Information Sciences 177 (2007) 2971–2984

www.elsevier.com/locate/ins

Solving linear programming problems under fuzziness

and randomness environment using attainment values
Nguyen Van Hop *

First Consulting Group Vietnam (FCGV), 111D Ly Chinh Thang Street, District 3, Ho Chi Minh City, Viet Nam

Received 27 October 2004; received in revised form 25 January 2007; accepted 31 January 2007

Abstract

In this paper, the author presents a model to measure attainment values of fuzzy numbers/fuzzy stochastic variables.
These new measures are then used to convert the fuzzy linear programming problem or the fuzzy stochastic linear
programming problem into the corresponding deterministic linear programming problem. Numerical comparisons are
provided to illustrate the effectiveness of the proposed method.
Ó 2007 Elsevier Inc. All rights reserved.

Keywords: Fuzzy random variable; Attainment value; Fuzzy linear programming; Fuzzy stochastic linear programming

1. Introduction

Solving fuzzy linear programming problems has received a great deal of attention [10,6,4,2,9,3,8]. Recently,
Inuiguchi and Ramik, [7] conducted a comprehensive review of fuzzy optimization. The most popular
approach to handle the challenge of solving fuzzy linear programming problems is to convert the fuzzy linear
program (FLP) into the corresponding deterministic linear program (LP). An example of this is the approach
that compares fuzzy numbers by using some ordering measures of fuzzy numbers, such as the area compen-
sation method [5], the expected mid-point approach [8], the grade of possibility and necessity measures [1], and
the signed distance method [3].
In addition, a new line of research in the fuzzy optimization area has been extended to the case of fuzzy
stochastic optimization, in which solving fuzzy stochastic linear programming problems is the main concern.
Together, stochastic and fuzzy aspects are considered to provide an efficient tool to describe real-life problems
where uncertainty and imprecise information are inherent. However, this kind of combination creates a great
challenge for researchers to find efficient methods to deal with both fuzzy and stochastic terms. Like the
approaches of fuzzy optimization, the general strategy to handle the fuzzy stochastic optimization problem

*
Tel.: +84 988 588 509.
E-mail address: [email protected]

0020-0255/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.ins.2007.01.032
2972 N. Van Hop / Information Sciences 177 (2007) 2971–2984

is to de-fuzzify and/or de-randomize fuzzy random variables so as to convert the problem into the correspond-
ing deterministic problem. There are two common methods to achieve this. The first is to perform the conver-
sions (de-fuzzify/de-randomize) in a sequential manner [16,17,15]. Although the obtained results of
Luhandjula’s approaches are LPs, the discretizing process of fuzzy sets via a-levels creates quite a large num-
ber of constraints and variables. The second method is to perform both actions at the same time by calculating
the expected value of fuzzy random variables [12,13,11,14]. Although the works of Liu proposed a promising
measure in terms of expected value of fuzzy random variables, the computation process of these expected val-
ues is quite complex and time consuming.
In this paper, the author proposes a new method to solve linear programming problems under the fuzziness
and randomness environment. In real-life, fuzzy stochastic linear programming arises in several situations.
The parameters of linear programs such as the right-hand-sides (RHSs) and the coefficients of the objective
and constraints could all be fuzzy random variables due to the fact that they depend upon many factors. Thus,
it is difficult to exactly determine the values of these parameters. Moreover, the factors, which are fluctuating
due to an uncertain environment, could cause these parameters to vary. These circumstances often happen
in situations wherein the described conditions (objectives, constraints, coefficients) cannot be determined pre-
cisely and certainly, such as long-term planning, development strategies [16], engineering design [20], and
financial modeling [21].
As an example of fuzzy stochastic linear programming, let’s consider the case of a production planning
problem where the objective is to minimize the total cost. This objective can be expressed as a fuzzy stochastic
variable because the total cost includes the cost of inventory holding, materials, and operation. Production
output depends on process parameters (cutting speeds and feed rates, for example) and machine running time.
However, machine running time is fluctuating and hard to estimate precisely. In addition, available resources,
demand, and constraints’ coefficients can also be modeled as fuzzy random variables because of the variances
in statistical data due to environmental conditions such as seasonal factors, market prices, and suppliers,
which contribute to constraint parameters.
Another example is the case of preventive maintenance. Factory equipment breaks down from time to time,
causing losses in production output. Preventive maintenance can be employed to reduce the frequency of such
breakdowns, and could involve activities such as inspections, repairs, and replacing components when neces-
sary. These activities are costly in terms of materials, wages, and the loss of production due to downtime while
the activities are being conducted. The length of the downtime is also uncertain due to the complexity of the
inspection, repair and replacement jobs, as well as the varying skills of maintenance staff. The problem is to
determine the frequency of preventative maintenance such that the total downtime, which includes downtime
due to breakdowns as well as preventive maintenance, is minimized while ensuring the associated costs do not
exceed the available budget. Here, the running time of the machine is also uncertain. Therefore, we would
rather consider these times and their associated costs as fuzzy random variables.
Examples such as these motivate the author to propose a new model for solving fuzzy stochastic linear pro-
gramming problems. In this paper, the attainment measure of fuzzy numbers and/or fuzzy random variables is
developed to convert the fuzzy stochastic linear programming problem into the corresponding deterministic
LP. The obtained LP, with few additional constraints, is easily solved by standard optimization packages such
as LINGO [18] or http://www.lindo.com.
This paper is organized as follows: Firstly, some important results of fuzzy random variables are summa-
rized. Then, in Section 3, the attainment model is developed. Section 4 utilizes the attainment measures to con-
vert the fuzzy linear programming problem into the corresponding deterministic LP. Section 5 presents the
solution method for the fuzzy stochastic linear programming problem. Finally, Section 6 presents the
conclusions.

2. Fuzzy random variable

In this section, we summarize some important concepts and results of the fuzzy random variable which is
the basis for the development of the attainment measures. There are several alternative approaches to define a
fuzzy random variable and its characteristics, and for this paper we utilize the results of Luhandjula [17] for
the definition.
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2973

Definition 1 [17]. Consider a probability space ðX; I; P Þ, a fuzzy random variable on this space is a fuzzy set-
valued mapping:
e : X ! F ðRÞ
X 0

w!X ew

such that for any Borel set B and for every a 2 ð0; 1Þ
e 1 ðBÞ ¼ fw 2 Xj X
X e a  Bg 2 I ð1Þ
a w

where F 0 ðRÞ and X e a stand for the set of fuzzy numbers with compact supports and the a-level set of the fuzzy
w
e w , respectively.
set X
e is a fuzzy random variable if and only if given w 2 X, X
Theorem 1 [17]. X e a is a random interval 8a 2 ð0; 1.
w

3. Attainment degree of fuzzy numbers and fuzzy random variables

Definition 2. Given two fuzzy numbers Pe ; Q e and Pe 6 Q,

e the lower-side attainment index of Pe to Q
e is defined
as
Z 1
e ¼
Dð Pe ; QÞ e
maxf0; supfs 2 R : Pe ðsÞ P ag  inffr 2 R : QðrÞ P agg da ð2Þ
0

e is defined as follows:
Definition 3. The both-side attainment index of fuzzy number Pe to fuzzy number Q

e ¼ maxf0; minðDð Pe ; QÞ;

V ð Pe ; QÞ e Dð Q;
e Pe ÞÞg ð3Þ

Let T be a collection of triangular fuzzy numbers ðT -numbers) with the following membership function:
8  
>
> max 0; 1  tx
l
; if x 6 t
>
< 1; if l ¼ 0; r ¼ 0; t ¼ x
T ¼ f~t; ~t ¼ ðt; l; rÞ; l; r P 0g and l~t ðxÞ ¼   ð4Þ
> max 0; 1  xt
> ; if x > t
>
: r
0; otherwise

where the scalars l; r P 0ðl; r 2 RÞ are called the left and right spreads, respectively.
e ¼ ðv; c; dÞ 2 Tðu 6 vÞ, the average lower-side
Proposition 1. Consider two fuzzy numbers Pe ¼ ðu; a; bÞ; Q
e is
attainment index of Pe to Q

e ¼ u vþ bþc:
Dð Pe ; QÞ ð5Þ
2

e at a-level.
Fig. 1. Attainment degree of fuzzy number Pe to Q
2974 N. Van Hop / Information Sciences 177 (2007) 2971–2984

e are (see Fig. 1)

Proof. The a-cuts of Pe and Q

½P l ðaÞ; P u ðaÞ ¼ ½u  að1  aÞ; u þ bð1  aÞ

½Ql ðaÞ; Qu ðaÞ ¼ ½v  cð1  aÞ; v þ dð1  aÞ
e at a-level:
We also have the lower-side attainment index of fuzzy number Pe to Q
e ¼ maxf0; Pe u  Q
Dð Pe ; QÞ elg
a a a
e ¼ maxf0; P u ðaÞ  Ql ðaÞg ¼ maxf0; ðu  vÞ þ ðb þ cÞð1  aÞg
Dð Pe ; QÞ a
Z 1
) Dð Pe ; QÞ e ¼ maxf0; ðu  vÞ þ ðb þ cÞð1  aÞg da
0
Z k ¼ð1bþc
vu
Þ
e ¼
() Dð Pe ; QÞ ½ðu  vÞ þ ðb þ cÞð1  aÞ da
0
Z k ¼ð1vu
bþcÞ
e ¼ 1
) Dð Pe ; QÞ ½ðu  vÞ þ ðb þ cÞð1  aÞ da
k 0
  2
k
e 1 a
() Dð Pe ; QÞ ¼  ðu  v þ b þ cÞa  ðb þ cÞ
k 2 0
  2   
e ¼ 1 ðu  v þ b þ cÞk  ðb þ cÞ k k
() Dð Pe ; QÞ ¼ ðu  v þ b þ cÞ  ðb þ cÞ
k 2 2
 
e ¼ ðu  v þ b þ cÞ  ðb þ cÞ 1  v  u ¼ u  v þ b þ c  ðb þ cÞ þ ðv  uÞ
() Dð Pe ; QÞ
2 ðb þ cÞ 2 2
e ¼ ðu  vÞ þ ðb þ cÞ
() Dð Pe ; QÞ 
2
From (3) and (5), the average both-side attainment index of Pe to Q e is defined as
 
e ¼ max 0; min u  v þ b þ c ; v  u þ a þ d
V ð Pe ; QÞ : ð6Þ
2 2
From (6) we derive the following result:
e 0 < k. Then
Proposition 2. Let Pe ; Q;
e Pk
V ð Pe ; QÞ ð7Þ
iff

e ¼ uvþbþc
Dð Pe ; QÞ Pk ð8Þ
2
e Pe Þ ¼ v  u þ a þ d P k
Dð Q; ð9Þ
2
e Pe 6 Q
Definition 4. The lower-side attainment index of fuzzy random variable Pe to fuzzy random variable Q; e
is defined as
Z 1
e e
Dð P ; QÞ ¼ e w ðrÞ P agg da
maxf0; supfs 2 R : Pe w ðsÞ P ag  inffr 2 R : Q ð10Þ
0

e is
Definition 5. The both-side attainment index of fuzzy random variable Pe to fuzzy random variable Q
defined as follows:

e ¼ maxf0; minðDð Pe ; QÞ;

V ð Pe ; QÞ e Dð Q;
e Pe ÞÞg ð11Þ
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2975

e w:
For any a 2 ð0; 1, consider the a-cut of the fuzzy sets Pe w ; Q
Pe aw ¼ ½P la ðwÞ; P ua ðwÞ e a ¼ ½Ql ðwÞ; Qu ðwÞ
Q w a a
e w are T-numbers, the a-cut of Pe w ; Q
By Theorem 1, these intervals are random intervals. If Pe w ; Q e w are: (see
Fig. 2)
Pe aw ¼ ½P la ðwÞ; P ua ðwÞ ¼ ½uðwÞ  aðwÞð1  aÞ; uðwÞ þ bðwÞð1  aÞ
Qe a ¼ ½Ql ðwÞ; Qu ðwÞ ¼ ½vðwÞ  cðwÞð1  aÞ; vðwÞ þ dðwÞð1  aÞ
w a a

Thus, the lower-side attainment index of Pe to Q e at a-level is

e ¼ maxf0; P u ðwÞ  Ql ðwÞg

Dð Pe ; QÞa a a

From these definitions, we have the following results:

e the average lower-side attainment
Proposition 3. Considering two triangular fuzzy random variables Pe 6 Q,
e is
index of Pe to Q
e ¼ uðwÞ  vðwÞ þ bðwÞ þ cðwÞ
Dð Pe ; QÞ ð12Þ
2

Proof

e ¼ maxf0; P u ðwÞ  Ql ðwÞg ¼ maxf0; ðuðwÞ  vðwÞÞ þ ðbðwÞ þ cðwÞÞð1  aÞg

Dð Pe ; QÞ a a a
Z 1
) Dð Pe ; QÞ e ¼ maxf0; ðuðwÞ  vðwÞÞ þ ðbðwÞ þ cðwÞÞð1  aÞg da
0

Z vðwÞuðwÞ
k ¼ 1bðwÞþcðwÞ
e ¼
() Dð Pe ; QÞ ½ðuðwÞ  vðwÞÞ þ ðbðwÞ þ cðwÞÞð1  aÞ da
0
Z k
e ¼ 1
) Dð Pe ; QÞ ½ðuðwÞ  vðwÞÞ þ ðbðwÞ þ cðwÞÞð1  aÞ da
k 0
  k
e e 1 a2
() Dð P ; QÞ ¼  ðuðwÞ  vðwÞ þ bðwÞ þ cðwÞÞa  ðbðwÞ þ cðwÞÞ
k 2 0
  
e e 1  k2
() Dð P ; QÞ ¼  ðuðwÞ  vðwÞ þ bðwÞ þ cðwÞÞk  ðbðwÞ þ cðwÞÞ
k 2
e ¼ uðwÞ  vðwÞ þ bðwÞ þ cðwÞ  ðbðwÞ þ cðwÞÞ þ ðvðwÞ  uðwÞÞ
() Dð Pe ; QÞ 
2 2

e at a-level.
Fig. 2. Lower-side attainment degree of Pe to Q
2976 N. Van Hop / Information Sciences 177 (2007) 2971–2984

e is defined as
From (8) and (9), the average both-side attainment index of Pe to Q
 
e ¼ max 0; min uðwÞ  vðwÞ þ bðwÞ þ cðwÞ ; vðwÞ  uðwÞ þ aðwÞ þ dðwÞ
V ð Pe ; QÞ ð13Þ
2 2
Similarly, from (10) we derive the following result:
e 0 < kðwÞ; w 2 X. Then
Proposition 4. Let Pe ; Q;
e P kðwÞ
V ð Pe ; QÞ ð14Þ
iff
uðwÞ  vðwÞ þ bðwÞ þ cðwÞ
P kðwÞ ð15Þ
2
vðwÞ  uðwÞ þ aðwÞ þ dðwÞ
P kðwÞ ð16Þ
2
The main difference of the proposed approach with traditional ones such as the signed distance method [19], is
the defuzzifying method used to obtain the attainment degree between fuzzy numbers and fuzzy stochastic
variables. The signed distance method defuzzifies fuzzy numbers by using the absolute relationship between
the converted points of fuzzy numbers, while the proposed approach uses the relative relationship between
fuzzy numbers/fuzzy stochastic variables via attainment degrees. To illustrate the advantages of the proposed
approach, in the next section the author will compare it with the signed distance method.

4. Fuzzy linear programming

Using the results obtained in Sections 2 and 3, the method for solving fuzzy linear programming problems
is now developed in this section. Consider the following fuzzy linear program:
8
< Max
Pn
cx
ðP 1Þ : aij Þxj 6 ð~
ð~ bi Þ; i ¼ 1; m
: j¼1
xj P 0
where c is a (1xn) matrix. The common way to solve this problem is to achieve the objective subject so as to
avoid any violation of the constraints. Consequently, any violation of the left-hand-side (LHS) to the right-
hand-side (RHS) of constraints is minimized. Thus, by applying the upper-side attainment index to
constraints, the problem (P1) is reformulated as:
8
> Pm
>
> Max cx  ki
>
>
>
>
i¼1
< s:t:
!
ðP 2Þ : P n
>
> ~
Di ðbi Þ; ð~ aij Þxj ¼ ki ; i ¼ 1; m
>
>
>
> j¼1
>
:
xj ; ki P 0
According the result of (5), (P2) is a standard LP and can be solved by traditional optimization software, such
as LINGO [18] or http://www.lindo.com.
To illustrate the advantages of the proposed approach, we consider the following production planning
problem to compare the proposed method with the signed distance method (modified from [7]):
Example 1. A factory produces two products P and Q. Products P and Q are manufactured by two mixing
processes M and N, respectively. It takes about 2 h at process M and about 6 h at process N to manufacture a
batch of Product P. On the other hand, it takes about 3 h at process M and about 4 h at process N to
manufacture a batch of product Q. Subject to many factors such as machine breakdown and time spent
waiting for materials, the working time of process M (resp. N) is substantially smaller than 9 h (resp. 18). The
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2977

profit rates ($1000/batch) of products P and Q are 7 and 9, respectively. What quantity of products P and Q
should be manufactured under such circumstances to maximize total profit?
Here we have the following FLP problem

Maximize 7x1 þ 9x2

Subject to ~
2x1 þ ~
3x2 6 ~
9
~ ~ f
6x1 þ 4x2 6 18
x1 ; x2 P 0

Assume all fuzzy numbers are in the form of ~t ¼ ðt; d1 ; d2 Þ 2 T with d1 ¼ d2 ¼ 1 for all coefficients. Convert-
ing this example to the form of a linear program (P2) by using (5), we have:

Maximize 7x1 þ 9x2  k1  k2

1
Subject to ½ð9  2x1  3x2 Þ þ ð1 þ x1 þ x2 Þ ¼ k1
2
1
½ð18  6x1  4x2 Þ þ ð1 þ x1 þ x2 Þ ¼ k2
2
x1 ; x2 P 0

The solution of this linear program is x1 ¼ 0:875; x2 ¼ 4:875 and objective value = 50.
If we use the signed distance method, the signed distance of fuzzy number l ~ ¼ ðl; D1 ; D2 Þ to ~01 (y axis)
is
1 1
l; ~
dð~ 01 Þ ¼ ð2l þ l  D1 þ l þ D2 Þ ¼ ð4l  D1 þ D2 Þ ð17Þ
4 4
Then, the linear program (P2) is converted to
Maximize 7x1 þ 9x2
Subject to 2x1 þ 3x2 6 9
6x1 þ 4x2 6 18
x1 ; x2 P 0

The solution of this linear program is x1 ¼ 1:8; x2 ¼ 1:8 and objective value = 28.8. Since our problem is
maximized problem, the obtained result demonstrates that our approach gives a better solution 50 versus
28.8.
To confirm the advantages of the proposed approach over the signed distance method, we designed an
experiment that utilizes real-life data collected from a detergent powder mixing process similar to the example
mentioned above. All processing times and the available times of machines are approximately measured in the
form of fuzzy numbers and given in Table 1. The profit rates per batch ($1000/batch) are also presented in
Table 1. Fifteen tests are studied.
The obtained results of the 15 cases are presented in Table 2. Row 1 is our method’s results. Row 2 is the
signed distance method’s results.
From the obtained results, it is seen that our method gives a better solution in terms of objective values. The
reason for this is that our method compares fuzzy numbers based on their relative relationships. This compar-
ison relaxes the constraints from the deterministic ones. Thus, the larger spreads of fuzzy numbers, the greater
the relaxation amount. Therefore, overlapping areas between fuzzy numbers will be increased and the
obtained results will be better. In contrast, the signed distance method uses the absolute relationship between
fuzzy numbers. They convert the fuzzy numbers into the corresponding deterministic numbers. This method
does not adjust the values of deterministic constraints, so it is similar to the deterministic case. In particular,
when two spreads of fuzzy numbers are equal, the converted constraints are exactly the same as the determin-
istic constraints. Therefore, the obtained results will be close to the deterministic solution.
2978 N. Van Hop / Information Sciences 177 (2007) 2971–2984

Table 1
Collected data
Case cj aij alij arij bi bli bri
1 8.03 6.81 0.76 0.63 14.71 0.83 0.65
9.07 2.24 0.11 0.52
6.82 0.87 0.41 5.89 0.90 0.80
3.86 0.34 0.86
2 1.73 0.78 0.52 0.07 5.84 1.17 1.23
0.98 8.49 0.49 0.48
2.95 0.69 0.71 12.10 1.12 0.65
7.02 0.61 0.75
3 3.53 8.67 0.03 0.10 9.39 0.99 0.75
4.69 0.72 0.94 0.84
5.46 0.85 0.23 14.08 0.94 1.14
6.57 0.91 0.47
4 9.86 3.49 0.42 0.45 13.80 0.77 0.93
4.36 9.61 0.59 0.36
6.69 0.76 0.38 10.77 1.13 0.68
5.23 0.10 0.52
5 0.50 7.11 0.58 0.95 14.01 0.87 1.23
7.28 2.26 0.23 0.36
8.26 0.06 0.71 14.11 1.04 0.72
3.39 0.61 0.44
6 5.42 6.65 0.08 0.02 14.89 0.56 1.30
8.29 3.78 0.02 0.92
0.69 0.67 0.98 7.23 0.73 0.65
4.15 0.93 0.22
7 0.11 5.89 0.17 0.99 5.62 0.92 0.64
6.68 7.78 0.07 0.56
3.41 0.96 0.31 12.95 0.97 0.89
3.81 0.27 0.70
8 8.36 9.03 0.75 0.12 14.81 1.12 1.13
0.16 3.85 0.28 0.26
5.02 0.66 0.38 10.80 1.33 0.68
2.85 0.57 0.78
9 2.97 8.46 0.15 0.04 10.27 0.82 1.20
1.52 0.94 0.39 0.41
4.63 0.22 0.94 12.42 0.56 1.48
3.03 0.38 0.68
10 7.79 6.85 0.82 0.28 9.01 1.36 0.65
3.78 2.37 0.96 0.81
8.58 0.82 0.42 9.98 1.11 1.29
0.86 0.70 0.16
11 9.52 0.64 0.77 0.69 5.50 0.58 0.64
1.68 3.57 0.35 0.76
5.41 0.79 0.46 7.31 1.23 0.56
1.72 0.45 0.09
12 3.90 4.02 0.46 0.24 12.55 0.71 1.24
3.71 3.89 0.91 0.46
9.64 0.82 0.12 14.16 0.91 1.31
6.20 0.51 0.72
13 6.2 9.8 0.58 0.85 13.82 1.05 1.24
8.34 2.26 0.7 0.64
4.74 0.68 0.68 6.36 0.55 0.91
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2979

Table 1 (continued)
Case cj aij alij arij bi bli bri
9.42 0.96 0.19
14 5.16 1.96 0.57 0.45 12.46 0.81 0.81
0.38 6.75 0.18 0.16
4.20 0.93 0.70 14.48 0.90 0.92
4.05 0.48 0.17
15 0.30 4.21 0.29 0.53 12.72 0.98 1.48
6.63 4.10 0.66 1.00
6.84 0.76 0.38 13.11 0.58 0.76
3.54 0.92 0.45

5. Fuzzy stochastic linear programming problem

Consider the following fuzzy stochastic linear program:

8
< Min
>
Pn
cx
ðP 3Þ : aij Þw xj 6 ð~
j¼1 ð~ bi Þw ; i ¼ 1; m
>
:
xj P 0; w 2 X; k ¼ 1; l
where c is a 1xn matrix. A; b are mxn and mx1 matrices of fuzzy random variable constraint coefficients defined
on a probability space ðX; F ; P Þ. Applying the lower-side attainment index to minimize the achievement of the
LHS to the RHS, problem (P3) is reformulated as follows:
8 Pm
>
> Min cx þ ki ðwÞ
>
>
>
> i¼1
>
< s:t:
!
ðP 4Þ : Pn
>
>
>
> Di aij Þw xj ; ð~
ð~ bi Þw ¼ ki ðwÞ; i ¼ 1; m
>
> j¼1
>
:
xj P 0; w 2 X
(P4) is then de-randomized by using stochastic programming techniques. The corresponding deterministic
program for this problem is:
8 m 
> P
>
> Min cx þ E ki ðwÞ
>
>
>
> i¼1
>
< s:t:
!
ðP 5Þ : Pn
>
> ~
>
> Di ðbi Þw ; ð~ aij Þw xj ¼ ki ðwÞ; i ¼ 1; m
>
> j¼1
>
>
:
xj ; ki ðwÞ P 0; w 2 X
where E denotes the mathematical expectation.
Example 2. We use this example to show that Luhandjula’s model solution can be improved by the proposed
method. Reconsider the example of [17]):

Table 2
Comparison between our method and signed method
Case 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 17.24 9.75 12.27 19.05 38.84 26.02 5.42 16.1 8.48 25.92 14.39 10.08 10.73 24.3 26.39
2 13.33 7.02 10.26 15.93 30.51 20.13 4.69 13.96 6.77 14.31 12.76 8.46 8.44 18.04 20.35
2980 N. Van Hop / Information Sciences 177 (2007) 2971–2984

Minimize 3x1 þ 2x2

e 6e
Subject to Ax b
x P 0; w 2 X
where
8 ( ! !)
>
> ~
1 ~
1 ~
3
>
> e ;~
ðA b Þ¼ ;
>
< w1 w1 ~ ~ ~
e ~ 2 1 4
ð A; 
bÞ ¼ ( ! !)
>
> ~
1 ~
3 ~
5
> ðA
> e ~
: w2 ; bw2 Þ ¼
> ~
1 ~
2
;
~
4
pðw1 Þ ¼ 0:25; pðw2 Þ ¼ 0:75; X ¼ ðw1 ; w2 Þ
and ~t denotes a triangular fuzzy number with the following membership function:
8
> 0; x 6 t  0:5
>
>
>
< x  ðt  0:5Þ; ðt  0:5Þ < x 6 t
l~t ðxÞ ¼
> x þ ðt þ 0:5Þ;
> t 6 x < ðt þ 0:5Þ
>
>
:
0; ðt þ 0:5Þ 6 x
The program (P5) corresponding to this example is then
Minimize 3x1 þ 2x2 þ ½ð0:25k11 þ 0:25k21 þ 0:75k12 þ 0:75k22 Þ
Subject to Dð~
1x1 þ ~
1x2 ; ~
3Þ ¼ k11
Dð~2x1 þ ~
1x2 ; ~
4Þ ¼ k21
Dð~1x1 þ ~
3x2 ; ~
5Þ ¼ k12
Dð~1x1 þ ~
2x2 ; ~
4Þ ¼ k22
xj ; k1h ¼ k1 ðwh Þ; k2h ¼ k2 ðwh Þ P 0
i ¼ 1; 2; j ¼ 1; 2; k ¼ 1; 2; h ¼ 1; 2
The corresponding LP is
Minimize 3x1 þ 2x2 þ ½ð0:25k11 þ 0:25k21 þ 0:75k12 þ 0:75k22 Þ
1
Subject to ðx1 þ x2  3 þ 0:5x1 þ 0:5x2 þ 0:5Þ ¼ k11
2
1
ð2x1 þ 1x2  4 þ 0:5x1 þ 0:5x2 þ 0:5Þ ¼ k21
2
1
ðx1 þ 3x2  5 þ 0:5x1 þ 0:5x2 þ 0:5Þ ¼ k12
2
1
ðx1 þ 2x2  4 þ 0:5x1 þ 0:5x2 þ 0:5Þ ¼ k22
2
xj ; k1h ¼ k1 ðwh Þ; k2h ¼ k2 ðwh Þ P 0
i ¼ 1; 2; j ¼ 1; 2; k ¼ 1; 2; h ¼ 1; 2
Using Lingo [18] or http://www.lindo.com, the obtained solution is x1 ¼ 0:85; x2 ¼ 0:92 and objective va-
lue = 4.9. The solution of Luhandjula is x1 ¼ 1:34; x2 ¼1 and objective value = 6.02, which is larger than
our solution. Since this is a minimized problem, our solution is better. In addition, the number of constraints
in our model is fewer than in the model of Luhandjula. Thus, for larger problems, the computation time of our
model will be relatively shorter.
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2981

Table 3
Comparison between our method and Luhandjula’s method
Case 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 15.2 25.27 17.85 35.35 11.61 0.60 14.30 32.94 25.54 3.77 3.80 18.79 8.16 27.65 6.72
2 23.34 88.2 21.17 57.38 26.91 0.72 16.33 118.42 33.19 6.89 8.56 25.28 9.83 32.81 8.43

Similar to the fuzzy linear programming case, we compare our approach with Luhandjula’s method by an
experiment in which all parameters are generated randomly in the form of uniform distributions. The param-
eters of Luhandjula’s method do not change, i.e., 0 ¼ a0 < 0:4 ¼ a1 < 0:6 ¼ a2 < 0:8 ¼ a3 < a4 ¼ 1;
0 ¼ d4 < 0:01 ¼ d3 < 0:02 ¼ d2 < 0:03 ¼ d1 ; and ga1 a2 a3
i ¼ 2; g i ¼ 3; g i ¼ 4. The generated data is given in
Appendix A and the obtained results are presented in Table 3. In Table 3, row 1 contains our method’s results,
and row 2 are the results of Luhandjula’s method.
From the obtained results, it is seen that our method gives a better solution in terms of objective values.
This improvement is a consequence of the primary difference between our method and the Luhandjula’s
method – our approach uses overlapping areas to convert the fuzzy stochastic linear programming problem
into the corresponding deterministic one, while Luhandjula’s method discretizes fuzzy stochastic variables
by a-levels. This discretizing process creates more constraints, causing the obtained deterministic LP to be clo-
ser to the deteriministic solution than ours. This also results in Luhandjula’s method requiring more time to
compute.

6. Conclusion

This paper has presented a simple method to convert the fuzzy and fuzzy stochastic linear programming
problems into the conventional LPs through the use of attainment values of fuzzy numbers and/or fuzzy ran-
dom variables. The proposed method provides simple deterministic LP models, which can then be solved eas-
ily using standard optimization packages. In addition, there is greater computation efficiency because the
obtained models have fewer constraints.

Appendix A

See Table 4.

Table 4
Data for the fuzzy stochastic experiment
Case cj p1 p2 aij alij arij bi bli bri
1 2.57 0.57 0.43 2.05 0.00 0.67 5.62 0.96 0.61
2.46 1.13 0.74 0.87
1.46 0.34 0.96 14.99 1.33 1.32
1.28 0.10 0.26
0.50 0.57 0.59 12.44 1.10 1.44
6.32 0.04 0.43
5.52 0.69 0.04 10.26 0.81 1.20
0.93 0.08 0.97
2 8.5 0.23 0.77 2.25 0.12 0.29 10.68 0.66 1.11
9.4 7.31 1 0.23
5.74 0.5 0.01 8.37 1.15 1.15
2.71 0.66 0.99
5.86 0.49 0.69 6.24 0.88 0.69
1.64 0.38 0.01
6.91 0.44 0.38 14.85 0.98 1.26
1.39 0.64 0.13
(continued on next page)
2982 N. Van Hop / Information Sciences 177 (2007) 2971–2984

Table 4 (continued)
Case cj p1 p2 aij alij arij bi bli bri
3 7.79 0.37 0.63 5.51 0.95 0.70 5.06 1.17 1.29
1.05 9.34 0.96 0.32
1.63 0.76 0.67 12.84 1.35 1.47
4.21 0.13 0.56
8.74 0.63 0.40 6.90 1.28 0.52
1.18 0.13 0.31
3.25 0.22 0.06 9.00 0.88 1.22
0.32 0.46 0.51
4 6.13 0.15 0.85 2.05 0.38 0.06 13.41 1.24 0.79
8.98 0.27 0.18 0.02
3.22 0.83 0.38 5.78 0.96 1.42
2.17 0.89 0.67
9.23 0.25 0.46 8.77 1.24 0.96
5.27 0.90 0.20
9.40 0.22 0.33 6.45 1.17 1.19
3.82 0.63 0.54
5 3.89 0.72 0.28 8.85 0.88 0.42 12.16 1.23 1.40
5.51 1.44 0.42 0.48
5.08 0.40 0.10 13.75 0.53 0.50
3.15 0.65 0.99
5.65 0.64 0.25 5.08 0.69 1.33
0.81 0.96 0.42
2.38 0.45 0.69 8.79 1.49 1.32
1.55 0.15 0.14
6 6.57 0.64 0.36 6.04 0.94 0.56 8.69 0.65 0.56
0.34 5.74 0.31 0.71
9.92 0.29 0.25 9.64 0.71 1.00
8.50 0.01 0.37
4.07 0.23 0.44 9.66 0.53 1.18
9.72 0.37 0.11
0.30 0.01 0.25 13.05 0.98 0.63
5.77 0.94 0.99
7 1.88 0.26 0.74 1.74 0.58 0.97 11.65 1.08 0.73
8.49 8.90 0.20 0.02
4.11 0.37 0.06 13.74 0.54 0.65
3.44 0.10 0.11
1.97 0.30 0.19 7.77 1.15 0.52
3.63 0.17 0.82
0.98 0.37 0.14 10.34 0.55 0.95
5.43 0.99 0.67
8 2.25 0.10 0.90 5.15 0.47 0.49 6.90 0.71 0.64
6.20 8.86 0.56 0.43
4.11 0.01 0.01 6.07 1.13 0.98
5.92 0.55 0.13
2.52 0.61 0.58 11.12 1.06 0.65
9.61 0.96 0.78
0.69 0.48 0.19 14.15 1.29 1.09
0.31 0.45 0.94
9 2.82 0.30 0.70 0.66 0.87 0.72 7.76 1.28 0.50
7.92 5.54 0.40 0.39
5.83 0.30 0.37 6.44 0.85 0.82
2.50 0.79 0.85
0.79 0.26 0.25 9.98 0.57 0.98
0.43 0.60 0.31
 N. Van Hop / Information Sciences 177 (2007) 2971–2984 2983

Table 4 (continued)
Case cj p1 p2 aij alij arij bi bli bri
9.18 0.04 0.37 5.16 1.05 0.75
4.43 0.16 0.43
10 1.75 0.21 0.79 7.09 0.80 0.20 9.09 0.61 1.44
3.54 8.70 0.82 0.95
5.77 0.18 0.78 14.79 0.65 0.90
7.51 0.79 0.16
8.15 0.39 0.67 7.56 0.75 0.94
3.83 0.57 0.76
5.92 0.50 0.42 5.93 1.27 1.12
5.24 0.51 0.08
11 7.69 0.67 0.33 9.91 0.35 0.01 13.14 1.48 0.59
3.19 9.40 0.87 0.40
8.72 0.47 0.86 10.66 1.06 0.93
7.22 0.90 0.96
3.93 0.10 0.29 5.75 1.16 0.75
4.68 0.82 0.91
6.25 0.92 0.57 8.00 0.66 0.65
9.98 0.88 0.05
12 5.43 0.23 0.77 5.15 0.75 0.93 14.22 0.58 0.70
5.74 8.94 0.61 0.68
0.04 0.05 0.59 7.92 1.22 0.67
6.85 0.97 0.73
7.39 0.56 0.01 10.37 0.53 0.63
2.32 0.10 0.97
2.25 0.50 0.52 9.36 0.92 1.37
0.64 0.61 0.53
13 2.91 0.51 0.49 3.69 0.69 0.66 12.82 1.44 1.04
7.70 7.14 0.19 0.71
3.40 0.53 0.08 10.60 0.84 0.74
0.44 0.04 0.46
5.37 0.24 0.97 11.00 1.39 1.02
7.19 0.04 0.35
5.46 0.19 0.61 13.54 0.77 0.93
8.54 0.56 0.15
14 9.62 0.44 0.56 6.77 0.69 0.34 11.48 0.95 1.16
8.71 4.22 0.43 0.21
7.36 0.36 0.25 13.11 1.03 0.76
6.45 0.48 0.30
0.51 0.76 0.83 12.84 0.84 1.44
5.94 0.69 0.36
4.29 0.82 0.44 10.08 1.16 1.38
1.31 0.80 0.05
15 5.65 0.70 0.30 6.45 0.42 0.19 8.74 0.98 1.42
6.54 8.72 0.84 0.23
6.18 0.10 0.91 9.75 1.32 0.91
0.86 0.91 0.89
5.95 0.79 0.16 6.13 0.87 1.27
1.71 0.23 0.83
7.41 0.85 0.48 9.15 0.96 0.94
1.58 0.50 0.80

References

[1] L. Campos, J.L. Verdegay, Linear programming problems and ranking of fuzzy numbers, Fuzzy Sets and Systems 32 (1989)
1–11.
2984 N. Van Hop / Information Sciences 177 (2007) 2971–2984

[2] S. Chanas, P. Zielinski, On the equivalence of two optimization methods for fuzzy linear programming problems, European Journal
of Operational Research 121 (2000) 56–63.
[3] J. Chiang, Fuzzy linear programming based on statistical confidence interval and interval-valued fuzzy set, European Journal of
Operational Research 129 (2001) 65–86.
[4] S.C. Fang, C.F. Hu, H.F. Wang, S.Y. Wu, Linear programming with fuzzy coefficients in constraints, Computers and Mathematics
with Applications 37 (1999) 63–76.
[5] Ph. Fortemps, M. Roubens, Ranking and defuzzification methods based on area compensation, Fuzzy Sets and Systems 82 (1996)
319–330.
[6] S.M. Guu, Y.K. Wu, Two-phase approach for solving the fuzzy linear programming problems, Fuzzy Sets and Systems 107 (1999)
191–195.
[7] M. Inuiguchi, J. Ramik, Possibilistic linear programming: a brief review of fuzzy mathematical programming and a comparison with
stochastic programming in portfolio selection problem, Fuzzy Sets and Systems 111 (2000) 3–28.
[8] K.D. Jamison, W.A. Lodwick, Fuzzy linear programming using a penalty method, Fuzzy Sets and Systems 119 (2001) 97–110.
[9] H.R. Maleki, M. Tata, M. Mashinchi, Linear programming with fuzzy variables, Fuzzy Sets and Systems 109 (2000) 21–33.
[10] D. Wang, An inexact approach for linear programming problems with fuzzy objective and resources, Fuzzy Sets and Systems 89
(1997) 61–68.
[11] B. Liu, Y.K. Liu, Expected value of fuzzy variable and fuzzy expected value models, IEEE Transactions on Fuzzy Systems 10 (4)
(2002) 445–450.
[12] B. Liu, Fuzzy random dependent-chance programming, IEEE Transactions on Fuzzy Systems 9 (5) (2001) 721–726.
[13] B. Liu, Fuzzy random chance-constrained programming, IEEE Transactions on Fuzzy Systems 9 (5) (2001) 713–720.
[14] Y.K. Liu, B. Liu, A class of fuzzy random optimization: expected value models, Information Sciences 155 (2003) 89–102.
[15] M.K. Luhandjula, Optimization under hybrid uncertainty, Fuzzy Sets and Systems 146 (2004) 187–203.
[16] M.K. Luhandjula, M.M. Gupta, On fuzzy stochastic optimization, Fuzzy Sets and Systems 81 (1996) 47–55.
[17] M.K. Luhandjula, Fuzziness and randomness in an optimization framework, Fuzzy Sets and Systems 77 (1996) 291–297.
[18] L.E. Schrage, Optimization Modeling with LINGO, LINDO Systems, Inc, Chicago, 1999.
[19] J.S. Yao, K. Wu, Ranking fuzzy numbers based on decomposition principle and signed distance, Fuzzy Sets and Systems (2001).
[20] C.J. Shih, R.A.S. Wangsawidjaja, Mixed fuzzy-probabilistic programming approach for multiobjective engineering optimization with
random variables, Computers and Structures 59 (2) (1996) 283–290.
[21] Z. Zmeskal, Value at risk methodology under soft conditions approach (fuzzy-stochastic approach), European Journal of Operational
Research 161 (2) (2005) 337–347.