Você está na página 1de 4

Vol. 23, Nos I-4, pp.

117-120, 1992
Printed in Great Britain. All fights reserved

0360-8352/92 $5.00+0.00
Copyright 1992 Pergamon Press Ltd

Computers a n d Industrial Engineering

Method for Solving Multiobjective Aggregate Production


Planning Problem with Fuzzy Parameters
M. Gen, Y. Tsujimura, & K. Ida
Dept. of Industrial & Systmns Engg.
Ashikaga inSL of Tech.
Ashikaga, 326 Japan

Abstract
We propose an efficient method that transforms a fuzzy multiple objective linear programming (MOLl') problem
model to crisp MOLP model, and an interactive solution proeedum that suggest the best compromise aggregate production
plans for the multi-period fuzzy multiple objective aggregate
production planning (APP) problem.
1. Introduction

present a realistic sppmach to handle multiple, conflicting, and


immonsurable objective&
In the past, deterministic demand data and resources available were used as inputs to the mAPP problem. However, in
the practical APP problem, forecasted demand for a pnxluct
and available resources in the future are imlnecise or fuzzy in
natlHe.
The ~
objective multi-period APP model with fuzzy
numbers can be formulated as follows:
(1) minimize totalrproduction cost:
min

The purposes of multi-period aggregate production


planning(APP) are to set up overall production levels for each
product category to meet the fluctuating" or uncertain demands
in the near future, and to set up decision and polices on the issues of hiring, layoff, overtime, backorder, subcontracting, and
inventory. Multi-period APP proNems usually involve multiple
conflicting objectives in which such objectives may be as follows:
(1) minimize production cost, (2) minimize backorders, (3)
minimize total inveatory, and (4) maintain a balanced work
force.
So, we need to decide making interactive trade-off decisions
to obtain the best compromise solution.
When using any of the multi-period APP models mentioned above, it is assumed that the goal and rite model inputs
(resources and demands) are deterministic, in a practical producfion system, the demand dAta~ resources, and costs ave imIm:cise or fuzzy. It is important to use the fuzzy sets theory for
solving multiple APP problem that it can be very helpful in
modeling the imprecision in goals, demands, and resources.
In this paper, we introduce a multiple objective aggregate production planning(f-mAPP) model with triangular positive fuzzy coefficients at Section 2. At Section 3, we propose
an efficient method that transforms a f-moAPP model to crisp
multiple objective linear programming(c-MOLP) model. At
Section 4, we propose a computational algorithm for interactively solving the fuzzy multiple objective linear programming
problem. Also, at Section 5 we demonstrate an interactive
solution result that suggest the best compromise aggregate production plans for the multi-period f-mAPP problem.
2. f.mAPP Model

Aggregate production planning problems usually involve


multiple conflicting objectives. Such problems involve making
trade.off decisions to obtain the best compromise solution. The
conflict arises because the improvement in one objective can
only be made to the impairment of One or move of the rest of
the objectives. The multiple objectives in an APP problem may
be:
(1) minimize production COSt,
(2) minimize beckorders,
(3) minimize total inventory,
(4) maintain a balanced work force,

zx = ~'.{c~(pn + pw) + c. w, + c~ k p ~ }

(2) minimize total invontory and backorder cost:


T

min

:2 = ~'~{ci~ i, + ct~ bi}


t=|

(3) minimize the changes in work force level:


T

min

% = ~{/~ + l,}

The constraints:
wt < wt,~

for all t

wt = wt-i + ~ - Ii

for all

k pn < 8 wf

for all t

k pm S ~,S wf
for all t
Pn +Po + it..i - bt--i > f mi,
for all t
ir - bt = if-i - bt..i + P n + Pot - f t
for all
Pn , Pot , wt , if , bf , ht , lt > 0
for all t

Parameters and constants :


c~ = production cost excluding labor cost in period t (S/trait)
c , = o v e ~ n e labor cost in period t (S/man-hour)
ca = labor cost in period t (S/man-day)
c. = inventory carrying cost in period t (S/unit-period)
c~ = cost to hire one worker in period t (S/man-day)
ca = cost to layoff one worker in period t (S/man-day)
c~ = stockout cost in period t (S/unit-period)
k = conversion factor in hours of labor per unit of production
8 = regular working hours per worker per day
1St = fraction of working hours available for overtime
production
t = planning horizon or number of periods
io = initial invontory level (units)
wo = initial work force level (man-day)
bo ffi initial backorder level (units)
w~ffi, = maximum work force available in period t (man-day)
ft = forecasted demand in period t (units)
f~,~ = minimum demand in period t (units)

etc.

Multiple objective decision making(MODM) methods

117

Decision Variables :
p~ ffi regular time production in period t (units)
p~ = overtime production in period t (units)
wt = work force level in period t (man-day)
if ffi inventory level in period t (units)
bt = beckorder level in period t (uni0
ffi worker hind in period t (man-day)
It = worker layoff in period t (man-day)

118

Proceedings o f the 14th Annual Conference on Computers and Industrial Engineering

3. Transfornmtinn Method
A multi-period aggregate production planning (f-mAPP)
model with triangular fuzzy coefficients is formulated as a fuzzy multiple objective linear programming (f-MOLP) problem.
In this section, we present transformation method from a fuzzy MOLl' problem to a crisp MOLP problem to solve a fMOLP problem. Without the loss of generality, we assume
that all the parameters of the f-MOLP problem are represented
by triangular fuzzy numbers.
The general f-MOLP problem maximizing ql objective
functions and minimizing q2 objective functions subject to ml
less than or equal type constraints, m2 greater than or equal
type constraints, m3 equality type constraints and n nonnegarive decision variables can be defined as follows:
f-MOLP:
max

zk

rain

zk = ~ . c t / x /

s.t.

~a"Ox / < b l ,

~. ct/xj
j=l

1, 2,

k=qt+l,

. .

. , ql

In this paper, we propose a generalized transformation


method which could treat with any type of constraints and
maximizing and/or minimizing objective functions of the fuzzy
MOLP problems.
The fuzzy MOLP problem can be reformulated into the
traditional MOLP problem consisting of q objective functions,
m constrains, and n decision variables as follows :

... ,q=ql+q2
c-MOLP :
max

i = 1,2, . - . ,mr

~ ~ijx) > bl , i = m r + l ,

fuzzy multiple objective linear programming problem consisting q objective functions, m constraints and n decision variables to the crisp MOLP problem consisting of q objective
functions, 3m+3q constraints, and n+q decision variables. As
said in Negi, the main difficulty in his transformation approach
is the increase in the number of constraints and decision variables
Recently, Okada et al[15] have improved the Negi's
transformation method and proposed that an efficient transformarion method of the fuzzy MOLP problem to the crisp
MOLP problem consisting q objective functions, m constraints,
and n decision variables.

zt = ~"~ { ( l - a ) c t i 3 + act./2}xj , k = 1,2, - . . ,qt


j=t

. . . ,m2

rain
~0xj
/=1

='bi,

i = m ~ + l , -.- , m = m t + m 2 + m 3
xj >- O,

j = 1, . . ,

where
'~+j = (ckj~ c+j2 ckja) is a fuzzy coefficient of the k-th objective function and j-th decision variable,
"~o = (ao~ alj2 ao9 is a fuzzy technical coefficient of the bth
constraint and the j-th decision variable,
~ = (b~ b~2 b~D is a fuzzy available resource of the i-th
constraint, and
x/ is the j-th decision variable.
The fuzzy parameters in the f-MOLP model are all the triangular positive fuzzy numbers '~=(rt, r2, r3) shown in Fig.l
in which their membership function is as follows :

IJ~(X) =

(x-r2)+ 1

(r~ ~ x ~ r2)

(X - r2) + 1

(r 2 < X < = r3)

I~

s.t.

~ {(I-or) a01 + aai/2}xj ~ ( 1 - a ) b~3 + abe2,


j=l
i = 1, . . . ,ml

~ {(I-or)a//3 + a a i j 2 } x i > ( 1 - a ) b , + ctbi2,


j=l
i = rtl I + 1, . . - ,m 2

)=1

{(1 - a) aol + a a/j2} xi < (1 - a) bi3 + Ct bi2 ,


i = M2 + t, "'" , m

~ { ( l - a ) aq3 + a a O 2 } x / >- ( 1 - a ) bll + abi2,


./=t
i = m2 + 1, . . . ,m
x/ > 0,
j = 1 , - . . ,n

where ct is a cutoff value such that 0 < a < 1.

(X < r i ,

r 3 < x)

1.0

r!

zt = ~ ( ( l - a ) ck/l + actj2}xj , k = ql, ' ' " ,q

I'2

I'3

Fig.1 A triangular fuzzy number


In the transformation approach, the conditional possibility
distribution for each objective function results in an objective
function and decision variable space. The joint conditional pnssibllity distribution of the objective functions is defined using
the fuzzy minimum operator. The joint conditional possibility
distribution of the objective functions is defined as the
minimum of the joint conditional possibility distribution function and the possibility of feasibility of each of the constraints.
Negi[13] has presented that a transformation method of the

4. Computational Algorithm

We show the computational algorithm for interactively


solving the fuzzy multiple objective linear programming fMOLP problem.
Step I: Formulate f-MOLP problem.
We formulate the fuzzy MOLP problem maximizing and/or
minimizing q objective functions subject to any type of constraints as defined at Section 3.
Step 2: Transform f-MOLP to c-MOLP.
By using the joint conditional possibility distribution, we
transform the f-MOLP problem to the crisp MOLP problem.
Step 3: Solve c-MOLP problem.
After setting ct parametrically, to get the positive and negative ideal solutions (PIS & N'IS), one of the nondominated
solutions,
and the value of each objective function we solve MOLP
problems by the Method of Global Criterion[10].
Step 4: Determine priority Wk of each objective function.
Decision maker determines a priority Wk of each objective
function by using the AHP technique[16].
Step 5: Determine compromise solution.
By using the TOPSIS method[7] as shown in below, we
determine a compromise solution.
ul

j = | , , " ' ' ,V

G E N et a l . :

MOLP Method

it i2 is -

where
z~ = ~ / ~ ,

k=l,2,...,q

: the best value of the k-th objective function


z~' : the worst value of the k-th objective function
and Wt is the weight of k-th objective function which is obtained in Step 4 .
We find the nearest noodominated solution to the best value
:ts using the following equation.

uf"

+ ,7'

119

0 < e7 < I ,

j = 1,2, "'" ,v

b t = io - bo +
b 2 = it - b t +
b3 = i2 - b2 +
i4 - b 3 = is - b s +
is - b s = i4 - b4 +
i ~ - b 6 = is - bs +
Pn, P..
w t , it, bt,

P,1
P,2
P,s
Pr4

+
+
+
+

Pot
P,2
Po3
Po4

(210, 240, 2 7 0 ) ,
(160, 200, 2 4 0 ) ,
(290, 320, 350),

(200.~$0,280),
P,s + Ps - (250,300,350),
p,.(; + Po6 - (220, 270, 320),
h,,

It ~ O .

t = 1,...,6

After transforming the fuzzy MOLP example to the crisp


MOLP problem by the transformation method proposed here,
we can get PIS and NIS at the level a =0 as a sample in
Table 1, and the values of the objective function for the various level a as shown in Table 2.

We select the nearest closeness e7 to 1, and stop as the best


compromise solution the nondominated solution is corresponding to its e7 .

Table 1. PIS and N'IS at file level a = 0

5. Numerical Example
To illustrate the proposed interactive f-MOLP approach, a
numerical example was solved. The parameters used are as follows:
(1) There is a six-period planning horizon with fuzzy demands
of (210,240,270), (160,200,240), (290,320,350), (200,240,280),
(250,300,350), and (220,270,320) units for periods 1 to 6
respectively. The decision maker also requires that the production level plus inventory level at any period has to he greater
than potential minimum demand f~,u The minimum demand in
periods 1-6 are 200, 180, 280, 220, 260, and 240 respectively.
(2) The production cost ether than labor cost in $20 per unit.
The hours of labor needed for each unit of production is three
and the regular work day is eight man-hour per day.
(3) The initial work force is 100 workers and the maximum
workers allowed are (60,80,100), (60,85,110) for periods 1 and
2, (60,90,120) for periods 3 to 6. The costs associated with the
reb'ular payroll, hiring, and firing are $64, $30, and $40 Per
worker per day respectively.
(4) Overtime production is limited to no more than 30% of
regular time production. The overtime charge is based on $15
per worker per hour.
(5) It is also assumed that there is no beginning inventory. The
inventory carrying cost is $2 per unit per period. The other
restriction is that the backorder must not be carried over more
than one period. The backorder cost is $10 per unit per period.
As a numerical example of the fuzzy MOLP, we treat with
the six periods and three objective APP model including positive triangular fuggy paramteters a follows:
f-mAPP:
6

min

zl = ~

{ (15, 20, 25) (p, +pot) + (59,64,69) wt

t=l

(10, 15,20).3p~ },

z2 = ~

[2ifi + 10bt },

f=l

zs = ~

{hi + I f } ,

=1

S.t.

wx < (60, 80, 100),


w2 ~ (60, 85, 110),
wt < (60,90,120),
t = 3,'-',6
3pn - 8wt < 0 ,
t = 1,'",6
3 p o t - (0.1,0.3,0.5)'8wt S 0 ,
t = 1,'.-,6
P,1 + Pol + io - bo ~ (170, 200,230),
P,2 + Po2 + il - bl > (150,180,210),
P,3 + P,s + i2 - b2 > (250, 280, 310),
Pr4 + P4 + is - bs 2 (190, 220, 250),
P,S + PoS + i4 - b4 ~ (230, 260, 290),
P,6 + Poe + is - b s > (210, 240, 270),

CAIE 23:1/~--I

:3

NIS
76590
1800
153

Table 2. Values of the objective functions


at the level of a

zI

Z2

i[ 3

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1

49755
56089
57923
59067
60392
61885
63647
65528
68146
71568
75170

160
15
8
16
25
34
43
55
80
104
131

18
4
4
6
8
10

12
14
16
18
20

By using Step 5 of the computational algorithm, we can


obtain the compromise solution ( a = 0.2 ) among the nondominated solutions as follows:
P n = [ 216, 172, 256, 208, 256, 230 ]
p o f = [ 0, 0, 36, 0, 4, 0]
w, = [ 9 6 , 9 6 , 9 6 , 9 6 , 9 6 , 9 6 ]
h~ = [ O, O, O, O, O, O]
If = [ 4, 0, 0, 0, 0, 0 ]
z l = 57923
z2 =
Z3 --

min

PIS
49005
0

it = [ 0, 4, 0, 0. 0, 0 ]
bt = [ 0, 0, 0, 0, 0, 0 ]

rain

zl
z2

8
4

6. Condusion
When we U'eat with the multi-period aggregate production
planning problem, the multiple conflicting objectives and fuzzy
parameters are important factors. In this paper, we formulated
the fuzzy multiple objective finear programming problem for
the f-moAPP model and proposed the u'ansfonnation method
the f-MOLP problem to the crisp MOLP problem.
Also, we combined AHP technique and TOPSIS method
for interactive by solving the crisp MOLP problem and gained
the compromise solution among the nondominated solutions.
We demonstrated the numerical example df the fuzzy moAPP
model and solved it by the computational algorithm.

120

A samlpe

Proceedings of the 14th Annual Conference on Computers and Industrial Engineering

of

input data

: 0.2)

app_alpha=O. 2
rain 16 p r l + 16 p r 2 + 16 p r 3 + 16 p r 4 + 16 p r 5 + 16
+ 4 9 p o 5 + 49 p o 6 + 60 w l + 60 w2 + 60 w3 + 60 w4
rain 2 i l + 2 i 2 + 2 i 3 + 2 i 4 + 2 i 5 + 2 i 6 + 10 b l
rain h l + h2 + h 3 + h4 + h5 + h6 + 1,1 + 1,2 + 1,3 + 1,4
$.t.
w l < g6 :
w2 < 1 0 5 :
w3 < 1 1 4 ;
prl
w4 < 114 ;
pr2
w5 < 114 ;
pr3
w6 < 114 ;
pr4
w l - wO - h l + 1,1 - 0 ;
pr5
w2 - w l - h2 + 1,2 = 0 ;
pr6
w3 - w2 - h 3 + 1,3 = 0 ;
iO w4 - w3 - h 4 + 1,4 = 0 ;
lO w5 - w4 - h5 + 1,5 - 0 ;
il w6 - w5 - h6 + 1,6 = 0 :
il 3 p r l - 8 wl < 0 ;
i2 3 p r 2 - 8 w2 < 0 ;
i2 3 p r 3 - 8 w3 < 0 ;
i3 3 p r 4 - 8 w4 < 0 ;
i3 3 p r 5 - 8 w5 < 0 ;
i4 3 p r 6 - 8 w6 < 0 ;
i4 3 p o l - 1. 12 wl < 0 :
i5 3 po2 - 1. t 2 w2 < 0 ;
i5 3 p o 3 - 1. 12 w3 < 0 ;
wO 3 p o 4 - 1. 12 w4 < 0 ;
iO 3 p o 5 - 1. 12 w5 < 0 :
bO 3 p o 6 - 1. 12 w6 < 0 ;

p r 6 + 4g p o l + 4 9 p o 2 + 49 p o 3 + 49 p o 4
+ 60 w5 + 60 w6 ;
+ 10 b2 + 10 b 3 + 10 b 4 + 10 b5 + 10 b6
+ 1,5 + 1,6 ;

+
+
+
+
+
+

pol +
po2 +
po3 +
po4 +
po5 +
po6 +
bO - i l
bO - i l
bl - i2
bl - i2
b2 - i 3
b2 - i 3
b3 - i 4
b3 - i 4
b4 - i 5
b4 - i 5
b5 - i 6
b5 - i 6
100 ;
0 ;
0 end

iO
il
i2
i3
i4
i5
+
+
+
+
+
+
+
+
+
+
+
+

- bO
- bl
- b2
- b3
- b4
- b5
bl +
bl +
b2 +
b2 +
b3 +
b3 +
b4 +
b4 +
b5 +
b5 +
b6 +
b6 +

> 176
> 156
> 256
> 196
> 236
> 216
prl +
prl
+
pr2 +
pr2 +
pr3 +
pr3 +
pr4 +
pr4 +
pr5 +
pr5 +
pr6 +
pr6 +

~ol
~ol
=o2
~o2
~o3
=o3
=o4
~o4
=o5
~o5
~o6
~o6

<
>
<
>
<
>
<
>
<
>
<
>

264
216
232
168
344
296
272
208
340
260
310
230

;
;
;
;
;
;
;
;
:
;
;
;

References

[1] Lawrence, K.D.& S.H. Zanakis eds: Producgon Planning [8] Goicoechea, A., D.R. Hansen, & L. Duckstein: Multiobjec~ve
and Scheduling: Ma~emagcal Progranmgng ApplicaDecision Analysis with Engineering and Business Applications, Industrial Engineering
and Management
gons, Wiley(1982).
Press(1984).
[9] Steuer, R.E.: Mulgple Criteria Optingzation: Theory, Corn[2] Masod, A.S.M. & C.L. Hwang: An aggregate production
pmagon, and Appli~agon, Wiley(1986).
planning model and application of three multiple objec- [10] Gen, M & K. Ida:/.d~ear Programming and Goal Programlive decision methods, Production Research, VoI.18,
ngng for Personal Computers, Denki-shoin(1984, in
No.6(1980).
Japanese).
[3] Alley, Rafik A.: Production Control on the Basis of Fuzzy [11] Zimmermmm, H.J.: Fuzzy Sets, Decision Making, and ExModels, Fuzzy Sets and Systems, VoL22, pp. 43pert Systems, Kluwer Academic Publishers(1987).
56(1987). [4] Anderson, EJ.: A Note on the Dynamic [12] Kaufinarm, A. & M.M. Gupta: FuszyMa:hemagcoJModels
Lot-Size Model With Uncertainty in Demand and Supply
,n Engineering and Management Science, Elsevier SciProcess, Management Science, Vol.35, No.5, pp. 635ence Publishers(1988).
640(1989).
[13] Nagi, D.S.: Fuzzy analysis and optimization, Ph.D. Thesis,
[5] Foote, B.L., A. Ravindran, and S. Lashine: Computational
Kansas State Univ.(1989).
Feasibility of Multi-criteria Models of Production, Plan- [14] Lee, Y.Y.: Fuzzy sets theory approach to aggregate lxOduculng and Scheduling, Int. Y. of Computers And lndusrrition planning and inventory control, Ph.D. Thesis, Kansas
al Engineering, VoLI5, Nos 1-4, pp. 129-138(1988).
State Univ.(1990).
[6] Gonzalez, Juan J., and Gary R. Reeves: Master Production [15] Okada, S.,Y. Nakahara, M. Gen, & K. Ida: A method for
Scheduling:a Mulliple-Objective Linear Programming Aptransforming multiple objective linear programming probproach, Int. J. of Producgon Research, VoL21, No.4, pp.
lems with triangular fuzzy coefficients, to appear(1991, in
553-562(1983).
Japanese).
[7] Hwang, C.L. & A.S.M. Masod: Multiple Objective Decision [16] Saaty, T.L." The Analytic Hiearchy Process, McGrawMaking: Methods and Applications, SpringerHili(1980).
Veriag(1980).

Você também pode gostar