Expert Systems with Applications 38 (2011) 42444253

Contents lists available at ScienceDirect

Expert Systems with Applications

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

Using multi-objective genetic algorithm for partner selection in green supply chain problems
Wei-Chang Yeh , Mei-Chi Chuang
Integration & Collaboration Laboratory, Department of Industrial Engineering and Engineering Management, National Tsing Hua University, P.O. Box 24-60, Hsinchu 300, Taiwan, ROC

a r t i c l e

i n f o

a b s t r a c t
Partner selection is an important issue in the supply chain management. Since environment protection has been of concern to public in recent years, and the traditional supplier selection did not consider about this factor; therefore, this paper introduced green criteria into the framework of supplier selection criteria. The aim of this research was to develop an optimum mathematical planning model for green partner selection, which involved four objectives such as cost, time, product quality and green appraisal score. In order to solve these conicting objectives, we adopted two multi-objective genetic algorithms to nd the set of Pareto-optimal solutions, which utilized the weighted sum approach that can generate more number of solutions. In experimental analysis, we introduced a {4, 4, 4, 4} supply chain network structure, and compared average number Pareto-optimal solutions and CPU times of two algorithms. 2010 Elsevier Ltd. All rights reserved.

Keywords: Green supplier selection Genetic algorithm Multi-objective optimization Pareto optimality

1. Introduction In a broad sense, green supply chain refers to the management between suppliers, their products and environment, that is to say, the environment protection principle is brought into suppliers management system. Its purpose is to add environment protection consciousness into original products and to improve competitive capacity in markets. The green supply chain known at present refers to supply chain effect brought about by green products proposed by European Community in the 21st century. Some companies, especially, small and medium enterprises, started to build cooperative corporations with supply chain partners, with the hope of promoting propagation of environment management initialization and designing new green products. With increasing public awareness in environmental protection, enterprises began producing more green products than last decade. Deans (1999) pointed out, environmental protection was initiated by American industry, and environmental considerations became a signicant factor to the procurement policy and selection of suppliers. WEEE and RoHS published by European Union in 2003 have exerted impact on the industries associated with electric and electronic equipment (EEE), since incompatible products are barred from the market of EU countries. Since environment protection was not taken into consideration in the traditional supply chain management, this paper introduced this concept to the green supplier selection mechanism which accords with the real situation.

The purpose of this paper was to construct an optimum mathematical planning model for defective supply chain system, and adopted two algorithms to solve this model, and obtained Pareto-optimal solutions for the supplier selection and product volume transportation problems. Four objectives were considered: (1) minimization of total cost comprised of product cost and transportation cost, (2) minimization of total time comprised of product time and transportation time, (3) maximization of average product quality, (4) maximization of green appraisal score. At last, the experimental study adopted two multi-objective genetic algorithms that were proposed by Murata, Ishibuchi, and Tanaka (1996) and Altiparmak, Gen, Lin, and Paksoy (2006) to solve the optimum mathematical planning model, and compared average number of Pareto-optimal solutions and CPU times of two algorithms to nd the most efcient algorithm. This paper is organized as follows: Section 2 is about literature review such as green supply chain management, green supplier selection criteria, WEEE/RoHS directives, defective supply chain system. Section 3 gives a problem statement and mathematical programming for supply chain. Section 4 gives a methodology. Section 5 gives computational results of three algorithms. Concluding remarks and future research are outlined in Section 6. 2. Literature review 2.1. Green supply chain management Noci (1997) pointed out that companies should construct efcient management environment and emphasized on integrating the relationship between customers and suppliers. In fact,

Corresponding author. Fax: +886 3 572 2204.

E-mail address: yeh@ieee.org (W.-C. Yeh). 0957-4174/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.09.091

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

4245

companies could produce green products together with suppliers as mutual cooperation relationship is developed. With this, companies could achieve the following effect: (1) reducing the amount of to-be-delivered accessories with low environment performance; (2) controlling the cost of suppliers green products effectively; (3) decreasing the time responding to the market. Sarkis (2003) advocated a policy and decision mechanism to assist administrators. The paper laid great stress on the components of green supply chain administration, and analyzed how these components became the basic elements of policy-making mechanism. The mechanism consisted of ve strategies: products life cycle policy, enforcing policy on the environment inuence on organization, efciency demands of organization and green supply chain program. Vachon and Klassen (2006) pointed out that through the interaction between suppliers and consumers, manufacturers could construct and practice an effective solution program, while facing the environment challenges. From the exploration of Canadian and American printing industries, they deduced that the partnership of green scheme with suppliers presented positive inuence on delivery time, while the partnership of green scheme with consumers exerts positive inuence on quality, exibility and environment protection. Thereby, the partnership with consumers provided much greater benet than the one with suppliers did. Kainuma and Tawara (2006) stated that USA issued a practical handbook named The Lean and Green Supply chain, with the hope of calculating the achievement rate of nancial improvement and environment protection, and briey reviewing special instruments and methods. Through the handbook, companies constructed an assessment mechanism to observe lifecycle assessment, customer satisfaction and average inventory amount. 2.2. Green supplier selection criteria In the traditional supply chain network, supplier selection criteria only consist of cost, time, and quality. However, environment protection issues were considered in the product process in the recent years, so enterprises hope that suppliers can produce products with environmental protection. Recently, researchers have developed a lot of supplier selection mechanisms which consider environmental protection issues in the following: Noci (1997) suggested that green products suppliers selecting consisted of the following 3 steps: (1) nding out the applicable green strategies the companies can fulll; (2) dening operation measure method to assess the performance of environment protection by the suppliers; (3) to select the most effective method to choose suppliers, to make sure that the suppliers could follow companies strategies of environment protection. Humphreys, Wong, and Chan (2003) held that traditional selecting process of suppliers only observed quality, exibility and so on; however, with environment pressure increasing, many companies started to observe environment protection and assess environment performance of suppliers. Thereby, he advocated policy-drawing supporting instrument, to help companies environment integrating principle enter the selecting process of suppliers. Handeld, Steven, Sroufe, and Melnyk (2002) interviewed 500 subjects from different companies, who were also the experts of environment management, about the signicance of environment performance index. The interviewees came from Mascotech, Cone Drive, IBM, Herman Miller, DLSC, Ford. Through the communication, he summarized the most important and the easiest ten supplier environment performance indexes to assess. Noci (1997) provided a selection system of green product suppliers. At the beginning, the paper dened environment performance index of assessing suppliers, including quantitative and qualitative indexes. These indexes could be separated into 4 major

types. First, companies should assess the environment efciency of suppliers at present; second, companies should assess green capacity of suppliers; third, companies should consider green image of suppliers; lastly, companies should observe the performance process in economic view. Finally, the paper introduced two procurement programs, calculated the weigh of each attribute with AHP, and selected adequate stocking programs. Handeld et al. (2002) introduced environment observation into stocking policies, and made policy-drawing process complex with quantitative and qualitative factors. However, a few companies performed analyzing with any random mechanism, and assessed suppliers by environment observation. The paper applied AHP as policy-drawing support mode and it further explained how AHP was used to assess relative signicance in different environment peculiarity. Through Deer method, the 500 interviewers, from Mascotech, Cone Drive, IBM, Herman Miller, DLSC, Ford assessed the signicance of environment performance index. According to theories of Lamming, Noci, Azzone, Sarkisr, and Handeld, Humphreys et al. (2003) summarized the principles of selecting suppliers. He suggested a policy-drawing support instrument to help company environment integrating principles enter the selecting process of suppliers. The system, based on computers, integrated standard database of industry and applied Visual Basic as its principal built environment. The cases including 8 suppliers were introduced to make quantitative assessment and then qualitative assessment. The quantitative assessment used multi-attribute analysis technique, MAA, while qualitative assessment applied a series of questionnaires. 2.3. WEEE and RoHS directives Manufacturing of electric and electronic equipment, EEE is one of the industries developing rapidest over the world. Simultaneously, with technique innovated and market of electric and electronic equipment, EEE expanded, the replacement of old equipments is speeded up, leading to apparent increasing of waste electric and electronic equipment, WEEE and arousing new environment challenges. In 1998, Western Europe witnessed 6 million tons of electric and electronic equipment, WEEE, and the increasing amount of electric and electronic equipment WEEE would at least reach 35% each year (EC). In 2004, over 3 billion computers would reach their using lifespan in America. From home appliances to computers, electric and electronic equipment, EEE covers vast range. Materials and accessories of waste electric and electronic equipment, WEEE are different and complex. Many materials are strongly toxic, such as bromide, toxic alloy, sensitive substance and plastic products. If these substances are not treated adequately, waste electric and electronic equipment, WEEE will cause serious environment pollution. From environment pollution caused by waste electric and electronic equipment, WEEE, companies should build management system for waste electric and electronic equipment, WEEE to prolong the lifecycle of electric and electronic equipment, EEE. The management system contained collecting, classifying and so on, and included ve common treating strategies (Rose, 2000; Rose, Beiter, & Ishii, 1999): (1) Reusing: to obtain the used products and their related accessories. (2) Repairing: to prolong usage term through repair and maintenance. (3) Reproducing: to detach certain accessories from waste products for future usage. (4) Recycling: to treat used products or their accessories, to reprocess the materials, and to replace the required new materials when producing new products. (5) Disposing: incinerating or burying.

4246

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

Because of canvassing and competing from enterprises and countries governments, WEEE Directive was separated into WEEE Directive and RoHS in 2000bans on toxic substances. Namely, one emphasized recycling, reusing and reproducing, and the other one prohibited certain substances from being contained in products. Willems, Dewulf, and Duou (2006) quantized the decrease of time for disassembling products, and realized economical applicability of systematic product disassembling. He used linear programming mode mechanism to select the nal strategy to optimize product lifecycle and to research the decreasing effect of expected time for disassembling and cost. He probed into the optimization with the view of nal treating of products. The linear programming mode suggested that it was different to perform the optimization for the products of WEEE type; however, for the medium and large products, the mode could be optimized well, if the decreasing of time for disassembling was realized. Queiruga, Walther, Gonzalez-Benito, and Spengler (2008) stated that recycling factories would be built in Spain, according to new laws and regulations of Europe. His article provided a method to rank Spanish autonomous cities according to their adequacy, in order to build cycling factories, PROMETHEE (Preference Ranking Organization Method for Enrichment Evaluations) would be used to rank these programs for establishing factories. The established factories were located in northern and eastern Spain. Judging from the ranking, we learned that the programs ranking in the top would be performed in southern and middle Spain. The method did not provide optimum mechanism for future recycling factories, but offered optimum choices for selecting potential recycling factories. Hammond and Beullens (2007) extended supply chain net where demand exceeded supply in the past into enclosed supply chain net, in order to calculate Equilibrium value obeying laws and regulations. The mode mentioned in their article, took WEEE Directive into account. The net mode was gambling games based on sufcient information, with manufacturers and customers attending. Equilibrium value provided certain important scopes and questions for future research. According to the result, recycling rate of certain minimum new products should be used to simulate reverse logistics activities of supply chain. 2.4. Defective supply chain system Most of former researches always assumed that supply chain was a balanced system. However, production loss may exist in production process. This kind of supply chain with production loss is

named defective supply chain system. There are usually two types of production loss, namely, process loss and resource loss. The early researches considered reject allowance (Klein, 1966; Levitan, 1960; White, 1967), while discussing process loss. In the research of reject allowance, the production equipment was regarded as reliable, but defective products could be produced in production process. The extra input amount required in the order form was called reject allowance. Most of early allowance researches concentrated on single-stage system, while the following researches focused on the system with different yield rate. And the majority of documentations concerning the system with different yield rate were about single-stage system (Lee & Rosenblatt, 1986; New & Mapes, 1984; Porteus, 1986). Lee and Yano (1988) suggested that the yield rate along production process was denite and discussed production batch policies under multi-stage single product system. Liu and Yang (1996) proposed an optimum production batch under a single-stage system with production loss. Chung and Hou (2003) built a mathematical mode to decide optimum production cycle period and minimize cost, under a system with production loss. Wang, Che, and Sha (2004) developed production planning for multi-stage ow production system with production loss, and strived to gure out the optimum machine amount and policies of production and delivery. Under the defective supply chain system, the case that input is not equal to output may occur. Therefore, this paper considered product yield rate and transportation loss rate in the optimum mathematical model to make production system more in accordance with the real situation. 3. Problem description and mathematical foundation The problem of this paper considered about a company which is producing electronic products in Taiwan. The company must make decisions such as green supplier selection and warehouse to be select, and must satisfy total market demand and capacity constraints. The problem is a multi-product, multi-stage supply chain network problem. Fig. 1 is the problem concept diagram. Recently, many researchers study multi-object optimization of supply chain network. Erol and Ferrell (2004) developed a multiobjective optimization modeling framework which can simultaneously solve two conicting objectives such as minimizing cost and maximizing customer satisfaction. And they adopted this model to select suppliers of the supply chain. Chan and Chung (2004) developed a multi-criterion genetic algorithm for an order distribution in a demand driven supply chain, which considered three objectives such as cost, delivery lead time and equity of utilization

Fig. 1. Problem concept map.

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

4247

ratios. At last, a set of Pareto solutions were obtained through this algorithm. Chan, Chung, and Wadhwa (2005) proposed a hybrid genetic algorithm for production and distribution problem, which utilized AHP to construct these criteria and calculate the tness value of chromosome, and considered three criteria such total cost, fulllment lead time and equity on utilization ratios. Farahani and Elahipanah (2008) adopted a hybrid non-dominated sorting genetic algorithm to optimize total cost and service level for JIT distribution in a supply chain, whose results were compared with Lingo software to evaluate the performance of proposed algorithm. The multi-objective optimization problems as given above usually contain a lot of optimal solutions which are Pareto-optimal solutions. The concept formulated by Coello, Veldhuizen, and Lamount (2001) is dened as follows: When a minimization problem and two vectors ~ u u1 ; u2 ; . . . ; um and ~ v v 1 ; v 2 ; . . . ; v m are considered. If ~ u would say to dominate ~ v (denoted by ~ u<~ v , if and only if

8i 2 fi; . . . ; mg;

ui 6 v i ^ 9i 2 f1; . . . ; mg : ui < v i

where m is the dimension of the objective space. All decision vectors are not dominated by another decision vector of a given set, which are called non-dominated set or Pareto-optimal set. There are many solution methods for solving multi-objective problems such as goal attainment, goal-programming, distancebased method, weighting method, direction-based method and so on. Recently, many researchers utilized GA to obtain Pareto-optimal solutions for multi-objective problems (Coello, Van Veldhuizen, & Lamount, 2002; Deb, 2001; Gen & Cheng, 2000). This paper adopted two weighting methods proposed by Murata et al. (1996) and Altiparmak et al. (2006) for solving mixed-integer non-linear problem, because the weighting sum method can generate more Pareto-optimal solutions than other algorithms. We formulated a multi-objective mixedinteger non-linear programming model that contained four objectives and nine constraints in this model. Four objectives are minimization of the total cost, minimization of the total time, maximization of average product quality and maximization of green appraisal scores. The assumptions used in this problem are: (1) supposing that demand amount of all demand sides is known, (2) maximum producing capacity and minimum boot-strap producing capacity exist in accessory ordering. The suppliers would accept order fore on condition that ordering amount is between maximum producing capacity and minimum boot-strap producing capacity, (3) transportation cost is determined by per transportation cost but transportation amount, (4) producing losses may result in inequality between accessory amount provided by suppliers and demand amount from customers in each phase of supply chain, (5) as for the losses, the model considers not only product yield rate but also transportation loss rate. First introduce parameters used in mathematical model below. Sets and indices: x Supply chain stage No., x = 1, 2, 3, . . ., X y Partner y of stage x, y = 1, 2, 3, . . ., Y z Partner z of stage x, z = 1, 2, 3, . . ., Z n The product number of the supply chain system, n = 1, 2, 3, . . ., N Variables:  FSNx,y 1 : if supply chain partner y of stagexhas operation 0 : others This variable determines whether of partner y of stage x will be selected or not. The quantity transported from of partner y of stage x Qn x;yx1;z to partner z of stage x + 1 among the n-th product

Parameters: K(x,y)(x+1,z) Transportation loss rate from of partner y of stage x to partner z of stage x + 1 Lx,y Minimum operational production capacity of partner y of stage x Ux,y Maximum operational production capacity of partner y of stage x kRk Integral value of value R Jx,y Product yield rate of partner y of stage x Partner y of stage x among the nth product unit PC n x;y manufacturing cost Partner y of stage x among the nth product unit PT n x;y manufacturing time TC(x,y)(x+1,z) Unit transportation cost of partner y of stage x to partner z of stage x + 1 BTx,y Batch transportation time of partner y of stage x Market demand of the nth product on the partner k of Dn x;k stage x AQx,y Partner y of stage x among the nth product average product quality Sx,y Green appraisal scores of partner y of stage x  1 : customers going on with purchase G1 x;y 0 : others Customers purchase or not of partner y of stage x. 8 1 : above 80% G2 > x;y > > < 2 : 60% to 80% 3 : 40% to 60% > > > : 4 : 20% to 40% 5 : less than 20% Green customers market share of partner y of stage x. 8 1 : pass G3 > x;y < 2 : more than 50% > : 3 : less than 50% 4 : Not yet Passing ISO14000 verication of partner y of stage x 8 1 : have done G4 > x;y < 2 : more than 50% > : 3 : less than 50% 4 : None Having environmental protection policies of partner y of stage x 8 1 : have done G5 > x;y < 2 : more than 50% > : 3 : less than 50% 4 : None Having environmental protection plans of partner y of stage x 8 1 : have done G6 > x;y < 2 : more than 50% > : 3 : less than50% 4 : None Having recycling product design of partner y of stage x 8 > G7 < 1 : have done x;y 2 : more than 50% > : 3 : less than 50% 4 : None Having renewable product design of partner y of stage x 8 1 : more than 80% > G8 > x;y > < 2 : 70% to 80% 3 : 60% to 70% > > > : 4 : 50% to 60% 5 : less than 50% Product recycling percentage of partner y of stage x 8 1 : have done > G9 > x;y < 2 : more than 50% > 3 : less than 50% > : 4 : None Having reverse logistics system of partner y of stage x

4248

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

G10 x;y G11 x;y G12 x;y G13 x;y G14 x;y G15 x;y G16 x;y G17 x;y G18 x;y G19 x;y G20 x;y G21 x;y G22 x;y G23 x;y G24 x;y

Solid wastes of partner y of stage x Energy consumption of partner y of stage x Air pollution of partner y of stage x Waste water of partner y of stage x Led content of electrical and electronic equipments of partner y of stage x Mercury content of electrical and electronic equipments of partner y of stage x Hexavalent chromium content of electrical and electronic equipments of partner y of stage x Polybrominated biphenyl (PBB) content of electrical and electronic equipments of partner y of stage x Polybrominated diphenyl ether (PBDE) content of electrical and electronic equipments of partner y of stage x Cadmium content of electrical and electronic equipments of partner y of stage x Solid wastes treatment costs of partner y of stage x Chemical wastes treatment costs of partner y of stage x Air pollution treatment costs of partner y of stage x Water pollution treatment costs of partner y of stage x Energy consumption costs of partner y of stage x

Y X y 1

kJ x;z 1 K x;yx1;z Q n xyx1;z k

L4 X l1

Qn x1;zx2;l 9 10

for x x 2; x 3; . . . 1; z 1; 2; . . . ; Z ; n 1; 2; . . . ; N G8 x;y < 4; for all x; y 6 1000; for all x; y

15 16 17 18 G14 x;y ; Gx;y ; Gx;y ; Gx;y ; Gx;y

11 12 13 14

G19 x;y 6 100; Qn x;yx1;z P

for all x; y 0 and in integer; for all x; y; z; n for all x; y

FSNx;y 2 f0; 1g;

Optimum mathematical programming model of defective supply chain network: Objective function:

Eqs. (2)(5) gives the objectives. In order to the experimental result was understood easily, we will transform the last two objectives of four objectives into minimization. While (2) denes the total cost of the green supply chain, (3) denes the total time of the green supply chain, (4) and (5) give the objectives about product average product quality and appraisal scores, respectively. Constraints (6) and (7) represent the limitation of partners production capacity, (8) and (9) give the satisfaction of customer and warehouses demands for the product, (10) imposes the product recycling rate of partner y of stage x must be less than 4, (11) imposes stage is supplier js electrical and electronic equipments led, mercury, cadmium and hexavalent chromium, PBB, PBDE contents shall be less than 1000 ppm, (12) imposes stage is supplier js electrical and electronic equipments cadmium contents shall be less than 100 ppm, (13) imposes the non-negative constraint on decision variable Q n x;yx1;z , (14) imposes the integrality constraint on decision variable FSNx,y. Since the rst and second objectives are non-linear, the model given is a mixed-integer non-linear programming model. 4. Proposed methodology 4.1. Develop the evaluation process for green supply chain partners

Min f 1

( " X 1 X Y N X X
x1 y1 n1

Z X N X TC x;yx1;z Q n x;yx1;z

# PZ x1;zn z1 Q x;y PC n x;y J x;y ) 2 # PZ n z1 Q x;yx1;z J x;y !) 3

Min f 2

( " X 1 X Y N X X
x1 y1 n1

z1 n1

PT n x;y

BT x;y Min f 3 Min f 4

Z X N X z1 n1

Qn x;yx1;z !

According to the result of the literature review, we developed the evaluation process for green supply chain partners. The selection criteria consist of two types such as quantity and quality principles. Quantity principles include capacity, yield, production cost etc.; Quality principles include average product quality, passing ISO14000 verication etc. 4.2. Collect relative information for supply chain partners

X X Y N X X x1 y1 X X Y X n1

AQ x;y FSNx;y

Sx;y FSNx;y
24 X q 1

Collect relative information of each supply partner includes production cost, transportation time, yield, air pollution treatment cost etc., and developed the database for customer information. 4.3. Use two algorithms to solve the optimum mathematical planning model This paper adopted two algorithms proposed by Murata et al. (1996) and Altiparmak et al. (2006) to solve the optimum mathematical planning model. We will refer to these algorithms with the abbreviations MOGA_1, MOGA_2, respectively. 4.3.1. MOGA_1 MOGA_1 was proposed by Murata et al. (1996), which must repeat eight steps in each generation. In the selection procedure, MOGA_1 used a weighted sum of multi-objective function to combine them into a scalar tness function. Therefore, the weights of the multiple objective functions were not constant but randomly specied for each selection. MOGA_1 can search various directions as shown in Fig. 2.

x1 y1

in which; Sx;y
Z X N X z1 n1

wq Gq x;y

s:t: Lx:y FSN x;y 6

Qn x;yx1;z 6 U x;y FSN x;y ; 6

for x 1; y 1; 2; . . . ; Y Lx;z FSNi;z 6

Y X N X y 1 n 1

Qn x;yx1;z 6 Usx;z FSN x;z ; 7

for x 2; 3; j 1; 2; . . . ; J
Y X y 1 n kJ x1;z 1 K x;yx1;z Q n x;yx1;z k Dx1;z ;

for x x 1; z 1; 2; . . . ; Z ; n 1; 2; . . . ; N

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

4249

Step 1: Initialization Generate an initial population which must satisfy all constraints (Eqs. (6)(14)). Step 2: Evaluation Calculate the objective function of each chromosome, for instance, Eqs. (2)(5). When the weighted sum approach was used, rstly, objective functions must be normalized because they have different measure units. Each objective used Eq. (15) to normalize.

where

fmin w minff xjx 2 wg

19

fi0

fi fimin max fi fimin

i 1; 2; . . . ; n

15

where fimax and fimin were the maximum and the minimum value of ith objective function. After rst iteration, must update a tentative set of Paretooptimal solutions. Step 3: Selection Calculate the tness value f(x) of each string by using Eqs. (16) and (17).

f x w1 f1 x w2 f 2x wn fn x

16

where x is a chromosome, f(x) is a combined tness function, fi(x) is the ith objective function, wi is a constant weight for fi(x), and n is the number of the objective function.

Step 4: Crossover Apply the crossover operation to generate an offspring with the crossover probability, as shown in Fig. 3. Step 5: Mutation Apply the mutation operation with the mutation probability, as shown in Fig. 3. Step 6: Elitist strategy In multi-objective optimization problem, the best value of each objective was regarded as an elitist solution. This paper hopes to preserve these solutions to the next generation. Therefore, the elitist strategy was to remove n chromosomes form current population, and added the same number of strings from a tentative set of Pareto-optimal solutions. Step 7: Termination test Until the iteration times was reached, or repeated Steps 2 to 6. Step 8: Obtain the nal set of Pareto-optimal solutions Through the multi-objective optimization algorithm, we can obtain the set of Pareto-optimal solutions for MOGA_1, which offer decision makers to select the best one.

rndi wi Pn ; j1 rndj

i 1; 2; . . . ; n

17

where rndi and rndj are non-negative random integers. Select a pair of chromosomes from the population according to the following selection probability. The selection probability was calculated by adopting Eqs. (18) and (19).

P x P

f x fmin w x2w ff x fmin wg

18

f 2(x) : To be minimized
Fig. 4. The crossover mechanism of MOGA_2 (Altiparmak et al., 2006).

Non-dominated solutions

Fig. 2. Directions of the search by MOGA_1 (Murata et al., 1996).

Fig. 5. The mutation mechanism of MOGA_2 (Altiparmak et al., 2006).

Fig. 3. The crossover and mutation mechanism of MOGA_1 (Murata et al., 1996).

4250

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

Fig. 6. {4, 4, 4, 4} Supply chain network.

Table 1 Demand volumes of each product of each customer. Type Customers 4.1 4.2 4.3 4.4 A 50 60 70 80 B 60 90 70 50 C 80 60 70 80 D 70 70 50 60

4.3.2. MOGA_2 MOGA_2 was proposed by Altiparmak et al. (2006), which also used the weight approach of MOGA_1 to determine the weight values, and proposed a new solution procedure to nd the set of Pareto-optimal solutions. Steps of MOGA_2 were illustrated in the following: Step 1: Initialization Generate an initial population which must satisfy all constraints (Eqs. (6)(14)). Step 2: Evaluation Calculate the objective function of each chromosome, for instance, Eqs. (2)(5). Step 3: Genetic operators Sub-step 3.1: Selection First, use Eq. (15) to normalize the objective function, and calculate the tness value f(x) of each string by using Eqs. (16) and (17). Select a pair of chromosomes according to the selection probability was calculated by adopting Eqs. (18) and (19).

Sub-step 3.2: Crossover MOGA_2 adopted a segment-based crossover operator. As shown in Fig. 4, 0 means that the parent 1 will transfer its gene to the offspring, then 1 means that the parent 2 will transfer its gene to the offspring. Sub-step 3.3: Mutation Similar to the crossover mechanism, this step also utilized the segment-based mutation mechanism as the mutation. Fig. 5 illustrated the mutation mechanism of MOGA_2. Step 4: Update the Pareto-optimal set of each generation. Step 5: Selection mechanism MOGA_2 utilized the (p+q) selection strategy as a selection mechanism where p and q represent the number of parents and offering in the current generation, respectively. Sub-step 5.1: Select two individuals from the set of Pareto-optimal solutions at random, and place them into the population. Sub-step 5.2: The population was lled by (p 2) different individuals from the evolving pool. If there were no (p 2) different individuals, the vacant pool were lled by generating individuals randomly. Step 6: Termination test Until the iteration times was reached, or repeated Steps 2 to 5. Step 7: Obtain the nal Pareto-optimal set. Through the multi-objective optimization algorithm, can obtain the set of Pareto-optimal solutions for MOGA_2, which offer decision makers to select the best one.

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

4251

Fig. 7. Quality and quantity criteria for green supplier selection.

5. Simulation results and discussion The proposed of this paper was to determine combinations of supply chain partners and production volumes transportation in

the supply chain. And adopting two multi-objective optimization algorithms to solve this problem, nally, can obtain the set of Pareto-optimal solutions offering to the decision to select. We introduced a {4, 4, 4, 4} supply chain network to illustrate the concept

Table 2 Comparison of MOGA_1 and MOGA_2. Parameters Type Problem 1 Problem 2 Problem 3 Problem 4 Algorithm MOGA_1 MOGA_2 MOGA_1 MOGA_2 MOGA_1 MOGA_2 MOGA_1 MOGA_2 Average number of Pareto-optimal solutions 5.90 2.60 7.80 4.90 5.70 2.70 3.00 1.50 CPU time (s) 35.47 134.18 48.00 141.83 34.88 127.00 94.70 194.10 Generation size 100 100 150 100 100 100 100 100 Population size 200 200 200 200 200 200 200 200 Crossover probability 0.7 0.8 0.5 0.8 0.8 0.8 0.8 0.9 Mutation probability 0.03 0.02 0.03 0.01 0.03 0.03 0.03 0.03

4252

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

Fig. 8. Pareto-optimal solutions of MOGA_1 and MOGA_2 for Problem 1.

Fig. 10. Pareto-optimal solutions of MOGA_1 and MOGA_2 for Problem 3.

Table 3 Examples for Pareto-optimal solutions of MOGA_1 for Problem 4. Solutions 1 2 3 4 (f1, f2, f3, f4) (196983, 158605, 71, 31674) (201814, 180232, 52, 24505) (225079, 202030, 40, 21660) (229443, 207179, 37, 17119)

Fig. 9. Pareto-optimal solutions of MOGA_1 and MOGA_2 for Problem 2.

of the system, as shown in Fig. 6. Each supplier has different yield, unit transportation cost, minimum and maximum capacity, and each route may be transportation routes. This supply chain network considered four products produced by the supply chain network, and Table 1 offers demand volumes of each product of each customer. Fig. 7 offers the framework of green supplier selection. In this paper, all experimental/design platforms were coded with Visual Basic 6.0 programming language and run on Intel Core 2, 1.86 GHz and 0.99 GB memory. In order to evaluate performances of MOGA_1 and MOGA_2 in the supply chain network problem with two and four objectives, we will divide the original problem into four problems. The four problems were different from the original problem, which only considered different objectives but all constrains were the same. The four problems were dened as follows: Problem 1. min f1 and min f3 Problem 2. min f1 and min f2 Problem 3. min f1 and min f4 Problem 4. min f1, min f2, min f3 and min f4

Experimental results are shown in Table 2. All parameters had been determined after preliminary experiments. MOGA_1 and MOGA_2 run 10 times according to these parameters for each problem. We utilized the indicator such as average number of Pareto-optimal solutions and CPU time to evaluate performances of MOGA_1 and MOGA_2. These results reected in Table 2 on each problem indicated that MOGA_1 not only obtained more number of Pareto-optimal solutions than MOGA_2, but also cost CPU time was faster than MOGA_2. As shown in Fig. 8, many solutions denoted by MOGA_2 are dominated by solutions by MOGA_1, so the set of Pareto-optimal solutions obtained by MOGA_1 is superior to the set of Paretooptimal solutions obtained by MOGA_2. Figs. 9 and 10 also support this result for Problems 2 and 3, respectively. Table 3 presents four Pareto-optimal solutions of MOGA_1 for Problem 4.

6. Conclusion In this paper, we utilized two multi-objective genetic algorithms for solving green supplier selection and production volumes transportation problems. Four objectives were considered: (1) minimization of total cost comprised of product cost and transportation cost, (2) minimization of total time comprised of product time and transportation time, (3) maximization of average product quality, (4) maximization of green appraisal score. Then we adopted the weighted sum approach to obtain the set of Paretooptimal solutions offering the decision maker to evaluate some better of alternative solutions. In order to evaluate the performance of two genetic algorithms, called as MOGA_1 and MOGA_2, we compared two indicators of each algorithm for four problems: (1) average number of Pareto-optimal solutions, (2) CPU time. Experimental results showed that MOGA_1 not only generates

W.-C. Yeh, M.-C. Chuang / Expert Systems with Applications 38 (2011) 42444253

4253

more number of Pareto-optimal solutions than MOGA_2 but also obtains better solutions than MOGA_2. Thence, MOGA_1 was superior to MOGA_2. In future, we hope to develop a modied weight sum approaching to obtain more Pareto-optimal solutions than previous studies. Additionally, uncertainty of costs and demands can be considered in this model, and a new solution methodology can be developed. References
Altiparmak, F., Gen, M., Lin, L., & Paksoy, T. (2006). A genetic algorithm approach for multi-objective optimization of supply chain networks. Computers and Industrial Engineering, 51(1), 197216. Chan, F. T. S., & Chung, S. H. (2004). A multi-criterion genetic algorithm for order distribution in a demand driven supply chain. International Journal of Computer Integrated Manufacturing, 17(4), 339351. Chan, F. T. S., Chung, S. H., & Wadhwa, S. (2005). A hybrid genetic algorithm for production and distribution. Omega, 33(4), 345355. Chung, K. J., & Hou, K. L. (2003). An optimal production run time with imperfect production process and allowable shortages. Computers and Operations Research, 30(4), 483490. Coello, C. A. C., Van Veldhuizen, D. A., & Lamount, G. B. (2002). Evolutionary algorithms for solving multi-objective problems. New York: Kluwer. Coello, C. A. C., Veldhuizen, D. A. V., & Lamount, G. B. (2001). Evolutionary algorithms for solving multi-objective problems. Kluwer Academic Publishers. Deans, I. (1999). An approach to the environmental management of purchasing in the utilities sector. Eco-Manage, 6, 1117. Deb, K. (2001). Multi-objective optimization using evolutionary algorithms. Chichester: Wiley. Erol, I., & Ferrell, W. G. Jr, (2004). A methodology to support decision making across the supply chain of an industrial distributor. International Journal of Production Economics, 89(2), 119129. Farahani, R. Z., & Elahipanah, M. (2008). A genetic algorithm to optimize the total cost and service level for just-in-time distribution in a supply chain. International Journal of Production Economics, 111(2), 229243. Gen, M., & Cheng, R. (2000). Genetic algorithms and engineering optimization. New York: Wiley. Hammond, D., & Beullens, P. (2007). Closed-loop supply chain network equilibrium under legislation. European Journal of Operation Research, 183(2), 895908. Handeld, R., Steven, V. W., Sroufe, R., & Melnyk, S. A. (2002). Applying environmental criteria to supplier assessment: A study in the application of the Analytical Hierarchy Process. European Journal of Operational Research, 141(1), 7087. Humphreys, P. K., Wong, Y. K., & Chan, F. T. S. (2003). Integrating environmental criteria into the supplier selection process. Journal of Materials Processing Technology, 138(13), 349356.

Kainuma, Y., & Tawara, N. (2006). A multiple attribute utility theory approach to lean and green supply chain management. International Journal of Production Economics, 101(1), 99108. Klein, M. (1966). Markovian decision models for reject allowance problems. Management Science, 12(5), 349358. Lee, H. L., & Rosenblatt, M. J. (1986). Economic production cycles with imperfect production processes. IIE Transactions, 18, 4855. Lee, H. L., & Yano, C. A. (1988). Production control in multistage systems with variable yield losses. Operations Research, 36(2), 269278. Levitan, R. E. (1960). The optimum reject allowance problem. Management Science, 6(2), 172186. Liu, J. J., & Yang, P. (1996). Optimal lot-sizing in an imperfect production system with homogeneous reworkable jobs. European Journal of Operation Researches, 91(3), 517527. Murata, T., Ishibuchi, H., & Tanaka, H. (1996). Multi-objective genetic algorithm and its applications to owshop scheduling. Computers and Industrial Engineering, 30(4), 957968. New, C., & Mapes, T. (1984). MRP with high uncertain yield losses. Journal of Operations Management, 4(4), 315330. Noci, G. (1997). Designing green vendor rating systems for the assessment of a suppliers environmental performance. European Journal of Purchasing & Supply Management, 3(2), 103114. Porteus, R. L. (1986). Optimal lot sizing, process quality improvement, and setup cost reduction. Operations Research, 34(1), 137144. Queiruga, D., Walther, G., Gonzalez-Benito, J., & Spengler, T. (2008). Evaluation of sites for the location of WEEE recycling plans in Spain. Waste Management, 28(1), 181190. Rose, M. (2000). Design for environment: A method for formulating product end-of-life strategies, Department of Mechanical Engineering, Stanford University. (pp. 19 144). Rose, C., Beiter, K., & Ishii, K. (1999). Determining of end-of-life strategies as a part of product denition. In Proceedings of the IEEE international symposium for electronics and the environment (pp. 219224). Sarkis, J. (2003). A strategic decision framework for green supply chain management. Journal of Cleaner Production, 11(4), 397409. Vachon, S., & Klassen, R. D. (2006). Green project partnership in the supply chain: The case of the package printing industry. Journal of Cleaner Production, 14(67), 661671. Wang, H. S., Che, Z. H., & Sha, D. Y. (2004). Two-stage model for ow-shop production planning with different machines, In Proceedings of the ninth annual international conference on industrial engineering Theory, applications and practice, The University of Auckland, Auckland, New Zealand (pp. 299304). White, L. S. (1967). Bayes Markovian decision models for a multi-period reject allowance problem. Operations Research, 15(5), 857865. Willems, B., Dewulf, W., & Duou, J. R. (2006). Can large-scale disassembly be protable? A linear programming approach to quantifying the turning point to make disassembly economically viable. International Journal of Production Research, 44(6), 11251146.