Fuzzy Sets and Systems 57 (1993) 295-311 295

North-Holland

Linear programming with fuzzy random

variable coefficients
Wang Guang-yuan
Harbin Architectural and Civil Engineering Institute, Harbin, Heilongjiang, China

Qiao Zhong
Hebei Institute of Architectural Engineering, Zhangjiakou, Hebei, China

Received September 1992

Revised January 1993

Abstract: Linear programming with fuzzy random variable coefficients is introduced by discussing a practical engineering problem.
In order to study the solution of the linear programming with fuzzy random variable coefficients, we discuss the simplex algorithm
for linear programming with random variable coefficients. Furthermore, the solution and distribution problem of this new fuzzy
random programming are studied.

Keywords: Fuzzy random variable; interval analysis; linear programming; stochastic programming; fuzzy programming; fuzzy
random programming.

I. Introduction

In the real world, many decision-making problems can be described by using a linear programming
model. Unfortunately, it is very difficult to clearly know all information in many practical systems. This
difficulty mainly comes from the uncertainty of many factors. Since R.E. Bellman and L.A. Zadeh [1]
studied decision-making in a fuzzy environment, H.-J. Zimmermann [23, 24], H. Tanaka, T. Okuda and
K. Asai [19] presented a concept of 'fuzzy mathematical programming' and discussed some approaches
to solve such a programming. Fuzzy mathematical programming has become a remarkable world in
mathematics, and there are many valuable works in this field (cf. [2-4, 6, 8, 11-13, 16, 18, 20, 22]).
Fuzzy programming is an effective tool to deal with decision-making problems in a fuzzy system.
Stochastic programming is a useful tool to treat decision-making problems in a stochastic system (cf.
[10, 17]). However in many practical systems, fuzzy information and random factors arise at the same
time. We are often faced with the case where fuzziness and randomness of a thing are fused with each
other, hence it is very difficult to distinguish between these two uncertain factors. Therefore, it is very
necessary to build a new kind of programming to make decisions in a fuzzy random system. H.
Kwakernaak [9], M.L. Puri and D.A. Ralescu [15], Wang Guangyuan and Zhang Yue [21] introduced
fuzzy random variables to describe the uncertain information that fuzziness and randomness are fused
with each other. Fuzzy random variable theory provides a fundament to make decisions in a fuzzy
random system. In this paper, we shall study a kind of linear programming with fuzzy random variable
coefficients. First, we shall discuss the practical background by a practical engineering problem in
Section 2. In Section 3, we shall give some basic concepts and conclusions on fuzzy random variables in
preparation for our discussion, We shall introduce two models of linear programming with fuzzy

Correspondence to: Dr. Qiao Zhong, Hebei Institute of Architectural Engineering, 33 Jiangou Road, Zhangjiakou 075024,
Hebei, China.

0165-0114/93/$06.00 © 1993--Elsevier Science Publishers B.V. All rights reserved

296 Guang-yuan Wang, Zhong Qiao / Linear programming

random variable coefficients, and point out the relation between these models and classical (stochastic)
programming (resp. fuzzy programming) in Section 4. In Section 5, we shall discuss the simplex
algorithm on linear programming with random variable coefficients. Finally, we shall present a solution
and a distribution problem on two models given in Section 4.

2. Background

Many decision-making problems in practical engineering systems may be referred to linear

programming problems with fuzzy random variable coefficients. We shall formulate the background of
our problem by a practical example.

Example (Economic model of polluted water treatment). Polluted water treatment (for example,
industry wastewater, town polluted water, etc.) is an important project on preserving the environment.
In order to provide science evidence, we need to decide the pollution degree of w a t e r - water class i
(where i = 1, 2, 3, 4, 5). i -- 1 denotes the ideal class; i = 2 denotes the sub-ideal class; i = 3 denotes the
light pollution class; i = 4 denotes the pollution class; i = 5 denotes the serious pollution class. Every
water class i is characterized by some allowable intervals [a,~, aft] (j = 1. . . . . 9) of densities of some
chemical matters, where j is the number of chemical matters, a/~, a~ E R ÷ = [0, ~). We often consider
the following chemical matters: Chemical oxygen demand, Biochemical oxygen demand, Dissolve
oxygen, Cyanide, Mercury, Arsenic, Cadmium, Chromium and Lead. If ai] E [a•, a~] (j = 1 , . . . , 9),
where aij is the density of the chemical matter ] (i.e. weight per unit volume), then water class is i. In
fact, the allowable intervals [a/7, a~] (j = 1 , . . . , 9) ought to be fuzzy subsets (particularly, bounded
closed fuzzy numbers) on R +: t~j:R+---> [0, 1], because there is a transition stage between water classes.
These fuzzy numbers ~j (i -- 1 , . . . , 5; j = 1. . . . . 9) should be decided by some experts in advance.
Let 12 be the set of test points of polluted water in a particular city. In order to treat polluted water in
this city by an economical and effective way, we choose some points, which form a set 12', from £2 at
random. We know that the density of chemical matter j in polluted water is xj(w) at the point w
(w ~ s'2', j --- 1 , . . . , 9) by test analysis. Thus we can know that the allowable scope of xj(w) is ffj(w)
(j = 1 , . . . , 9), (if akj(w)(xj) = m a x l ~ 5 ~ j ( x j ) , then we take aj(w) =akj(W)). Obviously, aj(w)
(] -- 1 , . . . , 9; w c £2) becomes a fuzzy random variable on £2. Put chemical doses into the polluted
water so that polluted water is purged. Assume that yj(w) denotes the weight of the chemical dose j
thrown corresponding to the chemical matter j in polluted water, and Y = ( y l ( w ) , . . . , yg(w)) x. For
every j, we structure an efficacy function fj(Y, ~j(w)), which shows purge degree of the chemical matter
j in polluted water when all chemical doses are put into the polluted water, and which is a fuzzy random
variable on 12.

Problem. Affirm the weight of all chemical doses such that fj(Y, aj(w)) <~~2j, ] = 1, . . . , 9, and such that
the expense of doses is the smallest.

Let us consider the inequality f~(Y, aj(w))<<-a2j (1 ~<]~<9). fj(Y, tij(w)) denotes the new scope of
chemical matter ] after we put all doses into polluted water. Hence fj(Y, ~j(w))~< azj, j = 1 , . . . , 9,
purport that the polluted water at the point w should be purged such that its class becomes or outstrips
sub-ideal class (i = 2) after we put all doses into polluted water.
Let Y(w) -- (yl(w) . . . . , y 9 ( w ) ) T be decision vector. Then objective function should be W(w):
9
W(w) = pj. yj(w)
j=l
where pj is the price per unit weight of chemical dose j.
Now we may state this problem as follows:
 Guang-yuan Wang, Zhong Qiao / Linear programming 297

Find
9
Y(w) to minimize W(w) = ~, pj" y](w) (2.1a)
]=l
subject to fj(Y(w), ~ij(wll<~62j, j = 1. . . . . 9, (2.1b)
where w ~ £2.
If (2.1b) are linear inequalities with respect to yl(w),... ,yg(W), then this problem is a linear
programming problem with fuzzy random variable coefficients.

3. Fuzzy random variable

H. Kwakernaak [9] and M.L. Puri and D.A. Ralescu [15] introduced the notion of a fuzzy random
variable. Wang and Zhang [21] gave another concept of a fuzzy random variable, and proved that the
concepts are equivalent under some conditions. It is reasonable that we describe uncertain information
where fuzziness and randomness are fused with each other by using fuzzy random variables. In this
section, we shall state some basic concepts and conclusions on fuzzy random variables in preparation of
our discussion.

3.1. Closed interval number and fuzzy number

R.E. Moore [14] gave the following definitions.

Definition 3.1.1. Let R = ( - ~ , ~). A closed interval [a-, a ÷] is called a closed interval number on R,
where a , a ÷ e R, are such that a - ~< a +.

Write l(R)={[a ,a+]la ,a+~ R,a-~<a+}. In particular, we shall not distinguish between the
degenerate interval [a, a] and the real number a.

Definition 3.1.2. Suppose that * is an algebraic operation on R. The operation * on I(R) is defined by
[ a - , a +]*[b , b + ] = { a * b l a E [ a - , a + ] , b e [ b ,b+]}, (3.1)
where [a , a+], [b-, b +] E I(R). Particularly, when * is ' + ', ' - ', '.', '/', ' v ' , ' ^ ' respectively, we
obtain
[a , a + ] + [ b - , b + ] = [ a - + b - , a ++b+], (3.2)
[a , a + ] - l b - , b + ] = [ a - - b + , a + - b ], (3.3)
[ a - , a + ] "[b , b ~ ] = [ m i n ( a b-,a+b +,a-b +,a+b ),max(a b-,a+b +,a-b +,a+b-)], (3.4)
[a , a+]/[b , b+] = [a-, a+] • [i/b +, 1/b-] (if 0 ¢ [b-, b+]), (3.5)
[a ,a + ] v [ b - , b + ] = [ a - v b ,a +vb+], (3.6)
[a , a + l ^ [ b -, b+l = [ a - A b - , a + ^b+]. (3.7)
Evidently, for any r e R, if let r __a[r, r], then
[ra-,ra ÷] if r/>0,
r'[a-,a+] = [ra+,ra ] ifr~<0. (3.8)

Definition 3.1.3. Let [a , a+], [b , b +] e I(R).

(1) The order relation ' ~<' is defined by
[a , a + ] ~ < [ b - , b +] if and only if a ~<b a n d a +~<b +.
(Remark. The definition is different from that of [14]).
298 Guang-yuan Wang, Zhong Qiao / Linear programming

(2) The order relation ' c ' is defined by

[a-,a+]c[b-,b +] i f a n d o n l y i f a-~b-anda+~<b +.
The following are some concepts on fuzzy numbers. Denote Of(R) = { f l f : R ~ [0, 1]}.

Definition 3.1.4. A fuzzy set f ( f E Of(R)) is called a fuzzy number on R, if it satisfies conditions (1) and
(2):
(1) There exists x E R such that f ( x ) = 1;
(2) For any c~ ~ [0, 1], f~ = {x If(x)/> a} is a convex set on R.
Furthermore, we have the following concepts:
(3) Let f be a fuzzy number on R. If f~ is a bounded set for any a E (0, 1], then f is said to be a
bounded fuzzy number on R.
(4) Let f be a fuzzy number on R. For any a E [0, 1], if whenever {x,} of,,, limn._.~Xn = X, we have
x ~ f~, then call f a closed fuzzy number on R.
Write Ofo(R) = {f I f is a bounded closed fuzzy number on R}. Obviously, if f c Of0(R), then for any
ol E (0, 1], f~ = [f£, f+] is a closed interval number on R, and f = U,, ~(o,11a [fS, f~+].

Remark. Here
{ c~ -+
ifx E [f~,f~],
0 ifx ~
which is different from the meaning given in Definition 3.1.2. Observe that R c I ( R ) c Ofo(R).

Definition 3.1.5. Let * be an algebraic operation on R, f, g ~ Ofo(R). The algebraic operation * on

Ofo(R) is defined by

(f*g)(z) = V (f(x)Ag(y)). (3.9)

z=x*y

Particularly, when * is ' + ', ' - ', '.', '/' respectively, we can obtain their correspondents on Ofo(R). Let
r e R. We define

r(x) = {~ i f x =r,
ifx ~r.
Then r E Ofo(R), and we have

(r'f)(z)= V (r(x)Af(y)). (3.10)

z=x .y

Theorem 3.1.6. Suppose f, g ~ Ofo(R). For any a ~ (0, 1] we have ( f *g)~ =f~ *g~, where * may be any
continuous algebraic operation. Particularly, (r . f ) ~ = r . f~, where r E R, r . f~ = [r, r] . f~.

Definition 3.1.7. Let {f [ t E T} c Ofo(R).

(1) V , ~ r f is defined by a fuzzy number g e Of0(R) such that

g~ = [,Yr (f)S' ,Yr(f)~+] for any a ~ (0, 1].

(2) /~,~Tf is defined by a fuzzy number g ~ ~o(R) such that

g~ = [ A (f)S, /~ (ftt)~+] for any a ~ (0, 1],

Lt~T t~T

where (f),, = [(f)~, ( f ) , ~+] , a n d Tis a n index set.

 Guang-yuanWang,ZhongQiao / Linearprogramming 299

Definition 3.1.8. Let f, g e o%0(R).

(1) f<~g if and only iff~ ~<g,~ for any a • (0, 1].
(2) f ~ g if and only if f(x) <~g(x) for any x • R.

Remark. This definition is identical with the ranking method given by [16, 20] and Dubois and Prade.

D. Dubois and H. Prade [5, 6] introduced the concept of L-R fuzzy number.

Definition 3.1.9. f • ~(R) is called an L-R fuzzy number, if its membership function can be expressed
by
~sl((m-x)/a) ifx<~m, a > 0 , (3.11)
f(x)= Ls2((x-m)/b) ifx ~>m, b > 0
where Sl and s2 are two reference functions (cf. [5,6]). Denote this L-R fuzzy number f by
f = (m, a, b)s/,. In this paper, we only consider L-R fuzzy numbers in ~ ( R ) .

Theorem 3.1.10. Let f = (m, a, b ),l, g = (n, c, d),tr be L-R fuzzy numbers. Then
(1) f + g = ( m + n , a + c , b + d ) a r . (3.12)
(2) f - g = ( m - n , a + d , b + c ) , l r ifsl=s2. (3.13)
(rm, ra, rb )~,r if r > 0
(3) (3.14)
r'f=l(rm,-rb,-ra)st, / f r <0, wherer•R.
(4) f<~g ifandonlyifm<.n, a>~c, b<~d. (3.15)
(5) f ^g--~(m^n, avc, b^d),t, (3.16)

3.2. Fuzzy random variable

H. Kwakernaak [9], Puri and Ralescu [15], Wang and Zhang [21] introduced the notion of a fuzzy
random variable under different conditions respectively. In this paper, we shall adopt the following
definition by using the results given in [21].

Definition 3.2.1. Let (~2, M, P) be a probability measure space. A mapping t~ :~2~ ~ ( R ) is called a
fuzzy random variable on (1"2,~/, P), if for any a E (0, 1], (w • ~2),
t/~(w) = {x Ix • n, a(w)(x) I> a} = [aS(w), a+(w)] (3.17)
is a random interval, namely, aS(w) and a+,,(w) are two random variables (or finite measurable
functions) on (12, ~/, P). Denote the set of all fuzzy random variables on (O, M, P) by FR(O).

Remark. We can prove that Definition 3.2.1 is equivalent to those concepts in [9, 15, 21] when the
value region of a fuzzy random variable is taken as ~(R).

The following is introduced from [21].

Definition 3.2.2. Let * be an algebraic operation on ~(R). The algebraic operation * on FR(12) may be
defined by
(a*/~)(w) a=~(w)*b(w) for any w • I2, (3.18)
where tT,/~ • FR(f2).
300 Guang-yuan Wang, Zhong Qiao / Linear programming

Furthermore, let {~, ] t e T} c FR(~). Then

(~d,)(w) -a- /~ ~,(w) for any w e£2. (3.19)

t tET

Definition 3.2.3. Suppose that ~ E F R ( ~ ) , ~ ( w ) --- [a~(w), a~+(w)], a E (0, 1].

(1) For any u ~ (0, 1], the function
eO~(x) ___aP{w ] w ~ ~2, x E ~i,(w) = [a~(w), a~+(w)]}
is called a projective distribution function of ~, where x ~ R.
(2) Let FS(x) (resp. F+~(x)) be the probability distribution function of aS(w) (resp. a+~(w)). Then call
F(x): U ~[F+(x),FS(x)]
aE(0,1]

the fuzzy probability distribution function of ~.

It is clear that F+~(x)<~F/~(x) and PD~(x) = FS(x + O) - F+~(x), where
FZ(x+O)=P{wlw~£Z, aS(w)<~x}, F+(x)=P{wlwef2, a+(w)<x}.
Definition 3.2.4. Let t~ E FR(12) such that 6 , ( w ) = [aS(w), a+(w)]. The expectation of ~ is defined by
E(ff) a= U alE(aS), E(a+)] (3.20)
a~(0,1]
where E(aS) and E(a +) are the expectations of a/~ and a~+ respectively.

4. Models of linear programming with fuzzy random variable coefficients

In this section, we shall introduce two models of linear programming with fuzzy random variable
coefficients and discuss the relation between them and the classical (resp. stochastic or fuzzy)
programming. Denote R(£2) = {h [ h is a random variable on (£2, M, P)}.
Suppose that 8q, /~i, ci ~ FR(I2), cj ~ R(£2), where i = 1. . . . . m; j = 1 . . . . . n. Write ~] = (~q),,×n,
/~ = (/~1. . . . . /~,~)V, ~ = (~1 . . . . , ~n) T, C = (cl . . . . . cn) v, X = (xl . . . . . xn) T, where xj ~ {x ] x : I 2 ~ R},
j=l,...,n.
We introduce the following two models.

Model 1 (Linear programming of constraints with fuzzy random variable coefficients).

Find X to minimize W(C, X) = c T x (4.1a)
subject to ,~X ~</~, (4.1b)
X>~0. (4.1c)
Remark. X / > 0 if and only if xj i> 0 for every j = 1 . . . . . n.

Model 2 (Linear programming of both objective and constraint containing fuzzy random variable
coefficients).
Find X to minimize W(C, X ) = c T x (4.2a)
subject to A X ~</~, (4.2b)
X ~> 0. (4.2c)
Distribution problem on Model 1. Find the probability distribution function F(x) and expectation of
W(w) = min c T x in Model 1.
Distribution problem on Model 2. Find the fuzzy probability distribution function, projective
distribution function and expectation of lYe(w) = min c T x in Model 2.
Let us discuss the relation between Model 1, 2 and classical (resp. stochastic or fuzzy) programming.
 Guang-yuan Wang, Zhong Qiao / Linearprogramming 301

(1) The stochastic linear programming (cf. [10, 17]) is a special case of Model 1 and 2.
Since R ~ ~o(R), if coefficients in Model 1, 2 are taken as particular fuzzy random variables such that
t~ : £ 2 ~ R, then Model 1 and 2 equal stochastic linear programming.
(2) The linear programming with fuzzy number coefficients is a special case of Models 1 and 2.
In fact, let £2 ={w} be the set of the single point w. Then FR(12) and {~i(w)[a E FR(£2)} are
isomorphic (f:~i---~i(w) is an isomorphic mapping). In this sense, Model 1 and 2 equal linear
programming with fuzzy number coefficients.
(3) The classical linear programming is a special case of Models 1 and 2.
Let £2 = {w} be the set of the single point w, and denote FR(£2, R) = {t] I ti : £ 2 ~ R is a fuzzy random
variable on £2}. Then FR(£2, R) and {ti(w) I a E FR(£2, R)} are isomorphic. Model 1 and 2 equal classical
linear programming in this sense.
(4) It is clear that Model 1 is a special case of Model 2.

5. Simplex algorithm of linear programming with random variable coefficients

In order to give a method for solving the linear programming with fuzzy random variable coefficients,
we shall discuss the simplex algorithm of linear programming with random variable coefficients in this
section.
Let us consider the following problem: (RP 1)
Find X to minimize W(C, X) = CTX (5.1a)
subject to A X = B, (5.1b)
X>~0 (5.1c)
where A = (aq(W))m×,, B = (b~(w) . . . . , bin(w)) T, C ~- ( C l ( W ) . . . . . On(W))T, X = (X 1. . . . . Xn)T, aij(W),
b~(w), Q(w) e R(£2) are random variables on the probability measure space (£2, ~/, P), i = 1. . . . , m;
j = l . . . . . n, we£2.

Remark. Each inequality constraint can be made an equation by adding a nonnegative variable (slack
variable) to a ' ~ ' constraint. Therefore we can consider constraint (5.1b) in (RP1). Furthermore, for a
given point w ~ £2, all random variables in (RP1) become some real constants, hence they will be
considered as some given parameters corresponding to any given point w E £2 in this section•
X is called a feasible solution of (RP1), if it satisfies (5.1b) in (RP1), and X/> 0.
Denote A = ( A 1 . . . . . A,), where Aj = (alj(w),..., ami(W)) T. Let Rank(A) be m, and suppose that
D -- (Aj, . . . . , Aj~) is a nonsingular submatrix of A.
Thus (5.1b) in (RP1) can be written as follows:

~ Ajxi= B or 2 Aj, xjk + ~ A~x, = B.

j= 1 k= 1 s ~ {Jl,-'-,Jm }

y r = D - 1 B , a n d x s0__ _0 i f
If D ~B >i 0, then the feasible solution Xo = (x~, • • • , x°) T such that (xj°, .. . , x 0Jm,
s ~ {jl . . . . , j m } , is called a basic feasible solution of (RPI), and the matrix D is said to be a base matrix
of (RP1). xj~ (k = 1 , . . . , m) is called a basic variable.
By using the theory of classical linear programming, we know that for a given w ~ £2, if (RP1) has a
feasible solution, then it must have a basic feasible solution, and the optimization (i.e. minimal) solution
can be found in the class of all basic feasible solutions of (RP1).
For convenience, we may assume that the base matrix D = (A~ . . . . . Am) such that D-~B > 0 , and
xl . . . . . Xm are basic variables.
Thus A = (D N), and in correspondence with this we can denote X = (XTo X,~) x, C T= (C~ C~),
where N = (Am+l . . . . . A,), X~ = ( X 1 . . . . . Xm) , X ~ V = ( X m + l . . . . . Xn), C T = ( e l ( W ) , . . . , Crn(W)),
c~ = (~,.+,(w) ..... ~.(w)).
302 Guang-yuan Wang, Zhong Qiao / Linear programming

Equation (5.1b) of (RP1) may be written as follows:

DXo + NX N = B.
It follows, from the nonsingularity of D, that
Xo = D-1B - D-1NXN . (5.2)
Therefore,
c T x = CT X o + CT X N = CTDD- aB + ( CTN - CT D - aN ) X N
= CToD-1B + (C T - C T D - 1 A ) X . (5.3)
Write
Wo(w) = CToD-IB, (5.4)
( c ' + l ( w ) . . . . , c',,(w)) a= C T _ C T D 1N' (5.5)

(
(btl(W) . . . . , b,~(w))T _a D _ a B '

.
"" n(' /
.
w

=a D -1N.
(5.6)

(5.7)
\a'm,m+l(W) "'" am,n(W)
' /[
By using (5.2)-(5.7) in this section, problem (RP1) may be expressed by (RP2):

Minimize W ( C , X ) = Wo(w) + ~ c;(w)xj (5.8a)

j=rn 4-1

subject to I X i + /=m~+l aij'(w)xj = b;(w), i = 1 . . . . , m,

I,xj/> 0, j = 1. . . . . n. (5.8b)
Evidently, X ' = (b'l(W),..., b'm(W), 0 , . . . , 0) T is a basic feasible solution of (RP2). We can prove the
following theorems.

T h e o r e m 5.1 (Judgement theorem of optimization solution). In (RP2), if c;(w) >-0 (j = m + 1 . . . . , n),

then the basic feasible solution X ' is an optimization solution o f (RP2) or (RP1), and min c T x = Wo(w).

We call c;(w), j = m + 1 , . . . , n, the test numbers with respect to the base matrix D.

Theorem $.2. I f there exists Jo (m + l ~ j 0 ~ < n ) such that c;o(w)<O, and if a~jo(w)<~O for every
i = 1 , . . . , m, then (RP2) or (RP1) has not any optimization solution.

Theorem 5.3. Let (RP2) satisfy the conditions

(1) there exists jo (m + 1 <~Jo<~n) such that c;o(w ) < 0;
(2) there exists io (1 <~io <- m ) such that ai'oJo(w) > 0;
then (RP2) has a new basic feasible solution X " such that c T x " < C T X '.

Remark. Let
0 = rain (b'(w)/a'jo(W)) = b'k(w)/a;jo(w), 1 <~k <~m.
a~0>0
Evidently, 0 > 0. Let us take
x;'=O, j=m+l .... ,n;j#jo, X;'o=O,
t
x['= b f ( w ) - Oaqo(w), i = 1 . . . . . m.
 Guang-yuan Wang, Zhong Qiao / Linearprogramming 303

It is easy to see t h a t x kIt = 0 a n d x ~rl 0~ , i = 1 , . . . , m ; i ~ k . Thus, X " = ( x ~It , . . . , x " ) x is a new basic
feasible solution of (RP2) such that CvX " < c T X '.

In the theory of stochastic programming, the following conclusions have been proved (cf. [10, 17]).
Assume that H = (A, B, C) is an r-random vector of all elements of A, B and C in (RP1), where
r = m + n + ran. Thus A, B and C in (RP1) may be denoted identically by A(H), B(H) and C(H), and
(RP1) may be expressed by (RP3):

W(H) = rain c T x (5.9a)

subject to A ( H ) X = B(H), (5.9b)
X >~0. (5.9c)

Remark. It is clear that H(w) e R r for any w E £2.

Definition 5.4. Let u be a Lebesgue measure on R ~, and let D(H) be a m-submatrix of A(H). If
u({H(w) I w e £2, det D(H(w)) = 0}) = 0, then D(H) is called almost everywhere nonsingular.

Remark. D(H) is almost everywhere nonsingular if and only if there exists w e£2 such that
Det D(H(w)) ~ O.

Theorem 5.5. (1) W(H) = min CTX is a Borel measurable function (or a random variable) on (£2, ~).
(2) If (RP3) has a feasible solution, then there exists an optimization solution which is a random
vector on (£2, ~I).

Definition 5.6. Let {D~(H)]i = 1. . . . . p} be the class of all almost everywhere nonsingular m-
submatrixes of A(H). If w E £2 such that Det Di(H(w)) = 0, we define D[1(H(w)) = 0 (1 ~<i ~<p). We
structure the following sets:

Ui = {H(w) ] w ~ 12, D[~(H)B(H)Iw >! O, (CT(H) - CVoD,~(H)A(H))Iw >t 0}, (5.10)

i-I
V~= U i - U Uj, i = 1,...,p. (5.11)
j-1

Then Ui or Vi (i -- 1 , . . . , p) are called decision regions of (RP3).

Theorem 5.7. Suppose that P ( { w I - ~ < W ( H ( w ) ) < + ~ } ) = I

and there is a w e£2 such that
Rank(A(H(w))) = m. If fH(s) is the probability density function of H, then the probability distribution
function Fw(n)(x) and the expectation EW(H) of W(H) may be expressed by

i=1 H(w)] H(w)~E,W(H(w))<x}

E W ( H ) = ~ fu W ( H ) . f , ( s ) d s , where x e R. (5.13)
i=l i

6. On the solution of Model 1

In this section, we shall discuss the solution of Model 1, and study its distribution problem.
For convenience, we adopt the following convention for ,4 a n d / ) given in Section 4.
304 Guang-yuan Wang, Zhong Qiao / Linear programming

Let a e (0, 1]. Since ffij, bij e FR(f2), it follows that their a-level cuts are closed random interval
numbers. Denote them by

(~ij)~ = [(aij)2, (aij)~+] for any w E £2,

(/~)~ = [(b~)~-, (b~)~+] for any w e £2,
that is, for any w ~ £2,

(aij)ot(W) = {X [ X ~ R, (aij)(w)(x) ~ O}
= [(aij)~(W), (aij)+(w)],
and (bi)~(w) = [(bi)~(w), (b;)~+(w)], where i = 1 . . . . , m; j = 1 , . . . , n.
Write

A~ = ((a0)2)m×,, AS = ((a~j)S),n×,,
B ~ = ((bl)~, • •., (bm)S) T, B ,+ = ((b,)~ . . . . , (bin)+) T.
It is clear that elements of these matrices or vectors are random variables on ~.
Let 5, b e FR(12). We define that a ~< b if and only if 5(w) ~< b ( w ) for any w E £2.
We have introduced the following Model in Section 4:

Model 1 (Linear programming of constraints with fuzzy random variable coefficients).

Find X to minimize W(C, X) = cTx (6.1a)
subject to ,4X ~</Y, (6.1b)
X I> 0 (6. lc)
where A = (a~j)m×,, B = (bl, . . . , 5,,)T, C = (cl, . . . , c,) T, X = (xl . . . . , x , ) T, aij, b~ ~ FR(t2), cj c R(O)
and x j ( w ) ~ R for w ~ g2.

For Model 1, we have the following theorem.

Theorem 6.1. M o d e l l is equivalent to the following M o d e l 1.1:

Find X to m i n i m i z e W(C, X) = CXX (6.2a)
subject to A £ X <~B £ , (6.2b)
A + X <~B +, (6.2c)
X>~O foranyac(0,1].

Remark. (1) It follows, from Theorem 3.1.6 and Definition 3.1.2, 3.1.3, 3.1.8, 3.2.2, that A X ~</~ if and
only if A S X <~B ~ and ._,__
~ A * ~ -<_~/~+ for any a E [0, 1]. This theorem is similar to the result given in [8]
(cf. [8]). In the following, write t~ = {X ] A X ~</~, X/> 0}, G~ = {X ] A l e X <~B ~ , A+~X <~B+~, X ~ 0}.
(2) For any given a E (0, 1], Model 1.1 is a linear programming with random variable coefficients
discussed by Section 5. Hence we may use the simplex algorithm given in Section 5 to solve the linear
programming with fuzzy random variable coefficients such as Model 1 by way of Theorem 6.1.

Definition 6.2. For any given a ~ (0, 1], an optimization solution X ( w , a) = (Xl(W , a ) . . . . , Xn(W , a ) ) x
of Model 1.1 is called an a-level optimization solution of Model 1.

In the following, we shall give the concepts of fuzzy random optimization solutions and fuzzy
pseudorandom optimization solutions.
 Guang-yuan Wang, Zhong Qiao / Linearprogramming 305

By using the coefficients in Model 1.1, we can build two programming problems.

Model 1.2.
Find X to minimize W(C, X ) = cTX (6.3a)
subject to A-~X <~B~,, (6.3b)
X~>0 for any a E (0,1].

Model 1.3.

Find X to minimize W(C, X) = c T x (6.4a)

subject to A+~X <~B +, (6.4b)
X/>0 for any a ~ (0,1].
For convenience, we write
S,dM1.2) = {X(w, a) IX(w, a)- is an optimization solution of Model 1.2 for some given
a ~ (0, 1]},
S,,(M1.3) = {X(w, a) + IX(w, a) + is an optimization solution of Model 1.3 for some given
a e (0, 1]}.

Definition 6.3. For any a ~ (0, 1] we take

X(w, a)- = (x~(w, a)-, . . . . xn(w, a ) - ) T e S,,(M1.2),
X(w, a) + = (x~(w, a) +. . . . , xn(w, a)+) T E S~(M1.3),
then ) ( ( w ) = (~l(w) . . . . , $,(w)) T is called a fuzzy pseudorandom optimization solution of Model 1,
where

Xi(w) = U ~[xj(w, a)- ^xj(w, a)+, xj(w, a) vxj(w, a)+], (6.5)

aE(0,1]
j = 1 . . . . n, w E E2 Furthermore, if X(w, a)- and X(w, a)+ are random vectors for any a E (0, 1],
then .,~'(w) is said to be a fuzzy random optimization solution of Model 1.

Theorem 6.4. If S,(M1.2) ~ 0 and S,~(M1.3) ~ 0 for any a ~ (0, 1], then there exists a fuzzy random
optimization solution of Model 1.

ProoL Let S,,(M1.2) ~ 0 and S,(M1.3) # 0 for any a ~ (0, 1], by Theorem 5.5(2). We know that Models
1.2 and 1.3 have optimization solutions which are random vectors on (~, ~/). Hence there exists a fuzzy
random optimization solution of Model 1.

Remark. Models 1.2 and 1.3 are two linear programming problems with random variable coefficients
discussed in Section 5. By using the simplex algorithm given in Section 5, we may find optimization
solutions of Models 1.2 and 1.3 for any a ~ (0, 1]. And therefore, we may obtain fuzzy pseudorandom
or random optimization solutions of Model 1.

Now we discuss the distribution problem of Model 1. Assume that

(A~) and B" = ( B i )

A'= A 2mXn B 2m
is, respectively, the matrix composed by A~, and A~+, and the 2m-random vector composed by B~ and
306 Guang-yuan Wang, Zhong Qiao / Linear programming

B~+, and H , -- (A', B',, C) is an r-random vector of all elements of A', B',, and C. By using these
appointments, Model 1.1 may be expressed by Model 1.4:
W(H,) = min c T x (6.6a)
subject to A'~X <~B'~, (6.6b)
X/>0 for any a ~ (0,1].
We have the following theorem on the distribution problem of Model 1 when the constraint is
.4X =/~ by adding slack variables, and there exists a such that 0 -- 0~.

Theorem 6.5. Suppose that P({w I -oo < W(H,~(w)) < +oo}) = 1 and there exists w E £2 such that
Rank(A'(w)) = 2m, f,~(s) is the probability density function of Ha. Then the probability distribution
function Fw(no)(x) and the expectation E(W(H~)) of W(H,) in Model 1.4 may be expressed by

FW(H~)(X)= ~ f n fn.(s) ds, (6.7)

i=1 ~(w) ]H~(w)E(V~)i,W(H~(w))<x}

E(W(H,)) = i=12J((~o), W(H,) "fH,(S) ds, (6.8)

where (V~)i is a decision region of Model 1.4 for any given a ~ (0, 1].

Proof. By using Theorem 5.7, this theorem is proved immediately.

Finally, let us see the case where t%j(w), bi(w) are L-R fuzzy numbers in ~o(R) for any given w ~ £2,
i = 1 , . . . , m; j = 1 . . . . , n. We may assume that

a~i(w) = (a~j(w), _a~j(w), dii(W))st, b~(w) = (bi(w), b_~(w), 6i(w))~,. (6.9)

In this case we can prove the following theorem.

T h e o r e m 6.6. If aij(w) and bi(w) (i = 1. . . . . m; j = 1. . . . , n) are L-R fuzzy numbers such that (6.9) for
any given w E £2, then Model 1 is equivalent to the following Model 1.5:
Find X to minimize W(C, X ) = c T x (6.10a)
subject to A X <~B, (6.10b)
_AX i> _B, (6.10c)
AX ~</~, (6. lOd)
X >i 0 for any given w E £2,
where A = (aij(w))m~,, B = (bl(w) . . . . , bin(w)) T, A = (aij(W))m×,, _B= ( b l ( w ) , . . . , b~(w)) T,
= (5,i(w))m×,, B = (b~(w) . . . . . b m ( w ) ) T.

Remark. By Definition 3.2.2 and Theorem 3.1.10, we easily prove this theorem. This conclusion is
similar to the result presented in [8, 20].

7. O n the solution o f M o d e l 2

In this section, we shall study the solution of Model 2, and discuss its distribution problem.
A~, A~, B~ and B~ used in this section are identical with those given in Section 6.
 Guang-yuan Wang, Zhong Qiao / Linearprogramming 307

Let

at~ At~+ 2m×n ' Bt3= Bt~ 2m

= ( 7 1 , . . . , ?,)T, where ?j E FR(£2), and
(?j),(w) = [(q)~(w), (q)~+(w)] for any w ~ £2, a e (0, 1], j = 1 , . . . , n.
Denote
C .+ = ((c,): . . . . . (cn):) T, C~ = ((c,);~ . . . . . ( c , ) J .
We have given the following model in Section 4.

Model 2 (Linear programming of both objective and constraint containing fuzzy random variable
coefficients).
Find X to minimize W(C, X) = c T x (7.1a)
subject to A X <~/3, (7.1b)
X~>0 (7.1c)
where ,4 = (~q)m×,, /] = ( / ) , , . . - , /)m)T, = ( ? t , . . - , ?,,)T, X = (xl . . . . . xn) T, aq, bi, ?j ~ FR(£2), and
Xj(W) e R for any w e £2.

We may prove the following theorem.

Theorem 7.1. Model 2 is equivalent to the following linear programming with random variable
coefficients: Model 2.1. for every a E (0, 1],
Find X to minimize W(C, X)2 = (C2)TX (7.2a)
W(C, X) + = (c+)Tx, (7.2b)
subject to A ~ X <~B~, (7.2c)
A ~ X <- B~, (7.2d)
X >10 for any/3 e (0, 1],
+ +
where A~ = ((aq)~)m×,, A~ = ((a0)n)m×,, B~ = ((b~)~ . . . . . (b,,)~) v, Bt3 = ((b,)~ . . . . . (b,,)~) T, C~ =
( ( c , ) ~ , . . . , ( c , ) S , C;~ = ((c,); . . . . . (co);~) ~.

Proof. (1) First of all, we prove that

( ~ T x ) ~ = (c~)Tx, (~Tx)+ = (c+)Tx (*)

for any w E £2, a e (0, 1], where ( c T x ) ~ = [(C'rX)~, (~Tx),+].

In fact, by Theorem 3.1.6 and Definition 3.1.2, for any w e £2, we have

[(CTx)=, ( U X ) : ] = ( U X ) o = ~ (ej(w)x3o = ~ (~3~(w).xj

j-1 j=l

= [(c~);(w), (c3~+(w)] .xj = (cj);(w) . x , (cj):(w) .xj

j=l j=l

= [(cJx, (c:yx].
Hence we obtain (C~X)~ = ( c , ) T x and ( c T x ) + = (C+)TX for any w e £2, a e (0, 1].
308 Guang-yuan Wang, Zhong Qiao / Linearprogramming

Now we prove that lg' = min C T x if and only if WS = min ( c 2 ) T x and W~+ = rain (C+)Tx for any
a ~ (0, 1], w ~ £2, where 17¢~= [W~, W~+].
Sufficiency: Let WS -- min (Cs)Tx, W~* = rain (c+)Tx. It follows, from the above result (*), that
1~,, = [W2, W +] -- [min (Cs)Tx, min ( c ~ ) T x ]
= [min ( c T x ) s , min ( c T x ) + ] , for any a e (0, 1], w ~ £2.
By using Definition 3.1.7, we obtain ff¢ = min CTx.
Necessity: Let if" = min c T x . By Definition 3.1.7 and (*), it follows that
[W2, W~+] = (min ~ T x ) ~
= [min ( c T x ) s , min (cTx)~+]
= [min (Cs)Tx, rain ( c + ) T x ] for any a ~ (0, 1], w ~ £2.
Hence we have W2 = min (c2)Tx, W~+ = min (C~+)Tx for any a E (0, 1], w ~ £2.
(2) Now we prove that
,z]X ~</~ if and only if A ~ X <<-B~ and A~X+ --~B~.-<
+
In fact, AX~</~ if and only if ~/=1 aijxj~bi for every i = 1 , . . . , m. By Definition 3.1.8, this is
equivalent to

( ~ a i j x i ) <~(bi)~, for any a ~ (0, a], WE£2, i = l , . . . , m .

'j"= 1 / oL

It follows, from Theorem 3.1.6 and Definition 3.1.2, that A X ~</) if and only if

(ao)2xj, (ao)+~xj ~ [(bi)2, (b~)~+] for any ~ e (0, 1], W e £2, i = 1 . . . . , m.

j=l

Consequently, A X ~/~ if and only if A 2 X <~B2 and A+~X <~B+~ for any ~ e (0, 1].
Synthesizing (1) and (2), we complete the proof of this theorem.

Remark. For any given ~ e (0, 1], Model 2.1 becomes a multiobjective linear programming with
random variable coefficients. We may obtain its c~-level (weak) Pareto solutions by using the theory of
multi-objective programming.
In the following, we will give concepts of induced fuzzy pseudorandom optimization solutions and
fuzzy random optimization solutions of Model 2.
We build two programming problems by using the coefficients in Model 2.1.

M o d e l 2.2. Let a e (0, 1].

Find X to minimize W(C, X)2 = (C~)TX (7.3a)
subject to A ~ X <~B~, (7.3b)
X/>0 for a n y / 3 e ( 0 , 1 ] .

M o d e l 2.3. Let a E (0, 1].

Find X to minimize W(C, X)+~ = (C~+)TX (7.4a)
subject to A ~ X ~<Bt3,+ (7.4b)
XI>0 for a n y / 3 ~ ( 0 , 1 ] .

Let S~(M2.2) (resp. S,,(M2.3)) be the set of all optimization solutions of Model 2.2 (resp. Model 2.3)
corresponding to any given a ~ (0, 1].
 Guang-yuan Wang, Zhong Qiao / Linearprogramming 309

Definition 7.2. For any a ~ (0, 1] we take

X(w, ~)- = (x~(w, ~)-, . . . , x~(w, cQ-) T e S,(M2.2),

X (w, c~) + = (x,(w, a) +. . . . , x,(w, c~)+)T e S,(M2.3).
Then ~'(w) = (£~(w) . . . . . £,(w)) T is said to be an induced fuzzy pseudo-random optimization solution
of Model 2, where

Yj(w) = U a[xj(w, o~)-^xj(w, a)+,xj(w, ~) vxj(w, c~)+], (7.5)

,~ E(0,1]

j = 1 , . . . , n, w ~ £2. If X(w, a)- and X(w, a) + are random vectors for any a ~ (0, 1], then call )((w)
an induced fuzzy random optimization solution of Model 2.

T h e o r e m 7.3. Suppose that S~(M2.2) ~ ~, S~(M2.3) ~ ~. Then Model 2 has an induced fuzzy random
optimization solution (~ E (0, 1]). (Suppose there is [3 such that G = Gt~.)

Proof. If S , ( M 2 . 2 ) ~ and S , ( M 2 . 3 ) ~ 9 for any a E (0, 1], using Theorem 5.5(2), it follows that
Models 2.2 and 2.3 have optimization solutions which are random vectors on (£2, M). And therefore,
Model 2 has an induced fuzzy random optimization solution.

R e m a r k . Models 2.2 and 2.3 are two linear programming problems with random variable coefficients
studied in Section 5. Hence we may use the simplex algorithm given in Section 5 to solve Models 2.2
and 2.3. Thus we get the induced fuzzy pseudorandom or random optimization solutions of Model 2.

Now we discuss the distribution problem of Model 2.

In the discussion on the distribution problem, we always assume that Models 2.2 and 2.3 possess the
same equality constraint A ~ ( = B~. Let H2 = (A~, B~, C2) be the r-random vector of all elements of
¢ i +
A m Bt~, C2, and let H S = (A~, B~, CS) be the r-random vector of all elements of A~, Bt~ and C~+,
where r = 2m + n + 2mn.
We have the following theorem. (Suppose there exists/3 ~ (0, 1] such that G = Gt~.)

T h e o r e m 7.4. Suppose that W(H2) = min(C,)Tx, W ( H +) = min(C,+)TX, and such that

P({w ] - ~ < W(H~) < +~}) = l, P({w I - ~ < W(HS) < + ~}) = 1,
and there exists w ~ £2 such that Rank(A~(w)) = 2m. If fH~(s) and fH+(s) are the probability density
functions of H~ and H+~ respectively, then we have the following conclusions:
(1) The probability distribution function Fw(Ho)(x) and the expectation of W(HS) may be expressed by

Fw(,,~)(x)= ~ f fHo(s)ds, (7.6)

i=1 H~(w) I H~(W)E(V'aa)" W(H~(w))'CX, wE~2}

E(W(HS)) = ~ f W(H2)'fHo(s)ds, (7.7)

i=1 V . ),

where (V~)i (i = 1 , . . . , p) is a decision region of Model 2.2, ~ E (0, 1], x E R.

(2) The probability distribution function Fw(m)(x) and the expectation of W ( H +) may be expressed by

Fw(nz)(x) = ~ f fH,:(s) ds, (7.8)

j= 1 H~,(w) I H+(w)E(Va+)j, W(H+~(w))<x, w E.O}

E(w(/-/.+)) = w ( . . +) .f.+(s) (7.9)

j = 1 a( w2 )j

where (V+~)j (j = 1. . . . . q) is a decision region of Model 2.3, c~ ~ (0, 1], x E R.

310 Guang-yuan Wang, Zhong Qiao / Linear programming
(3) The fuzzy probability distribution function of if" = rain CTX in Model 2 may be expressed by

For(x) -= [._) a[Fw(w~)(x), Fw(t4~)(x)], where x • R. (7.10)

aE(0,1]

(4) The projective distribution function of 17V = min c T x may be expressed by

PD,,(x) = Fw(n:)(x + O) - Fw(u+)(x), x • R, (7.11)
where
Fw(t-t~)(x + O) = P({w I w • £2, W(HS(w)) ~<x}).
(5) The expectation o f g¢ = rain CTX may be expressed by

E(ff)= a e(0,1]
U (7.t2)

Proof. It is easy to see that (1) and (2) are true by Theorem 5.7. Now we prove (3)-(5).
(3) From the proof of Theorem 7.1 we may see that
V¢ = min C T x if and only if W~ = W(HS), W+~ = W(H+~), where g¢~ = [W~, W~].
It follows, from W~ --= W,,, that {w [ W+~(w) < x} ~ {w ] W~,(w) < x}, and therefore
Fwtw~)(x) <- Fw(,~)(x). By Definition 3.2.3(2), we have

P v(x)= LJ
ae(0,1]

(4) By using the result given in Section 3.

(5) Since ff¢ = rain c T x if and only if WS = W(H2), W+~ = W(H+~), using Definition 3.2.4, we know
that (5) is true.

Finally, we discuss the case that elements of A, B, C are L-R fuzzy numbers in ,~o(R) for any given
we£2.
Let aij(w), cj(w), bi(w ) be L-R fuzzy numbers in ~o(R) for any given w • £2. Write

~o(w) -- (aq(w), _aq(w), dq(W))st, bi(w) = (bi(w), b_i(w), bi(w))slr,

Cj(W) -= (Cj(W), Cj(W), ~(W))slr, A = (aij(W))m×n, _m= (aij(W))m×n, m = (aij(W))m×n,
B = (b~(w) . . . . , b,,(w)) T, _B = (b~(w) . . . . . b m(w)) w, # = (b~(w) . . . . , ~m(W))T, (7.13)
c = (c,(w) .... , c°(w))L _c = (_c,(w) . . . . . _c.(w))L = (e,(w) .....

We have the following theorem.

Theorem 7.5. I f ~ii(w), 6j(w), bi(w) (i = 1. . . . . m; j = 1. . . . . n) are L-R fuzzy numbers in ~o(R) with
(7.13) for any given w e I2, then Model 2 is approximately equivalent to the following Model 2.4:
Find X to minimize W x = CTX, We = - - _C T X , W3 = (~Tx (7.14)
subject to A X ~ B, A X >! B_, J t X <~B, (7.15)
X>~O.

Proof. Using Theorem 3.1.10, we know that

rain CTX ~ (min CTX, max c T x , min CTX)str,
therefore, approximately, min c T x if and only if min CTX, min(-C_TX), min CTX.
 Guang-yuan Wang, Zhong Qiao / Linear programming 311

Furthermore, from Theorem 3.1.10, ,4X ~</3 if and only if

aq(w)xj, ~ _aq(w)xj,~ ffq(w)xj sir ~ <hi(w), b-i(w),bi<w))slr

j=l j=l

namely,

~aq(w)xj<-bi(w), ~_aq(w)x~>~b_i(w), ~ffq(w)x,<~bi(w)

j-1 j=l j-1

where i = 1,... ,m. This is equivalent to AX <~B, AX >IB, fiiX <~B. We complete the proof of this
theorem.

Acknowledgement

The authors are very much indebted to Dr. Zhang Yue, Prof. H.-J. Zimmermann and referees for
their valuable suggestions for improvements and critical reading of the manuscript.

References

[1] R.E. Bellman and L.A. Zadeh, Decision-making in a fuzzy environment, Management Sci. 17 (1970) B141-BI64.
[2] J.J. Buckley, Multiobjective possibilistic linear programming, Fuzzy Sets and Systems 35 (1990) 23-28.
[3] S. Chanas, The use of parametric programming in fuzzy linear programming, Fuzzy Sets and Systems 11 (1983) 243-251.
[4] C. Carlsson and P. Korhonen, A parametric approach to fuzzy linear programming, Fuzzy Sets and Systems 20 (1986) 17-30.
[5] D. Dubois and H. Prade, Operations on fuzzy numbers, J. Systems Sci. 9 (1978) 613-626.
[6] D. Dubois and H. Prade, Systems of linear fuzzy constraints, Fuzzy Sets and Systems 3 (1980) 37-48.
[7] L.R. Foulds, Optimization Techniques: An Introduction (Springer-Verlag, New York, 1981).
[8] M. Inuiguchi, H. Ichihashi and H. Tanaka, Fuzzy programming: A survey of recent developments, in: R. Sloinski and J.
Teghem (Eds.) Stochastic versus Fuzzy Approaches to Multiobjective Mathematical Programming under Uncertainty (Kluwer
Academic Publishers, Dordrecht, 1990) 45-68.
]9] H. Kwakernaak, Fuzzy random variables- Definitions and theorems, Inf. Sci. 15 (1978) 1-29,
[10] P. Hall, Stochastic Linear Programming (Springer-Verlag, Berlin-Heidelberg-New York, 1976).
]11] M.K. Luhandjula, Satisfying solutions for a possibilistic linear program, Inf. Sci. 40 (1986) 247-265.
[12] M.K. Luhandjula, Fuzzy optimization: An appraisal, Fuzzy Sets and Systems 30 (1989) 257-282.
[13] Y.J. Lai and C.L. Hwang, Interactive fuzzy linear programming, Fuzzy Sets and Systems 45 (1992) 169-183.
[14] R.E. Moore, Methods and Applications of Interval Analysis (SIAM, Philadelphia, 1979).
[15] M.L, Puri and D.A. Ralescu, Fuzzy random variables, J. Math. Anal. Appl. 114 (1986) 409-422.
[16] J. Ramik and J. Rimanek, Inequality relation between fuzzy numbers and its use in fuzzy optimization, Fuzzy Sets and
Systems 16 (1985) 123-138.
[17] J.K. Sengupta, Stochastic Programming - Methods and Applications (North-Holland, Amsterdam, 1972).
]18] M. Sakawa and H. Yano, Feasibility and Pareto optimality for multiobjective nonlinear programming problems with fuzzy
parameters, Fuzzy Sets and Systems 43 (1991) 1-15.
[19] H. Tanaka, T. Okuda and K. Asai, On fuzzy mathematical programming, J. Cybernetics 3 (1974) 37-46.
[20] H. Tanaka, H. Ichihashi and K. Asai, A formulation of fuzzy linear programming problem based on comparison of fuzzy
numbers, Control and Cybernetics 13 (1984) 186-194.
[211 Wang Guangyuan and Zhang Yue, The theory of fuzzy stochastic processes, Fuzzy Sets and Systems 51 (1992) 161-178.
[22] Wang Guangyuan and Wang Wenquan, Fuzzy optimum design of structure, Engineering Optimization 8 (1985) 291-300.
[23] H.-J. Zimmermann, Fuzzy programming and linear programming with several objective functions, Fuzzy Sets and Systems 1
(1978) 45-55.
[24] H.-J. Zimmermann, Description and optimization of fuzzy systems, J. General Systems 2 (1976) 209-215.