Professional Documents
Culture Documents
H.-J. Z I M M E R M A N N
Lehrstuhlfiir Unternehmensforschung (Operations Research),
R WTH Aachen, 51 Aachen, Federal Republic of Germany
In the recent past numerous models and methods have been suggested to solve the
vectormaximum problem. Most of these approaches center their attention on linear programming
problems with several objective functions. Apart from these approaches the theory of fuzzy sets has
been employed to formulate and solve fuzzy linear programming problems. This paper presents the
application of fuzzy linear programming approaches to the linear vectormaximum problem. It shows
that solutions obtained by fuzzy linear programming are always efficient solutions. It also shows the
consequences of using different ways of combining individual objective functions in order to
determine an "optimal" compromise solution.
1. Introduction
In the recent past it has become more and more obvious that comparing different
ways of action as to their desirability, judging the suitability of products, determining
"optimal" solutions in decision problems can in many cases not be done by using a
single criterion or a single objective function. This leads to numerous evaluation
schemes (for instance in the areas of cost benefit analysis and marketing) and to the
formulation of the vectormaximum problem in mathematical programming.
Another way of describing decision problems, particularly in case of ill-defined
problems, is the use of fuzzy sets. Decision problems can be formulated as fuzzy
decision models. Crisp models can then be designed which are equivalent to the fuzzy
models and which can be solved by using existing standard algorithms. This approach
can, of course, be used in particular for decision problems which have the structure of
linear programming.
In the following, the assumptions and results of some approaches of multicriteria
programming will be critically compared with those of fuzzy linear programming.
The vectormaximum problem was first mentioned by K uhn and Tucker in their
*Paper presented at the Joint ORSA/TIMS Meeting, Miami, November 3-5, 1976.
45
46 H.-J. Z immermanr'
"max" Z ( x ) =
(' 2
xl)
x2
(effect on balance of trade),
(profit),
xl +3x2 <27
4xt + 3x2 <45 capacity constraints.
3x~ + X2~30
xl,x2 >0
Fig. 1 shows the solution space of this problem. The "complete solution" is the edge
x I -x2-x3-x4. x I is optimal with respect to objective function zl ( x ) = - x ~ + 2x2
(i.e. best improvement of balance of trade), x 4 is optimal with respect to objective
function z2(x)=2xl + x2 (profit). The "optimal" values are zl (x 1)= 14 (maximum net
export) and z2 (x 4)= 21 (maximum profit), respectively. For x l = (7;0) T total profit is
z2(x ~)= 7 and x4= (9, 3)x yields zl (x 4)= - 3, i.e. a net import of 3.
Solution x s = (3.4; 0.2) T is the solution which yields z~ (x 5)= - 3, z2 (x 5)= 7 that is
Fuzzy programming and linear programming with several objectivefunctions 47
4 x2
1
x
z1
2
C
~z2
Xv v ~
i • i • ! w I • •
0 2 3 4 5 6 7 8 9 10 x1
Fig. 1. Vectormaximum problem.
the lowest "justifiable" value of the objective functions in the sense that a further
decrease of the value of one objective function can not be "balanced" or even
counteracted by an increase of the value of the other objective function. 1
Three major approaches are known, to single out one specific solution from the set of
efficient solutions which qualifies as an "optimal" compromise solution:
(1) The utility approach e.g.: [7,8].
(2) Goal Programming e.g.: [3].
(3) Interactive Approaches e.g.: 1-2, 4, 5, 17].
The first two of these approaches assume that the decision maker can specify his
"preference-function" with respect to the combination of the individual objective
functions in advance, either as "weights" (utilities) or as "distance-functions"
(concerning the distance f.i. from the "ideal solution"2). Generally they assume that the
combination of the individual objective functions which arrives at the compromise
solution with the highest overall utility, is achieved by linear combinations (i.e. adding
the weighted individual objective functions).
The third approach uses only local information in order to arrive at an acceptable
compromise solution.
3. Fuzzyapproaches
3ol. Fuzzy linear programming using the mimoperator
In a fuzzy environment a decision can be defined as follows [l, 12]"
1For more than 2 objective functions see p. 52.
2Se¢ [5, 6b].
48 H.-J. Z immermann
The fuzzy objective function is characterized by its membership function, and so are
the constraints. Since we want to satisfy (optimize) the objective function as well as the
constraints, a decision in a fuzzy environment is defined in analogy to nonfuzzy
environments as the selection of activities which simultaneously satisfy Objective
function(s) and constraints. In fuzzy set theory the intersection of sets normally
corresponds to the logical "and". The "'decision" in a fuzzy environment can therefore
be viewed as the intersection of fuzzy constraints and fuzzy objective function{s). The
relationship between constraints and objective functions in a fuzzy environment is
therefore fully symmetric, i.e. there is no longer a difference between the former and
latter. Applied to linear programming the fuzzy "decision" can therefore be defined as
the intersection of the fuzzy sets describing the constraints and the objective functions.
If one defines the solution with the highest degree of membership to the fuzzy "decision-
set" as the maximizing decision, then the following approach can be used [14, 15]:
Starting from the problem
Ax~b,
x~O. (4)
Here Zo means an aspiration level of the decision maker.
We now define membership functions p4 such that
Using the simplest kind of membership functions, i.e. linear functions of the type
1 for (Bx)i<-b~
(Bx )i - b~
/~(Bx)= f 1 for bi < (Bx )i < b; + d i, (8)
di
0 for (Bx )i > b~ + di,
substituting
b'; b~
di
and
B~ Bi
di
componentwise, and dropping the "1" (which does not change the problem) we arrive
at the following problem:
Max 2. (10)
such that
2<b~'-(B'x)i, i=0(1)m,
x>0.
We shall now apply this approach to the vectormaximum problem of Example 2.1
and make the following assumptions:
The membership functions/~l(x) and /~z(x) of the fuzzy sets characterizing the
objective functions rise linearly from 0 to 1 at the highest achievable values of zt Ix) = 14
and Zz(x)= 21, respectively.
That means that we assume that the level of satisfaction with respect to the
improvement of the balance of trade rises from 0 for imports of 3 units or more to 1 for
exports of 14 and more and the satisfaction level with respect to profit from 0 if the
profit is 7 or less to 1 if total profit is 21 or more.
5In the sensethat the optimal solution to (10)is also optimalto (9).
6See [14] p. 213.
50 H.-J. Z immermann
Thus
0 for z l ( x ) - - 3 ,
zl(x)+3
m(x)=~ f o r - 3 < z l ( x ) < 14,
17
1 for 14<zx(x),
(ll)
0 for z2(x)<7,
z2(x)-7
~2(X)-- 4 for 7 < zz(x) <__21,
14
1 for 21 < Ze(X).
In analogy to formulation (4) and including the unfuzzy constraints of Example 2.1 we
arrive at the following problem-formulation
Max 2,
such that
2__< -0.05882xl +0.117 x2 +0.1764,
2<_ +0.1429 xl +0.0714x2 -0.5,
21> --X 1 + 3X 2,
27> Xl + 3X2,
45> 4Xt + 3X2,
30> 3Xl + x2,
x> 0, (12)
depicted in Fig. 2.
The maximum degree of "overall satisfaction" (2max=0.74) is achieved for the
solution Xo=(5.03," 7.32)T. This is the "maximizing solution" which yields in our
example a profit of $17.38 and an export contribution of $4.58. The basic solution x 2
= (3; 8)T reaches a level of 2 = 0.5 and, by assumption, the (efficient) solutions x 1 and x 4
yield ;t = 0.
In contrast to the usual vectormaximum models in the fuzzy approach the efficient
solutions contained in the "complete solution" are distinguishable by their different
degrees of satisfaction.
X2
In addition Hamacher [6] has shown that the conr~ective D corresponding to the
logical "and" has to be
A 1" D is associative.
A2: D is continuous.
A 3: D is injective in each argument.
A 4" A D(x, x )= x.c~x = I,
x e ( O . 1}
If we apply 114) in Example 2.1 we get f o r - 3 < z l ( x ) < 1 4 and 7<z2(x)<21 the
objective function
(--xt+2x2+3) (2x2+x2 - 7 )
17 14
such that
- x l + 3x2 <21,
xl + 3x2 =<27,
4Xl + 3x2 _--<45,
3xl + X2-----30,
x~ >0.
X2
10
X1
such that
Ax<b,
x>_O.
The question arises whether the optimal solutions of problems of this type are always
efficient solutions of the corresponding vectormaximum problem:
In the following we a~sume p~ to be equal to Min.i,~ z~(x~,t). (This corresponds to the
"lowest justifiable solution" defined on page 45). We also assume d~ >0. It can be
shown that under some weak assumptions there exists a set of parameters d*, i = 1 (1)K
for almost each efficient solution x~ such that x~ is the optimal solution of problem (16).
This statement can be formulated as the following theorem:
Theorem 3.1. Let E be the set of all efficient solutions in (1). Let zl (x), .... zh., (x), K'
< K be a minimal generating system s of set E (i.e. redundant objective lunctions being
omitted) with the individual optima Xopt,...,Xop1 h t. Let F~ be equal to E - t XIo.1o t,..., Xoptj.
K )
Then f o r each ,~ ~ E there exists d~{, . . . d*, such that ~ solves (16).
K (ci)T ~ - 0 ,
il-Ii d* = 1,
and
KN (ci) I x - - P i <
1 for all x ~ X.
i=lll d? --
Thus
Corollary 3.2. Variations ofdffor di > (c i)T X oip t - - Pi > 0 do not influence the "max imi =in g
solution" of problem (16) but only the resulting values of the objective.function.
c~ x - Pi
#i(x) = elsewhere, ( 17 )
di
for (el) ~ x > p i + d i .
aSee [ I0].
54 H.-J. Z immermann
K' l K'
~ Max I-I (c')T xi-p,, (18)
i= i=1
such that
Ax<b,
x>O.
In (18) the optimal solution is ~bviously independent of the first term, 1-I~'_-'1 1/di, of the
objective function.
This means that for the optimal solution of (16) it is irrelevant whether the "anchor
points", i.e. the aspiration levels for the individual objective functions at which #,-(x)
= 1, are set at the individual optima or at any higher level. This is not true when using
the minimum operator instead of the product operator!
4. Conclusions
References
[1] R. Bellman and L.A. Zadeh, Decision making in a fuzzy environment, Management Sci. 17B (4) (1970)
141-164.
[2] R. Benayoun, J. De Montgolfier, J. Tergny and O. Laritchev, Linear programming with multiple
objective functions: STEP method (STEM), Math. Programming 1 (1971) 366-375.
[3] A. Charnes and W.W. Cooper, Management Models and Industrial Applications of Linear
Programming (Wiley, New York, 1961).
[4] J.S. Dyer, Interactive goal programming, Management Sci. 19 (1972/73) 62-70.
Fuzzy programming and linear programming witk several objectivefunctions 55
[5] I. Fandel, Optimale Entscheidung bei mehrfacher Zielsetzung, Lecture Notes in Economics and
Mathematical Systems 76 (Springer, Berlin, 1971).
[6a] A.M. Geoffrion, J.S. Dyer and A. Feinberg, An interactive approach for multi-criterion optimization
with an application to the operation of an academic department, Management Sci. 19 (1972/73) 357-
368.
[6b] A.M. Geoffrion, A parametric programming solution to the vector maximum problem with
applications to decisions under uncertainty, Stanford University, CA (1965) 2.
[7] H. Hamacher, Ober logische Verkniipfungen unscharfer Aussagen und deren zugeh6rigcn
Bewertungsfunktionen, working paper No. 75/14, Lehrstuhl fiir Unternehmensforschung, RWTH
Aachen (1975).
[8] H.W. Kuhn and A.W. Tucker, Nonlinear programming, in" J. Neyman ed., Proc. of 2nd Berkeley
Syrup. on Mathematical Statistics and Probabilities (1951).
[9] H.W. Kuhn, A.W. Tucker and A.M. Geoffrion, A parametric programming solution to the vector
maximum problem, Techmca! Repui t No. I1, ~tanibrd Universay, LA (1965).
[10] H. Leber|;:,ng, Zfir Theorie der Linearen Vektormaximumprobleme, Dissertation, RWTH Aachen
(1977).
[1 I1 W. R~dder. On 'and' and 'or' connectives in fuzzy set theory, working paper No. 75/07, Institut f~il
Wirtschafts~ issenschaften. Lehrstuhl fi.ir Untemehmem;forschung R WTH Aachen ! !975 ~.
[12] W. R6dder a.t,,d H.J. ~ m e r m a n n , Analyse, Beschreibung und Optimierung yon unscharf
formulierten Problemen, Z. Operations Res. 271.¢,~a.7-7
,~.~.:: 1-18..
[t3] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338--353.
[14] H.-J. Zi.mmermann, Description and optimization of fuzzy systems, Int. J. General Syst. 2 (4) 1975.
[i 5] H.-J. Zimmermann, Opiitn~,!e Entscheidungen bei unscharfen Problembeschreibungen,
7-. lbF 27 (1075j 785- 795.
[16] H.-J. Zimmermann, et al., DFG-Forschungsbericht 1975/76" Unscharfe Entscheidungsprob|eme,
Aachen (1976).
[17] S. Zionts and J~ Wallenius: An interactive linear programming method for solving the multiple criteria
problem, working paper 74-i0, European Institute for Advanced Studies in Management, Brussels
(i975).