A General Model For Fuzzy Linear Programming
A General Model For Fuzzy Linear Programming
North-Holland
Abstract: In the current literature there are several models of fuzzy linear programming
problems. The aim of this paper is, t~rst, to describe the more important problems in fuzzy
linear programming and, second, to give a general model of fuzzy linear programming
problems involving all of the above ones. Finally, a resolution method for that general model is
proposed.
1. Introduction
Since the pioneer work on fuzzy linear programming (FLP) by Tanaka et al. [8]
and Zimmermann [~_2], in the last past years several kinds of FLP problems have
appeared in the literature [9, 7, 3, 5 , . . . ] and obviously with them d~fferent
approaches of resolution have been proposed too.
Perhaps by the different origins or contexts in which those problems were
formerly considered it seems there is no link among them. However all of these
problems correspond to feasible fuzzifications of a same problem: The conven-
tional linear programming (LP) problem
Max cx
s.t. A x ~ b , (1)
x~>O,
As is well known, the first approach to the FLP problem was introduced by
Bellman and Zadeh in [2]. With X e R ~ a set of possible alternatives, then fuzzy
objective and fuzzy constraints are, respectively fuzzy s e t s / ~ , ~c e IF(X). Thus a
fuzzy decision is the fuzzy set/Jr> ¢ IF(X) defined by
o(x) = ^ x x.
If we want to optimize the decisio~h that is, to find the best x e X which
simultaneously satisfies both objective and constraint (or constraints), we can see
the above problem like a FLP one i~ a wide sense.
Let it be noted that in this basic framework ~of a FLP problem, there are no
additional assumptions about the nature of objective and constraints. Then,
according to the different hypotheses being considered, distinct FLP problems
shall be obtained,
In order to describe the different possible FLP problems, we will consider that
the fuzziness of (1) can appear either in the constraint set or in the coefficients
taking part in the definition of this set.
(a) F L P p r o b l e m s with f u z z y consuaint~: Consider the case in which a decision
maker assumes he can tolerate violations in the accomplishment of the con-
straiats, that is, he permits the constraints to be satisfied 'as well as possible'. For
each constraint in the constraint set this assumption can be represented by
a~x ~ b , i = l, 2, . . . , m,
and modeled by means of a membership function
i if aix <<.b,
~(x) = (asx) if b~ <~a~x <~bt + fi~, (2)
if a~x >~ b~ + b~.
This membership function expresses that the decision maker tolerates violations
in each constraint up the value b~ + / ; , i = I , . . . , m. On the other hand, the
functions ~ are assumed to be non-decreasing and continuous.
Oraphically~ we obtain a picture as shown in Figure 1.
Let it be noted that the functions ~ are defined for every x ¢ X and give the
b.
i
b. ÷~.
.~ a.x
I
Fig. 1.
A general model for fuzzy linear programming 23
accomplishment degree of the i-th constraint for x e R", but this value is
calculated by means of the functions~ which are defined over R.
N o w the associated problem is usually represented by
Max cx
xj>~O, j = l, . . . , n,
This kind of problems was introduced and studied by Tanaka et al. [9]. In [9]
fuzzy costs was also supposed in (6) but here, for the sake of simplicity, crisp
costs are considered.
The main feature of (6) is that there are no fuzzy constraints like in (3). Here
the fuzziness is in the coefficients but the decision maker does not permit any
violation in the accomplishment of the constraints. Thus (3) and (6) are
completely different problems.
To solve (6) the following auxiliary conventional LP problem is proposed in [9]:
Max cx
S.t. [(1 - ½h)(a~ + ~) + ½h(a~ - gi)]x
~<(1 - ~h)(b, + b~) + ~h(b, - b~), (7)
[½h(a, + ~,) + (1 - ~h )(a, - o,)]x
<<.½h(b, + 6,) + (1 - ½h)(b, - b,),
x ~>0,
where triangular fuzzy numbers are supposed and h ¢ (0, 1] is a level or degree of
optimism being specified by the decision maker a priori.
To obtain (7) a particular order relation between fuzzy numbers is considered:
for ~, 6 ¢ N(R), ~ >h 6 if
(a + a)k (b + b)k, (a - (b - b)k (8)
hold for all k ¢ [h, 1], where (a + ~)k and (a-_a)k denote the upper and lower
limits of the k-lesel set of ~.
la order to avoid further mistakes, let it be observed that although the relation
(8) is denoted >h, no fuzziness is assumed in the accomplishment of the
constraints in (6).
So far we have described the main FLP problems when crisp costs are
supposed. Obviously some combination of them may be considered, that is, FLP
problems with fuzzy constraints and fuzzy coefficients in such constraints. The
next section is devoted to propose a general model for FLP which shall consider
this last case.
The aim of this section is to study a model of FLP which permits one: (1) to
involve the particular models so far described, and (2) to give a resolution
method for them and all ~articular problems tha~~,may be deduced from it.
The next situation is considered. Suppose that a decision maker in order to
establish some linear constraint
ailX 1 "~ ai2x 2 " 4 ' ' ' . @ ~linXn ~ b~ (9)
has some information about the constraint as follows: coefficients have values
about a#, j ffi 1 , . . . , n, and never are greater than aq + ~q nor less than aq - aq.
A general model jor fuzzy linear programming 25
Also the value of the right hand side is about b, i ffi 1 , . . . , m, with limits b~ + be
and b i - b~.
On the other hand, the decision maker tolerate.s violations in the accomplish-
ment of the constraints, some imprecise value d~ being the maximum violation
that he permits.
Obviously all of these imprecisions about the coett~cients can be modeled by
means of fuzzy numbers like in (6). But now, another membership function giving
the accomplishment degree of (9) for each fuzzy number a~lxl + ai2X2 " 4 " ' . . +
as,x,, should be considered. This last membership function will also represent a
fuzzy number giving the violation that the decision maker permits.
Then the problem to consider is,
Max cx
s.t. A x ,~ b, (10)
x~>0,
ranking them,
a, b eN(R) ~ a©b,
which preserves the ranking when fuzzy n,.imbers are multiplied by positive
scalars.
Let T, B • N(R) be fuzzy numbers fixed by the decision maker. We define the
fuzzy set
VA eN(R), /~(A)-sup{a~e[0, 1]lAmB + T ( 1 - a~)}. (14)
/~ • IF[N(I~)] gives, for each A e I~I(R), the degree to which the fuzzy number A
verifies the property A @ B + T ( 1 - ~).
For instance, suppose B = (15, 14, 16) and T - (4, 3.5, 5) are triangular fuzzy
numbers fixed by the decision maker. Consider two different relations ~) defined
VX, Y • N(R) by
X~)t Y ¢~ X °<~ yo, Xt~2 Y ¢~ ,f~< Y
where X = (X °, ~, ~') and Y ffi (yo, y, 17'). Then, if A ffi (16, 13, 17), according to
(14) one has
A~)t B + T ( 1 - a~) ¢:~ 16~<15+4(1-a¢) ¢~ a~<~4.
Thus, for (~)l,/~(A) ffi ~.
On other hand,
A(~zB+ T ( 1 - ~ ) ¢~ 17<~14+3.5(1-a¢) ¢~ a~<~
and therefore, in this ease,/A(A) = ~.
Now, by considering .u e F[0=(R)] tv be a convex fuzzy set if and only if for each
A, B e 0:(R) and each ~, • [0, 1],
/412A + (1 - ~.)B] ~>/~(A) ^/~(B), (15)
it is clear that this definition does not introduce any perturbations if A and B are
fuzzy numbers, because addition of fuzzy numbers and its product by a scalar are
well defined operations. Moreover (15) generalizes the classic definition of fuzzy
convex.
Then we can show that (14) is a fuzzy convex set like (15). In fact, if
M, N • N(R) then/~(M) - o¢ and/~(N) - / 3 , that is
4. Numerical example
X1, X2 ~ O,
where
= (3, 2, 4), 4 = (4, 2.5, 5.5), 1"8= (18, 16, 19),
2=(2,1,3), i=(1,0.5,2), 7=(7,6,9),
and the decision maker permits violations in the accGmplishment of each
28 M. Delgado, J.L. Verdegay, M.A. Vila
5. Conclusions
In this paper a general model of FLP has been presented which simultaneously
involves in the constraint set both fuzzy numbers and fuzzy constraints.
Also, an approach has been given to solve it. This approach is based on the
concept of a comparison relation between fuzzy numbers and it has been proved
that for different relations, different auxiliary models and solutions are obtained.
The study of those several auxiliary models and other ones that may be
obtained by considering other comparison relations will be the aim of a
forthcoming paper.
A general model for fuzzy linear programming 29
Achow|edgment
The attthors would like to thank the referees for helpful suggestions on an
earlier version of this paper°
References
11] G. Bortolan and R. Degani, A review of some methods for ranking fuzzy subsets, Fuzzy Sets and
Systems 1,5 (1985) 1-20.
[2] R. Bellman and L.A. Zadeh, Decision making in a fuzzy environment, Management Sci. B 17
(1970) 141-164.
[3] M. Delgado, J.L. Verdegay and M.A. Vila, Mathematical programming problems with fuzzy
costs, Presented at 1st IFSA Congress, Palma de Mallorca, Spain (1985).
[41 Dubois and H. Prade, Fuzzy Sets and System,t Theory and Apptications (Academic Press,
D.
New York, 1980).
[5] M.K. Luhandjula, Linear programming under randomness and fuzziness, Fuzzy Sets and Systems
~.0 (1983) 57-63.
[6] C.V. Negoita and D. Ralescu, Applications of Fuzzy Sets to Systems Analysis (Birkhauser,
Basel-Boston, 1975).
[7] M. Oh'Eigeartaigh, A fuzzy transportation algorithm, Fuzzy Sets and Systems $ (1982) 235-245.
[8] H. Tanaka, T. Okuda and K. Asai, On fuzzy mathematical programming,/. Cybernet. 3 (1974)
37-46.
[9] H. Tanaka, H. Ichihashi and K. Asai, A formulation of fuzzy linear programming problems
based on comparison of fuzzy numbers, Control and Cybernet. 13 (1984) 185-194.
[1o1 I.L. Verdegay, Fuzzy mathematical programming, in: M.M. Gupta and E. Sanchez, Eds., Fuzzy
Information and Decision Processes (North-Holland, Amsterdam, 1982) 231-237.
[11] L.A. Zadeh, Fuzzy sets as a basis for a theory of possibility, Fuzz), Sets and Systems 1 (1978)
3-28.
In] H.-J. Zimmermann, Optimization in fuzzy environment, Presented at XX! Int. TIMS and 46th
ORSA Conference, San Juan, Puerto Rico (1974).
[13] H.-J. Zimmermann, Description and optimization o~ fuzzy systems, lnw~nat. J. General Systems
2 (1975) 209-215.