𝑥𝑙26 + 𝑥𝑙27 + 𝑥𝑙28 = 72 ⎪
𝑥𝑙34 = 0 𝑥𝑚34 = 0 𝑥𝑢34 = 0 𝑥
̃34 = (0 0 0) ⎪
𝑥𝑙35 = 23 𝑥𝑚35 = 23 𝑥𝑢35 = 25 𝑥
̃35 = (23 23 25) 𝑥𝑙31 + 𝑥𝑙32 + 𝑥𝑙33 + 𝑥𝑙34 + 𝑥𝑙35 + 𝑥𝑙36 + 𝑥𝑙37 + 𝑥𝑙38 = 73 ⎪
𝑥𝑙36 = 0 𝑥𝑚36 = 0 𝑥𝑢36 = 0 𝑥
̃36 = (0 0 0) ⎪
𝑥𝑙37 = 24 𝑥𝑚37 = 26 𝑥𝑢37 = 32 𝑥
̃37 = (24 26 32)
𝑥𝑙11 + 𝑥𝑙21 + 𝑥𝑙31 = 26 ⎪

𝑥𝑙38 = 0 𝑥𝑚38 = 0 𝑥𝑢38 = 0 𝑥
̃38 = (0 0 0) 𝑥𝑙12 + 𝑥𝑙22 + 𝑥𝑙32 = 22⎪

𝑥𝑙13 + 𝑥𝑙23 + 𝑥𝑙33 = 32⎪

𝑥𝑙14 + 𝑥𝑙24 + 𝑥𝑙34 = 37⎪
+168𝑥𝑢22 + 75𝑥𝑢23 + 30𝑥𝑢24 + 160𝑥𝑢25 + 151𝑥𝑢26 + 148𝑥𝑢27 ⎪
𝑥𝑙15 + 𝑥𝑙25 + 𝑥𝑙35 = 50⎪
+115𝑥𝑢28 + 78𝑥𝑢31 + 145𝑥𝑢32 + 81𝑥𝑢33 + 80𝑥𝑢34
𝑥𝑙16 + 𝑥𝑙26 + 𝑥𝑙36 = 22⎪⎪
+131𝑥𝑢35 + 140𝑥𝑢36 + 130𝑥𝑢37 + 140𝑥𝑢38 ) (65)
𝑥𝑙17 + 𝑥𝑙27 + 𝑥𝑙37 = 24⎪

𝑥𝑙18 + 𝑥𝑙28 + 𝑥𝑙38 = 37⎪
⎫ ⎪
subject to 𝑥𝑢11 + 𝑥𝑢12 + 𝑥𝑢13 + 𝑥𝑢14 + 𝑥𝑢15 + 𝑥𝑢16 + 𝑥𝑢17 + 𝑥𝑢18 = 140 𝑥𝑙𝑖𝑗 ≥ 0 ∀ 𝑖, 𝑗 ⎪
⎪ ⎭
𝑥𝑢21 + 𝑥𝑢22 + 𝑥𝑢23 + 𝑥𝑢24 + 𝑥𝑢25 + 𝑥𝑢26 + 𝑥𝑢27 + 𝑥𝑢28 = 108 ⎪

𝑥𝑢31 + 𝑥𝑢32 + 𝑥𝑢33 + 𝑥𝑢34 + 𝑥𝑢35 + 𝑥𝑢36 + 𝑥𝑢37 + 𝑥𝑢38 = 83 ⎪ (68)

𝑥𝑢11 + 𝑥𝑢21 + 𝑥𝑢31 = 33 ⎪ M - CSOLPP :
⎪ min 𝑢𝑚 + min 𝑢𝑚 + min 𝑢𝑚
𝑥𝑢12 + 𝑥𝑢22 + 𝑥𝑢32 = 35 ⎪ min 𝜆𝑚 = 11 12 13
⎪ 𝑥
̃ 1 𝐻𝑀1 𝑚
𝑥𝑢13 + 𝑥𝑢23 + 𝑥𝑢33 = 40 ⎪
⎬ = 0.0738𝑥𝑚 + 0.1022𝑥𝑚 + 0.1119𝑥𝑚 + 0.0949𝑥𝑚
𝑥𝑢14 + 𝑥𝑢24 + 𝑥𝑢34 = 49 ⎪ 11 12 13 14
⎪ +0.1237𝑥𝑚 + 0.0937𝑥𝑚 + 0.1384𝑥𝑚 + 0.1001𝑥𝑚
𝑥𝑢15 + 𝑥𝑢25 + 𝑥𝑢35 = 60 ⎪ 15 16 17 18
𝑥𝑢16 + 𝑥𝑢26 + 𝑥𝑢36 = 29 ⎪
⎪ +0.1208𝑥𝑚
21
+ 0.1302𝑥𝑚
22
+ 0.0576𝑥𝑚
23
+ 0.0097𝑥𝑚
24
𝑥𝑢17 + 𝑥𝑢27 + 𝑥𝑢37 = 40 ⎪
⎪ +0.1273𝑥𝑚
25
+ 0.1246𝑥𝑚
26
+ 0.1190𝑥𝑚
27
+ 0.0912𝑥𝑚
28
𝑥𝑢18 + 𝑥𝑢28 + 𝑥𝑢38 = 45 ⎪
⎪ +0.0681𝑥𝑚 + 0.1184𝑥𝑚 + 0.0709𝑥𝑚 + 0.0430𝑥𝑚
𝑥𝑢𝑖𝑗 ≥ 𝑥𝑚 ⎪ 31 32 33 34
𝑖𝑗 ∀ 𝑖, 𝑗 ⎭
+0.1145𝑥𝑚
35
+ 0.1111𝑥𝑚
36
+ 0.1093𝑥𝑚
37
+ 0.1030𝑥𝑚
38
(69)
(66)

Now, the individual optimal solution of each objective function 𝑢𝑠𝑟𝑘 , subject to 𝑥𝑚
11
+ 𝑥𝑚12
+ 𝑥𝑚13
+ 𝑥𝑚14
+ 𝑥𝑚15
+ 𝑥𝑚16
+ 𝑥𝑚17
+ 𝑥𝑚18
= 120 ⎫
𝑥𝑚 + 𝑥 𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 85 ⎪
(for 𝑟 = 1, 𝑘 = 1, 2, … , 𝑝 𝑎𝑛𝑑 𝑠 = 𝑙, 𝑚, 𝑢) involved in respective 21 22 23 24 25 26 27 28 ⎪
three-levels are given in Table 4, 𝑥𝑚31
+ 𝑥𝑚32
+ 𝑥𝑚33
+ 𝑥𝑚34
+ 𝑥𝑚35
+ 𝑥𝑚36
+ 𝑥𝑚37
+ 𝑥𝑚38
= 75 ⎪
𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 30 ⎪
The fuzzy harmonic mean of individual optimal solution obtained 11 21 31 ⎪
in Table 4 for each of these three-level is 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 25 ⎪
12 22 32 ⎪
( ) 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 35 ⎪
𝐹 𝐻𝑀1 (𝜓𝑟𝑡𝑠 ) = 𝐻𝑀1𝑙 , 𝐻𝑀1𝑚 , 𝐻𝑀1𝑢 13 23 33 ⎪
𝑥𝑚
14
+ 𝑥𝑚
24
+ 𝑥𝑚34
= 40 ⎬
= (693.5504 , 1232.6571 , 2097.1704) ⎪
𝑥𝑚
15
+ 𝑥25 + 𝑥𝑚
𝑚
35
= 55 ⎪
Where, ⎪
3 𝑥𝑚 + 𝑥𝑚 + 𝑥𝑚 = 25 ⎪
𝐻𝑀1𝑙 = 1∕1607.3+1∕275.2+1∕14350 = 693.5504 (Harmonic mean of 16 26 36

optimal solutions obtained in lower level), 𝑥𝑚
17
+ 𝑥𝑚
27
+ 𝑥 𝑚
37
= 30 ⎪
3 ⎪
𝐻𝑀1𝑚 = 1∕2026+1∕526+1∕25613 = 1232.6571 (Harmonic mean of 𝑥𝑚
18
+ 𝑥𝑚
28
+ 𝑥𝑚38
= 40 ⎪
optimal solutions obtained in Middle level), ⎪
𝑥𝑚
𝑖𝑗 ≥ 𝑥𝑙𝑖𝑗 ∀ 𝑖, 𝑗 ⎭
3
𝐻𝑀1𝑢 = 1∕2684+1∕971.9+1∕34473 = 2097.1704 (Harmonic mean of
optimal solutions obtained in Upper level). (70)

11
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 6 Table 7
Data for the given problem. Final optimal values table.
𝐷1 𝐷2 𝐷3 𝐷4 Supply Lower level Middle level Upper level Composite fuzzy
optimal solution
𝑆1 (1, 1.5, 2) (1, 2, 3) (5, 7, 9) (4, 6, 8) (7, 8, 9)
(3, 4, 5) (2, 4, 6) (2, 3, 4) (1, 3, 5) 𝑥𝑙11 = 0 𝑥𝑚11 = 0 𝑥𝑢11 = 0 𝑥
̃11 = (0 0 0)
𝑥𝑙12 = 2 𝑥𝑚12 = 3 𝑥𝑢12 = 4 𝑥
̃12 = (2 3 4)
𝑆2 (1, 1.5, 2) (7, 8.5, 10) (2, 4, 6) (3, 4, 5) (17, 19, 21)
𝑥𝑙13 = 0 𝑥𝑚13 = 0 𝑥𝑢13 = 0 𝑥
̃13 = (0 0 0)
(4, 5, 6) (7, 8, 9) (7, 8.5, 10) (9, 10, 11)
𝑥𝑙14 = 5 𝑥𝑚14 = 5 𝑥𝑢14 = 5 𝑥
̃14 = (5 5 5)
𝑆3 (7, 8, 9) (7, 9, 11) (3, 4, 5) (5, 6, 7) (16, 17, 18) 𝑥𝑙21 = 10 𝑥𝑚21 = 11 𝑥𝑢21 = 12 𝑥
̃21 = (10 11 12)
(4, 6, 8) (1, 2, 3) (3, 4.5, 6) (1, 1.5, 2) 𝑥𝑙22 = 0 𝑥𝑚22 = 0 𝑥𝑢22 = 0 𝑥
̃22 = (0 0 0)
Demand (10, 11, 12) (2, 3, 4) (13, 14, 15) (15, 16, 17) 𝑥𝑙23 = 7 𝑥𝑚23 = 8 𝑥𝑢23 = 9 𝑥
̃23 = (7 8 9)
𝑥𝑙24 = 0 𝑥𝑚24 = 0 𝑥𝑢24 = 0 𝑥
̃24 = (0 0 0)
𝑥𝑙31 = 0 𝑥𝑚31 = 0 𝑥𝑢31 = 0 𝑥
̃31 = (0 0 0)
𝑥𝑙32 = 0 𝑥𝑚32 = 0 𝑥𝑢32 = 0 𝑥
̃32 = (0 0 0)
𝑥𝑙33 = 6 𝑥𝑚33 = 6 𝑥𝑢33 = 6 𝑥
̃33 = (6 6 6)
U - CSOLPP :
𝑥𝑙34 = 10 𝑥𝑚34 = 11 𝑥𝑢34 = 12 𝑥
̃34 = (10 11 12)
min 𝑢𝑢11 + min 𝑢𝑢12 + min 𝑢𝑢13
min 𝜆𝑢1 =
𝑥
̃ 𝐻𝑀1𝑙
= 0.1549𝑥𝑢11 + 0.1960𝑥𝑢12 + 0.2141𝑥𝑢13 + 0.1888𝑥𝑢14
Table 8
Comparison of fuzzy compromised solution.
+0.2364𝑥𝑢15 + 0.1845𝑥𝑢16 + 0.2631𝑥𝑢17 + 0.1932𝑥𝑢18
DT Loss Profit
+0.2311𝑥𝑢21 + 0.2530𝑥𝑢22 + 0.1268𝑥𝑢23 + 0.0656𝑥𝑢24 DEA based (1793.7, 2233, 2893) (595, 792, 1285.4) (26715, 31741, 39256)
approach [23]
+0.2479𝑥𝑢25 + 0.2343𝑥𝑢26 + 0.2285𝑥𝑢27 + 0.1881𝑥𝑢28 Proposed (1796.9, 2255, 2967) (544.5, 775.7, 1294.5) (26762, 31787, 40184)
approach
+0.1355𝑥𝑢31 + 0.2270𝑥𝑢32 + 0.1369𝑥𝑢33 + 0.1376𝑥𝑢34

+0.2148𝑥𝑢35 + 0.2213𝑥𝑢36 + 0.2069𝑥𝑢37 + 0.2263𝑥𝑢38 (71)

Table 9
subject to 𝑥𝑢11 + 𝑥𝑢12 + 𝑥𝑢13 + 𝑥𝑢14 + 𝑥𝑢15 + 𝑥𝑢16 + 𝑥𝑢17 + 𝑥𝑢18 = 140 ⎫ Comparison of crisp solution.

⎪ DT Loss Profit
𝑥𝑢21 + 𝑥𝑢22 + 𝑥𝑢23 + 𝑥𝑢24 + 𝑥𝑢25 + 𝑥𝑢26 + 𝑥𝑢27 + 𝑥𝑢28 = 108
⎪ DEA based approach [23] 2306.566 890.8 32570.666
𝑥𝑢31 + 𝑥𝑢32 + 𝑥𝑢33 + 𝑥𝑢34 + 𝑥𝑢35 + 𝑥𝑢36 + 𝑥𝑢37 + 𝑥𝑢38 = 83 ⎪ Proposed approach 2339.6 871.567 32911

𝑥𝑢11 + 𝑥𝑢21 + 𝑥𝑢31 = 33 ⎪

𝑥𝑢12 + 𝑥𝑢22 + 𝑥𝑢32 = 35 ⎪
⎪ using Data envelopment analysis (DEA) based approach [23] and our
𝑥𝑢13 + 𝑥𝑢23 + 𝑥𝑢33 = 40 ⎪
⎬ proposed approach (FHM). Similarly for Example 2, Table 10 shows
𝑥𝑢14 + 𝑥𝑢24 + 𝑥𝑢34 = 49 ⎪ the respective comparison between the fuzzy compromise solution and

𝑥𝑢15 + 𝑥𝑢25 + 𝑥𝑢35 = 60 ⎪ crisp solution obtained by our proposed method and the method used
𝑥𝑢16 + 𝑥𝑢26 + 𝑥𝑢36 = 29 ⎪ by R Gowthami and K Prabakaran [37]. From Tables 8–10, it is clear

⎪ that the optimal solution obtained for Examples 1 and 2 by using the
𝑥𝑢17 + 𝑥𝑢27 + 𝑥𝑢37 = 40
⎪ Fuzzy harmonic mean method (Proposed method) is better than the
𝑥𝑢18 + 𝑥𝑢28 + 𝑥𝑢38 = 45 ⎪ optimal solution obtained by using the fuzzy DEA-based approach

𝑥𝑢𝑖𝑗 ≥ 𝑥𝑚 ⎪ [23] and the method used by R Gowthami and K Prabakaran [37].
𝑖𝑗 ∀ 𝑖, 𝑗 ⎭ Since, due to the conflicting nature of multiple objectives, the time
(72) objective is a little high, but the cost and profit objectives are relatively
more optimized than the other existing methods. In sum, our proposed
After solving each of the three-level single-objective linear program- method gives a better fuzzy compromise solution.
ming problem individually, we will get the following solution Table 5:
6. Advantages and disadvantages
Example 2. Consider a fully fuzzy multi-objective transportation
problem in which the data for transportation cost, transportation time 6.1. Advantages:
together with supplies and demands are measured as triangular fuzzy
numbers and is given in Table 6, It has been noticed that both the fuzzy DEA method and our sug-
gested method reduce FFMOTP (2)–(5) to SOTP. The following are the
Solution: By following the above steps describe in our proposed significant benefits of our proposed strategy over the fuzzy DEA-based
method, the final solution comes out to be (see Table 7), approach [23]:

5. Results and discussion • The main advantage of this proposed method is its computational
simplicity. In the Fuzzy DEA-based approach, a three-step pro-
This section compares the results obtained for Example 1 and Exam- cedure is used to find the fuzzy efficiency score, later splitting
ple 2 by using the proposed method with the other existing methods the entire problem into a three-level crisp linear programming
(M.Bagheri et al. [23], R.Gowthami and K. Prabhakaran [37]). How- problem. However, in our proposed method, the entire problem
ever, to better understand the final results, the ranking function is directly splits into three-level crisp MOLPP. Then, by using fuzzy
defined in Section 2.2.1 is used to defuzzify the fuzzy compromised harmonic mean, it reduces into three-level crisp single-objective
solutions obtained by using the proposed methods and the methods linear programming problems in a single step only. Thus, our
used in [23,37]. proposed method requires less number of steps for calculation
Tables 8 and 9 respectively shows the comparison of final fuzzy and therefore it has less computational complexity in the solution
compromised solutions and crisp solution obtained in Example 1, by process.

12
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

Table 10
Comparison of results.
Objectives Fuzzy compromised solution Crisp solution
R Gowthami and Proposed method R Gowthami and Proposed method
K Prabakaran [37] K Prabakaran [37]
Cost (187.5, 189.5, 191.5) (114, 174.5, 244) 189.5 177.5
Time (176.5, 178.5, 180.5) (126, 193.5, 271) 178.5 196.8

• Using the fuzzy harmonic mean (FHM) operator 2.2.3 as a con- References
verter tool is much simpler than using any other tool.
• Model requirement (6)–(8) of this proposed method converted the [1] G. Monge, The Founding Fathers of Optimal Transport, Springer, 1971.
[2] A.N. Tolstoĭ, On the history of the transportation and maximum flow problems,
entire problem into a linear programming problem (LPP) which
Math. Program. Ser. B 91 (1930) 23–55.
can be easily solved using the optimization tool TORA. [3] L.V. Kantorovich, Mathematical methods of organizing and planning production,
• The fuzzy compromise solution is as a non-negative fuzzy trian- Manag. Sci. 6 (4) (1960) 366–422.
gular number. It means that, the fuzzy quantities of products and [4] F.L. Hitchcock, The distribution of a product from several sources to numerous
the fuzzy cost are all positive. localities, J. Math. Phys. 20 (1941) 224–230.
[5] Tjalling, C. Koopmans, S. Reiter, A model of transportation, in: Activity Analysis
of Production and Allocation, Wiley, New York, 1951, pp. 222–259.
6.2. Disadvantages: [6] A. Charnes, W.W. Cooper, The stepping stone method of explaining linear
programming calculations in transportation problems, Manag. Sci. 1 (1) (1954)
49–69.
• Converting all the objectives of the problem into a minimization [7] G.B. Dantzig, Application of the simplex method to a transportation problem, in:
type is one of the premier requirements of this method. Activity Analysis and Production and Allocation, Wiley, New York, 1951.
[8] H.J. Zimmermann, Fuzzy programming and linear programming with several
• The proposed method is efficient for solving only those type
objective functions, Fuzzy Sets Syst. 1 (1) (1978) 45–55.
of FFMOTP in which the fuzzy parameters are considered as [9] A. Gupta, A. Kumar, A. Kaur, Mehar’s method to find exact fuzzy optimal solution
non-negative triangular fuzzy numbers. of unbalanced fully fuzzy multi-objective transportation problems, Optim. Lett.
6 (8) (2012) 1737–1751.
[10] P. Kundu, S. Kar, M. Maiti, Multi-objective multi-item solid transportation
7. Conclusions and future direction problem in fuzzy environment, Appl. Math. Model. 37 (4) (2013) 2028–2038.
[11] P.K. Giri, M.K. Maiti, M. Maiti, Fully fuzzy fixed charge multi-item solid
transportation problem, Appl. Soft Comput. 27 (2015) 77–91.
This research article introduced a new approach to solve a fully [12] A. Khoshnava, M.R. Mozaffari, Fully fuzzy transportation problem, J. New Res.
fuzzy multi-objective transportation problem (FFMOTP) using the fuzzy Math. 1 (3) (2015) 41–54.
harmonic mean technique. This technique uses fuzzy arithmetic to [13] A. Ebrahimnejad, An improved approach for solving fuzzy transportation problem
with triangular fuzzy numbers, J. Intell. Fuzzy Syst. 29 (2) (2015) 963–974.
decompose FFMOTP into a three-level crisp multi-objective linear pro-
[14] S. Dhanasekar, S. Hariharan, P. Sekar, Fuzzy Hungarian MODI algorithm to solve
gramming problem (TL-CMOLPP). Each of these crisp problem’s ob- fully fuzzy transportation problems, Int. J. Fuzzy Syst. 19 (5) (2017) 1479–1491.
jective functions can be solved independently to get the optimum [15] P. Singh, S. Kumari, P. Singh, Fuzzy efficient interactive goal programming
values that can be used to calculate the fuzzy harmonic means (FHM) approach for multi-objective transportation problems, Int. J. Appl. Comput. Math.
of these values. This FHM is used as a tool to convert TL-CMOLPP 3 (2) (2017) 505–525.
[16] S.K. Bharati, S.R. Singh, A computational algorithm for the solution of fully fuzzy
into TL-CSOLPP. Finally, we solved this TL-CSOLPP independently to
multi-objective linear programming problem, Int. J. Dyn. Control 6 (3) (2018)
get the required fuzzy compromise solution. Moreover, two numerical 1384–1391.
examples are given to test the validity of our proposed approach, and [17] G. Maity, S.K. Roy, Solving fuzzy transportation problem using multi-choice
we have found that our proposed approach provides a better fuzzy goal programming, Discrete Math. Algorithms Appl. 9 (6) (2017) 1750076,
http://dx.doi.org/10.1142/S1793830917500768.
compromise solution for both examples.
[18] V. Vidhya, K. Ganesan, Efficient solution of a multi objective fuzzy transportation
In the future scope of this study, one can try to modify the proposed problem, J. Phys. Conf. Ser. 1000 (1) (2018) 012132, http://dx.doi.org/10.1088/
approach for solving fully fuzzy multi-objective transportation prob- 1742-6596/1000/1/012132.
lems without converting its objective functions into minimization types. [19] A. Mahmoodirad, T. Allahviranloo, S. Niroomand, A new effective solution
method for fully intuitionistic fuzzy transportation problem, Soft Comput. 23
Also, one can generalize this proposed method for solving FFMOTP
(12) (2019) 4521–4530.
with other types of fuzzy parameters. [20] P.U. Maheswari, K. Ganesan, Solving fully fuzzy transportation problem using
pentagonal fuzzy numbers, J. Phys. Conf. Ser. 1000 (1) (2018) 012014, http:
//dx.doi.org/10.1088/1742-6596/1000/1/012014.
Declaration of competing interest [21] A. Baidya, U.K. Bera, Solid transportation problem under fully fuzzy
environment, Int. J. Math. Oper. Res. 15 (4) (2019) 498–539.
[22] P. Anukokila, B. Radhakrishnan, Goal programming approach to fully fuzzy
The authors declare that they have no known competing finan-
fractional transportation problem, J. Taibah Univ. Sci. 13 (1) (2019) 864–874.
cial interests or personal relationships that could have appeared to [23] M. Bagheri, A. Ebrahimnejad, Solving the fully fuzzy multi-objective transporta-
influence the work reported in this paper. tion problem based on the common set of weights in DEA, J. Intell. Fuzzy Syst.
39 (3) (2020) 3099–3124.
[24] G. Krishnaveni, K. Ganesan, A fully fuzzy multi objective fuzzy transportation
Acknowledgments problem under fuzzy environment, AIP Conf. Proc. 2277 (1) (2020) 090009,
http://dx.doi.org/10.1063/5.0025267.
[25] A. Mishra, A. Kumar, Mehar method to find a unique fuzzy optimal value of
The authors would like to acknowledge many helpful comments balanced fully triangular fuzzy transportation problems, in: Aggregation Opera-
and suggestions from anonymous reviewers and editors of this paper to tors for Various Extensions of Fuzzy Set and Its Applications in Transportation
improve the manuscript. Also, the first author would like to thank the Problems, Springer, 2021, pp. 87–118.
[26] A. Mishra, A. Kumar, JMD approach for solving unbalanced fully trapezoidal
Council of Scientific and Industrial Research (CSIR), Govt. of India for
intuitionistic fuzzy transportation problems, in: Aggregation Operators for Var-
providing the financial support as JRF under the HRDG(CSIR) sanction ious Extensions of Fuzzy Set and Its Applications in Transportation Problems,
letter-number/file number 09∕1032(0019)∕2019 − 𝐸𝑀𝑅 − 𝐼. Springer, 2021, pp. 143–203.

13
Y. Kacher and P. Singh Journal of Computational Science 63 (2022) 101782

[27] A.K. Nishad, Abhishekh, A new ranking approach for solving fully fuzzy [37] R. Gowthami, K. Prabakaran, Solution of multi objective transportation problem
transportation problem in intuitionistic fuzzy environment, J. Control Autom. under fuzzy environment, J. Phys. Conf. Ser. 1377 (1) (2019) 012038, http:
Electr. Syst. 31 (2020) 900–911. //dx.doi.org/10.1088/1742-6596/1377/1/012038.
[28] X. Wang, E.E. Kerre, Reasonable properties for the ordering of fuzzy quantities
(I), Fuzzy Sets Syst. 118 (3) (2001) 375–385.
[29] S. Midya, S.K. Roy, V.F. Yu, Intuitionistic fuzzy multi-stage multi-objective fixed-
Mr. Yadvendra received his M.Sc. degree from Guru Ghasi-
charge solid transportation problem in a green supply chain, Int. J. Mach. Learn.
das Central University Bilaspur, India. Currently He is a
Cybern. 12 (3) (2021) 699–717.
Ph.D. scholar in the Department of Mathematics, Motilal
[30] S. Ghosh, S.K. Roy, Fuzzy-rough multi-objective product blending fixed-charge
Nehru National Institute of Technology Allahabad, Praya-
transportation problem with truck load constraints through transfer station,
graj, India. His research interest includes multiobjective
RAIRO Oper. Res. 55 (2021) S2923–S2952.
transportation optimization and its applications.
[31] A. Mondal, S.K. Roy, Multi-objective sustainable opened-and closed-loop supply
chain under mixed uncertainty during COVID-19 pandemic situation, Comput.
Ind. Eng. 159 (2021) 107453, http://dx.doi.org/10.1016/j.cie.2021.107453.
[32] Y. Kacher, P. Singh, A comprehensive literature review on transportation
problems, Int. J. Appl. Comput. Math. 7 (5) (2021) 1–49.
[33] D. Mardanya, G. Maity, S.K. Roy, The multi-objective multi-item just-in-
time transportation problem, Optimization (2021) http://dx.doi.org/10.1080/ Dr. Pitam Singh received his Ph.D. from Department of
02331934.2021.1963246. Mathematics, Motilal Nehru National Institute of Technol-
[34] S. Ghosh, S.K. Roy, A. Ebrahimnejad, J.L. Verdegay, Multi-objective fully ogy Allahabad, India. Currently, he is holding the post
intuitionistic fuzzy fixed-charge solid transportation problem, Complex Intell. of Associate Professor in the same Department. He has
Syst. 7 (2) (2021) 1009–1023. published more than 40 research papers in International
[35] S. Midya, S.K. Roy, G.W. Weber, Fuzzy multiple objective fractional optimization Journals of repute. His research interest includes Opti-
in rough approximation and its aptness to the fixed-charge transportation mization theory, Fuzzy Sets and Systems, GIS and Remote
problem, RAIRO Oper. Res. 55 (3) (2021) 1715–1741. Sensing and its Applications.
[36] E. Fathy, A.E. Hassanien, Fuzzy harmonic mean technique for solving fully fuzzy
multilevel multi-objective linear programming problems, Alex. Eng. J. 61 (10)
(2022) 8189–8205.

14

You might also like