Vous êtes sur la page 1sur 8

Computers & Industrial Engineering 62 (2012) 616–623

Contents lists available at SciVerse ScienceDirect

Computers & Industrial Engineering


journal homepage: www.elsevier.com/locate/caie

Multi-coefficients goal programming q


Ching-Ter Chang a,⇑, Huang-Mu Chen b,1, Zheng-Yun Zhuang b,2
a
Department of Information Management, Chang Gung University, 259 Wen-Hwa 1st Road, Kwei-Shan, Tao-Yuan 333, Taiwan, ROC
b
Graduate Institute of Business and Management, Chang Gung University, 259 Wen-Hwa 1st Road, Kwei-Shan, Tao-Yuan 333, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: In today’s increasingly competitive business environment, maintaining profit margins or quantities of
Received 2 February 2011 goods sold is an important issue for businesses. Accordingly, more and more industries use group pricing
Received in revised form 11 September discrimination strategy to attract potential customers in order to increase competitive advantage. Hence,
2011
to find ways maximize profit and to minimize total cost, group pricing discrimination strategy has
Accepted 22 November 2011
Available online 1 December 2011
become an important issue for decision makers. Unfortunately, these types of problems cannot be solved
by any current goal programming models. The objective of this study is to deduce a new method, which
we call the multi-coefficients goal programming, for group pricing discrimination problems. In addition,
Keywords:
Group pricing
an example is given to illustrate the correctness and usefulness of the proposed model.
Goal programming Ó 2011 Elsevier Ltd. All rights reserved.
Multi-segment goal programming

1. Introduction allowing a firm to extract more consumer surplus/profits than it


would be charging a single price. In order to implement PD strat-
In today’s increasingly competitive business environments, in egy, enterprises can hold some promotion activities that will di-
order to maximize profit or maximize quantities of products sold rectly or indirectly incur additional costs (e.g., the costs of how
and to attain or sustain competitive advantage, companies must to attract general customers to become member customers and
customize products or their prices to satisfy different customers’ how to enhance the customer loyalty of those member customers,
needs. For example, Dell offers different prices for the same com- etc.). In addition, a company may sell many products, and each
puter to different types of customers, such as general customers, product is endowed with two or more different market prices.
enterprises, governmental institutions and small business; simi- Therefore, based on the combinations of these variables, it is hard
larly, a movie theater usually offers at least two kinds of ticket work for decision makers (DMs) to determine what are the exact
prices (adult ticket and student ticket) for the same movie; hotels amounts of products which should be sold to different types of cus-
or car rental agencies also offer special discounts to their loyal cus- tomers. In consideration of the facts stated above, DMs would be
tomers (Schön, 2010). From the above examples, it can be seen that glad to have a method to balance the goals of controlling cost
price discrimination (PD) plays an important role in today’s busi- and selling products. It would also be useful to have a method to
ness environment; in fact, the strategy of ‘‘adjusting price’’ or determine the proper quantities and to be sold to general and
‘‘price discrimination’’ for profit is also the easiest and fastest member customers. Such a method would thereby contribute to
way to increase enterprises’ competitiveness (Dolgui, 2010; Liao, the total sales plan/forecast of an enterprise. In fact, the problem
2011). Therefore, more and more industries use PD strategy to en- on balancing multiple objectives can be grounded in the field of
hance their competitiveness. goal programming (GP).
Group pricing is one type of PD. It refers to how to sell the same GP is a decision tool for DMs to solve multiple objective decision
product to different types of customer at different prices, thus, making problems. It was first introduced by Charnes and Cooper
(1961), and further developed by Lee (1972), Ignizio (1985), Tamiz,
q Jones, and Romero (1998), and Romero (2001), among others
This manuscript was processed by Area Editor Imed Kacem.
⇑ Corresponding author. Address: Department and Graduate Institute of Information (Chang, 2007; Liao, 2009). GP has been applied to solve many
Management, Chang Gung University, 259 Wen-Hwa 1st Road, Kwei-Shan, Tao-Yuan real-world decision making problems such as those of supply
333, Taiwan, ROC. chain, human resources, production planning, healthcare and so
E-mail addresses: chingter@mail.cgu.edu.tw (C.-T. Chang), jckychen@gmail.com on (Adeyeye, 2008; Amiri, 2009; Caballero, Gómez, & Ruiz, 2009;
(H.-M. Chen), waynemcgwire@yahoo.com (Z.-Y. Zhuang).
1 Giannikos, 2009; Ku, Chang, & Ho, 2010; Oddoye, Jones, Tamiz, &
Mobile: +886 939567580.
2
Mobile: +886 910249241. Schmidt, 2009; Pati, 2008).

0360-8352/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2011.11.027
C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623 617

A classical GP model can be expressed as follows: straints, however, there are nonetheless some drawbacks, as
(GP Model) follows. Let us consider the decision-making problem depicted in
X
n Fig. 1. Assume that there is a large rectangle, S, representing the
min jfi ðXÞ  g i j global market and four small rectangles, A, B, C and D, representing
i¼1 the domain marketing corresponding to four different companies,
s:t: X 2 F ðF is a feasible setÞ CA, CB, CC, CD, respectively. Each company sells two products and
each product can be sold to two types of customers (e.g., general
where fi(X) is the function of the ith goal, and gi is the aspiration
customer and member customer). Based on group pricing discrim-
level of the ith goal.
ination strategy (GPDS), each product has two different prices (one
The purpose of GP is to minimize the deviations between the
price is for general customers; the other is for member customers).
achievement of goals and their aspiration levels. There are many
Let us take the small rectangle (A) in Fig. 1 as an example to illus-
minimization processes that can be accomplished with various
trate this idea. Suppose that g1 represents the sale goal of CA, and
types of methods, such as MINMAX GP, weighted GP and lexico-
the DM of CA would like to find out the best quantities of products
graphic GP, integer GP, nonlinear GP, stochastic GP, fractional GP,
sold to the two types of customers under goal g1.
interactive GP, GP with intervals, extended interval GP, fuzzy GP,
To solve this problem, we can let the decision variable xi (i = 1, 2)
fuzzy extended lexicographic GP (Aouni & Kettani, 2001; Arenas-
be the individual quantity for selling each of the two products.
Parra, Bilbao-Terol, Pérez-Gladish, & Rodrı́guez-Urı´a, 2010; Daim,
Then, we can define the coefficients pij (i, j = 1, 2) as the two differ-
Kayakutlu, & Cowan, 2010; Romero, 2001; Stoyan & Kwon, 2011;
ent prices for the two types of customers for the ith product. That
Vitoriano & Romero, 1999). Unfortunately, all of these methods
is, every decision variable (x1 and x2) in rectangle (A) has two asso-
allow only one aspiration level on the right-hand-side of each goal.
ciated coefficients. By referring to the MSGP model, this problem
In some special situations, multiple aspiration levels on the right-
can be expressed as follows:
hand-side of each goal might exist in many real-world problems.
For example, an insurance company sets a policy stating that its ½p11 b1 þ p12 ð1  b1 Þx1 þ ½p21 b2 þ p22 ð1  b2 Þx2 ¼ g 1 ð1Þ
salespersons’ annual sales must exceed the previous year’s sales
by 10% in order to receive that year’s bonus. The result is that even where bi (i = 1, 2) are binary variables, pij (i, j = 1, 2) are two possible
if a salesperson can reach a sale of 1 million, he/she may hold back coefficients for two products, x1 and x2 are the quantities of the two
to only 500,000 in order not to create too much pressure in the fol- products which are sold.
lowing year. In turn, this prevents the company from maximizing As seen in the Eq. (1), if b1 = 1 then the term [p11
its overall performance. However, the aspiration level of the right- b1 + p12(1  b1)]x1 becomes p11x1, and if b1 = 0 then the term
hand-side in the GP model allows only one aspiration level. Hence, [p11b1 + p12(1  b1)]x1 becomes p12 x1. This implies the possible
in order to help the DM find the optimal aspiration level of each goal, coefficients for each decision variable is actually a mutually exclu-
Chang (2007) has proposed multi-choice GP (MCGP) to solve this sive relationship. That is, only one coefficient will be chosen and
problem. The MCGP model can be expressed as follows: other coefficients are excluded for each decision variable in the
X
n MSGP model. Briefly speaking, this relationship can be regarded
min jfi ðXÞ  g i1 or g i2 or . . . or g im j as an exclusive-or (XOR) relationship (Hempel, 1966). Render the
i¼1 result of Eq. (1) into Table 1 according to the behavior of the two
s:t: X 2 F ðF is a feasible setÞ; binary variables b1 and b2.
From above descriptions, if the DM would like to solve a prob-
Recently, following the idea of MCGP, Liao (2009) proposed a lem in which double coefficients might co-exist and co-affect each
multi-segment goal programming (MSGP) model as follows: other at the same time, the solution of the MSGP model could not
  þ   þ   þ  
min g 1 d1 ; d1 ; g 2 d2 ; d2 ; . . . ::; g n dn ; dn fully meet the DM’s requirements. To the best of our knowledge, no
Xn previous studies have been conducted to solve such a problem.
 þ
s:t: Sij xi þ di  di ¼ g i ; j ¼ 1; 2; . . . ; m Thus, the purpose of this study is to solve the problems of multiple
i¼1 coefficients existing in the GPDS and to establish a newly articu-
Sij ¼ Si1 or Si2 or . . . or Sim ; i ¼ 1; 2; . . . ; n lated model.
X 2 F ðF is a feasible setÞ In sum, the above discussion reveals that we need to develop a
new model to solve such problem where the relationship between
where Sij are the coefficients attached to the decision variable xi of the multiple coefficients and decision variables can be an ‘‘AND’’
þ 
the ith goal; di and di are the over-and under-achievement of the relationship not only a ‘‘XOR’’ relationship. This significant
ith goal, respectively. improvement is the major contribution of this study. Throughout
Though the MSGP model allows DMs to set multiple coefficients this study, we are generalizing, proposing and deducting our brand
for mapping each decision variable in the left-hand-side of the con- new model based on the above observation. Our hypothesis con-
tributes to the proposed model, in as much as it solves our target
Rectangle S problem in a practical way. The rest of this paper is organized as
follows: the problem description and proposed model are de-
{( p11 , p12 )x1} g1 {( p31 , p32 )x3} g2 scribed in Section 2, illustrative examples are given in Section 3,
{( p21 , p22 )x2} g1 {( p41 , p42 )x4} g2 and finally conclusions are made in Section 4.

Table 1
Rectangle (A) Rectangle (B)
The result of MSGP.

{( p51 , p52 )x5} g3 {( p71 , p72 )x7} g4 b2 b1

{( p61 , p62 )x6} g3 {( p81 , p82 )x8} g4 0 1


1 p12x1 + p21x2 = g1 p31x1 + p41x2 = g2
Fall in Rectangle (A) Fall in Rectangle (B)
Rectangle (C) Rectangle (D)
0 p52x1 + p62x2 = g3 p71x1 + p82x2 = g4
Fall in Rectangle (C) Fall in Rectangle (D)
Fig. 1. Example of MSGP coefficients.
618 C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623

2. Modeling multi-coefficients goal programming problem Table 2


The result of DCGP coefficients.

In the GPDS, a DM sets different prices for one single product in b2 b1


accordance with customers’ properties in order to maximize prof- 0 1
its or maximize the quantities of products sold (Schön, 2010). In
1 p12t11 + p11t12 + p21t21 + p22t22 = g1 p31t31 + p32t32 + p41t41 + p42t42 = g2
fact, many enterprises, such as airline companies, hotels, retailers Fall in Rectangle (A) Fall in Rectangle (B)
and car rental agencies, as well as companies in the travel, hospi- 0 p52t51 + p51t52 + p62t61 + p61t62 = g3 p71t71 + p72t72 + p82t81 + p81t82 = g4
tality and entrainment industries, often take this policy to enhance Fall in Rectangle (C) Fall in Rectangle (D)
their competitiveness (Belobaba, 1987, 1989; Bitran & Mondsch-
ein, 1995; Bitran & Gilbert, 1996; Belobaba & Wilson, 1997; Bern-
stein & Federgruen, 2003; Chen, Federgruen, & Zheng, 2001; Schön,
MSGP DCGP
2010). In order to implement the ‘‘AND’’ relationship for multiple
coefficients of GPDS, we try to derive a double-coefficients goal xi
xi
programming (DCGP) model as follows. t i1 ti2
Fig. 2 is revised from Fig. 1. We take the rectangle (A) in Fig. 2 as
an example to demonstrate the ‘‘AND’’ relationship and to repre-
sent the multiple coefficients of GPDS. First, we introduce an addi-
tional decision variable tij, which indicates the quantity of product i pi1 pi2 pi1 pi2 pi1 pi2
sold to the specified customer group j. Other notations are defined
as follows: Fig. 3. The difference between MSGP and DCGP model.

pij: Selling price of ith product sold to the jth type of customer
b2 = 0 then Eq. (2) becomes ‘‘p12t11 + p11t12 + p22t21 + p21t22 = g1’’.
(i = 1, 2; j = 1, 2)
As can be seen, though bi is changed, finally, each tij can easily be
tij: Quantity Sold of ith product to the jth type of customer
mapped to a proper coefficient (pij). That is, through this method,
(i = 1, 2; j = 1, 2)
we can reach the ‘‘AND’’ relationship for the DCGP problem.
xi: total amount of ith product sold (i = 1, 2)
From previous descriptions, we know that the relationship be-
tween all the possible coefficients of each decision variable can
Second, we show how to find out the proper coefficient of each
actually be an ‘‘AND’’ relationship, not only an ‘‘XOR’’ relationship.
decision variable in the DCGP model. Following the idea of Chang
Subsequently, the general DCGP model can be expressed as
(2007) and Liao (2009), we introduce two extra binary variables,
follows:
b1 and b2, into the DCGP model to deal with this problem; the
new model can be expressed as follows: X
n
þ 
min z ¼ di þ di
½p11 b1 þ p12 ð1  b1 Þt 11 þ ½p11 ð1  b1 Þ þ p12 b1 t12 i¼1
þ½p21 b2 þ p22 ð1  b2 Þt 21 s:t: ½pi1 bi þ pi2 ð1  bi Þti1 þ ½pi1 ð1  bi Þ þ pi2 bi t i2
þ½p21 ð1  b2 Þ þ p22 b2 t 22 ¼ g 1 ð2Þ þ 
 di þ di ¼ g i ; i ¼ 1; 2; . . . ; n
P
2
X
2
tij ¼ xi ði ¼ 1; 2Þ
j¼1
t ij ¼ xi ; i ¼ 1; 2; . . . ; n
j¼1
where bi (i = 1, 2) are binary variables; x1 and x2 are the quantities þ 
di ; di P 0; i ¼ 1; 2; 3 . . . ; n
sold of the two products, respectively.
X 2 F ðF is a feasible setÞ
Render the result of Eq. (2) into Table 2 according to the two
binary variables, b1 and b2. where bi is the binary variable; xi represents the total amount of the
Fig. 3 shows the difference between the MSGP and DCGP þ 
ith product sold; di and di are the over- and under-achievements of
models. As can be seen in Fig. 3, each decision variable (xi) in the the ith goal, respectively.
DCGP model can be decomposed into two ‘sub decision variables’ In general, if there are more than two prices mapping one prod-
(ti1 and ti2), which represent two different types of customers, uct, it becomes a multi-coefficient GP problem. Furthermore, based
and each sub decision variable maps two different prices. In order on the idea of the proposed DCGP model, we can add extra binary
to find the proper price of each sub decision variable, we can add variables for modeling multi-coefficient GP; thus, which can be ex-
binary variables (bi) to deal with this problem. Eq. (1) is then trans- pressed as follows:
formed to Eq. (2). For example, if b1 = 1 and b2 = 1 then Eq. (2)
becomes ‘‘p11t11 + p12t12 + p21t21 + p22 t22 = g1’’, and if b1 = 0 and X
n
þ 
min z ¼ di þ di
i¼1
X
n X
m
þ 
Rectangle S s:t: pik cik ðBÞtij  di þ di ¼ g i ; i ¼ 1; 2; . . . ; n;
j¼1 k¼1
{[( p11 , p12 )t11; ( p11 , p12 )t12]x1 } g1 {[( p31 , p32 )t31; ( p31 , p32 )t32]x3 } g2 X
m
{[( p21 , p22 )t21; ( p11 , p12 )t22]x2 } g1 {[( p41 , p42 )t41; ( p41 , p42 )t42]x4 } g2 t ij ¼ xi ; i ¼ 1; 2; . . . ; n;
j¼1
Rectangle (A) Rectangle (B) þ 
di ; di P 0; i ¼ 1; 2; . . . ; n;
{[( p51 , p52 )t51; ( p51 , p52 )t52]x5 } g3 {[( p71 , p72 )t31; ( p71 , p72 )t72]x7 } g4 cik ðBÞ 2 Ri ðxÞ; i ¼ 1; 2; . . . ; n;
{[( p61 , p62 )t21; ( p61 , p62 )t62]x6 } g3 {[( p81 , p82 )t41; ( p81 , p82 )t82]x8 } g4 X 2 F ðF is a feasible setÞ;

Rectangle (C) Rectangle (D) where cik(B) is a function of the binary serial number, referring to
Chang (2007); Ri(x) is the function of resource limitations; other
Fig. 2. Example of DCGP coefficients. variables are defined as in DCGP.
C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623 619

By referring to Chang (2000), the binary terms cik(B) can be x2: Total sales amount of Radios.
linearized as follows. Let y = cik(B), where y must satisfy the follow- x3: Total sales amount of Cellular Phones.
ing inequalities t1j: Quantities sold of TVs to member customers or general cus-
! ! tomers (j = 1, 2).
X
m X
m
cik ðBÞ  m þ 1 6 y 6 m cik ðBÞ þ 1 ð3Þ t2j: Quantities sold of Radios to member customers or general
k¼1 k¼1 customers (j = 1, 2).
y 6 cik ðBÞ; for all k ð4Þ t3j: Quantities sold of Cellular Phones to member customers or
general customers (j = 1, 2).
y P 0 ð5Þ
p1j: Selling price of TVs sold to member customers or general
The above inequalities can be checked as follows: customers by referring to Table 3; the p11 = 8 and p12 = 10.
p2j: Selling price of Radios sold to member customers or general
P
(i) if m c ðBÞ ¼ m then y = 1 (from Eq. (3)), customers by referring to Table 3; the p21 = 6 and p22 = 7.
Q k¼1 ik
(ii) if m
k¼1 cik ðBÞ ¼ 0 then y = 0 (from Eqs. (4), (5)).
p3j: Selling price of Cellular Phones sold to member customers
or general customers by referring to Table 3, the p31 = 5 and
One may have a question concerning ‘‘whether the number of p32 = 6.
group types impacts the efficiency of multi-coefficient GP model u1j: Purchasing cost of TVs sold to member customers or general
or not’’. In fact, in real business cases as mentioned in Section 1, customers by referring to Table 3; the u11 = 5 and u12 = 7.
there usually may not be too many group types. Hence, to solve u2j: Purchasing cost of Radios sold to member customers or gen-
the moderate problem, efficiency of linearized method has been eral customers by referring to Table 3; the u21 = 4 and u22 = 5.
provided in the study of Chang (2000). u3j: Purchasing cost of Cellular Phones sold to member custom-
ers or general customers by referring to Table 3, the u31 = 3 and
3. An illustrative example u32 = 4.
c1j: Marketing cost of selling TVs to member customers or gen-
A wholesaler sells three products: TV, radio, and cellular phone. eral customers by referring to Table 3; the c11 = 0.5 and c12 = 0.3.
Each product can be sold to two types of customers, namely gen- c2j: Marketing cost of selling Radios to member customers or
eral customers and member customers. The DM of the wholesaler general customers by referring to Table 3; the c21 = 0.8 and
company decided to deploy the GPDS for the three products. For c22 = 0.3.
each product, the selling price and related costs are described in c3j: Marketing cost of selling Cellular Phones to member cus-
Table 3. tomers or general customers by referring to Table 3; the
The DM wants to find out the best combination of quantities of c31 = 0.6 and c32 = 0.1.
the three products for general customers and member customers
and satisfy the following pre-defined goals. According to the above mentioned definitions, the problem can
be expressed as follows.
 Goal 1 (sales goal): The DM wishes the revenue to be greater Goals:
than or equal to 4500.
 Goal 2 (purchasing cost goal): The DM wishes the total purchase ðg 1 Þf8; 10gx1 þ f6; 7gx2 þ f5; 6gx3 P 4500
cost of acquiring the three products to be smaller than or equal ðg 2 Þf5; 7gx1 þ f4; 5gx2 þ f3; 4gx3 6 2800
to 2800. ðg 3 Þf0:5; 0:3gx1 þ f0:8; 0:3gx2 þ f0:6; 0:1gx3 6 400
 Goal 3 (marketing cost goal): The DM wishes the total market-
ing cost of the three products to be smaller than or equal to 400. Constraints:

Except the abovementioned three major goals, the DM also x1 ¼ 0:3ðx1 þ x2 þ x3 Þ; ð6Þ
hopes that among the total sales, 30% is in TV sales, 40% is in Radio x2 ¼ 0:4ðx1 þ x2 þ x3 Þ; ð7Þ
sales, and 30% is in Cellular Phone sales for risk management. x3 ¼ 0:3ðx1 þ x2 þ x3 Þ; ð8Þ
In order to demonstrate how to use the proposed model to solve
the above problem, the following two cases are taken by using
DCGP and MSGP models. The correctness and usefulness of the t11 þ t 12 ¼ x1 ; ð9Þ
DCGP are also provided in the two cases. t21 þ t 22 ¼ x2 ; ð10Þ
t31 þ t 32 ¼ x3 ; ð11Þ
3.1. Case I. Using DCGP to solve this problem
where the {} braces enclose a set of some possible coefficients in an
To illustrate the modeling and solving of the problem by DCGP, OR relationship (each stated {} set includes two coefficients that ap-
the related notations are defined as follows: pear in a XOR relationship or in an AND relationship), as mentioned
in Section 1. For example, for g1, there is a possible coefficient set of
x1: Total sales amount of TVs. {8, 10} beneath x1 as stated. In this case, rather than taking 8 or 10

Table 3
Selling price, purchasing cost and marketing cost of each product.

Item Product name


TV Radio Cell phone
Member customer General customer Member customer General customer Member customer General customer
Selling price $8 $10 $6 $7 $5 $6
Purchasing cost $5 $7 $4 $5 $3 $4
Marketing cost $0.5 $0.3 $0.8 $0.3 $0.6 $0.1
620 C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623

individually, either 8, 10, or a mixture of both, are the possible Table 5


coefficients of x1 which are supposed to affect the result. The goal deviations of Case I.

Based on the proposed DCGP model, this problem can be formu- Deviation variable Goal
lated as the following program: g1 g2 g3
(P1) þ  þ  þ 
d1 d1 d2 d2 d3 d3
þ  þ  þ 
min z ¼ d1 þ d1 þ d2 þ d2 þ d3 þ d3 Deviation value 0 0 90 0 10 0
s:t: ð8b1 þ 10ð1  b1 ÞÞt11 þ ð10b1 þ 8ð1  b1 ÞÞt 12 Total deviation 0 0 90 0 10 0

þ ð6b2 þ 7ð1  b2 ÞÞt21 þ ð7b2 þ 6ð1  b2 ÞÞt22


þ  þ  þ  þ 
þ ð5b3 þ 6ð1  b3 ÞÞt31 þ ð6b3 þ 5ð1  b3 ÞÞt32  d1 þ d1 ¼ 4500; ð12Þ min z ¼ d1 þ d1 þ d2 þ d2 þ d3 þ d3
ð5b1 þ 7ð1  b1 ÞÞt11 þ ð7b1 þ 5ð1  b1 ÞÞt12 þ ð4b2 þ 5ð1  b2 ÞÞt21 s:t: ð8b1 þ 10ð1  b1 ÞÞx1 þ ð6b2 þ 7ð1  b2 ÞÞx2 þ ð5b3 þ 6ð1  b3 ÞÞ
þ ð5b2 þ 4ð1  b2 ÞÞt22 þ ð3b3 þ 4ð1  b3 ÞÞt31 þ 
þ 
 d1 þ d1 ¼ 4500; ð5b1 þ 7ð1  b1 ÞÞx1 þ ð4b2 þ 5ð1  b2 ÞÞx2
þ ð4b3 þ 3ð1  b3 ÞÞt32  d2 þ d2 ¼ 2800; ð13Þ
þ 
þ ð3b3 þ 4ð1  b3 ÞÞ  d2 þ d2 ¼ 2800;ð0:5b1 þ 0:3ð1  b1 ÞÞx1
ð0:5b1 þ 0:3ð1  b1 ÞÞt 11 þ ð0:3b1 þ 0:5ð1  b1 ÞÞt12
þ
þ ð0:8b2 þ 0:3ð1  b2 ÞÞt21 þ ð0:3b2 þ 0:8ð1  b2 ÞÞt22 þ ð0:8b2 þ 0:3ð1  b2 ÞÞx2 þ ð0:6b3 þ 0:1ð1  b3 ÞÞx3  d3
þ  
þ ð0:6b3 þ 0:1ð1  b3 ÞÞt31 þ ð0:1b3 þ 0:6ð1  b3 ÞÞt32  d3 þ d3 ¼ 400 þ d3 ¼ 400;
ð14Þ Eqs:ð6Þ  ð8Þ
Eqs: ð6Þ  ð11Þ þ 
di ; di P 0; i ¼ 1;2; 3
þ 
di ;di P 0; i ¼ 1;2;3
x1 ;x2 ;x3 P 0 and integer
t11 ;t12 ;t21 ;t22 ;t31 ;t32 ;x1 ;x2 ;x3 P 0 and integer

where b1, b2 and b3 are binary variables; x1, x2 and x3 are the total where all variables are defined as in P1.
sales quantities of TVs, Radios and Cellular Phones, respectively; Similarly, we solve the program using LINGO (Scharge, 2002) on
{t11, t12}, {t21, t22}, {t31, t32} denote the actual quantities of products a PC586 with 2 GHz CPU to obtain the optimal solutions
sold to general customers OR member customers for TVs (x1), (x1, x2, x3, b1, b2, b3) = (213, 284, 213, 1, 1, 1) as shown in Table 6. The
þ 
Radios (x2) and Cellular Phones (x3); di and di are the positive goal deviations are shown in Table 7.

and negative deviation variables, respectively. From Table 7, the g1 has a value ðd1 ¼ 27Þ below the sales goal
þ
Solving the above DCGP modeled problem using LINGO of 4500; g2 has a value ðd2 ¼ 40Þ over the purchase cost goal of
þ
(Scharge, 2002) on a PC 586 with 2 GHz CPU, we obtain the optimal 2800, while g3 has a value ðd3 ¼ 61:5Þ over the marketing cost goal
solutions (x1,x2,x3,t11, t12,t21,t22,t31,t32,b1,b2,b3,g1,g2,g3) = (210,280, of 400. The total deviation of the objective function z is 128.5. In
210,0,210,45,235,45,165,0, 0,0) as shown in Table 4. The goal devia- addition, we can also observe that the relationship of coefficients
tions are shown in Table 5. of the decision variables in this model (P2) is ‘‘XOR’’.
þ 
As can be seen in Table 5, the deviations (d1 and d1 ) of g1 are Based on the results obtained from both Cases I and II, Table 8
zero, which means we reach the sales goal of 4500 exactly; g2 compares the details derived from the DCGP and MSGP models.
þ
has a value ðd2 ¼ 90Þ over the purchase cost goal of 2800, while As seen in Table 8, DCGP has the total deviation value of 100
þ
g3 has a value ðd3 ¼ 10Þ over the marketing cost goal of 400. The that is better than MSGP has the total deviation value of 128.5.
total deviation of the objective function z is 100. Moreover, DCGP can obtain the quantities of each product sold to
Although it is not proper to apply any current goal program- both member customers and general customers. However, MSGP
ming models to the problem that we have pointed out above, we can only obtain the quantities of product sold to either member
still use the MSGP model to solve this problem again as follows. customers or general customers. Obviously, the utilization of MSGP
has been improved by the proposed DCGP.
To illustrate the superiority of the DCGP model, we made a mar-
3.2. Case II. Using MSGP to solve the same problem
ginal benefit analysis diagram, as shown in Fig. 4. In Fig. 4, the prof-
it of DCGP and MSGP are 4500 and 4473, respectively. That is,
Goals:
against MSGP, the profit increased by 27 (see mark A in Fig. 4) in
ðg 1 Þf8 XOR 10gx1 þ f6 XOR 7gx2 þ f5 XOR 6gx3 P 4500 DCGP model. Moreover, purchasing cost and marketing cost be-
ðg 2 Þf5 XOR 7gx1 þ f4 XOR 5gx2 þ f3 XOR 4gx3 6 2800 came (2890, 410) and (2840, 461.5), respectively. Notably, the pur-
ðg 3 Þf0:5 XOR 0:3gx1 þ f0:8 XOR 0:3gx2 þ f0:6 XOR 0:1gx3 6 400 chasing cost increased by 50 (see mark B in Fig. 4), but the
marketing cost had dropped 51.5 (see mark C in Fig. 4) in the DCGP
Constraints: Eqs. (6)–(8) model. Consequently, by adopting the proposed DCGP model, the
Based on the MSGP method proposed by Liao (2009), this prob- marginal benefit earned by the company became 28.5 (mark
lem can be formulated as the following program: A + mark C-mark B), which would be more than that obtained by
(P2) MSGP.

Table 4
Solutions of Case I.

Item Product name


TV Radio Cell phone
Member customer General customer Member customer General customer Member customer General customer
Quantities sold 210 0 235 45 165 45
Total selling amount 210 280 210
C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623 621

Table 6
Solutions of Case II.

Item Product name


TV Radio Cell phone
Member customer General customer Member customer General customer Member customer General customer
Quantities sold 213 0 284 0 213 0
Total selling amount 213 284 213

Table 7 Table 8
The goal deviations of Case II. Comparisons of DCGP model and MSGP model.

Deviation variable Goal Observations Percentage Observations Percentage


value of goal value of goal
g1 g2 g3
(DCGP) achievement (MSGP) achievement
þ  þ  þ 
d1 d1 d2 d2 d3 d3 (%) (%)

Deviation value 0 27 40 0 61.5 0 Quantities sold 210 213


Total deviation 0 27 40 0 61.5 0 (x1)
Member 210 213
customers
General 0 0
3.3. Case III. Using DCGP to solve the same problem with fixed ‘‘ratio of customers
selling’’ Quantities sold 280 284
(x2)
In order to suit the real case, the DM might wish that a fixed ra- Member 235 284
customers
tio of selling 60% (40%) can be kept to the member customers (the General 45 0
general customers). According to this condition, the constraints of customers
Case I can be reformulated as follows. Quantities sold 210 213
Constraints: (x3)
Member 165 213
x1 ¼ 0:3ðx1 þ x2 þ x3 Þ; customers
General 45 0
x2 ¼ 0:4ðx1 þ x2 þ x3 Þ; customers
x3 ¼ 0:3ðx1 þ x2 þ x3 Þ; Goal g1 4500 100 4473 99.4
t11 þ t 12 ¼ x1 ; Goal g2 2890 97 2840 98.57
Goal g3 410 97.5 461.5 84.6
t21 þ t 22 ¼ x2 ;
Total deviation 100 128.5
t31 þ t 32 ¼ x3


0:6x1 if b1 ¼ 1
t 11 ¼ value
1  0:6x1 if b1 ¼ 0

0:6x2 if b2 ¼ 1
t 21 ¼
1  0:6x2 if b2 ¼ 0
 4500
0:6x3 if b3 ¼ 1 A 27 (Revenue difference)
t 31 ¼ 4473
1  0:6x3 if b3 ¼ 0
To solve this problem, we can add the ‘‘ratio of selling con-
DCGP model
straints (RoS)’’ to the DCGP model, which can be expressed as
follows: 2890
50 (Purchasing
MSGP model B
þ  þ  þ  Cost difference)
min z ¼ d1 þ d1 þ d2 þ d2 þ d3 þ d3 2840
s:t Eqs: ð6Þ  ð14Þ
b1 t11 þ ð1  b1 Þt 12 ¼ ð0:6b1 x1 Þ þ ð0:6ð1  b1 Þx1 Þ
461.5
b2 t21 þ ð1  b2 Þt 22 ¼ ð0:6b2 x2 Þ þ ð0:6ð1  b2 Þx2 Þ 51.5 (Marketing
C
410 Cost difference)
b3 t31 þ ð1  b3 Þt 32 ¼ ð0:6b3 x3 Þ þ ð0:6ð1  b3 Þx3 Þ
þ 
di ; di P 0; i ¼ 1; 2; 3
t 11 ; t12 ; t21 ; t 22 ; t 31 ; t 32 ; x1 ; x2 ; x3 P 0 and integer g1 g2 g3 goal
Revenue Purchasing Marketing
where all variables are defined as in P1. Cost Cost
Again, solve the above problem by using LINGO (Scharge, 2002)
on a PC 586 with 2 GHz CPU, we obtain the optimal solutions Fig. 4. Marginal benefit analysis of two models.
(x1, x2, x3, t11, t12, t21, t22, t31,t32, b1, b2, b3) = (195, 260, 195, 78, 117, 104,

156, 78, 117, 0, 0, 0) as shown in Table 9. The goal deviations are 2800, while g3 has a value ðd3 ¼ 84:1Þ below the marketing cost
shown in Table 10. goal of 400. The total deviation of the objective function z is 289.1.

From Table 10, the g1 has a value ðd1 ¼ 67Þ below the sales goal Notably, as seen in Table 9, each product can be sold to both
þ
of 4500; g2 has a value ðd2 ¼ 138Þ over the purchase cost goal of member customers and general customers at the same time. How-
622 C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623

Table 9
Solutions of Case III.

Item Product name


TV Radio Cell phone
Member customer General customer Member customer General customer Member customer General customer
Quantities sold 117 78 156 104 117 78
Total selling amount 195 260 195

Table 10 Table 12
The goal deviations of Case III. The goal deviations of Case IV.

Deviation variable Goal Deviation variable Goal


g1 g2 g3 g1 g2 g3
þ  þ  þ  þ  þ  þ 
d1 d1 d2 d2 d3 d3 d1 d1 d2 d2 d3 d3

Deviation value 0 67 138 0 0 84.1 Deviation value 0 67 138 0 0 84.1


Total deviation 0 67 138 0 0 84.1 Total deviation 0 67 138 0 0 84.1

ever, the RoS constraints cannot be added to the MSGP model. This Table 13
can also imply the possibly potential advantage and superiority of Comparisons of DCGP model and MINMAX-DCGP model.
the proposed DCGP model. Observations Percentage Observations Percentage
value of goal value of goal
(DCGP) achievement (MINMAX- achievement
3.4. Case IV. Using MINMAX-DCGP to solve the same problem with (%) DCGP) (%)
fixed RoS
Quantities sold 195 195
(x1)
Since there are many minimization processes that can be Member 117 117
accomplished with various types of extension models in the field customers
of GP. MINMAX is a common technique among them. Thus, we will General 78 78
customers
demonstrate how to adopt MINMAX-DCGP to solve the GPDS prob-
Quantities sold 260 260
lem. A MINMAX-GP can be modeled as follows (Yaghoobi & Tamiz,
(x2)
2007). Member 156 156
(MINMAX-GP) customers
General 104 104
min z ¼ D customers
þ 
fi ðXÞ  di þ di ¼ g i i ¼ 1; 2; 3; . . . ; n Quantities sold 195 195
þ  (x3)
D P di þ di i ¼ 1; 2; 3; . . . ; n Member 117 117
s:t: X 2 F ðF is a feasible setÞ customers
General 78 78
customers
By referring to the MINMAX-GP and DCGP models, the problem of
Goal g1 4433 98.5 4433 98.5
Case III can be transformed as the following MINMAX-DCGP model.
Goal g2 2938 95 2938 95
(MINMAX-DCGP) Goal g3 315.9 121 315.9 121

min z ¼ D Total deviation 289.1 289.1

s:t: Eqs: ð6Þ  ð14Þ


þ 
D P di þ di i ¼ 1; 2; 3
Again solving the above problem using LINGO (Scharge, 2002)
b1 t11 þ ð1  b1 Þt 12 ¼ ð0:6b1 x1 Þ þ ð0:6ð1  b1 Þx1 Þ on a PC 586 with 2 GHz CPU, we obtain the optimal solutions
b2 t21 þ ð1  b2 Þt 22 ¼ ð0:6b2 x2 Þ þ ð0:6ð1  b2 Þx2 Þ (x1, x2, x3, t11, t12, t21, t22, t31, t32, b1, b2, b3) = (195, 260, 195, 117, 78,104,
b3 t31 þ ð1  b3 Þt 32 ¼ ð0:6b3 x3 Þ þ ð0:6ð1  b3 Þx3 Þ 156, 117, 78, 1, 0, 1) as shown in Table 11. The goal deviations are
þ  shown in Table 12.Based on the results obtained from Cases III
di ; di P 0; i ¼ 1; 2; 3
and IV, we can compare the details for the DCGP and MINMAX-
t 11 ; t12 ; t21 ; t 22 ; t 31 ; t 32 ; x1 ; x2 ; x3 P 0 and integer DCGP models in Table 13. As seen in Table 13, DCGP and
MINMAX-DCGP obtain the same optimal solutions.
where all variables are defined in P1.

Table 11
Solutions of Case IV.

Item Product name


TV Radio Cell phone
Member customer General customer Member customer General customer Member customer General customer
Quantities sold 117 78 156 104 117 78
Total selling amount 195 260 195
C.-T. Chang et al. / Computers & Industrial Engineering 62 (2012) 616–623 623

4. Conclusions Belobaba, P. P., & Wilson, J. L. (1997). Impacts of yield management in competitive
airline markets. Journal of Air Transport Management, 3, 3–9.
Bernstein, F., & Federgruen, A. (2003). Pricing and replenishment strategies in a
The traditional GP model allows each decision variable on the distribution system with competing retailers. Operations Research, 51, 409–426.
left-hand-side of each goal mapping to have only one coefficient. Charnes, A., & Cooper, W. W. (1961). Management model and industrial application of
linear programming. New York: Wiley.
Practically, in some cases, one decision variable mapping multiple
Chen, F., Federgruen, A., & Zheng, Y. S. (2001). Near-optimal pricing and
coefficients may exist in the real world, such as in cases with GPDS replenishment strategies for a retail/distribution system. Operations Research,
problems. Previous studies have proposed the MSGP model, in 49(6), 839–853.
Chang, C. T. (2000). An efficient linearization approach for mixed integer problems.
which each decision variable can map multiple coefficients. How-
European Journal of Operational Research, 123, 652–659.
ever, the relationship among these coefficients is an ‘‘XOR’’ rela- Caballero, R., Gómez, T., & Ruiz, F. (2009). Goal programming: Realistic targets for
tionship, not an ‘‘AND’’ relationship. Therefore, the MSGP model the new future. Journal of Multi-Criteria Decision Analysis, 16, 79–110.
cannot solve problems with multiple coefficients. Chang, C. T. (2007). Multi-choice goal programming. Omega, 35(4), 389–396.
Daim, T. U., Kayakutlu, G., & Cowan, K. (2010). Developing Oregon’s renewable
In this study, we propose a new DCGP model to solve the ‘‘AND’’ energy portfolio using fuzzy goal programming model. Computers & Industrial
relationship among multiple coefficients. In addition, we take a Engineering, 59, 786–793.
simple group pricing problem as an example to show the correct- Dolgui, A. (2010). Pricing strategies and models. Annual Reviews in Control, 34,
101–110.
ness and usefulness of the proposed model. In this example, the Giannikos, I. (2009). A fuzzy goal programming model for task allocation in
proposed model not only allows a DM to set different prices for teamwork. International Journal of Human Resources Development and
one single product, but also it obtains a properly satisfactory solu- Management, 9(1), 97–115.
Hempel, C. G. (1966). Philosophy of natural science (1st ed.). Englewood Cliffs, New
tion by finding a balance for all goal targets. Finally, the superiority Jersey.
of the proposed approach is observed from the proposed illustra- Ignizio, J. P. (1985). Introduction to linear goal programming. Beverly, Hills, CA: Sage.
tive example, whereas, the result also reveals that the solution of Ku, C. Y., Chang, C. T., & Ho, H. P. (2010). Global supplier selection using fuzzy
analytic hierarchy process and fuzzy goal programming. Quality and Quantity,
the DCGP model is much better than that of MSGP model.
44(4), 623–640.
The model proposed in this study is driven by the operational Lee, S. M. (1972). Goal programming for decision analysis (1th ed.). Philadelphia, PA:
facts which render better results in comparison to the previous Auerbach.
Liao, C.-N. (2009). Formulating the multi-segment goal programming. Computer &
MSGP method. Thus, our study might connote a more complete
Industrial Engineering, 56, 138–141.
range of problem-solving and decision-making capability. We hope Liao, C.-N. (2011). Fuzzy analytical hierarchy process and multi-segment goal
the promising results can advise us and spur further research in the programming applied to new product segmented under price strategy.
field of multi-coefficient GP. Computers & Industrial Engineering, 61, 831–841.
Oddoye, J. P., Jones, D. F., Tamiz, M., & Schmidt, P. (2009). Combining simulation and
goal programming for healthcare planning in a medical assessment unit.
References European Journal of Operational Research, 193(1), 250–261.
Pati, R. K. (2008). A goal programming model for paper recycling system. Omega,
Aouni, B., & Kettani, O. (2001). Goal programming model: A glorious history and a 36(3), 405–417.
promising future. European Journal of Operational Research, 133, 225–231. Romero, C. (2001). Extended lexicographic goal programming: A unifying approach.
Adeyeye, A. D. (2008). Goal programming model for production planning in a Omega, 29, 63–71.
toothpaste factory. South African Journal of Industrial Engineering, 19(2), Scharge, L. (2002). LINGO release 8.0., LINDO System, Inc.
197–209. Schön, C. (2010). On the optimal product line selection problem with price
Amiri, M. (2009). Developing and solving a new model for the location problems: discrimination. Management Science, 56(5), 896–902.
Fuzzy-goal programming approach. Journal of Applied Sciences, 9(7), 1344–1349. Stoyan, S. J., & Kwon, R. H. (2011). A stochastic-goal mixed-integer programming
Arenas-Parra, M., Bilbao-Terol, A., Pérez-Gladish, B., & Rodrı́guez-Urı́a, M. V. (2010). approach for integrated stock and bond portfolio optimization. Computers &
A new approach of Romero’s extended lexicographic goal programming: Fuzzy Industrial Engineering, 61, 1285–1295.
extended lexicographic goal programming. Soft Computing, 14(11), 1217–1226. Tamiz, M., Jones, D., & Romero, C. (1998). Goal Programming for decision making:
Belobaba, P. P. (1987). Airline yield management: An overview of seat inventory An overview of the current state-of-the-art. European Journal of Operational
control. Transportation Science, 21, 63–73. Research, 111, 567–581.
Belobaba, P. P. (1989). Application of a probabilistic decision model to airline seat Vitoriano, B., & Romero, C. (1999). Extended interval goal programming. Journal of
inventory control. Operations Research, 37, 183–197. the Operational Research Society, 50(12), 1280–1283.
Bitran, G. R., & Mondschein, S. V. (1995). An application of yield management to the Yaghoobi, M. A., & Tamiz, M. (2007). A method for solving fuzzy goal programming
hotel industry considering multiple day stays. Operations Research, 43, 427–443. problems based on MINMAX approach. European Journal of Operational Research,
Bitran, G. R., & Gilbert, S. M. (1996). Managing hotel reservations with uncertain 177, 1580–1590.
arrivals. Operations Research, 44, 35–49.

Vous aimerez peut-être aussi