Applied Mathematics and Computation 188 (2007) 786–800
www.elsevier.com/locate/amc
A probabilistic bi-level linear multi-objective
programming problem to supply chain planning
E. Roghanian *, S.J. Sadjadi, M.B. Aryanezhad
Department of Industrial Engineering, Iran University of Science and Technology, Narmak, Tehran P.C. 16844, Iran
Abstract
Bi-level programming, a tool for modeling decentralized decisions, consists of the objective(s) of the leader at its first
level and that is of the follower at the second level. Three level programming results when second level is itself a bi-level
programming. By extending this idea it is possible to define multi-level programs with any number of levels.
In most of the real life problems in mathematical programming, the parameters are considered as random variables.
The branch of mathematical programming which deals with the theory and methods for the solution of conditional extremum problems under incomplete information about the random parameters is called ‘‘stochastic programming’’.
Supply chain planning problems are concerned with synchronizing and optimizing multiple activities involved in the
enterprise, from the start of the process, such as procurement of the raw materials, through a series of process operations,
to the end, such as distribution of the final product to customers.
Enterprise-wide supply chain planning problems naturally exhibit a multi-level decision network structure, where for
example, one level may correspond to a local plant control/scheduling/planning problem and another level to a corresponding plant-wide planning/network problem. Such a multi-level decision network structure can be mathematically represented by using ‘‘multi-level programming’’ principles.
In this paper, we consider a ‘‘probabilistic bi-level linear multi-objective programming problem’’ and its application in
enterprise-wide supply chain planning problem where (1) market demand, (2) production capacity of each plant and (3)
resource available to all plants for each product are random variables and the constraints may consist of joint probability
distributions or not. This probabilistic model is first converted into an equivalent deterministic model in each level, to
which fuzzy programming technique is applied to solve the multi-objective nonlinear programming problem to obtain a
compromise solution.
2006 Published by Elsevier Inc.
Keywords: Bi-level programming; Multi-objective decision-making; Multi-level multi-objective decision-making; Fuzzy decisionapproach; Stochastic programming; Supply chain management
*
Corresponding author.
E-mail addresses: E-roghanian@iustarak.ac.ir (E. Roghanian), Sjsadjadi@iust.ac.ir (S.J. Sadjadi), Mirarya@iust.ac.ir (M.B.
Aryanezhad).
0096-3003/$ - see front matter 2006 Published by Elsevier Inc.
doi:10.1016/j.amc.2006.10.032
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
787
1. Introduction and literature review
1.1. Bi-level programming
A bi-level programming problem is formulated for a problem in which two decision-makers make decisions
successively. For example, in a decentralized firm, top management makes a decision such as budget of the
firm, and then each division determines a production plane in the full knowledge of the budget [1].
Research on multi-level mathematical programming to solve organizational planning and decision-making
problems has been conducted widely. The research and application have concentrated mainly on bi-level programming [1]. In the BLP problem, each decision maker tries to optimize its own objective function(s) without
considering the objective(s) of the other party, but the decision of each party affects the objective value(s) of
the other party as well as the decision space.
The general formulation of a bi-level programming problem (BLPP) is [2]:
min
F ðx; yÞ
s:t:
Gðx; yÞ 6 0;
min
f ðx; yÞ
s:t:
gðx; yÞ 6 0;
x
y
ð1Þ
where x 2 Rn1 and y 2 Rn2.The variables of problem are divided into two classes, namely the upper-level
variables x 2 Rn1 and the lower-level variables y 2 Rn2. Similarly, the functions F : Rn1 · Rn2 ! R and
f : Rn1 · Rn2 ! R are the upper-level and lower-level objective functions respectively, while the vector-valued
functions G : Rn1 · Rn2 ! Rm1 and g : Rn1 · Rn2 ! Rm2 are called the upper-level and lower-level constraints
respectively. All of the constraints and objective functions may be linear, quadratic, nonlinear, fractional,
etc.
1.2. Stochastic programming
In most of the real life problems in mathematical programming, the parameters are considered as random
variables. The branch of mathematical programming which deals with the theory and methods for the solution
of conditional extremum problems under incomplete information about the random parameters is called ‘‘stochastic programming’’. Most of the problems in applied mathematics may be considered as belonging to any
one of the following classes [7]:
1. Descriptive Problems, in which, with the help of mathematical methods, information is processed about the
investigated event, some laws of the event being induced by others.
2. Optimization Problems in which from a set of feasible solutions, an optimal solution is chosen.
Besides the above division of applied mathematics problems, they may be further classified as deterministic
and stochastic problems. In the process of the solution of the stochastic problem, several mathematical methods have been developed. However, probabilistic methods were for a long time applied exclusively to the solution of the descriptive type of problems. Research on the theoretical development of stochastic programming
is going on for the last four decades. To the several real life problems in management science, it has been
applied successfully [13]. The chance constrained programming was first developed by Charnes and Cooper
[4]. Subsequently, some researchers like Sengupta [12], Contini [5], Sullivan and Fitzsimmons [14], Leclercq
[9], Teghem et al. [15] and many others have established some theoretical results in the field of stochastic programming. Stancu-Minasian and Wets [13] have presented a review paper on stochastic programming with a
single objective function.
The fuzziness occurs in many of the real life decision making problems. Decision making in a fuzzy environment was first developed by Bellman and Zadeh [3]. Zimmermann [16] presented an application of fuzzy
linear programming to the linear vector-maximum problem and showed that the solution obtained by fuzzy
788
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
linear programming is always efficient. Hanan [6], Narasimhan [10], Leberling [8] and many others have made
contributions in fuzzy goal programming and fuzzy multi-objective programming.
1.3. Supply chain and supply chain management
In recent years there has been a great interest in enterprise-wide supply chain planning problems because of
their impact on substantially improving the overall competitiveness of economic potential of individual organizations. Supply chain planning problems are concerned with synchronizing and optimizing multiple activities involved in the enterprise, from the start of the process, such as procurement of the raw materials, through
a series of process operations, to the end, such as distribution of the final product to customers [17].
Supply chain planning has thus obtained a great deal of attention in the open literature. A number of outstanding issues deserve some further attention [17]:
First, supply chain planning problems naturally incorporate multiple decision modeling steps, which are
connected in a ‘‘hierarchical’’ way. Since individual activities are often governed by separate supply chain
components which have their own, often mutually conflicting, objectives, the operation and control of the
entire networks is based on multi-perspectives (see [18]). Most of the planning models are however, based
on the assumption that ‘‘all’’ activities of supply chain networks are governed by a ‘‘global organizer’’ neglecting such multiple perspectives (for example [19–22]).
Second, different participant components in the supply chain network may not operate with the same level
of information. Some may possess more information, while other may possess less information; this may lead
to information distortion as discussed in [23]. Recent works where this important issue is being discussed
include Gjerdrum et al. [24] and Zhou et al. [25].
Finally, the presence of uncertainty in supply chain planning models further amplifies the complexity of
the problem. Uncertainty typically exists in supply chain parameters, such as processing times, performance
coefficients, utility coefficients, delivery and inventory costs, supply of raw materials, etc. Here, approaches
which have started to appear in the open literature include scenario-based multi-period formulations [5],
stochastic programming formulations [20,21] and supply chain dynamics and control formulations [26–
28].
In this paper, we consider a ‘‘probabilistic bi-level linear multi-objective programming problem’’ and its
application in enterprise-wide supply chain planning problem, where market demand and warehouse capacity
are random variables and the constraints follow a joint probability distributions or not. This probabilistic
model is first converted into an equivalent deterministic model in each level, to which fuzzy programming technique is applied to solve the multi-objective nonlinear programming problem to obtain a compromise
solution.
The paper is further organized as follows: in the next section we will present an example for motivating
reader, in Section 3 multi-objective chance constrained programming will be described, in Section 4 we will
explain a multi-level nonlinear multi-objective decision-making under fuzziness, in Section 5 model formulation will be present, in Section 6 a solution method will be presented, in Section 7 an illustrative numerical
example is given to demonstrate the obtained results and finally in Section 8, conclusions are drawn regarding
the model.
2. A motivating example [17]
Consider a manufacturing enterprise involving two plants P1, P2 and one warehouse or distribution centre
(DC), with two products, A and B. The enterprise aims to minimize the overall cost which corresponds to a
production cost and a distribution cost. Based on the notation in Table 1, the supply chain model may be
mathematically formulated into the following constraints (see Fig. 1).
The production part is subject to the following constraints:
(a) Common resources are shared by both plants
Y 1A þ Y 1B þ Y 2A þ Y 2B 6 500:
ð2Þ
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
789
Table 1
Notation for motivating example [17]
Description
Variable
Y1A
Y1B
Y2A
Y2B
XA
XB
Production amount A in plant P1 (ton)
Production amount B in plant P1 (ton)
Production amount A in plant P2 (ton)
Production amount B in plant P2 (ton)
Inventory holding of A in DC (ton)
Inventory holding of B in DC (ton)
Parameter
hA
hB
Demand amount A (ton)
Demand amount B (ton)
P1
DC
PRODUCT A,
B
CUSTOMER
P2
Fig. 1. Process configuration of an enterprise in motivating example [17].
(b) Some resources may be controlled by individual plant conditions:
Y 1A þ Y 1B 6 200;
Y 2A þ Y 2B 6 250:
ð3Þ
(c) Production levels achieved by both plants should not be lower than levels required by the inventory
Y 1A þ Y 2A P X A ;
Y 1B þ Y 2B P X B :
ð4Þ
The objective of the production part is to minimize a production cost which can be typically formulated as
follows:
min Z PC ¼ 1:5X A þ 2X B þ 7Y 1A þ 3Y 1B þ 10Y 2A þ 6Y 2B :
ð5Þ
On the other hand, distribution centers are subject to the following constraints:
(a) Inventory levels are limited by their overall capacity
3X A þ 2X B 6 500:
ð6Þ
(b) Inventory levels should meet demands
X A P hA ;
X B P hB :
ð7Þ
The objective of the distribution centre is to minimize a distribution cost which may be formulated as follows:
min
Z DC ¼ 15X A þ 13X B þ 3Y 1A þ 2Y 1B þ 3:5Y 2A þ 2:5Y 2B :
ð8Þ
A way to pose this problem is formulating the problem as a bi-level optimization model which is as follows:
min
X A ;X B
s:t:
Z DC ¼ 15X A þ 13X B þ 3Y 1A þ 2Y 1B þ 3:5Y 2A þ 2:5Y 2B
3X A þ 2X B 6 500;
X A P hA ;
790
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
X B P hB ;
Y 1A þ Y 2A P X A ;
min
Y 1A ;Y 1B ;Y 2A ;Y 2B
s:t:
Y 1B þ Y 2B P X B ;
Z PC ¼ 1:5X A þ 2X B þ 7Y 1A þ 3Y 1B þ 10Y 2A þ 6Y 2B
Y 1A þ Y 1B þ Y 2A þ Y 2B 6 500;
Y 1A þ Y 1B 6 200;
Y 2A þ Y 2B 6 250;
Y 1A þ Y 2A P X A ;
Y 1B þ Y 2B P X B ;
X A ; X B ; Y 1A ; Y 1B ; Y 2A ; Y 2B P 0:
ð9Þ
The formulation in (9) consists of two sub-problems: the higher level decision problem and the lower level
decision problem. The objective function for the higher level problem is ZDC and for the lower level is
ZPC.1 The higher level problem is also known as the leader’s problem or the outer problem, the lower level
problem is sometimes called a follower’s problem or an inner problem. The two problems are connected in
a way that the leader’s problem sets parameters influencing the follower’s problem and the leader’s problem,
in turn, is affected by the outcome of the follower’s problem. To compare the two problems in terms of scope
of information, a follower makes decisions using only its local information while a leader does using the complete information including the follower’s possible reaction to the leader’s decision. This is an important feature which can contribute to address the complicated industrial situations which may be difficult to model
using other modeling methodologies. Therefore, BLPPs have been applied in extensive and diverse areas.
Nevertheless (9) deals with following weaknesses:
(I) Each level consists of ‘‘one’’ objective function while in real situations it is possible to have ‘‘more than
one’’.
(II) Input data of parameters are often stochastic or fuzzy in most real-world situations.
(III) Objective functions and constraints are often nonlinear (quadratic, geometric, etc.) rather than linear in
real-world situations.
3. Multi-objective chance constrained programming problem with a joint constraint [30]
A multi-objective chance constrained programming problem with a joint probability constraint can be stated as
n
X
ðkÞ
min zðkÞ ðxÞ ¼
cj xj ; k ¼ 1; 2; :::; K
j¼1
s:t:
pr
"
n
X
j¼1
xj P 0
1
a1j xj P b1 ;
n
X
a2j xj P b2 ; . . . ;
j¼1
j ¼ 1; 2; :::; n;
n
X
j¼1
#
amj xj P bm P 1 a;
ð10Þ
The issue of dominance between multiple elements of supply chains, such as production and distribution, needs to be addressed. The
illustrated mathematical model takes the distribution problem as the outer problem which is ultimate optimization problem and the
production operation is regarded as a subordinate optimization problem. An importance is placed on distribution since actual profits are
thought to be obtained through distribution inventory manipulation. However, there may be other types of bi-level formulation model
encapsulating various types of supply chain operations. For instance, some special industries involving difficult processes may pose their
outer problem as maintaining process operability and inner problem as distribution problem due to the critical importance of keeping their
process condition safe. It may be also possible that transportation problem can be an outer problem for industries where delivery needs
special care and corresponding high costs [17].
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
791
where bi’s are independent normal random variables with known means and variances. Constraint is a joint
probabilistic constraint and 0 < a < 1 is a specified probability. We assume that the decision variables xj’s are
deterministic.
Let the mean and standard deviation of the normal independent random variable bi be given by E(bi) and
r(bi) respectively. Sinha et al. [30] have shown equivalent deterministic model of probabilistic problem (10) can
be presented as
n
X
ðkÞ
cj xj ; k ¼ 1; 2; :::; K
min zðkÞ ðxÞ ¼
j¼1
s:t:
rffiffiffi
2
3bi
bi
p
exp
ð2y þ 1Þ;
P
2 i
2
3 b2i
m
Y
y i P 1 a;
ð11Þ
i¼1
n
X
aij xj bi rðbi Þ ¼ Eðbi Þ;
j¼1
0 6 y i 6 1; i ¼ 1; 2; :::; m;
xj P 0; j ¼ 1; 2; :::; n:
When bi’s are independent uniform random variables with known minimum and maximum, problem (10)
should be presented as
n
X
ðkÞ
cj xj ; k ¼ 1; 2; :::; K
min zðkÞ ðxÞ ¼
j¼1
s:t:
m
Y
i¼1
Pn
mini
maxi mini
j¼1 aij xj
ð12Þ
P 1 a;
xj P 0; j ¼ 1; 2; :::; n:
4. A multi-level nonlinear multi-objective decision-making under fuzziness [29]
Let xi 2 Rni (i = 1, 2) be a vector variables indicating the first decision level’s choice and the second decision
level’s choice ni P 1 (i = 1, 2).
Let F1 : Rn1 · Rn2 ! RN1 be the first level objective functions and F2 : Rn1 · Rn2 ! RN2 be the second level
objective functions.
Let the FLDM and SLDM have N1 and N2 objective functions, respectively. Let G be the set of feasible
choices {(x1, x2)}. So the BLN-MODM problem may be formulated as follows:
½1st level
Max F 1 ðx1 ; x2 Þ ¼ Max ðf11 ðx1 ; x2 Þ; . . . ; f1N 1 ðx1 ; x2 ÞÞ; where x2 solves;
x1
x1
½2nd level
ð13Þ
Max F 2 ðx1 ; x2 Þ ¼ Max ðf21 ðx1 ; x2 Þ; . . . ; f2N 2 ðx1 ; x2 ÞÞ
x2
x2
s:t: G ¼ fðx1 ; x2 Þjgi ðx1 ; x2 Þ 6 0; i ¼ 1; 2; . . . ; mg;
where G is the bi-level convex constraint set. F1, F2 are nonlinear and concave functions. The decision mechanism of BLN-MODM is that the FLDM and SLDM adopt the two-planer Stackelberg game.
Osman et al. [29] propose a solution method for solving above problem. For more information refer to [29].
5. Model formulation
Production and distribution models of an enterprise can be mathematically formulated as follows (see
Table 2 for the notation used).
792
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
Table 2
Notation for supply chain planning model
Indices
i
l
w
r
Product (1, . . . , I)
Plant (1, . . . , L)
Warehouse (1, . . . , W)
Market (1, . . . , R)
Constant parameters
ali
bli
cwi
ali
blwi
hwri
trwri
Mri
Pl
Q
0 6 c, k, p 6 1
Capacity coefficient of product i at plant l
Resource coefficient of product i at plant l
Resource coefficient of product i at warehouse w
Production cost coefficient for of product i at plant l
Transportation cost coefficient for product i from plant l to warehouse w
Inventory holding cost coefficient for product i at warehouse w for market r
Transportation cost coefficient for product i from warehouse w to market r
Demand of product i at market r (random variable)
Production capacity of plant l (random variable)
Resources available to all the plants for product i (random variable)
Probability
Variables
ZPC
ZDC
Z 0DC
Ylwi
Xwri
Objective function of a production part (cost)
Objective function of a distribution part (cost)
Objective function of a distribution part (capacity)
Production amount of i at plant l for warehouse w
Inventory of product i at warehouse w for market r
5.1. A production model
A production part of supply chains is typically subject to the following constraints:
(a) Production amounts from the plants should meet the levels required at the warehouses
L
X
Y lwi P
R
X
X wri
8w; i:
ð14Þ
r¼1
l¼1
(b) Production levels at the plants are limited by individual plant capacities
!
W
I X
X
P
ali Y lwi 6 P l ; ð8lÞ P 1 kl ð8lÞ2 :
ð15Þ
i¼1 w¼1
(c) Common used resources may be shared by all the plants
!
W
L X
X
bli Y lwi 6 Qi ; ð8iÞ P 1 ci ð8iÞ:
P
ð16Þ
l¼1 w¼1
An operating objective of production parts is to minimize their costs, which typically consists of its manufacturing cost and distribution cost between plants and warehouses
min
Z PC ¼
Y lwi
2
P
l¼1 w¼1 i¼1
P P
I
W
i¼1
L X
W X
I
X
w¼1 ali Y lwi
ali Y lwi þ
L X
W X
I
X
blwi Y lwi :
ð17Þ
l¼1 w¼1 i¼1
P P
I
W
6 P l ; 8l P 1 kl refer to joint distribution of P’s in other hand P
i¼1
w¼1 ali Y lwi 6 P l P 1 kl ; 8l [11]
refer to disjoint distribution of them, these two states are one at time.
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
793
Operation of the production part can thus be formulated as the following mathematical programming
problem:
min
Y lwi
Z PC ¼
ali Y lwi þ
l¼1 w¼1 i¼1
L
X
s:t:
L X
W X
I
X
Y lwi P
X wri
8w; i;
W
I X
X
L X
W
X
!
P 1 kl
ð8lÞ;
!
P 1 ci
ð8iÞ;
ali Y lwi 6 P l ; ð8lÞ
i¼1 w¼1
P
blwi Y lwi
l¼1 w¼1 i¼1
r¼1
l¼1
P
R
X
L X
W X
I
X
bli Y lwi 6 Qi ; ð8iÞ
l¼1 w¼1
X wri P 0
8w; r; i;
Y lwi P 0
8w; l; i:
ð18Þ
5.2. A distribution model
A distribution part is typically subject to the following constraints:
(a) Sums of individual warehouses’ holding should meet demands in markets
!
W
X
X wri P M ri ; ð8r; iÞ P 1 p ð8r; iÞ:
P
ð19Þ
w¼1
(b) The first objective function indicates capacity of all warehouses
min
X wri
Z 0DC ¼
W X
R X
I
X
w¼1 r¼1
ð20Þ
cwi X wri :
i¼1
(c) The following indicates an objective function for the distribution part of the supply chain:
min
X wri
Z DC ¼
I
R X
W X
X
w¼1 r¼1
hwri X wri þ
I
R X
W X
X
w¼1 r¼1
i¼1
trwri X wri ;
ð21Þ
i¼1
where the first term denotes inventory holding cost including material handling cost at warehouses and the
second indicates transportation cost from warehouses to markets:
Operation of the inventory part can thus be formulated as the following mathematical programming
problem:
min
X wri
min
X wri
s:t:
Z DC ¼
W X
R X
I
X
w¼1 r¼1
Z 0DC ¼
P
W
X
W X
R X
I
X
w¼1 r¼1
i¼1
W X
R X
I
X
w¼1 r¼1
hwri X wri þ
trwri X wri
i¼1
cwi X wri
i¼1
X wri P M ri ; ð8r; iÞ
w¼1
X wri P 0
8w; r; i;
Y lwi P 0
8w; l; i:
!
ð22Þ
P1p
ð8r; iÞ;
794
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
Note that the decisions of the distribution part are based on those of the production part: for example,
inventory policies are made using the outcome of production decisions. Similarly, decisions on the production
part are affected by parameters which are decided by the distribution part: for example, production levels are
decided from given information regarding the inventory conditions. Therefore the overall supply chain planning model can be posed as the following bi-level optimization problem:
min
X wri
min
X wri
s:t:
Z DC ¼
I
R X
W X
X
w¼1 r¼1
Z 0DC ¼
P
Y lwi
s:t:
Z PC ¼
cwi X wri
X wri P M ri ; ð8r; iÞ
L X
W X
I
X
!
ali Y lwi þ
l¼1 w¼1 i¼1
L
X
Y lwi P
X wri
P1p
L X
W X
I
X
blwi Y lwi
l¼1 w¼1 i¼1
ð23Þ
8w; i;
I X
W
X
W
L X
X
!
P 1 kl
ð8lÞ;
!
P 1 ci
ð8iÞ;
ali Y lwi 6 P l ; ð8lÞ
i¼1 w¼1
P
ð8r; iÞ;
r¼1
l¼1
P
R
X
trwri X wri
i¼1
i¼1
w¼1
min
I
R X
W X
X
w¼1 r¼1
i¼1
I
R X
W X
X
w¼1 r¼1
W
X
hwri X wri þ
bli Y lwi 6 Qi ; ð8iÞ
l¼1 w¼1
X wri P 0
8w; r; i;
Y lwi P 0
8w; l; i;
where the inner problem corresponds to the production optimization problem and the outer problem to the
distribution optimization problem.
6. Solution method
In this section we will present a solution method in flowchart frame work for solving model (23) (see Fig. 2).
Transform model (23) to a deterministic program
according to sec. 3.
Solve deterministic program according to sec. 4
Adjust
tolerances of
objective
functions and
decision
variables
No
If solution is acceptable for
all DM’s
yes
End
Fig. 2. Solution method.
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
795
7. Numerical examples
Example 1. We have changed Eq. (9) of Section 2 such that
min
Z DC ¼ 15X A þ 13X B þ 3Y 1A þ 2Y 1B þ 3:5Y 2A þ 2:5Y 2B
min
Z 0DC ¼ 3X A þ 2X B
X A ;X B
X A ;X B
s:t:
min
P ðX A P hA Þ P 1 p; P ðX B P hB Þ P 1 p;
Z PC ¼ 1:5X A þ 2X B þ 7Y 1A þ 3Y 1B þ 10Y 2A þ 6Y 2B
s:t:
P ðY 1A þ Y 1B þ Y 2A þ Y 2B 6 QÞ P 1 k;
Y 1A ;Y 1B ;Y 2A ;Y 2B
ð24Þ
P ðY 1A þ Y 1B 6 P 1 ; Y 2A þ Y 2B 6 P 2 Þ P 1 c;
Y 1A þ Y 2A P X A ;
Y 1B þ Y 2B P X B ;
X A ; X B ; Y 1A ; Y 1B ; Y 2A ; Y 2B P 0;
where
p ¼ k ¼ c ¼ 0:05;
hA N ð400; 50Þ;
Z 0:05 2;
hB N ð500; 60Þ;
Q U ð4000; 6000Þ;
P 1 U ð1000; 1500ÞhA U ð1500; 2000Þ:
According to the procedure in Section 4, individual best solutions and individual worst solutions for above
models are
This data can then be formulated as the following membership functions of fuzzy set theory [31]:
lZ DC ½Z DC ¼
lZ 0 ½Z 0DC ¼
DC
lZ PC ½Z PC ¼
8
>
< 1;
43202:7Z DC
> 25092:7
:
0;
8
1;
>
<
6112:5Z 0DC
>
: 3372:5
0;
8
>
< 1;
26906:2Z PC
> 19696:2
:
0;
;
Z DC 6 18 110;
18 110 6 Z DC 6 43202:7;
43202:7 6 Z DC ;
Z 0DC 6 2740;
; 2740 6 Z 0DC 6 6112:5;
6112:5 6
Z 0DC ;
Z PC 6 7210;
; 7210 6 Z PC 6 26906:2;
26906:2 6 Z PC :
ð25Þ
796
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
According to [29] and by using Lingo-8, solutions of FLDM and SLDM are
FLDM problem solution
Objective value:
Variable
LANDADC
XA
XB
Y1A
Y1B
Y2A
Y2B
1.000000
Value
1.000000
500.0000
620.0000
0.000000
500.0000
0.000000
620.0000
SLDM problem solution
Objective value:
Variable
LANDAPC
XA
XB
Y1A
Y1B
Y2A
Y2B
1.000000
Value
1.000000
500.0000
620.0000
0.000000
500.0000
0.000000
620.0000
Z DC ¼ 18 110;
Z 0DC ¼ 2740;
Z PC ¼ 7210:
Solution is complete and it is not necessary to continue.
Example 2. Consider problem (24) with following changes:
min
Z DC ¼ 15X A þ 13X B þ 3Y 1A þ 2Y 1B þ 3:5Y 2A þ 2:5Y 2B
min
Z 0DC ¼ 3X A þ 2X B
X A ;X B
X A ;X B
s:t:
min
Y 1A ;Y 1B ;Y 2A ;Y 2B
s:t:
P ðX A P hA Þ P 1 p; P ðX B P hB Þ P 1 p;
Z PC ¼ 1:5X A þ 2X B þ 7Y 1A þ 3Y 1B þ 10Y 2A þ 6Y 2B
P ðY 1A þ Y 1B þ Y 2A þ Y 2B 6 QÞ P 1 k;
ð26Þ
P ð5Y 1A þ 3Y 1B 6 P 1 ; 2Y 2A þ 7Y 2B 6 P 2 Þ P 1 c;
Y 1A þ Y 2A P X A ;
Y 1B þ Y 2B P X B ;
X A ; X B ; Y 1A ; Y 1B ; Y 2A ; Y 2B P 0:
According to the procedure in Section 4, individual best solutions and individual worst solutions for above
models are
797
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
This data also can then be formulated as the following membership functions of fuzzy set theory [31]:
8
Z DC 6 18 160;
< 1;
DC
lZ DC ½Z DC ¼ 72 715Z
;
18
160 6 Z DC 6 72 715;
: 54 555
0;
72 715 6 Z DC ;
8
1;
Z 0DC 6 2740;
>
<
lZ DC ½Z 0DC ¼
11 680Z 0DC
8940
2740 6 Z 0DC 6 11 680;
11 680 6 Z 0DC ;
Z PC 6 7311:5;
450Z PC
lZ PC ½Z PC ¼ 4740138:5
; 7311:5 6 Z PC 6 47 450;
:
0;
47 450 6 Z PC :
ð27Þ
;
>
:
0;
8
1;
<
According to [29] and by using Lingo-8, solutions of FLDM and SLDM are
FLDM problem solution
Objective value:
Variable
LANDADC
XA
XB
Y1A
Y1B
Y2A
Y2B
0.9306607
Value
0.9306582
500.0000
620.0000
1094.782
0.000000
0.000000
1239.437
SLDM problem solution
Objective value:
Variable
LANDAPC
XA
XB
Y1A
Y1B
Y2A
Y2B
0.814852
Value
0.8148886
500.0000
620.0000
0.000000
2089.142
0.000000
1080.695
798
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
Z DC ¼ 21 943;
Z 0DC ¼ 2740;
Z PC ¼ 14741:6:
As we noted before, two solutions are usually different because of nature between two levels objective functions. We assume the FLDM’s control decisions XA, XB with the tolerance 10,20, respectively.
With FLDM solution
With SLDM solution
ZDC
Z 0DC
ZPC
21 943
22 440
2740
2740
17090.1
14741.6
According to [29] following problem should be solved:
Max d
X A ð500 10Þ
P d;
10
ð500 þ 10Þ X A
P d;
10
X B ð620 20Þ
P d;
20
ð620 þ 20Þ X B
P d;
20
22 440 ð15X A þ 13X B þ 3Y 1A þ 2Y 1B þ 3:5Y 2A þ 2:5Y 2B Þ
P d;
22 440 21 943
17090:1 ð1:5X A þ 2X B þ 7Y 1A þ 3Y 1B þ 10Y 2A þ 6Y 2B Þ
P d;
17090:1 14741:6
P ðX A P hA Þ P 1 p; P ðX B P hB Þ P 1 p;
P ðY 1A þ Y 1B þ Y 2A þ Y 2B 6 QÞ P 1 k;
P ðY 1A þ Y 1B 6 P 1 ; Y 2A þ Y 2B 6 P 2 Þ P 1 c;
Y 1A þ Y 2A P X A ;
Y 1B þ Y 2B P X B ;
X A ; X B ; Y 1A ; Y 1B ; Y 2A ; Y 2B P 0:
According to [29] and by using Lingo-8, solution is
Objective value
Variable
DELTA
XA
XB
Y1A
Y1B
Y2A
Y2B
Z DC ¼ 20856:4;
1.000000
Value
1.000000
500.0000
620.0000
104.1667
395.8333
483.4933
1000.000
Z 0DC ¼ 2740;
Z PC ¼ 14741:6:
Obtained solution is better than or equal to prior solution and we assume it is a compromise solution.
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
799
8. Conclusions
In this paper, we considered a ‘‘probabilistic bi-level linear multi-objective programming problem’’ and its
application in enterprise-wide supply chain planning problem where (1) market demand, (2) production capacity of each plant and (3) resource available to all plants for each product were random variables and the constraints included joint and disjoint probability distributions. This probabilistic model was first converted into
an equivalent deterministic model in each level, to which fuzzy programming technique was applied to solve
the multi-objective nonlinear programming problem to obtain a compromise solution.
Two simplified example was used to illustrate the process of interaction. Method can be applied to explicit
situations by changing certain assumptions to solve the specific problem properly. Although the optimal solution is rarely possible, a compromise solution, which is acceptable for all parties with conflicting objectives,
provides conflict resolution.
For future research we will propose following topics:
(1) Robust optimization in BLP problems.
(2) Stochastic BLP with all random coefficients.
(3) Stochastic BLP with multi follower in level 2.
References
[1] O.E. Emam, A fuzzy approach for bi-level integer non-linear programming problem, Appl. Math. Comput. (2005).
[2] Benoıˆt Colson, Patrice Marcotte, Gilles Savard, Bilevel Programming: A Survey, A Quarterly Journal of Operations Research,
Springer-Verlag, 2005.
[3] R.E. Bellman, L.A. Zadeh, Decision making in a fuzzy environment, Manage. Sci. 17 (1970) 141–164.
[4] A. Charnes, W.W. Cooper, Chance constrained programming, Manage. Sci. 6 (1959) 73–79.
[5] B. Contini, A stochastic approach to goal programming, Oper. Res. 16 (1978) 576–586.
[6] E.L. Hanan, On fuzzy goal programming, Decision Sci. 12 (1981) 522–531.
[7] V.V. Kolbin, Stochastic Programming, D. Reidel Publishing Company, Dordrecht, 1977.
[8] H. Leberling, On finding compromise solution in multi-criteria problems using the min-operator, Fuzzy Sets Syst. 6 (1981) 105–118.
[9] J.P. Leclercq, Stochastic programming: an interactive multi-criteria approach, Eur. J. Oper. Res. 10 (1982) 33–41.
[10] R. Narasimhan, Goal programming in a fuzzy environment, Decision Sci. 11 (1980) 325–336.
[11] F.B. Abdelaziz, B. Aouni, R. El Fayedh, Multi-objective stochastic programming for portfolio selection, Eur. J. Oper. Res. (2005)
4–6.
[12] J.K. Sengupta, Stochastic Programming: Methods and Applications, North-Holland, Amsterdam, 1972.
[13] I.M. Stancu-Minasian, M.J. Wets, A research bibliography in stochastic programming 1955–1975, Oper. Res. 24 (1976) 1078–1119.
[14] R.S. Sullivan, J.A. Fitzsimmoms, A goal programming model for readiness and optimal deployment of resources, Socio-Economic
Planning Sci. 12 (1978) 215–220.
[15] J. Teghem Jr., D. Dufrance, M. Thauvoye, P. Kunch, Strange: an interactive method for multi-objective linear programming under
uncertainty, Eur. J. Oper. Res. 26 (1986) 5–82.
[16] H.J. Zimmermann, Fuzzy programming and linear programming with several objective functions, Fuzzy Sets Syst. 1 (1978) 45–55.
[17] Jun-Hyung Ryu, Vivek Dua, Efstratios N. Pistikopoulos, A bilevel programming framework for enterprise-wide process networks
under uncertainty, Comput. Chem. Eng. 28 (2004) 1121–1129.
[18] T. Backx, O. Bosgra, W. Marquardt, Towards intentional dynamics in supply chain conscious process operation, in: Proceedings of
the Third International Conference on Foundations of Computer-aided Process Operations, 1998.
[19] P. Tsiakis, N. Shah, C.C. Pantelides, Design of multiechelon supply chain networks under demand uncertainty, Ind. Eng. Chem. Res.
40 (2001) 3585–3604.
[20] A. Gupta, C.D. Maranas, A two-stage modeling and solution framework for multisite midterm planning under uncertainty, Ind. Eng.
Chem. Res. 39 (2000) 3799–3813.
[21] A. Gupta, C.D. Maranas, C.M. McDonald, Mid-term supply chain planning under demand uncertainty: customer demand
satisfaction and inventory management, Comput. Chem. Eng. 24 (2000) 2613–2621.
[22] C.M. McDonald, I.A. Karimi, Planning and scheduling of parallel semicontinuous processes. Part 1. Production planning, Ind. Eng.
Chem. Res. 36 (1997) 2691–2700.
[23] H.L. Lee, V. Padmanabhan, S. Whang, The bullwhip effect in supply-chains, Sloan Manage. Rev. 38 (1997) 93–102.
[24] J. Gjerdrum, N. Shah, L.G. Papageorgiou, Transfer prices for multienterprise supply chain optimization, Ind. Eng. Chem. Res. 40
(2001) 1650–1660.
[25] Z. Zhou, S. Cheng, B. Hua, Supply chain optimization of continuous process industries with sustainability considerations, Comput.
Chem. Eng. 24 (2000) 1151–1158.
800
E. Roghanian et al. / Applied Mathematics and Computation 188 (2007) 786–800
[26] E. Perea-Lopez, I.E. Grossmann, E. Ydstie, T. Tahmassebi, Dynamic modeling and classical control theory for supply chain
management, Comput. Chem. Eng. 24 (2000) 1143–1149.
[27] S. Bose, J.F. Pekny, A model predictive framework for planning and scheduling problem: a case study of consumer goods supply
chain, Comput. Chem. Eng. 24 (2000) 329–335.
[28] M.E. Flores, D.E. Rivera, V. Smith-Daniels, Supply chain management using model predictive control, in: AIChE Annual Meeting,
Los Angeles, 2000.
[29] M.S. Osman, M.A. Abo-Sinna, A.H. Amer, A multi-level non-linear multi-objective decision-making under fuzziness, Appl. Math.
Comput. 153 (2004) 239–252.
[30] S.B. Sinha, Suwarna Hulsurkar, M.P. Biswal, Fuzzy programming approach to multi-objective stochastic programming problems
when bi’s follow joint normal distribution, Fuzzy Sets Syst. 109 (2000) 91–96.
[31] M. Sakawa, Fuzzy Sets and Interactive Multi-Objective Optimization, Plenum Press, New York, 1993.