Sunteți pe pagina 1din 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220473073

A Fuzzy Approach to the Multiobjective Transportation Problem

Article  in  Computers & Operations Research · January 2000


DOI: 10.1016/S0305-0548(99)00007-6 · Source: DBLP

CITATIONS READS

139 508

2 authors:

Lushu Li Kin Keung Lai


City University of Hong Kong
13 PUBLICATIONS   291 CITATIONS   
737 PUBLICATIONS   9,494 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

National Natural Science Foundation of China (NSFC) under grant No. 71473155 View project

All content following this page was uploaded by Kin Keung Lai on 17 January 2019.

The user has requested enhancement of the downloaded file.


Computers & Operations Research 27 (2000) 43}57

A fuzzy approach to the multiobjective transportation problem


Lushu Li , K.K. Lai *
Faculty of Administration, University of New Brunswick, Fredericton, New Brunswick, Canada E3B 5A3
Department of Management Sciences, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong
Received December 1997; received in revised form June 1998

Abstract

The aim of this paper is to present a fuzzy compromise programming approach to multiobjective
transportation problems. A characteristic feature of the approach proposed is that various objectives are
synthetically considered with the marginal evaluation for individual objectives and the global evaluation for
all objectives. The decision-maker's preference is taken into account by his/her assigning the weights of
objectives. With the global evaluation for all objectives, a compromise programming model is formulated.
This model covers a wide spectrum of methods with Zimmermann's fuzzy programming approach essentially
equivalent to one of its special cases. Using ordinary optimization technique, we solve the fuzzy compromise
programming model to obtain a non-dominated compromise solution at which the synthetic membership
degree of the global evaluation for all objectives is maximum. A numerical example is given to demonstrate
the e$ciency of the proposed fuzzy compromise programming approach.

Scope and purpose

In many real-world situations for transportation problems, decisions are often made in the presence of
multiple, con#icting, incommensurate objectives. Intensive investigations on multiobjective linear transpor-
tation problem have been made. Among them Zimmermann's fuzzy programming appears to be an ideal
approach for obtaining the optimal compromise solution to a multiobjective transportation problem.
However, due to the ease of computation, the aggregate operator used in Zimmermann's fuzzy programming
is the &&min'' operator, which does not guarantee a non-dominated solution. In this paper, we propose a new
fuzzy compromise programming approach to multiobjective transportation problems. It is shown that the
approach proposed can give a compromise solution that is not only non-dominated but also optimal in
a certain sense. The proposed approach is robust enough to cover a wide spectrum of methods. Zimmer-
mann's fuzzy programming approach is essentially equivalent to one of the proposed method's special cases.
 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Transportation problem; Multiobjective optimization; Fuzzy compromise programming

*Corresponding author. Tel.: #852-2788-8563; fax: #852-2788-8560.


E-mail address: mskklai@cityu.edu.hk (K.K. Lai)

0305-0548/99/$ - see front matter  1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 0 5 - 0 5 4 8 ( 9 9 ) 0 0 0 0 7 - 6
44 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

1. Introduction

A commonly occurring problem in the distribution of goods from manufacturer to customer can
be described by the general title of &&transportation problem'' (TP), originally developed by
Hitchcock [1]. The TP is well known because its model can be used when a "rm is trying to decide
where to locate a new facility. Good "nancial decisions concerning facility location also attempt to
minimize total transportation and production costs for the entire system. Moreover, there are
many problems which are not exactly the TP, that can also be modeled in the same way. In
addition, although the transportation problem can be solved using the simplex method [2], some
tremendous computational shortcuts can be obtained by exploiting the special structure of the TP
model. More e$cient special purpose algorithms have been developed. Besides the introductory
text book approaches of stepping stone (SS) method [3] and the modi"ed distribution (MODI)
method, the traditional TP often can be solved as minimum cost #ow problem without capacities
[4, 5]. A short}out potential reduction algorithm for TP is also introduced in [6].
In actual transportation problems, multiobjective functions are generally considered, which
includes average delivery time of the commodities, minimum cost, and so forth. Intensive investiga-
tions on multiobjective linear transportation problems (MOLTP) have been made by several
researchers. Aneja and Nair [7] presented a bicriteria transportation problem model. Lee and
Moore [8] studied the optimization of transportation problems with multiobjectives. Diaz [9, 10]
and Isermann [11] proposed the procedures to generate all non-dominated solutions to the
MOLTP. Current et al. [12, 13] did a review of multiobjective design of transportation
networks. Climaco et al. [14] and Ringuest et al. [15] developed interactive algorithms for the
MOLTP.
For a MOLTP with K objective functions, the interactive algorithms developed by Climaco
et al. [14] and Ringuest et al. [15] give more than K non-dominated and dominated solutions.
Thus the decision maker has to determine a preferred compromise solution from the set of
non-dominated solutions. The multi-objective analysis of transportation problems is further
complicated by the fact that the number of non-dominated solutions may grow exponentially
with the problem size [16, 17]. Consequently, it is often impossible or impractical to generate
the entire set of non-dominated solutions, and it is not easy to "nd a preferred compromise
solution.
Fuzzy set theory was proposed by Zadeh [18] and has been found extensive in various "elds. The
early applications of fuzzy set theory to TP include Prade [19], Chanas et al. [20] and Verdegay
[21], etc. Bit et al. [22, 23] advocated the application of Zimmermann's fuzzy programming
[24}27] and presented an additive fuzzy programming model for the MOLTP [28]. The most
important aspect in the fuzzy approach is the compensatory or non-compensatory nature of the
aggregate operator. Several investigators [29, 30, 25] have discussed this aspect. Due to the ease of
computation, the most frequently used aggregation operator in Zimmermann's fuzzy programming
approach is the &&min'' operator which has also been employed by Bit et al. [22, 23], Lai and Hwang
[31] and Chang et al. [32]. The biggest disadvantage of the aggregation operator &&min'' is that it is
non-compensatory in the sense of Yager [33]. In other words, the results obtained by the &&min''
operator represent the worst situation and cannot be compensated by other members which may
be very good. Lee and Li [34] pointed out that the aggregation operator &&min'' does not guarantee
non-dominated solutions for multiple objective programming problems.
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 45

To overcome the drawbacks mentioned above, we propose a fuzzy compromise programming


approach to MOLTP. In the fuzzy compromise programming approach proposed, various
objectives are synthetically considered by marginally evaluating individual objectives and globally
evaluating all objectives. The decision-makers' preferences among the objectives are re#ected in
their global subjective evaluation, taking into consideration the various respective objectives. It is
shown that the approach proposed here can give a compromise solution that is not only
non-dominated but also optimal in the sense of that decision-maker's global subjective evaluation
value is maximum. It is also shown that the approach proposed covers a wide spectrum of methods
with Bit's approach [22, 23, 28] essentially equivalent to one of its special cases under certain
conditions.

2. Multiobjective linear transportation porblem

The well-known traditional TP is concerned with the distribution of goods (products) from
several sources (supply points) to several destinations (demand points) at minimal total transporta-
tion cost. The multiobjective linear transportation problem (MOLTP), on the other hand, deals
with the distribution of goods with the consideration of several objectives, such as transportation
cost, delivery time and quantity of goods delivered, simultaneously.
Consider m sources S , S , 2 , S , n destinations D , D , 2 , D and K objectives
  K   L
Z , Z , 2 , Z . Without loss of generality, we assume that all K objectives are to be minimized.
  )
Suppose that the source S has a given available supply a (i"1, 2, 2 , m) and the destination
G G
D has a given required level of demand b ( j"1, 2, 2 , n) . For each objective Z , a penalty cI is
H H I GH
associated with transporting a unit of the goods from the source S to the destination D . Let
G H
variable x represent the unknown quantity of goods to be transported from source S to
GH G
destinations D , (i"1, 2, 2 , m; j"1, 2, 2 , n).
H
It is usual to assume that the balance condition K a " L b holds (i.e., the total demand is
G G H H
equal to the total supply, since any imbalance can be corrected by introducing a &&"ctitious'' source
or destination). With this assumption, the multiobjective linear transportation problem (MOLTP)
can be formulated as the following multiobjective linear programming problem:

K L
MOLTP: Minimize Z (+x ,)" cI x , k"1, 2, 2 , K (1)
I GH GH GH
G H


K
x "b ∀j"1, 2, 2 , n,
GH H
G
L
Subject to x "a ∀i"1, 2, 2 , m, (2)
GH G
H
x *0 ∀i"1, 2, 2 , m; j"1, 2, 2 , n.
GH

Notice that the balance condition K a " L b is both a necessary and su$cient condition for
G G H H
the existence of a feasible solution of the MOLTP (1)}(2). We denote by X the set of all feasible
46 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

solutions of the MOLTP which is formulated via Eq. (2). Moreover, we have the following two
concepts (cf. [24}27]):

De5nition 2.1. A feasible solution x*"+x*,3X is said to be a non-dominated (e$cient or


GH
Parteo-optimal) solution of the MOLTP (1) and (2) if there exists no other feasible solution
x"+x ,3X such that
GH
K L K L
cI x ) cI *
GH xGH for all k"1, 2, 2 , K (3)
GH GH
G H G H
and

K L K L
cI x ( cI x* for at least one k"1, 2, 2 , K. (4)
GH GH GH GH
G H G H
The set of all non-dominated solution is generally called the complete solution.

De5nition 2.2. An optimal compromise solution of the MOLTP (1)}(2) is a feasible


solution x"+x ,3X at which decision maker's preferences value (decision-maker's global
GH
subjective evaluation value), taking into consideration the various respective objectives, is
maximum.

It is generally accepted that an optimal compromise solution has to be a non- dominated


according to the de"nition of non-dominated solution and it is an optimal solution in a certain
sense. For most real-world MOLTP, the number of non-dominated solutions may grow exponenti-
ally with the problem size [16, 22]. Consequently, it is often impossible or impractical to generate
the entire set of non-dominated solutions. We need only a procedure which "nds an optimal
compromise solution.
In this paper, we present an e$cient fuzzy compromise programming approach to solving
multiobjective linear, as well as non-linear, programming problems. Applying the fuzzy compro-
mise programming approach proposed to the MOLTP model (1)}(2), we can get an optimal
compromise solution by the ordinary technique of linear programming, quadratic programming or
other non- linear programming method.

3. Fuzzy compromise programming

Consider the multiobjective programming problem:


Minimize Z(x)"[Z (x), Z (x), 2 , Z (x)]2,
  )
(5)
Subject to x3X,
where x3RL is an n-dimensional decision variable and X is the set of feasible solutions (feasible
solution space).
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 47

Notice that the multiobjective programming problem (5) often consists of a set of con#icting
goals that cannot be achieved simultaneously. Instead of trying to "nd an optimal solution such
that every objective is optimal (usually this is impossible), we try to "nd an optimal compromise
solution at which the global evaluation of the synthetic membership degree of optimum for all
objectives is maximum. The global evaluation employed here re#ects the decision maker's consid-
eration of all criteria contained in the multiobjective functions.
In this section, we present an approach to obtain the marginal evaluation for each objective and
to aggregate these marginal evaluations into the global evaluation of the synthetic membership
degree of optimum for whole objectives. Based on the global evaluation obtained, we can formulate
a fuzzy compromise programming approach to multiobjective programming problems.

3.1. Marginal evaluation for single objective

For each particular objective Z in the multiobjecitve programming problem (5)


I
(k"1, 2, 2 , K), two values ; and ¸ (; '¸ ) can be assigned as its upper and lower bounds.
I I I I
That is, ; and ¸ can be viewed as the highest acceptable level and the aspired level of achievement
I I
for objective Z , respectively. With the highest acceptable level ; and the aspired level ¸ , we can
I I I
obtain a marginal evaluation for each objective Z . By a marginal evaluation of objective Z , we
I I
mean a mapping
: XP[0, 1] which tells us to what degree the decision x3X makes the
I
objective Z close to its aspiration level ¸ . In other word,
(x)3[0, 1] is the degree of
I I I
compatibility between the realized value Z (x) and the aspiration level ¸ for objective Z . Accord-
I I I
ing to the de"nition of fuzzy sets,
is just a fuzzy subset describing the fuzzy concept
I
of &&optimum'' for objective Z on feasible solution space X, and
(x)3[0, 1] is the member-
I I
ship degree of x3X to which x is compatible with the &&optimal solution'' considering only the
objective Z .
I
Once the highest acceptable level ; and the aspired level ¸ have been speci"ed, the following
I I
formula can be employed to de"ne the marginal evaluation mapping
: XP[0, 1] for the
I
objective Z (k"1, 2, 2 , K)
I


1 if Z (x))¸ ,
I I
Z (x)!;

(x)" I I if ¸ (Z (x)(; , (6)
I ¸ !; I I I
I I
0 if Z (x)*; .
I I

In most practical applications, the highest acceptable level ; and the aspired level ¸ can be
I I
determined from the ideal solution [35, 19] of the multiobjective programming problem (5) in the
following way:
Let (x*, x*, 2 , x)*) be the ideal solution of the multiobjective programming problem (5),
i.e. xI* is the optimal solution of the single objective programming problem

min Z (x) (k"1, 2, 2 , K).


I
VZ6
48 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

Then the values of all the K objective functions can be calculated at all these K optimal solution
xI* (k"1, 2, 2 , K) to form a payo! matrix

Z (x*) Z (x*) 2 Z (x)*)


  
Z (x*) Z (x*) 2 Z (x)*)
   . (7)
$ $ $
Z (x*) Z (x*) 2 Z (x)*)
) ) )

The diagonal of the matrix (7) just constitutes the K ideal values Z>"Z (xI*) of objectives
I I
(k"1, 2, 2 , K).
From the payo! matrix (7), the highest acceptable level ; and the aspired level ¸ can be
I I
determined for each objective Z (k"1, 2, 2 , K) via
I
; " max +Z (xH*),
I I
)H)) (k"1, 2, 2 , K). (8)
¸ "Z>"Z (xH*)
I I I
Alternately, the aspired level ¸ and the highest acceptable level ; can also be determined by
I I
solving 2K single-objective programming problems to obtain the ideal value Z> and the anti-ideal
I
value Z\ [19, 35] for each objective Z . That is
I I
; "Z\"max Z (x)
I I I
VZ6 (k"1, 2, 2 , K) (9)
¸ "Z>"min Z (x)
I I I
VZ6
3.2. Global evaluation for multiple objectives

Having all the marginal evaluations


(x),
(x), 2 ,
(x) for a given decision x3X, the
  )
problem is then to determine a global evaluation of x with respect to all the objectives. Thus, we
want to de"ne a mapping (score function) k : XP[0, 1] which tells us to what degree each decision
x3X satis"es all of the objectives, or in other words, what is the global subjective evaluation value
of decision x. Here k : XP[0, 1] is just a fuzzy subset describing the fuzzy concept of &&optimum''
for all the objectives in feasible solution decision space X, and k(x)3[0, 1] is the subjective
membership degree of x3X to which x is compatible with the &&optimal solution'' considering all
the objectives. According to De"nition 2.2, an optimal compromise solution is just the point in the
feasible solution space X at which the global subjective evaluation value (subjective membership
degree of &&optimum'') is maximum.
The word subjective has been employed here intentionally to stress that, unlike as the marginal
evaluation mechanism, the subjectivity of the decision-maker must be taken into account, for
di!erent decision-makers with the same objectives will not reach the same decision. In this paper,
we assumed that the subjectivity of the decision-maker is re#ected by his/her judgment on the
relative importance of each objective. The relative importance of objectives is usually given by a set
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 49

of weights which are normalized to sum to one, i.e.


)
w"(w , w , 2 , w ), w *0 and w "1. (10)
  ) I I
I
Many procedures for the determination of weights have been proposed over the past 20 years.
There are a vast literature on methods and algorithms of determining the weights [35}39]. We
understand that the choice of weights will a!ect the decision result. The intensive investigation on
the e!ects of various weighting methods should be undertaken. For the limitation of this paper,
however, the methods and algorithms of assigning weights and the e!ects of various weighting
methods will not be discussed further. For details of assigning weights see, for example,
[35, 37}39]. In this paper, we assumed that the weights were assigned by the decision maker using
direct assessing method or Saaty's eigenvector method in AHP [28].
For given weights w"(w , w , 2 , w ) and marginal evaluations
(x) ( j"1, 2, 2 , K) of
  ) I
decision x3X, we are looking for a suitable K-place aggregation operator 'w : [0, 1])P[0, 1]
such that for ∀x3X,
k(x)"'w (
(x),
(x), 2 ,
(x)). (11)
  )
The mapping k : XP[0, 1] de"ned via Eq. (11) is just the global evaluation for all the objectives,
and k(x) is the membership degree of x3X to which x is compatible with the ideal solution
(x*, x*, 2 , x)*) of the multiobjective programming problem (5).
An important family of aggregation operators we can employ is the weighted root- power mean
operator M? w :

 
) ?
Mw? (a , a , 2 , a )" w a? (0(" a "(R), (12)
  ) G G
G
because it covers a wide spectrum of aggregation operators which are often used in the areas of
multicriteria decision making and multiple objective programming problems. For example, the
weighted root-power mean operator includes
(1) The weighted arithmetic mean (a"1):
)
Mw(a , a , 2 , a )" w a (13)
  ) G G
G
(2) The weighted quadratic mean (a"2):

 
) 
Mw (a , a , 2 , a )" w a . (14)
  ) G G
G
and the three limit cases:
(3) The weighted geometric mean (aP0):
)
Mw (a , a , 2 , a )" “ aUG (15)
  ) G
G
(4) The disjunctive mean (aPR):
M(a , a , 2 , a )" max a (w "w "2"w "1/K). (16)
  ) G   )
)G))
50 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

(5) The conjunctive mean (aP!R):


M\ (a , a , 2 , a )" min a (w "w "2"w "1/K). (17)
  ) G   )
 G )
))

We note that Mw produces the weighted additive model used in goal programming problems,
whereas M\ is the most frequently used aggregation operator in fuzzy multicriteria decision
making [40] and Zimmermann's fuzzy programming problems [24}27].

3.3. Fuzzy compromise programming

Having chosen a suitable aggregating operator 'w and obtained the global evaluation k:
XP[0, 1] for all the objectives, we can convert the multiobjective programming problem (5) into
the following fuzzy compromise programming problem:
Maximize k(x)"'w (
(x),
(x), 2 ,
(x))
  )
(18)
Subject to x3X.
The fuzzy compromise programming problem (18) is a single objective programming problem.
We can employ the ordinary optimization technique to solve it. Let x*3X be the optimal solution
of the fuzzy compromise programming problem (18), i.e.
k(x*)"max k(x). (19)
VZ6
The following theorem then indicates that the x* obtained from Eq. (19) is a non-dominated
compromise solution at which the synthetic membership degree of optimum for all objectives is
maximum.

Theorem 3.1. Suppose w '0 and the marginal evaluation


(x) of objective Z (x) is strictly
I I I
decreasing with respect to objective value Z (x) for all k"1, 2, . . . K. If we employ the weighted
I
root-power mean M? w (0(" a "(R) as an aggregating operator to de,ne the global evaluation
mapping, then the optimal solution x* of fuzzy compromise programming problem (18) is also a non-
dominated (e.cient or Pareto-optimal) solution of the multiobjective programming problem (5). That
is, there is no other solution x3X such that
Z (x))Z (x*) for all k"1, 2, 2 , K (20)
I I
and
Z (x)(Z (x*) for at least one k "1, 2, 2 , K. (21)
I I 
Proof. We prove this theorem by contradiction. Suppose x* is a dominated decision of the
multiobjective programming problem (5), i.e., there exists a decision x3X such that Eqs. (20) and
(21) hold. Since each marginal evaluation mapping
(x) decreases strictly with respect to the
I
corresponding objective value Z (x), we have
I

(x)*
(x*) for all k"1, 2, 2 , K
I I
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 51

and


(x)'
(x*) for at least one k "1, 2, 2 , K.
I I 
w (0(" a "(R)
Hence, it follows from the strict monotonicity of the weighted root-power mean M?
that

k(x)"M?w (
(x),
(x), 2 ,
(x))
  )
'M?w (
(x*),
(x*), 2 ,
(x*))"k(x*). (22)
  )
This inequality is contrary to the fact that x* is the optimal solution of the fuzzy compromise
programming problem (18). We end the proof. )

4. Fuzzy compromise programming for MOLTP

The multiobjective linear transportation problem (MOLTP) consisting of Eqs. (1) and (2) is
a multiobjective linear programming problem. Therefore, it can be solved with the fuzzy compro-
mise programming approach described above. The details of foregoing steps may be presented as
follows:

Step 1: Solve the following single transportation problem independently for each objective
Z (k"1, 2, 2 , K) (ignore all other objectives each time):
I
K L
min Z (+x ,)" cI x (23)
+ ,
I GH GH GH
VGH Z6 G H
to obtain the optimal solution xI*"+xI*,3X, where X is the set of all feasible solutions
GH
formulated by Eq. (2).
Step 2: Calculate the values of all the K objective functions at all these K optimal solution
xI* (k"1, 2, 2 , K), and then let

; " max +Z (xH*),


I I
) H)) (k"1, 2, 2 , K). (24)
¸ "Z (xI*)
I I

Step 3: De"ne the marginal evaluation


(+x ,) corresponding to the kth objective via
I GH


1 if Z (+x ,))¸ ,
I GH I
Z (+x ,)!;

(+x ,)" I GH I , if ¸ (Z (+x ,)(; , (25)
I GH ¸ !; I I GH I
I I
0 if Z (+x ,)*; .
I GH I
52 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

Step 4: Determine the weight w"(w , w , 2 , w ) and choose a suitable weighted root-power
  )
mean M?
w to formulate a fuzzy compromise programming problem as follows:

w (
(+x ,),
(+x ,), 2 ,
(+x ,)).
min k(+x ,)"M? (26)
+ GH  GH  GH ) GH
VGH+Z6
Step 5: Solve the fuzzy compromise programming problem (26) using the ordinary optimization
technique to obtain the optimal compromise solution of the MOLTP.
In above algorithm, the Steps 1 and 2 can be replaced by the following Steps 1 and 2,
respectively:

Step 1: Solve the following single transportation problem and its anti-problem independently
for each objective Z (k"1.2, 2 , K) (ignore all other objectives each time)
I
K L
min Z (+x ,)" cI x , (27)
+ I GH GH GH
VGH+Z6 G H
K L
max Z (+x ,)" c Ix (28)
+ I GH GH GH
VGHZ6+ G H
to obtain the optimal solutions xI*"+xI*,3X for the model (27) and xI>"+xI>,3X for
GH GH
model (28), where X is the set of all feasible solutions formulated by Eq. (2).
Step 2: Let

K L
; " cI xI>
I GH GH
G H (k"1, 2, 2 , K). (29)
K L
¸ " c x I I*
I GH GH
G H
From a practical standpoint, the most important values for the parameter a are 1, 2 and !R.
We discuss the details as follows
Case 1: a"1, then

) Z (+x ,)!;
k(+x ,)" w I GH I
GH I ¸ !;
I I I
) K L cI x !;
" w G H GH GH I
I ¸ !;
I I I

 
K L ) w cI ) w;
" I GH x ! I I .
¸ !; GH ¸ !;
G H I I I I I I
Thus, the fuzzy compromise programming problem (26) is just Bit's additive fuzzy programming
model for multiobjective transportation problem [28]. It is an ordinary single objective linear
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 53

transportation problem and we can solve it by any standard linear programming package, such as
the linear interactive and discrete optimizer (LINDO) and the optimization toolbox of MATLAB.
Case 2: a"2, then

  
) Z (+x ,)!; 
k(+x ,)" w I GH I "( y2Hy#P2y#R,
GH G ¸ !; 
I I I
where
y"(x , 2 , x , x , 2 , x , 2 , x , 2 , x )2,
 L  L K KL
) w
H"2 I C2 C ,
(¸ !; ) I I
I I I
) w;
P"!2 I I C2 ,
¸ !; I
I I I

 
) ; 
R" w I
I ¸ !;
I I I
and
C "(cI , 2 , cI I I I I
L, c , 2 , cL , 2 , cK , 2 , cKL)2.
I 
Thus, the fuzzy compromise programming problem (26) is equivalent to a quadratic program-
ming problem and we can solve it by any standard quadratic programming package.
Case 3: a"!R and w "w "2"w "1/K, then
  I

 
Z (+x ,)!; Z (+x ,)!; Z (+x ,)!;
k(+x ,)"M\  GH  ,  GH  , 2 , ) GH )
GH ¸ !; ¸ !; ¸ !;
    ) )
) Z (+x ,)!;
" R I GH I (30)
¸ !;
) I I
where "min is the minimization operator. Therefore, if we let
) Z (+x ,)!;
j" R I GH I,
¸ !;
I I I
then the fuzzy compromise programming problem (26) is essentially equivalent to the following
fuzzy programming problem:
Maximize j


Z (+x ,)!;
Subject to I GH I*j, k"1, 2, 2 , K
¸ !;
I I (31)
+x ,3X,
GH
Thus, the fuzzy programming approach to MOLTP developed by Bit et al. [22, 23] is a special
case of the fuzzy compromise programming approach proposed in this paper.
54 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

5. Numerical example

Consider the following two-objective transportation problem [7, 22, 15]:

Min Z "x #2x #7x #7x #x #9x #3Xx #4x


        
#8x #9x #4x #6x ,
   
Min Z "4x #4x #3x #3x #5x #8x #9x #10x
        
#6x #2x #5x #x
   
Subject to

  
x "8, x "19, x "17,
H H H
H H H
   
x "11, x "3, x "14, x "16,
G G G G
G G G G
x *0, i"1, 2, 3; j"1, 2, 3, 4. (32)
GH
Ringuest et al. [15] computed (156, 200) as the most preferred value of the objective vector
(Z , Z ). Bit et al. [22] obtained (160, 195) (integer solution) or (160.8591, 193.9260) as the optimal
 
compromise value of the objective vector (Z , Z ).
 
Applying the fuzzy compromise programming with weighted root-power mean M? w as an
aggregation operator and taking the aspired level ¸ and the highest acceptable level ; for each
I I

Table 1
The optimal compromise objective values for Example (32)

Weights Optimal fuzzy compromise objective values: (Z , Z )


 
(w , w )
 
a"1 a"2 a"!R

(0.0, 1.0) (208, 167) (208.00, 167.00)


(0.1, 0.9) (208, 167) (176.00, 175.00)
(0.2, 0.8) (186, 171) (175.30, 175.88)
(0.3, 0.7) (176, 175) (171.62, 180.48)
(0.4, 0.6) (176, 175) (167.84, 185.20)
(0.5, 0.5) (176, 175) (163.97, 190.04) (163.33, 190.83)
(0.6, 0.4) (156, 200) (159.99, 195.01)
(0.7, 0.3) (156, 200) (156.00, 200.00)
(0.8, 0.2) (156, 200) (156.00, 200.00)
(0.9, 0.1) (143, 256) (156.00, 200.00)
(1.0, 0.0) (143, 256) (143.00, 256.00)
(Z>, Z>)"(143, 167); (Z\, Z\)"(265, 310)
   
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 55

objective Z as its ideal value Z> and anti-ideal value Z\, respectively, we obtain the results of this
I I I
example with a calculation program based on MATLAB 5.0 platform. Various weights assigned
and the corresponding optimal compromise objective values for a"1, 2 and !R are shown in
Table 1. The corresponding fuzzy compromise solutions +x*, are omitted for reasons of space
GH
limitations.

6. Conclusions

For multiobjective linear transportation problem with K objective functions, the interactive
algorithms developed by Climaco et al. [14] and Ringuest et al. [15] give more than K non-
dominated and dominated solutions. Thus the decision maker has to determine a preferred compro-
mise solution from the set of non-dominated solutions. For the larger MOLTP, it is not easy to "nd
a preferred compromise solution [22, 23]. However, using the fuzzy compromise programming
approach proposed in this paper, one can easily "nd a preferred compromise solution which is not
only non-dominated but also optimal in the sense of that the decision-maker's global subjective
evaluation value, taking into consideration the various respective objectives, is maximum. In the
fuzzy compromise programming approach proposed, various objectives are synthetically considered
by marginally evaluating individual objectives and globally evaluating all objectives. The decision-
makers' preferences among the objectives are re#ected in their assignments to the weights of
objectives. Diversi"ed weights of objectives result in various non-dominated optimal compromise
solutions. The fuzzy compromise programming covers a wide spectrum of approaches including the
weighted average method, the quadratic programming method and Bit's fuzzy programming ap-
proach [22, 23, 28], and it is applicable to all types of multiobjective transportation problems, linear
or non-linear vector minimum problems and vector maximum problems. On the whole, the fuzzy
compromise programming approach proposed here is a more suitable method for the multiobjective
transportation problem and other multiobjective programming problems.

Acknowledgements

Both authors would like to thank the anonymous referees for helpful and constructive
comments.

References

[1] Hitchcock FL. The distribution of a product from several sources to numerous localities. Journal of Mathematical
Physics 1941;20:224}30.
[2] Dantzig GB. Linear programming and extensions. Princeton, NJ: Princeton University Press, 1963.
[3] Charnes A, Cooper WW, Henderson A. An introduction to linear programming. New York: Wiley, 1953.
[4] Bazaraa MS, Jarvis, JJ. Linear programming and network #ows. New York: Wiley, 1977.
[5] Ruhe G. Algorithmic aspects of #ows in networks. Dordrecht. Kluwer, 1991.
[6] Kaliski JA, Ye YY. A short-out potential reduction algorithm for linear programming. Management Science
1993;39:757}71.
[7] Aneja YP, Nair KPK. Bicriteria transportation problems. Management Science, 1979;25:73}8.
56 L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57

[8] Lee SM, Moore LJ. Optimizing transportation problems with multiple objectives. AIEE Transactional
1973;5:333}8.
[9] Diaz JA. Solving multiobjective transportation problem. Ekonomicko-Matematicky Obzor 1978;14:267}74.
[10] Diaz JA. Finding a compelte description of all e$cient solutions to a multiobjective transportation problem.
Ekonomicko-Matematicky Obzor 1979;15:62}73.
[11] Isermann H. The enumeration of all e$cient solutions. Naval Research Logistic Quarterly 1979;26:123}39.
[12] Current J, Min H. Multiobjective design of transportation networks: Taxonomy and annotation. European Journal
of Operations Research 1986;26:187}201.
[13] Current J, Marish M. Multiobjective transportation network design and routing problems: Taxonomy and
annotation. European Journal of Operations Research 1993;65:4}19.
[14] Climaco JN, Antunes CH, Alves MJ. Interactive decision support for multiopbjective transportation problems.
European Journal of Operations Research 1993;65:58}67.
[15] Ringuest J-L, Rinks DB. Interactive solutions for the linear multiobjective transportation problem. European
Journal of Operations Research 1987;32:96}106.
[16] Hansen P. Bicriterion path problems. Lecture Notes in Economics and Mathematical Systems, vol. 177, Berlin:
Springer, 1980.
[17] Ruhe G. Complexity results for multicriterial and parametric network #ows using a pathological graph of Zadeh.
Zeitschrift fur Operations Research 1988;32:9}27.
[18] Zadeh LA. Fuzzy sets. Information Control 1965:8:338}53.
[19] Prade H. Operations research with fuzzzy data. In Wang PP, Chang SK, editors. Fuzzy sets and applications to
policy analysis and information systems. New York: Plenum Press 1980.
[20] Chanas S, Kolodziejczyk W, Machaj A. A fuzzy approach to the transportation problem. Fuzzy Sets and Systems
1984;13:211}21.
[21] Verdegay JL. Applications of fuzzy optimization in operation in operational research. Control Cybernetics
1984;13:229}39.
[22] Bit AK, Biswal MP, Alam SS. Fuzzy programming approach to multicriteria decision making transportation
problem. Fuzzy Sets and Systems 1992;50:35}41.
[23] Bit AK, Biswal MP, Alam SS. Fuzzy programming approach to multiobjective solid transportation problem. Fuzzy
Sets and Systems 1993;57:183}94.
[24] Zimmermann H-J. Fuzzy programming and linear programming with several objective functions. Fuzzy Sets and
Systems 1978;1:45}55.
[25] Zimmermann H-J. Applications of fuzzy set theory to mathematical programming, Information Science
1985;34:29}58.
[26] Zimmermann H-J. Fuzzy sets, decision making and expert systems. Dordrecht: Kluwer Academic Publishers, 1987.
[27] Zimmermann H-J. Fuzzy set theory } and its applications. Boston: Kluwer Academic Publishers, 1996.
[28] Bit AK, Biswal MP, Alam SS. An additive fuzzy programming model for multiobjective transportation problem.
Fuzzy Sets and Systems 1993;57:313}19.
[29] Fodor J. Fuzzy reference modeling and multicriteria decision support. Dordrecht: Kluwer Academic, 1994.
[30] Luhandjula MK. Compensatory operators in fuzzy linear programming with multiple objectives. Fuzzy Sets and
Systems 1982;8:245}52.
[31] Lai YJ, Hwang CL. Fuzzy mathematical programming} methods and applications. Lecture notes in economics and
mathematical systems, vol. 394. Berlin: Springer, 1992.
[32] Chang NB, Wen CG, Chen YL. A fuzzy multi- objective programming approach for optimal management of the
reservoir watershed. European Journal of Operations Research 1997;99:289}302.
[33] Yager RR. Competitiveness and compensation in decision making. Iona College Report, RRY 78-14, New York:
New Rochelle, 1978.
[34] Lee ES, Li RJ. Fuzzy multiple objective programming and compromise programming with Pareto optimum. Fuzzy
Sets and Systems 1993;53:275}88.
[35] Hwang CL, Yoon KS. Multiple attribute decision making: methods and applications. Berlin: Springer, 1981.
[36] Diagoulaki DM, Papayannakis L. determining objective weights in multiple criteria problems: the CRITIC
method. Computers & Operations Research 1995;22:763}70.
L. Li, K.K. Lai / Computers & Operations Research 27 (2000) 43}57 57

[37] Keeney RL, Rai$a H. Decision with multiple objectives: perferences and value trade-o!s. New York: Wiley, 1976.
[38] Saaty TL. Multicriteria decision making: the analytic hierarchy process. Pittsburgh, USA, RWS Publications, 1988.
[39] Schenkerman S. Use and abuse of weights in multiple objectives decision support models. Decision Science
1991;2:368}78.
[40] Ribeiro RA. Fuzzy multiple attribute decision making: a review and new preference elicitation techniques. Fuzzy
Sets and Systems 1996;78:155}81.
[41] Chankong V, Haimes YY. Multiobjective decision making theory and methodology, Series in System Science and
Engineering. Amsterdam: North-Holland, 1983.
[42] Zeleny M. Multiple criteria decision making. New York: McGraw-Hill, 1982.

Lushu Li received his Ph.D degree in System Engineering from Southeast University, China. Currently, he is
a Post-doctoral Fellow at Faculty of Administration, University of New Brunswick, Canada. He has scienti"c
publications in EJOR, Fuzzy Sets and Systems and J. Fuzzy Mathematics. His main area of research interests includes
multiobjective optimization, MCDM, intelligent decision-making modeling, fuzzy logic and fuzzy neural networks.
K.K. Lai received his Ph.D. at Michigan State University, and is currently a Professor of Operations Research and
Management Sciences at the City University of Hong Kong. He is also the Associate Dean of the Faculty of Business.
Prior to his current post, he was an Operational Research Analyst at Cathay Paci"c Airways and an Area Manager for
Marketing Information Systems at Union Carbide Eastern. Professor Lai's main area of research interests is logistics and
operations management, computer simulation and business decision modeling.

View publication stats

S-ar putea să vă placă și