ARTICLE IN PRESS

Int. J. Production Economics 98 (2005) 328–341

www.elsevier.com/locate/ijpe

Applying possibilistic linear programming to aggregate

production planning
Reay-Chen Wanga,, Tien-Fu Liangb
a
Department of Industrial Management, Tung Nan Institute of Technology, 152 Pei Shen Rd., Sec. 3, Taipei 222, Taiwan
b
Department of Industrial Management, Hsiuping Institute of Technology, 11 Gungye Rd., Dali City, Taichung, 412, Taiwan
Received 24 October 2003; accepted 30 September 2004
Available online 24 November 2004

Abstract

This work presents a novel interactive possibilistic linear programming (PLP) approach for solving the multi-
product aggregate production planning (APP) problem with imprecise forecast demand, related operating costs, and
capacity. The proposed approach attempts to minimize total costs with reference to inventory levels, labor levels,
overtime, subcontracting and backordering levels, and labor, machine and warehouse capacity. The proposed
approach uses the strategy of simultaneously minimizing the most possible value of the imprecise total costs,
maximizing the possibility of obtaining lower total costs, and minimizing the risk of obtaining higher total costs.
An industrial case demonstrates the feasibility of applying the proposed approach to real APP decision problems.
Consequently, the proposed PLP approach yields an efficient APP compromise solution and overall degree of
decision maker (DM) satisfaction with determined goal values. Particularly, several significant management
implications and characteristics of the proposed PLP approach that distinguish it from the other APP decision
models are presented.
r 2004 Elsevier B.V. All rights reserved.

Keywords: Aggregate production planning; Possibilistic linear programming; Fuzzy sets theory

1. Introduction intermediate future, often from 3 to 18 months

Aggregate production planning (APP) deter-
mines the best way to meet forecast demand in the
Corresponding author. Tel.: +886 2 86625930/5931 Ext.
918; fax: +886 2 5932.
E-mail addresses: rcwang@mail.tnit.edu.tw (R.-C. Wang),
farmer@mail.hit.edu.tw (T.-F. Liang).
ahead, by adjusting regular and overtime produc-
tion rates, inventory levels, labor levels, subcon-
tracting and backordering rates, and other
controllable variables. APP has attracted consid-
erable interest from both practitioners and aca-
demics (Shi and Haase, 1996). Since Holt et al.
(1955) proposed the HMMS rule in 1955,
researchers have developed numerous models to

0925-5273/$ - see front matter r 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2004.09.011
 ARTICLE IN PRESS
R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341
help to solve APP problems, each with its own
advantages and disadvantages. HMMS model is
one of the extensively adopted frameworks for
solving APP decision problems. Its linear decision
rules attempt to specify an optimum production
rate and labor levels that minimize the total costs
of regular payoff, overtime, hiring, layoffs, and
inventory through a series of quadratic cost
curves. Saad (1982) classifies all conventional
decision models for solving APP problems into
six categories—(1) linear programming (LP)
(Charnes and Cooper, 1961), (2) linear decision
rule (Holt et al., 1955), (3) transportation method
(Bowman, 1956), (4) management coefficient
approach (Bowman, 1963), (5) search decision
rule (Taubert, 1968), and (6) simulation (Jones,
1967). Subsequent APP decision-related works
include Shi and Haase (1996), Baykasoglu (2001),
Buxey (2003), Dobos (2003), Gnoni et al. (2003),
Stephen et al. (2003), and Jolayemi and Olorunni-
wo (2004). However, the goals and model inputs
when any of these APP models are used generally
are assumed to be deterministic/crisp.
In real-world APP problems, input data or
related parameters, such as market demand,
available resources and capacity, and relevant
operating costs, frequently are imprecise/fuzzy
owing to some information being incomplete or
unobtainable. Traditional mathematical program-
ming techniques clearly cannot solve all fuzzy
programming problems. Zimmermann (1976) first
introduced fuzzy set theory into conventional LP
problems in 1976. His study considered LP
problems with fuzzy goal and constraints. Follow-
ing the fuzzy decision-making method of Bellman
and Zadeh (1970), the same study confirmed the
existence of an equivalent single-goal LP problem.
Since then, fuzzy mathematical programming
has developed into several fuzzy optimization
methods for solving APP problems. Fung et al.
(2003) presented a fuzzy multi-product aggregate
production planning (FMAPP) model to cater to
different scenarios under various decision-making
preferences by applying integrated parametric
programming, best balance and interactive meth-
ods. This model can also effectively enhance
the ability of an aggregate plan to provide feasi-
ble disaggregate plans under varying circum-
329
stances with fuzzy demands and fuzzy capacities.
Wang and Liang (2004a) more recently developed
a fuzzy multi-objective linear programming
model with the piecewise linear membership func-
tion to solve multi-product APP decision problems
in a fuzzy environment. The model can yield an
efficient compromise solution and the decision
maker’s overall levels of satisfaction. Additional
fuzzy APP problem solving studies include Lee
(1990), Wang and Fang (1997), Tang et al. (2000),
Wang and Fang (2001), and Tang et al. (2003).
Moreover, Zadeh (1978) presented the theory of
possibility, which is related to the theory of fuzzy
sets by defining the concept of a possibility
distribution as a fuzzy restriction, which acts as
an elastic constraint on the values that can be
assigned to a variable. Zadeh (1978) demonstrated
that the importance of the theory of possibility
stems from the fact that much of the information
on which human decisions is based on is possibi-
listic rather than probabilistic in nature. Buckley
(1988) formulated a mathematical programming
problem in which all parameters may be fuzzy
variables specified by their possibility distribution,
and moreover illustrated this problem using the
possibilistic linear programming (PLP) approach.
Furthermore, Buckley (1989) described a proce-
dure for solving existing PLP problems in a
standard form with no equality constraints. Lai
and Hwang (1992b) developed an auxiliary multi-
ple objective linear programming (MOLP) model
for solving a PLP problem with imprecise objective
and/or constraint coefficients. Tang et al. (2001)
designed two types of PLP with general possibi-
listic distribution, including LP problems
with general possibilistic resources (GRPLP) and
general possibilistic objective coefficients
(GOPLP). Hsu and Wang (2001) applied Lai and
Hwang’s PLP approach to manage production
planning decision problems in an assemble-to-
order environment. Other studies on PLP pro-
blems include Inuiguchi and Sakawa (1996),
Hussein (1998), Tanaka et al. (2000), and Jensen
and Maturana (2002).
This work develops a novel interactive PLP
approach for solving the multi-product APP
decision problem with imprecise forecast demand,
related operating costs and capacity. The proposed
 ARTICLE IN PRESS

330 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341

approach attempts to minimize total costs in production rates, inventory levels, labor levels,
terms of inventory levels, labor levels, overtime, subcontracting and backordering rates, and other
subcontracting and backordering levels, and labor, controllable variables.
machine and warehouse capacity. The proposed The following notation is used here.
approach simultaneously minimizes the most
possible value of the imprecise total costs, max- N types of products
imizes the possibility of obtaining lower total T planning horizon
costs, and minimizes the risk of obtaining higher z~ total costs
total costs. D~ nt forecast demand of nth product in
The rest of this paper is organized as follows: period t (units)
Section 2 describes the problem, details the assum- a~ nt regular time production cost per unit of
ptions, and formulates the problem. Section 3 nth product in period t ($/unit)
then develops the interactive PLP approach and Qnt regular time production of nth product
algorithm for solving PM decision problems. in period t (units)
Subsequently, Section 4 presents an industrial case b~nt overtime production cost per unit of nth
designed to implement the feasibility of applying product in period t ($/unit)
the proposed PLP approach to real APP decision Ont overtime production of nth product in
problems. Next, Section 5 discusses the results and period t (units)
findings for the practical application of the c~nt subcontracting cost per unit of nth
proposed PLP approach. Conclusions are finally product in period t ($/unit)
drawn in Section 6. Snt subcontracting volume of nth product in
period t (units)
d~nt inventory carrying cost per unit of nth
2. Problem formulation product in period t ($/unit)
I nt inventory level in period t of nth product
2.1. Problem description and notation (units)
e~nt backorder cost per unit of nth product in
The multi-product APP decision problem ex- period t ($/unit)
amined here can be described as follows. Assume Bnt backorder level of nth product in period
that a company manufactures N types of products t (units)
to fulfill market demand over planning horizon T. k~t cost to hire one worker in period t ($/
Generally, the environmental coefficients and man-hour)
related parameters are uncertain in a medium time Ht workers hired in period t (man-hour)
horizon. Therefore, the forecast demand, related ~t
m cost to layoff one worker in period t ($/
operating costs, and labor and machine capacity man-hour)
are imprecise over the planning horizon. Assigning Ft workers laid off inperiod t (man-hour)
a set of crisp values for the environmental int hours of labor per unit of nth product in
coefficients and related parameters is inappropri- period t (man-hour/unit)
ate for dealing with such ambiguous APP decision r~nt hours of machine usage per unit of nth
problems. Fortunately, possibility distribution product in period t (machine-hour/unit)
provides an effectual alternative for proceeding vnt warehouse spaces per unit of nth
with inherent ambiguous phenomena in determin- product in period t (ft2 /unit)
ing environmental coefficients and related para- ~ t max
W maximum labor level available in period
meters (Zadeh, 1978; Inuiguchi and Sakawa, 1996; t (man-hour)
Hsu and Wang, 2001). This APP problem focuses ~ t max
M maximum machine capacity available in
on developing an interactive PLP approach to period t (machine-hour)
determine the optimum aggregate plan for meeting V t max maximum warehouse space available in
forecast demand by adjusting regular and overtime period t ðft2 Þ:
 ARTICLE IN PRESS

R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341 331

2.2. Possibilistic linear programming model
2.2.1. Objective function
Most practical decisions made to solve APP
problems usually consider total costs. The pro-
posed PLP model selected total costs as objective
function, after reviewing the literature and con-
sidering practical situations (Masud and Hwang,
1980; Saad, 1982; Wang and Fang, 2001). The
total costs are the sum of the production costs and
the costs of changes in labor levels over the
planning horizon T. The related cost coefficients in
the objective function frequently are imprecise in
nature because some information is incomplete or
where D~ nt denotes the imprecise forecast demand
of the nth product in period t. In real-world
APP decision problems, the forecast demand D~ nt
cannot be obtained precisely in a dynamic market.
The sum of regular and overtime production,
inventory levels, and subcontracting and back-
order levels essentially should equal the market
demand, as in Eq. (2). Demand over a particular
period can be either met or backordered, but a
backorder must be fulfilled in the subsequent
period.
Constraints on labor levels:
X
N

unobtainable in a medium time horizon. Accord- int1 ðQnt1 þ Ont1 Þ þ H t  F t

n¼1
ingly, the objective function of the proposed model
is as follows: X
N
 int ðQnt þ Ont Þ ¼ 0 8t; ð3Þ
X
N X
T n¼1
Min z~ ¼ ða~ nt Qnt þ b~nt Ont þ c~nt S nt
n¼1 t¼1
X
N
X
T ~ t max
int ðQnt þ Ont ÞpW 8t; (4)
þ d~ nt I nt þ e~nt Bnt Þ þ ðk~t H t þ m
~ t F t Þ; n¼1
t¼1
ð1Þ where Eq. (3) represents a set of constraints in
which the labor levels in period t equal the labor
where a~ nt ; b~nt ; c~nt ; d~nt ; e~nt ; k~t ; and m ~ t are imprecise levels in period t  1 plus new hires less layoffs in
coefficients with triangular possibility distribu- period t. Actual labor levels cannot exceed the
PN PT
tions. The terms n¼1 t¼1 ða~ nt Qnt þ b~nt Ont þ maximum available labor levels in each period, as
~
c~nt Snt þ d nt I nt þ e~nt Bnt Þ are used to calculate total in Eq. (4). Maximum available labor levels are
production costs. The total production costs imprecise, owing to uncertain labor market de-
include five components, as following: mand and supply.
PN PT Constraints on machine capacity and warehouse
n¼1 a~ Q
nt nt are the regular time production
Pt¼1
N PT
space:
costs, n¼1 b~nt Ont are the overtime produc-
PN t¼1PT X
N
tion costs, n¼1 t¼1 c~nt Snt are the subcontracting ~ t max
P PT ~ r~nt ðQnt þ Ont ÞpM 8t; (5)
costs, N n¼1 d nt I nt are the inventory carrying
Pt¼1N PT
n¼1
costs, and n¼1 e~nt Bnt are the backorder
PTt¼1 ~
costs. The term t¼1 ð k tH t þ m ~ t F t Þ specifies the X
N
costs of changes in labor levels, including the costs vnt I nt pV t max 8t; (6)
to hire and layoff workers. n¼1

where r~nt and M ~ nt max denote the imprecise

hours of machine usage per unit of the nth
2.2.2. Constraints product and the imprecise maximum avail-
Constraints on carrying inventory: able machine capacity in period t, respectively.
Eqs. (5) and (6) represent the limits of actual
I nt1  Bnt1 þ Qnt þ Ont þ S nt  I nt þ Bnt ¼ D~ nt machine and warehouse capacity in each
8n; 8t; ð2Þ period.
 ARTICLE IN PRESS

332 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341

Non-negativity constraints on decision variables: distributions, as follows:

Qnt ; Ont ; S nt ; I nt ; Bnt ; H t ; F t X0 8n; 8t: (7) a~ nt ¼ ðapnt ; am o
nt ; ant Þ 8n; 8t;
b~nt ¼ ðbpnt ; bm o
nt ; bnt Þ 8n; 8t;
3. Model development c~nt ¼ ðcpnt ; cm o
nt ; cnt Þ 8n; 8t;

3.1. Model the imprecise data with triangular d~nt ¼ ðd pnt ; d m o

nt ; d nt Þ 8n; 8t;
possibility distribution e~nt ¼ ðepnt ; em o
nt ; ent Þ 8n; 8t;

k~t ¼ ðkpt ; km o
t ; kt Þ 8t;
This work assumes the decision maker (DM) to
have already adopted the pattern of triangular ~ t ¼ ðmpt ; mm
m o
t ; mt Þ 8t;
possibility distribution for all imprecise coeffi- r~nt ¼ ðrpnt ; rm o
nt ; rnt Þ 8n; 8t;
cients. The possibility distribution can be stated
as the degree of occurrence of an event with D~ nt ¼ ðDpnt ; Dm ; Do Þ 8n; 8t;
nt nt
imprecise data. Fig. 1 presents the triangu- ~ t max ¼ ðW pt max ; W m ; W o Þ 8t;
W t max t max
lar possibility distribution of imprecise number ~ p m o
M t max ¼ ðM t max ; M t max ; M t max Þ 8t:
a~ nt ¼ ðapnt ; am o
nt ; ant Þ:
In practice, a DM can construct the triangular
possibility distribution of a~ nt based on the three 3.2. An auxiliary multiple objective linear
prominent data, as follows. programming model

1. The most pessimistic value ðapnt Þ that has a very
low likelihood of belonging to the set of
available values (possibility degree ¼ 0 if nor-
malized).
2. The most possible value ðam nt Þ that definitely
belongs to the set of available values (possibility
degree ¼ 1 if normalized).
3. The most optimistic value ðaont Þ that has a very
low likelihood of belonging to the set of
available values (possibility degree ¼ 0 if nor-
malized).
The imprecise data for the previous PLP model
can thus be modeled with triangular possibility
1 minimizing zp ; zm ; and zo : That is, the proposed
nt
3.2.1. Strategy for solving the imprecise objective
function
The imprecise objective function of the PLP
model in the previous section has a triangular
possibility distribution. Geometrically, this impre-
cise objective is fully defined by three prominent
points ðzp ; 0Þ; ðzm ; 1Þ; and ðzo ; 0Þ: The imprecise
objective can be minimized by pushing the three
prominent points towards the left. Because of the
vertical coordinates of the prominent points being
fixed at either 1 or 0, the three horizontal
coordinates are the only considerations. Conse-
quently, solving the imprecise objective requires
minimizing zp ; zm ; and zo simultaneously. Using Lai
and Hwang’s (1992b) approach, the approach
developed here minimizes zm ; maximizes ðzm  zp Þ;
and minimizes ðzo  zm Þ; rather than simultaneously
approach simultaneously involves minimizing the

most possible value of the imprecise total costs, zm ;

a

maximizing the possibility of obtaining lower total

costs, ðzm  zp Þ; and minimizing the risk of obtain-
ing higher total costs, ðzo  zm Þ: The last two
0 antp antm anto objectives actually are relative measures from zm ;
a~nt the most possible value of the imprecise total costs.
Fig. 2 presents the strategy for minimizing the
Fig. 1. The triangular possibility distribution of a~ nt : imprecise objective function.
 ARTICLE IN PRESS

R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341 333

In real APP decision problem, Eqs. (8)–(10) are

equivalent to simultaneously minimizing the most
1
possible value of total costs, maximizing the
possibility of obtaining lower total costs (region I
of the possibility distribution in Fig. 2), and

z

~
~ A minimizing the risk of obtaining higher total costs
B
(region II of the possibility distribution in Fig. 2).
I II
3.2.2. Imprecise technological coefficients and
0 zp zm zo available resources
~z
Recalling Eq. (2) of the PLP model, consider the
Fig. 2. The strategy to minimize the total costs. situation in which available resource (the right-
hand side), D~ nt ; is imprecise and has triangular
possibility distributions with the most possible and
As indicated in Fig. 2, possibility distribution B~ least possible values. In real-world APP decision
is preferred to possibility distribution A: ~ This
problems, the DM can estimate a possible interval
result is presented below for three new crisp for imprecise demand based on experience and
objective functions. knowledge. The main problem is obtaining a crisp
Min z1 ¼ zm representative number for the imprecise demand.
This work adopts the weighted average method
N X
X T
½am m m proposed by Lai and Hwang (1992b) to convert
¼ nt Qnt þ bnt Ont þ cnt S nt
n¼1 t¼1 D~ nt into a crisp number. If the minimum accep-
þ dm m table possibility, b; is given, then the auxiliary crisp
nt I nt þ ent Bnt 
equality constraints can be presented as follows:
X
T
þ ðkm m
t H t þ mt F t Þ; ð8Þ I nt1  Bnt1 þ Qnt þ Ont þ S nt  I nt þ Bnt
t¼1
¼ w1 Dpnt;b þ w2 Dm o
nt;b þ w3 Dnt;b 8n; 8t; ð11Þ
Max z2 ¼ ðzm  zp Þ
where w1 þ w2 þ w3 ¼ 1; w1 ; w2 ; and w3 denote the
N X
X T
weights of the most pessimistic, most possible and
¼ ½ðam
nt  apnt ÞQnt þ ðbm
nt  bpnt ÞOnt
n¼1 t¼1
most optimistic value of the imprecise demand,
respectively. The weights w1 ; w2 ; and w3 can be
þ ðcm
nt  cpnt ÞSnt þ ðd m p
nt  d nt ÞI nt determined subjectively by the experience and
þ ðem
nt  epnt ÞBnt  knowledge of the DM.
XT
p
Similarly, if the minimum acceptable possibility,
þ ½ðkm p m
t  k t ÞH t þ ðmt  mt ÞF t ; b; is given, then the auxiliary crisp inequality
t¼1 constraints on Eq. (4) can be presented as follows:
ð9Þ
X
N
o m
Min z3 ¼ ðz  z Þ int ðQnt þ Ont Þ
n¼1
N X
X T
¼ ½ðaont  am o m
nt ÞQnt þ ðbnt  bnt ÞOnt pw1 W pt max;b þ w2 W m o
t max;b þ w3 W t max;b 8t:
n¼1 t¼1
o m
ð12Þ
þ ðcont  cm
nt ÞS nt þ ðd nt  d nt ÞI nt
þ ðeont  em To solve the imprecise right-hand sides in
nt ÞBnt 
Eqs. (11) and (12), this work applies the concept
XT
of the most likely values proposed by Lai and
þ ½ðkot  km o m
t ÞH t þ ðmt  mt ÞF t :
t¼1
Hwang’s (1992b) approach, assuming w2 ¼ 46 and
ð10Þ w1 ¼ w3 ¼ 16: The reason for using the most likely
 ARTICLE IN PRESS

334 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341

8
values here is that the most possible values usually >
> 1; z2 ozPIS
2 ;
are the most important ones, and thus should be >
< z2  zNIS
2 pz2 pz2 ;
2
assigned more weights. f 2 ðz21 Þ ¼ ; zNIS PIS
(18)
>
> z PIS
 z NIS
Moreover, to solve Eq. (5) with imprecise >
: 2 2

technological coefficient ð~rnt Þ and available re- 0; z2 4zNIS

2 ;
sources M ~ t max ; the approach proposed here
8
converted these imprecise inequality constraints >
> 1; z3 ozPIS
3 ;
>
< zNIS  z
into a crisp one using the fuzzy ranking concept 3
3 pz3 pz3 ;
3
(Tanaka et al., 1984; Rimik and Rimanek, 1985; f 3 ðz3 Þ ¼ ; zPIS NIS
(19)
>
>
NIS
z3  z3PIS
Lai and Hwang, 1992b). Consequently, the aux- >
:
iliary inequality constraints on Eq. (5) can be 0; z3 4zNIS
3 :
presented as follows: Each linear membership function can be deter-
X N
mined by asking the DM to specify the imprecise
rpnt;b ðQnt þ Ont ÞpM pt max;b 8t; (13)
n¼1
objective value interval. Figs. 3 and 4 show the
graphs of linear membership functions for Eqs.
X
N
(17)–(19).
rm m
nt;b ðQnt þ Ont ÞpM t max;b 8t; (14)
n¼1
Using the fuzzy decision-making of Bellman and
Zadeh (1970) and Zimmermann’s (1978) fuzzy
X
N
programming method, the complete equivalent
ront;b ðQnt þ Ont ÞpM ot max;b 8t: (15)
n¼1
single-goal LP model for solving the APP decision
problem can be formulated as follows:
3.3. Solving the auxiliary MOLP problem
Max L
Furthermore, the auxiliary MOLP problem can s:t:
be converted into an equivalent single-goal LP Lpf i ðzi Þ; i ¼ 1; 2; 3;
problem using the fuzzy decision-making of Bell-
man and Zadeh (1970) and Zimmermann’s fuzzy I nt1  Bnt1 þ Qnt þ Ont þ S nt  I nt þ Bnt
programming method (1978). The positive ideal ¼ w1 Dpnt;b þ w2 Dm o
nt;b þ w3 Dnt;b 8n; 8t;
solutions (PIS) and negative ideal solutions (NIS) P
N
of the three objective functions can be specified as int1 ðQnt1 þ Ont1 Þ þ H t  F t
n¼1
follows, respectively (Hwang and Yoon, 1981; Lai
and Hwang, 1992b). P
N
 int ðQnt þ Ont Þ ¼ 0 8t;
n¼1
zPIS
1 ¼ Min z ; m
zNIS
1 ¼ Max z ; m
(16a)

zPIS
2 ¼ Max ðzm  zp Þ; zNIS
2 ¼ Min ðzm  zp Þ;
(16b)
f1(z1) (f3(z3))

zPIS
3 ¼ Min ðzo  zm Þ; zNIS
3 ¼ Max ðzo  zm Þ: 1
(16c)
For each objective function, the corresponding
linear membership function is defined by
8 PIS NIS
>
> 1; z1 ozPIS
1 ;
0 z1 z1
>
< zNIS  z PIS NIS
1 (z3 ) (z3 )
1 pz1 pz1 ;
1
f 1 ðz1 Þ ¼ ; zPIS NIS
(17)
>
> z NIS
 z PIS z1(z3)
>
: 1 1
0; z1 4zNIS
1 ; Fig. 3. Linear membership functions of z1 and z3 :
 ARTICLE IN PRESS

R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341 335

Step 4: Given the minimum acceptable possibi-

lity, b; convert the imprecise constraints into crisp
ones using the weighted average method or the
1 fuzzy ranking concept.
f2(z2)

Step 5: Specify the linear membership functions

for three new objective functions, and then convert
the auxiliary MOLP problem into an equivalent
LP model using the fuzzy decisions of Bellman and
Zadeh (1970) and Zimmermann’s (1978) fuzzy
0 z2NIS z2PIS
z2 programming method.
Step 6: Solve and modify the model interac-
Fig. 4. Linear membership function of z2 : tively. If the DM is not satisfied with the initial
solution, then the model must be modified until a
P
N satisfactory solution is found.
int ðQnt þ Ont Þpw1 W pt max;b
n¼1

þw2 W m o
t max;b þ w3 W t max;b 8t;
4. Model implementation
P
N
rpnt;b ðQnt þ Ont ÞpM pt max;b 8t; 4.1. Case description
n¼1

P
N
m Daya Technologies Corporation was used as a
rm
nt;b ðQnt þ Ont ÞpM t max;b 8t;
n¼1 case study to demonstrate the practicality of the
P
N proposed methodology (Wang and Liang, 2004b).
ront;b ðQnt þ Ont ÞpM ot max;b 8t; The Daya Technologies Corporation is the first
n¼1 ballscrew manufacturer worldwide to have
P
N achieved ISO 9001, ISO 14001, and OHSAS
vnt I nt pV t max 8t; 18001 certification, and moreover is the main
n¼1
global producer of super precision ballscrew,
0pLp1; linear bearing, and guideways. The APP decision
Qnt ; Ont ; Snt ; I nt ; Bnt ; H t ; F t X0 8n; 8t: problem for Daya’s ballscrew plant presented here
focuses on developing an interactive possibilistic
programming approach for minimizing total costs.
The planning horizon is 4 months long, including
3.4. Algorithm May, June, July, and August. The model includes
two types of standard ballscrew, namely the
The algorithm of the proposed PLP approach external recirculation type (product 1) and the
for solving the APP decision problem is as follows. internal recirculation type (product 2). According
Step 1: Formulate the PLP model for the multi- to the preliminary environmental information,
product APP decision problem. Tables 1–3 summarize the forecast demand,
Step 2: Model the imprecise coefficients ða~ nt ; b~nt ; related operating cost, and capacity data used in
c~nt ; d~nt ; e~nt ; k~t ; m
~ t ; r~nt Þ and right-hand sides ðD~ nt ; W
~ t;
the Daya case. The forecast demand, relevant
M ~ t Þ using triangular possibility distributions.
operating cost, and maximum labor and machine
Step 3: Develop three new crisp objective capacity are imprecise numbers with triangular
functions of the auxiliary MOLP problem that possibility distributions from period to period.
are equivalent simultaneously minimizing the most Other relevant data are as follows.
possible total cost value, maximizing the possibi-
lity of obtaining lower costs, and minimizing the 1. Initial inventory in period 1 is 400 units of
risk of obtaining higher costs. product 1 and 200 units of product. End
 ARTICLE IN PRESS

336 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341

inventory in period 4 is 300 units of product 1 Step 4: Formulate the auxiliary crisp constraints
and 200 units of product 2. using Eqs. (11)–(15) at b ¼ 0:5:
2. Initial labor level is 300 man-hours. The costs Step 5: The PIS of the three new objective
associated withhiring and layoff are ($8, $10, functions is ($300,000, $80,000, $20,000) and the
$11) and ($2.0, $2.5, $3.2) per worker per hour, NIS is ($600,000, $30,000, $50,000). The corre-
respectively. sponding linear membership function of the three
3. Hours of labor per unit for any periods are fixed new objective functions can be defined according
to 0.05 man-hours for product 1 and 0.07 man- to Eqs. (17)–(19). Furthermore, the complete
hours for product 2. Hours of machine usage equivalent single-goal LP model for the Daya case
per unit for each of the four planning periods can be formulated using the fuzzy decisions of
are (0.09, 0.10, 0.11) machine-hours for product Bellman and Zadeh (1970) and Zimmermann’s
1 and (0.07, 0.08, 0.09) machine-hours for fuzzy programming method (1978).
product 2. Warehouse spaces required per unit Step 6: Solve and modify the model interac-
are two square feet for product 1 and three tively. LINDO computer software is used to run
square feet for product 2. this ordinary LP model and yields the following
results. Initial total cost is imprecise and has a
triangular possibility distribution of ($282,329,
4.2. Solving procedures $338,571, $372,329), and overall degree of DM
satisfaction with multiple goal values is 0.6242.
The solution procedure using the proposed Moreover, the DM may try to modify the results
interactive PLP approach for the Daya case is
described as follows.
Step 1: Formulate the PLP model for the APP Table 3
decision problem according to Eqs. (1)–(7). Maximum labor, machine, and warehouse capacity data
Step 2: Model the imprecise data with triangular ~ t max ~ t max
Item W M V t max ðft2 Þ
possibility distributions. Tables 1–3 list the trian- (man-hours) (machine-hours)
gular possibility distribution of the imprecise
coefficients and right-hand sides. 1 (175, 300, 320) (360, 400, 430) 10000
2 (175, 300, 320) (450, 500, 540) 10000
Step 3: Develop three new crisp objective
3 (175, 300, 320) (540, 600, 650) 10000
functions of the auxiliary MOLP problem accord- 4 (175, 300, 320) (450, 500, 540) 10000
ing to Eqs. (8)–(10).

Table 1
Forecast demand data

Item Period

1 2 3 4

D~ 1t (900, 1000, 1080) (2750, 3000, 3200) (4600, 5000, 5300) (1850, 2000, 2100)
D~ 2t (900, 1000, 1080) (450, 500, 540) (2750, 3000, 3200) (2300, 2500, 2650)

Table 2
Related operating cost data

Product a~ nt ($/unit) b~nt ($/unit) c~nt ($/unit) d~nt ($/unit) e~nt ($/unit)

1 (17, 20, 22) (26, 30, 33) (22, 25, 27) (0.27, 0.30, 0.32) (35, 40, 44)
2 (8, 10, 11) (12, 15, 17) (10, 12, 13) (0.13, 0.15, 0.16) (16, 20, 23)
 ARTICLE IN PRESS

R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341 337

Table 4
Initial and improved APP plan for the Daya case

Item Initial compromise solutions Improved compromise solutions

Period Period

1 2 3 4 1 2 3 4

Product 1
Q1t (units) 598 2996 4992 2296 598 235 3824 2296
O1t (units) 0 0 0 0 0 0 0 0
S1t (units) 0 0 0 0 0 3929 0 0
I 1t (units) 0 0 0 300 0 1167 0 300
B1t (units) 0 0 0 0 0 0 0 0

Product 2
Q2t (units) 1439 2140 714 1320 3852 0 0 1320
O2t (units) 0 0 0 0 0 0 0 0
S2t (units) 0 0 0 1376 0 0 441 1376
I 2t (units) 640 2282 0 200 3054 2555 0 200
B2t (units) 0 0 0 0 0 0 0 0

H t (man-hours) 0 169 0 0 0 0 179 16

F t (man-hours) 169 0 0 92 0 288 0 0
W t (man-hours) 131 300 300 208 300 12 191 207

Objective values L ¼ 0:6242; z1 ¼ $338;571 L ¼ 0:9660; z1 ¼ $326;499

z2 ¼ $56;242; z3 ¼ $33;758 z2 ¼ $49;660; z3 ¼ $30;340
z~ ¼ ð$282;329; $338;571; $372;329Þ z~ ¼ ð$276;839; $326;499; $356;839Þ

PIS ($300,000, $80,000, $20,000) ($320,000, $60,000, $30,000)

NIS ($600,000, $30,000, $50,000) ($600,000, $48,000, $40,000)

Note: The total cost is imprecise and has a triangular possibility distribution of z~ ¼ ðz1 ; z1  z2 ; z1 þ z3 Þ:

interactively by adjusting the linear membership
functions and related model parameters until a
satisfactory solution is obtained. Consequently,
the improved total cost is imprecise and has a
triangular possibility distribution of ($276,839,
$326,499, $356,839), and overall degree of DM
satisfaction is 0.9660. Table 4 lists the initial and
improved APP plan for the Daya case based on the
present information.
5. Results and findings
Significant management implications for the
practical application of the proposed PLP ap-
proach are as follows. First, the proposed PLP
approach yields an efficient compromise solution.
The APP decision problem presented in the Daya
case was solved using the crisp ordinary LP model,
as summarized in Table 5. This work assumed that
the DM specified the most possible value of the
possibility distribution of each imprecise data as
the precise numbers. Consequently, the optimal
value when applying LP to minimize the total costs
was $326,499. In contrast with the proposed PLP
approach, the improved results were ($276,839,
$326,499, $356,839). These figures indicate that the
PLP solutions are an efficient compromise solu-
tion, compared to the optimal goal value obtained
by the LP model. Thereafter, an improved APP
 ARTICLE IN PRESS
338 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341
plan is obtained by the proposed PLP approach
under an acceptable degree of DM satisfaction in a
fuzzy environment. Notably, the goal values using
the proposed approach should be imprecise
because the unit cost and related coefficients
are always imprecise in nature. On the other
hand, the analytical results demonstrate that
replacing the imprecise data with the most possible
value or average of the possibility distribution is
unreasonable.
Second, the proposed PLP approach provides
the overall degree of DM satisfaction under the
proposed strategy of minimizing the most possible
value and the risk of obtaining higher total costs,
and maximizing the possibility of obtaining lower
total costs. If the solution is L ¼ 1; then each
goal is fully satisfied; if 0oLo1; then all of the
goals are satisfied at the level of L, and if L ¼ 0;
then none of the goals are satisfied. For example,
the overall degree of DM satisfaction with
the imprecise goal values ($282,329, $338,571,
$372,329) initially was generated as 0.6242.
Furthermore, the L value was adjusted to seek a
set of better APP compromise solutions since the
DM did not accept the initial overall degree of this
satisfaction value. Consequently, the improved
total cost is imprecise and has a triangular
possibility distribution of ($276,839, $326,499,
$356,839), and the overall degree of DM satisfac-
tion is up to 0.9660. Fig. 5 shows the change in the
triangular possibilistic distributions of the impre-
cise total costs for the Daya case.
Third, comparing the initial and improved
solutions reveals that the changes in the PIS and
NIS of the objective functions, influences the
Table 5
Comparison of solutions
Item LP The proposed PLP
approach
Objective function Min z Max L
L (DM’s overall 100% 96.60%
degree of
satisfaction)
z (Total costs) $326,499 ($276,839, $326,499,
$356,839)
objective and L values. From Table 4, the L
values rapidly increased from 0.6242 to 0.9660
when the PIS changed from ($300,000, $80,000,
$20,000) to ($320,000, $60,000, $30,000), and NIS
changed from ($600,000, $30,000, $50,000) to
($600,000, $48,000, $40,000). Conversely, total
costs sharply decrease from ($282,329, $338,571,
$372,329) to ($276,839, $326,499, $356,839). These
findings imply that the DM must specify an
appropriate set of PIS and NIS of the objective
functions for making APP decisions, to effectively
seek the right linear membership function for each
objective function. In practice, the LP solutions
often were used as a starting point of the PIS and
NIS, and both intervals must cover the LP
solutions. For example, three crisp objective
functions of the auxiliary MOLP problem pre-
sented in the Daya case were respectively solved
using the crisp single-goal LP model, and the
corresponding PIS and NIS of the initial solution
are specified as summarized in Table 6.
Finally, the proposed PLP approach is the most
practical for solving APP decision problems and
can generate better decisions than other models.
Table 7 compares the proposed PLP approach
presented here to the LP (Saad, 1982), fuzzy goal
programming (FGP) (Wang and Liang, 2004a),
and FLP (Lee, 1990) models. Several features
distinguish the proposed PLP approach from
other APP models.
1. The proposed PLP approach can solve most
real-world APP problems involving imprecise
Improved solutions Initial solutions
1

z
0 276,839 326,499 356,839
282,329 338,571 372,329
~z
Fig. 5. The possibilistic distribution of the optimal total costs.
 ARTICLE IN PRESS

R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341 339

parameters through an interactive decision- making process. The proposed approach also
making process. The proposed model constitu- provides information on alternative strategies
tes a systematic framework that facilitates the for overtime, subcontracting, inventory, back-
decision-making process, enabling the DM orders, and hiring and layoffs workers, in
interactively to modify the imprecise data and response tovariations in forecast demand. Ad-
related model parameters until a satisfactory ditionally, the proposed model considers the
solution is found. actual limitations in labor, machine, and ware-
2. The proposed PLP approach outputs more house capacity.
wide-ranging decision information than other 3. It is important to distinguish the meaning of the
models. Notably, the optimal goal values using PLP approach studied here from FLP and
the proposed approach should be imprecise stochastic linear programming approaches.
since the forecast demand and related operating Basically, the FLP is based on the subjective
costs and capacity are always imprecise in preferred concept for establishing membership
nature. The proposed PLP approach focuses functions with fuzzy data, while the PLP is
on the multi-periods and multi-products (pro- based on the objective degree of event occur-
duct family) problems in an APP decision- rence required to obtain possibilistic distribu-
tions with imprecise data. PLP thus usually is
Table 6 distinguished from FLP and some FLP techni-
Comparison of solutions ques may not be applicable for PLP. On the
Item LP-1 LP-2 LP-3 ðzPIS NIS other hand, the main problem in applying the
i ; zi Þ
stochastic programming models is the lack of
Objective Min z1 Max z2 Min z3 — computational efficiency and the inflexible
function probabilistic doctrines in which the real im-
z1 $326,499 — — ($300,000, $600,000)
z2 — $56,260 — ($80,000, $30,000)
precise meaning of the DM might be impossible
z3 — — $30,333 ($20,000, $50,000) to model. Alternatively, the PLP approach not
only provides more computational efficiency

Table 7
Comparison of the APP models

Factor LP (Saad, 1982) FGP (Wang and Liang, FLP (Lee, 1990) The proposed PLP
2004a)

Objective function Single Multiple Single or multiple Single, linear

Objective value Deterministic Fuzzy Fuzzy Imprecise
Constraint Deterministic Deterministic Fuzzy Imprecise
Product item Single item Product family Product family Product family
Degree of satisfaction Not presented Presented Presented Presented
Planning horizon Multi-period Multi-period Multi-period Multi-period
Market demand Deterministic Deterministic Fuzzy Imprecise
Machine capacity Infinite Limited Infinite Limited
Warehouse space Infinite Limited Infinite Limited
Costs of regular time, Deterministic Deterministic Fuzzy Imprecise
overtime production,
and inventory carrying
Subcontracting Considered Considered Considered Considered
Backordering Considered Considered Not considered Considered
Hiring cost Deterministic Deterministic Fuzzy Imprecise
Layoff cost Deterministic Deterministic Fuzzy Imprecise
Labor level Deterministic Deterministic Fuzzy Imprecise
 ARTICLE IN PRESS
340 R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341
and more flexible doctrines, but also supports
possibilistic decision-making in an uncertain
environment (Lai and Hwang, 1992a; Tang
et al., 2001).
6. Conclusions
The APP decision aims to set overall production
levels for each product category to meet fluctuat-
ing future demand, frequently from 3 to 18 months
ahead, such that APP also determines the appro-
priate resources to be used. This work presents a
novel interactive PLP approach for solving multi-
product APP decision problems with the imprecise
forecast demand, related operating costs, and
capacity. The proposed approach attempts to
minimize total costs with reference to inventory
levels, labor levels, overtime, subcontracting and
backordering levels, and labor, machine, and
warehouse capacity. The proposed approach
simultaneously minimizes the most possible value
of the imprecise total costs, maximizes the
possibility of obtaining lower total costs, and
minimizes the risk of obtaining higher total costs.
An industrial case demonstrates the feasibility
of applying the proposed approach to real APP
decision problems. The proposed PLP approach
yields an efficient APP compromise solution and
overall degree of DM satisfaction with determined
goal values. Moreover, the proposed approach
provides a systematic framework that facilitates
the decision-making process, enabling a DM to
interactively modify the imprecise data and related
model parameters until a satisfactory solution is
obtained. Particularly, several significant manage-
ment implications and features of the proposed
PLP approach that distinguish it from the LP,
FGP, FLP, and stochastic linear programming
models are presented. Consequently, the proposed
approach is the most suitable for making real-
world APP decisions.
References
Baykasoglu, A., 2001. MOAPPS 1.0: Aggregate production
planning using the multiple-objective tabu search. Interna-
tional Journal of Production Research 39, 3685–3702.
Bellman, R.E., Zadeh, L.A., 1970. Decision-making in a fuzzy
environment. Management Science 17, 141–164.
Bowman, E.H., 1956. Production scheduling by the transporta-
tion method of linear programming. Operations Research 4,
100–103.
Bowman, E.H., 1963. Consistency and optimality in managerial
decision making. Management Science 9, 310–321.
Buckley, J.J., 1988. Possibilistic linear programming with
triangular fuzzy numbers. Fuzzy Sets and Systems 26,
135–138.
Buckley, J.J., 1989. Solving possibilistic linear programming
problems. Fuzzy Sets and Systems 31, 329–341.
Buxey, G., 2003. Strategy not tactics drives aggregate planning.
International Journal of Production Economics 85 (3),
331–346.
Charnes, A., Cooper, W.W., 1961. Management Models and
Industrial Applications of Linear Programming. Wiley,
New York.
Dobos, I., 2003. Optimal production–inventory strategies for a
HMMS-type reverse logistics system. International Journal
of Production Economics 81–82, 351–360.
Fung, R.Y.K., Tang, J., Wang, D., 2003. Multiproduct
aggregate production planning with fuzzy demands and
fuzzy capacities. IEEE Transactions on Systems, Man, and
Cybernetics—Part A: Systems and Humans 33 (3), 302–313.
Gnoni, M.G., Iavagnilio, R., Mossa, G., Mummolo, G., Leva,
A.D., 2003. Production planning of a multi-site manufac-
turing system by hybrid modelling: A case study from the
automotive industry. International Journal of Production
Economics 85 (2), 251–262.
Holt, C.C., Modigliani, F., Simon, H.A., 1955. Linear decision
rule for production and employment scheduling. Manage-
ment Science 2, 1–30.
Hsu, H.M., Wang, W.P., 2001. Possibilistic programming in
production planning of assemble-to-order environments.
Fuzzy Sets and Systems 119, 59–70.
Hussein, M.L., 1998. Complete solutions of multiple objective
transportation problems with possibilistic coefficients.
Fuzzy Sets and Systems 93, 293–299.
Hwang, C.L., Yoon, K., 1981. Multiple Attribute Decision
Making: Methods and Applications. Springer, Berlin.
Inuiguchi, M., Sakawa, M., 1996. Possible and necessary
efficiency in possibilistic multiobjective linear programming
problems and possible efficiency test. Fuzzy Sets and
Systems 78, 231–241.
Jensen, H.A., Maturana, S., 2002. A possibilistics decision
support system for imprecise mathematical programming
problems. International Journal of Production Economics
77 (2), 145–158.
Jolayemi, J.K., Olorunniwo, F.O., 2004. A deterministic model
for planning production quantities in a multi-plant, multi-
warehouse environment with extensible capacities. Interna-
tional Journal of Production Economics 87 (2), 99–113.
Jones, C.H., 1967. Parametric production planning. Manage-
ment Science 13, 843–866.
Lai, Y.J., Hwang, C.L., 1992a. Fuzzy Mathematical Program-
ming: Methods and Applications. Springer, Berlin.
 ARTICLE IN PRESS
R.-C. Wang, T.-F. Liang / Int. J. Production Economics 98 (2005) 328–341
Lai, Y.J., Hwang, C.L., 1992b. A new approach to some
possibilistic linear programming problems. Fuzzy Sets and
Systems 49, 121–133.
Lee, Y.Y., 1990. Fuzzy set theory approach to aggregate
production planning and inventory control. Ph.D. Disserta-
tion, Department of I.E., Kansas State University.
Masud, S.M., Hwang, C.L., 1980. An aggregate production
planning model and application of three multiple objective
decision methods. International Journal of Production
Research 18, 741–752.
Rimik, J., Rimanek, J., 1985. Inequality relation between fuzzy
numbers and its use in fuzzy optimization. Fuzzy Sets and
Systems 16, 123–138.
Saad, G., 1982. An overview of production planning model:
Structure classification and empirical assessment. Interna-
tional Journal of Production Research 20, 105–114.
Shi, Y., Haase, C., 1996. Optimal trade-offs of aggregate
production planning with multi-objective and multi-capa-
city-demand levels. International Journal of Operations and
Quantitative Management 2 (2), 127–143.
Stephen, C.H.L., Wu, Y., Lai, K.K., 2003. Multi-site aggregate
production planning with multiple objectives: A goal
programming approach. Production Planning and Control
14 (5), 425–436.
Tanaka, H., Ichihashi, H., Asai, K., 1984. A formulation of
fuzzy linear programming problem based on comparison of
fuzzy numbers. Control and Cybernetics 13, 185–194.
Tanaka, H., Guo, P., Turksen, B., 2000. Portfolio selection
based on fuzzy probabilities and possibilisty distributions.
Fuzzy Sets and Systems 111, 387–397.
Tang, J., Wang, D., Fung, R.Y.K., 2000. Fuzzy formulation for
multi-product aggregate production planning. Production
Planning and Control 11, 670–676.
341
Tang, J., Wang, D., Fung, R.Y.K., 2001. Formulation of
general possibilistic linear programming problems for
complex systems. Fuzzy Sets and Systems 119, 41–48.
Tang, J., Fung, R.Y.K., Yong, K.L., 2003. Fuzzy modell-
ing and simulation for aggregate production planning.
International Journal of Systems Science 34 (12–13),
661–673.
Taubert, W.H., 1968. A search decision rule for the aggregate
scheduling problem. Management Science 14, 343–359.
Wang, D., Fang, S.C., 1997. A genetics-based approach for
aggregate production planning in a fuzzy environment.
IEEE Transactions on Systems, Man, and Cybernetics—
Part A: Systems and Humans 27 (5), 636–645.
Wang, R.C., Fang, H.H., 2001. Aggregate production planning
with multiple objectives in a fuzzy environment. European
Journal of Operational Research 133, 521–536.
Wang, R.C., Liang, T.F., 2004a. Application of fuzzy multi-
objective linear programming to aggregate production
planning. Computers and Industrial Engineering 46 (1),
17–41.
Wang, R.C., Liang, T.F., 2004b. Aggregate production
planning with multiple fuzzy goals. International Journal
of Advance Manufacturing Technology, in press.
Zadeh, L.A., 1978. Fuzzy sets as a basis for a theory of
possibility. Fuzzy Sets and Systems 1, 3–28.
Zimmermann, H.-J., 1976. Description and optimization of
fuzzy systems. International Journal of General Systems 2,
209–215.
Zimmermann, H.-J., 1978. Fuzzy programming and linear
programming with several objective functions. Fuzzy Sets
and Systems 1, 45–56.