a dynamic decision approach for supplier selection using ant colony system

Expert Systems with Applications 37 (2010) 8313–8321

Contents lists available at ScienceDirect

Expert Systems with Applications

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

A dynamic decision approach for supplier selection using ant colony system

Ya Ling Tsai a,1, Yao Jung Yang b,*, Chi-Hsiang Lin c,2

a Department of Marketing and Logistics Management No. 1, Nan Tai St., Yung Kang City, Tainan County 71005, Taiwanb Department of Applied Information, Hsing-Kou University, No. 600, Sec. 3, Taijiang Blvd., Annan District, Tainan 709, Taiwan, ROCc Logistics Management Department, Southern Taiwan University, Department of Information Management, Taiwan, ROC

a r t i c l e i n f o

Keywords:Supplier selectionACS

0957-4174/$ - see front matter � 2010 Elsevier Ltd. Adoi:10.1016/j.eswa.2010.05.053

* Corresponding author.E-mail addresses: [email protected] (Y.L.

edu.tw, [email protected] (Y.J. Yang), chlin@m1 Address: No. 26, Lane 350, Yu-Nung Road, Tainan2 Address: No. 46, Wei-Guo Street, Tainan City 7014

a b s t r a c t

Purpose: This study based on the attribute-based ant colony system (AACS) to construct a platform toexamine the critical factors for decision making in a dynamic business environment in order to selectthe appropriate suppliers.Design/methodology/approach: This study focuses on how to search for optimal suppliers in a similar fash-ion to how the optimal route can be found. The AACS is based on the ant colony system (ACS) algorithm,which is then modified to achieve the adaptive optimal system used to set the policy for companies toselect their suppliers, as the researcher (as like source node) and chosen supplier’s attributes to be con-ditions of research (destination node).Findings: At first, we provide the development of policy model and can effective and immediately tochoose the best suppliers from the company’s policy and the attribute of suppliers. Secondly, this policysystem is based on the platform of AACS and also modifies the new heuristics algorithm.Research limitations/implications: There are two limitations with this study. First, the criteria for the policyand attribute numbers and sequence for suppliers must be same. Secondly, the score has evaluated by thebuyer company before the decision group to use which one policy.Practical implications: The value of this study divides two points; the parameters of AACS platform areadjustment for the buyer decision policy from dynamically business environment and the AACS can findan optimal solution from the decision policy.Originality/value: AACS according to the decision group’s policy to enter parameters in order to find theadaptive solution for buyer business firm to find their finest suppliers.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Over the past decade, the need to gain global competitivenesson the supply side has increased substantially. Particularly forcompanies that spend a high portion of their sales revenue onraw material and component parts, savings from reduction in unitprices became much more important as their material costs take alarger percentage of total costs. Obviously selection of the rightsuppliers plays a key role in any organizations because it signifi-cantly reduces the unit prices and improves corporate price com-petitiveness. Selection of the right suppliers can improve a firm’scompetitive advantage, as suppliers are key participants within asupply chain channel, able to affect the quality and the price ofthe final goods that a business offers its customers. Consequently,the issue of supplier selection has attracted much attention within

ll rights reserved.

Tsai), [email protected] (S.C.-H. Lin).

City 701, Taiwan, ROC.8, Taiwan, ROC.

the field of ‘supply chain management, and most approachesexamine the problem based on several criteria, such as quality,price, service, performance and so on. However, emphasis on qual-ity and timely delivery, in addition to the cost consideration, in to-day’s globally competitive marketplace adds a new level ofcomplexity to supply selection decisions. In practice, there couldbe several criteria used by a firm for its supplier selection decision,such as price offered, part quality, on-time delivery, after-sales ser-vices, supplier location and supplier’s financial status. Apparently,supplier selection is a multi-criteria problem, which includes bothquantitative and qualitative factors. For the firm to select the bestsuppliers, it is necessary to make trade-off between these tangibleand intangible factors. Traditionally, decision group (purchasingteams) used such methods as supplier rating or supplier assess-ments in order to choose suppliers from the candidate supplier list.These methods assessed suppliers based on a selected number ofcriteria in a linear manner. Facing the new challenges in the supplychains, however, a buyer now faces multiple objectives to achievesimultaneously in its purchasing decision. Quality of parts, deliveryreliability, financial status and other criteria as well as price shouldnow be taken into account in selecting the best suppliers. In this

http://dx.doi.org/10.1016/j.eswa.2010.05.053

mailto:[email protected]





http://dx.doi.org/10.1016/j.eswa.2010.05.053

http://www.sciencedirect.com/science/journal/09574174

http://www.elsevier.com/locate/eswa

8314 Y.L. Tsai et al. / Expert Systems with Applications 37 (2010) 8313–8321

paper, an approach to model development and analysis of the sup-plier selection problem is presented. The proposed approach,which based on AACS to implement a framework for help buyerschoose the most appropriate suppliers in a dynamic environment.

The paper is organized as follow. First, we provide a review ofsupplier selection literature to show how the research fits withexisting research. Next, we proposed an algorithm based on the cri-teria attributes and AACS to apply optimal supplier selectionsearch platform, and then, computational experiences on theimplementation of proposed approach along with a case studyare presented. Finally, make a conclusion.

2. Relative work

Supplier selection has been discussed for more than 30 years(Micheli, Cagno, & Zorzini, 2008) and is also a popular topic withinthe field of supply chain management, as firms aim to choose theright suppliers in order to raise their competitive abilities (Hsu,Kannan, Leong, & Tan, 2006). Additionally, buyers choose morethan one supplier to match their requirements (Ndubisi, Jantan,Hing, & Ayub, 2005; Ting & Cho, 2008), and these are chosen notonly based on specific orders but also on their abilities to helpthe buyers achieve their future goals (Ting & Cho, 2008). In addi-tion to this, Lo and Yeung (2006) also point out that buyers needto consider customer requirements when selecting a supplier, asthis will lead to increased customer satisfaction. However, thereis a gap in the current literature with regard to making such deci-sions based on maximizing customer satisfaction, and this studywill address this issue by considering how the right supplier canbe chosen on this basis in a dynamic and rapidly changing businessenvironment?

2.1. Decision policy

Although many studies have discussed the issue of supplierselection (Bhutta & Huq, 2002; Chou & Chang, 2008; Garfamy,2006; Kirytopoulos, Leopoulos, & Voulgaridou, 2008; Ramanathan,2007; Sevkli, Koh, Zaim, Demirbag, & Tatoglu, 2008; Teng & Jara-millo, 2005; Ting & Cho, 2008; Wang, 2008; Yang & Chen, 2006),most of them focus on price, quality, services, delivery time, sup-plier location, supplier financial statues and performance.

From the late 1970s, American firms tended to focus on priceand quality to increase their competitive abilities (Huang, Uppal,& Shi, 2002). Meanwhile, since the 1980s, Japanese firms have beenusing the principles of total quality management (TQM) to monitortheir product quality (Huang et al., 2002) and to improve their sup-ply chain operations (Ndubisi et al., 2005).

Those of settings are able to explain the decision policy of sup-plier selection. The reason is that buyers need to consider manyconditions for supplier selection. It is estimation for buyer to verifytheir supplier performance. On the other hand, it is also can assistthem to do a good customer service and get the advantage for thecompetitive market.

Gill and Ramaseshan (2007) indicated that few scholars discussthe performance during the purchasing processes or consider it asa significant factor in supplier selection. They divided this perfor-mance into five parts: (1) relationship commitment, (2) productquality, (3) price, (4) payment facilities, and (5) brand recognition.Moreover, they found that supplier commitment can strengthenthe business relationship and this can then be reflected in im-proved product quality, price, and payment facilities, and thus in-crease the likelihood of repurchases.

The total cost of ownership (TCO) is a way to measure the com-plete purchase cost (Bhutta & Huq, 2002; Garfamy, 2006), and itcan be broken down into four main parts: manufacturing costs

(material, labour, and so on), quality costs (quality control, andso on), technology costs (design and engineering) and after-salesservice costs. In addition to this, it is also necessary to considerother costs, such as those associated with research, transportation,and order-placement. Consequently, in this study, we divide thetotal cost into manufacturing, research, transportation, and or-der-placement costs.

Yang and Chen (2006), noted that ‘‘Manufacturers, therefore, re-quire suppliers to have effective systems in production managementand quality control”. These effective systems are needed to managethe quality and quantity of the products manufactured, and firmsnow use technology and diagrams and make a process for theproducts management. Lo and Yeung (2006) note that supplierquality management is a essential part of TQM that can make sup-ply channels more effective, enhance the relationships among re-lated firms and improve their performance. It is also can expandthe supplier development within supply chain management fromthe supplier quality (Nwankwo, Obidigbo, & Ekwulugo, 2005).With regard to supplier service, this includes the factors of deliverytime, after-sales service and customer relationship service. Therelationship between a buyer and supplier should be like a partner-ship, and the ideal supplier should be able to help their buyer todesign new products from the goods that they provide.

2.2. Ant colony optimization system

Ant colony optimization (ACO) is a metaheuristic in which a col-ony of artificial ants cooperates in finding good solutions to difficultoptimization problems. A metaheuristic is a set of algorithmic con-cepts that can be used to define heuristic methods applicable to awide variety of problems. The use of metaheuristics has signifi-cantly increased the possibility of finding high quality solutions todifficult, practically relevant combinatorial optimization problemswithin a reasonable time, and the first ACO algorithm, called theant colony system (ACS), was successfully applied in tacking thewell-known traveling salesman problem (Dorigo & Gambardella,1997).

The ACS is based on agents that simulate the natural behav-iour of ants, develop mechanisms for cooperation, and assist themin using experience (Dorigo & Gambardella, 1997) to find theshortest path between a food source and the nest. ACS is a pop-ulation-based heuristics that exploits something similar to thepositive feedback that takes place when ants are able to commu-nicate information concerning food sources via pheromones, in aprocess of indirect communication that is called stimergy in bothant and technological contexts. Ants lay a pheromone and heuris-tic information to mark trails. As the paths are visited by otherants, some of the trails may be reinforced and other paths maybe allowed to evaporate. Pheromone trails can be observed viathe number of ants passing through the trail. When there aremore pheromones on a path, there is larger probability that otherants will use that path and therefore, the pheromone trail on sucha path will grow faster and attract more ants to follow (so calledpositive feedback). An iterative local search algorithm tries tosearch the current paths to neighboring paths until a better solu-tion is found.

Many researchers have worked with ACO to extend its algo-rithm and develop more sophisticated models. The elitist ant sys-tem (EAS) was introduced in Dorigo (1992) and Dorigo,Maniezzo, and Colorni (1991, 1996), based on the principle of pro-viding strong additional reinforcement to the arcs belonging to thebest tour found since the start of the algorithm. Note that this extrafeedback can be viewed as supplementary pheromone depositedby an additional ant called the best-so-far ant. Arank-based ant sys-tem was proposed by Bullnheimer, Hart1, and Strauss (1999), inwhich each ant deposits a pheromone along with its rank. Cordon,

Y.L. Tsai et al. / Expert Systems with Applications 37 (2010) 8313–8321 8315

Herrera, Fernandez de Viana, and Moreno (2000) presented a sys-tem which uses the transition rule and pheromone evaporationmechanism to improve the ants’ solutions. In addition, Semet, Ja-mont, Biojout, Lutton, and Collet (2003) Semet, Lutton, and Collet(2003) applied the ACO heuristics to an e-learning pedagogic mate-rial navigation problem, and Jamont et al. (2005) experimentedwith an ‘‘ant-hill” method which laid the pheromone dependingon how students validated an item. More recently, Yang and Wu(2008) proposed an AACS for optimal learning object recommenda-tion, so as to optimize learning paths with different students whohave different views. These varied applications demonstrate thatthe ACO method is well suited for tackling adaptive learning objectsearches in a dynamic learning environment.

3. Methodology

Based on the literature review, this study considers the decisionpolicy for supplier selection by including the following factors:price, costs, transportation costs, order-placement costs, productquality, JIT and TQM, after-sales service, customer relationship ser-vice, supplier location, supplier financial status, payment facilities,

Fig. 1. The decision flow fro AACS pl

brand recognition, and performance. Of these, cost, service, andqualities are considered to be the three main elements. The deci-sion flow we propose for the AACS platform is shown in Fig. 1.The buyers rate the suppliers on the attributes listed above, withthe following scores: Very Bad (VB), Bad (B), Good (G), Very Good(VG), and Excellent (E). The decision group then sets the parame-ters within the AACS platform in order to obtain the adaptive solu-tion for supplier selection.

The duty of the decision group is to put the criteria into a policypool for a dynamic environment, set weight to policy and get a pol-icy for AACS to search for the optimal suppliers.

Step 1: To chose the best criteria for supplier selection. The bestcriteria come from the conditions of supplier selection. They areincluding cost, quality. . ., and service. The criteria can change inorder to face the dynamic environment.Step 2: This step is to decide which criteria and then set theweight to them. The weight to divide to five parts as: verybad, bad, good, very good, and excellent. Each of them has thescore weight (for example: very bad for 1, bad for 2, etc.). Thedecision group need to decide the score weight for each crite-rion and then put in the policy pool.

atform to get optimal suppliers.


Step 3: The final step is to get a best police for AACS in order tosearch for the optimal suppliers. It is a kind standard for AACSto choose the optimal suppliers. For example, the police decidecost is good, service is very good, and quality is excellent. It is acondition for the AACS to choose the optimal suppliers forbuyer business firms.

The function of AACS is to find the optimal suppliers. Every sup-plier has their own score from the supplier score list by the purchasedepartment of buyer business. The AACS search the suppliers’ scorewhich is more approach the optimal suppliers’ policy score.

3.1. Search rule and pheromone update rule for supplier search

The ‘‘attributes” ant mainly uses an adaptive rule to search forthe optimal suppliers in a dynamically change environment. Fol-lowing this rule, the system can improve the quality of the phero-mone as well as help learners easily find their own adaptivelearning object. In this paper, we redefine two adaptive rules fromAACS to reinforce the pheromone update, as follows:

Rule 1: IF ((Policy’s criteria MATCH the Supplier’s attributes)OR(Policy’s criteria PARTIALLY MATCH the Supplier’s attributes))THEN Daemon action with MRCriteria.Rule 2: IF ((Policy’s criteria NOT MATCH the Supplier’s attributes)AND (Policy’s criteria NOT PARTIALLY MATCH the Supplier’sattributes )) THEN Do Nothing.

There is a formal definition for the adaptive rule of a ‘‘MATCH”and ‘‘PARTIALLY MATCH” that we define for the algorithm:

1. IF (Policy’s criteria = Supplier’s attribute) AND (Policy’s criteriaweight = Supplier’s attribute weight) THEN we call it a ‘‘MATCH”.

2. IF ((Policy’s criteria = Supplier’s attribute) AND (Policy’s criteriaweight <> Supplier’s attribute weight)) OR ((Policy’s crite-ria <> Supplier’s attribute) AND (Policy’s criteria weight = Sup-plier’s attribute weight)) THEN we call it a ‘‘PARTIALLY MATCH”.

In AACS, whether or not the heuristic information can be rein-forced relies on the ‘‘attribute” ant, which is formally defined as fol-lows: k ant, which has one of the policy criteria and its own policycriteria weight, generates a fixed amount of pheromone when ittravels along a node. If a k ant’s attributes ‘‘MATCH” or ‘‘PARTIALLYMATCH” the node attributes (supplier’s attributes), then the nodeswhich are on the path created by the k ant may obtain an extrapheromone after each time unit.

3.1.1. The decision model based on the AACS algorithmThe decision model and algorithm we propose are extensions of

AACS, and the parameters and functions used in this paper are thesame as those defined earlier for AACS. However, we applied twonew rules for heuristic information and pheromone trail updating,as follows.

3.2. The heuristic information

In AACS, the heuristic value gij is a normalized value function ofthe queue length qij (the supplier waiting to be processed) on thenode connecting node i with its neighbor j. The heuristic informa-tion is defined as follows:

gij ¼ 1� ðMRij � dijÞXNi

l¼1

qil

, !

where MRij is the match ratio between the node i with its neighbor j.In this paper, we replaced the heuristic information with the short-

en distance (given in Eq. (1)) between node i with its neighbor j. Thenew heuristic information is defined as follows:

dij ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxi1 � yj1Þ

2 þ ðxi2 � yj2Þ2 þ � � � þ ðxin � yinÞ

2q

ð1Þ

gij ¼ 1� ðqij � dijÞXNi

l¼1

qil

, !ð2Þ

where (xi1,xi2, . . .,xin) 2 P, (yj1,yj2, . . .,yjn) 2 S, gij gives a quantitativemeasure associated with the node waiting time and the distance be-tween the nodes, and a higher value of gij means there is a nodewith higher probability of being chosen, and to get the approxi-mately optimal suppliers.

3.3. Pheromone trail updating

The relational strength between the ith node and the jth node isthe pheromone trail intensity sij. The incremental intensity Dsij(t)is that which locates pheromone trail value at the time t, and it isupdated as the following formula:

sijðtÞ ¼ qsijðt � 1Þ þ DsijðtÞ ð3Þ

where q is the evaporation ratio of the trail at an interval time unit.If an ‘‘attribute” ant has chosen the jth node after locating node iand laid its pheromone trail, the pheromone levels on the jth nodeshould be updated and the contributions of all ‘‘attribute” ants, andthe amounts of the pheromone laid by the ants are defined asbelow,

sijðtÞ ¼ qsijðt � 1Þ þ Ds Criteriaij ðtÞ þ DsWeight

ij ðtÞ ð4Þ

where DsCriteriaij (t) and DsWeight

ij (t) are the variable amounts of phero-mone deposited in the arc (i, j), and these represent the candidatesupplier’s attribute and attribute weight ‘‘MATCH” or ‘‘PartiallyMatch” the policy criteria and criteria weight. Thus, the numberof attributes ants has to be considered, and the adaptive solutionis as follows:

Dsk;Criteriaij ðtÞ ¼

Xm

n¼0

ðm� nÞ � Q �MRk;Criteriaij ðtÞ ð5Þ

where m is the number of MRCriteriaij ðtÞ on the arc (i, j)

Dsk;Weightij ðtÞ ¼

Xm

n¼0

ðm� nÞ � Q �MRk;Weightij ðtÞ ð6Þ

where c is a constant number to adjust the actual learning objectsearch situation, m is the number of MR Criteria

ij ðtÞ on the arc(i, j),MRCriteria is the match ratio for the policy criteria and supplier’s attri-bute, and MRWeight is the match ratio for the policy criteria weightand supplier attribute weight, and the definition of match ratio isas follows:

LetX8j2Ni

MRij ¼X8j2Ni

MRCriteriaij þ

X8j2Ni

MR Weightij ð7Þ

where

MR Weightij ¼ ðXi � YjÞX8r2Ni

ðXi � YrÞ þ 1

!,

MR Criteriaij ¼ ðUi � VjÞX8r2Ni

ðUi � VrÞ þ 1

!,

and

MRCriteriaij ¼ e�MR Criteriaij ð8Þ

MRWeightij ¼ e�MR Weightij ð9Þ

!,Y.L. Tsai et al. / Expert Systems with Applications 37 (2010) 8313–8321 8317

where X is the current policy criteria weight, Y is the current supplierattribute weight, U is the current policy criteria, and V is the currentsupplier attribute.

3.4. Algorithm for supplier selection

The proposed AACS algorithm has five main procedures forresolving the problem of adaptive paths, and the pseudo-code ofthe attribute-based ant colony system is shown as follows:

Procedure Attribute_ACS_for_Supplier_Selection/* mainprocedure*/Initialize_Parameter;set Policy pool P = {p1,p2, . . .,pn}, which is decided by thedecision group.get pi from Policy pool P = {p1,p2, . . .,pn} for i = 1, . . .,n;set criteria weight for policy pi

pi{c1 = wk,c2 = wk, . . .,cm = wk} where i = 1, . . .,n; k = 1, . . .,5and wk 2W;get policy pn and policy’s criteria weight from Policy pool P;get candidate supplier’s attribute and weight from supplierattribute set N;

while (condition not terminate) doConstruct_SolutionHeuristic_Decision_RuleUpdate_Pheromone_Trailsif (MATCH or PARTIALLY MATCH) then Daemon_Actionsend while

Get_Optimal_SupplierEnd procedure

procedure Initial_Parameterset the Q for trail constant intensity, heuristic factor, a, band evaporation rate q, and setting the pheromone trails toa initial value s0 > 0;set criteria weight set W = {VB,B,G,VG,E}Initialize the weight of the policy criteria and supplier’sattribute as follows;set VB = 1, B = 2, G = 3, VG = 4, E = 5;set criteria set C = {c1,c2, . . .,cm};

end Initial_Paramete

procedure Construct_SolutionSelect a start node s1 from supplier candidate set Sc and putit into the construct node list sL and get candidate the

supplier j from the neighbor nodes Nki

while (sL R Sc and Nki –/Þ do

j Select_Next_Node (sL,Sc)put j into sL;

end whileif sL 2 Sc then return sL

else abortend-ifend Construction_Solution

procedure Heuristic_Decision_Rule/* set variable and values for the weight */let X = Current_policy_criteria_weight for X # W,let Y = Current_supplier_attribute_weight for Y # W,let U = Current_policy_criteria where U # C.let V = Current_supplier_attribute for V # C.

/* Calculate the Match Ratio between the source node (thepolicy node) and */

/* destination node (the supplier node). */

Computing MR Weightij ¼ ðXi�YjÞ=P8r2NiðXi�YrÞþ1

� �

MR Criteriaij ¼ ðUi � VjÞX8r2Ni

ðUi � VrÞ þ 1

MRCriteriaij ¼ e�MR Criteriae;

MRWeightij ¼ e�MR Weight

letP8j2Ni

MRij ¼P8j2Ni

MRCriteriaij þ

P8j2Ni

MRWeightij

for each sij do

get candidate supplier j from neighbor nodes Nki , compute

the heuristic information gij. /* first calculate the shortestlength between the node i to its neighbor node j. */

dij ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxi1 � yj1Þ

2 þ ðxi2 � yj2Þ2 þ � � � þ ðxin � yinÞ

2q

where (xi1,xi2, . . .,xin) 2 P, (yj1,yj2, . . .,yjn) 2 S,

gij ¼ 1� ððqij � dijÞ=PNi

l¼1qilÞ/* Define the decision probability model */

PkijðtÞ ¼

½sijðtÞ�a ½gijðtÞ�bP

l2Nki½silðtÞ�a ½gilðtÞ�

b ; 8j 2 Nki ; k 2 1 � m

end for eachend Heuristic_Decision_Rule

procedure Update_Pheromone_Trailsset the adding pheromone with evaporate factorDsij (1 � q) � sij "(i, j) 2 S where 0 < q 6 1,computing all ants deposit pheromonesij sij þ

Pmk¼1Dsij; 8ði; jÞ 2 S

end Update_Pheromone_Trails

procedure Daemon_Actionsfor each (attributes ant MATCH the rule)computing global pheromone update with attribute matchratio

Dsk;Criteriaij ðtÞ ¼

Xm

n¼0

ðm� nÞ � Q �MRk;Criteriaij ðtÞ

Dsk;Weightij ðtÞ ¼

Xm

n¼0

ðm� nÞ � Q �MRk;Weightij ðtÞ

sijðtÞ ¼ qsijðt � 1Þ þ DsCriteriaij ðtÞ þ DsWeight

ij ðtÞ

end for eachend Daemon_Actions

procedure Get_Optimal_SupplierList the supplier candidates from node list sL

Return Optimal Suppliers listend Get_Optimal_Supplier

3.5. Transition probability of attribute ants

The transition probability PkijðtÞ given in procedure Heuris-

tic_Decision_Rule is defined as a compromise value of sij and gij,where thek ant moves forward from a supplier node i to the nextsupplier node j, j 2 Nk(i). The role of parameter a and b is as fol-lows: If a = 0, this corresponds to a classic stochastic greedy algo-rithm. If b = 0, only pheromone amplification is at work, that is,only a pheromone is used, without any heuristic bias. This gener-ally leads to poor results, in particular, for values of a > 1 it leadsto the rapid emergence of a stagnation situation, that is, a situationin which all the ants follow the same path and construct the sametour, which, in general, is strongly suboptimal (Dorigo, 1992; Dor-igo et al., 1996).


Below are brief descriptions of the main procedure in the AACSalgorithm pseudo-code and an example of its search mechanism.

Initialize_Parameter: the procedure which initializes the param-eters of AACS, Q,a,b,q, policy attribute, supplier attribute, policy’sattribute weight, and supplier’s attribute weight.

Construct_Solution: the procedure for constructing the solu-tion, where Sc is the set of candidate optimal supplier solutions,and Nk

i is set of the candidate nodes next to i for ant k, and the

Table 1A case for policy p1(Cost, Service,Quality ) = (VB,VG,E) vs supplier attributes.

Policy pool Policy 1: p1(Cost,Service,Quality) = p1(VB,VG,E) ? p1(1,4,

Item Policy criteria Supplier at

Criteria/attribute Cost Service Quality Cost

Supplier 1 1 4 5 1Supplier 2 1 4 5 3Supplier 3 1 4 5 1Supplier 4 1 4 5 2Supplier 5 1 4 5 2

Supplier 1(Cost,Service,Quality) = (VB,G,G) ? S1(1,3,3),Supplier 3(Cost,Service,Quality) = (VB,VG,E) ? S3(1,4,5),Supplier 5(Cost, Service, Quality) = (B,VG,B) ? S5(2,4,2)

Fig. 2. Example for AACS se

random walk of ants is biased by pheromone trails which areassociated with a connection between nodes i and j. The ant kselects the next node j from the neighbor list based on the Se-lect_Next_Node function, which mainly focuses on the transitionprobability value Pk

ij (as shown in the Heuristic_Decision_Rule pro-cedure, below), after that, the next node j from the candidatesupplier nodes is chosen and then put into the optimal solutionsupplier node list sL.

5)

tribute Heuristic Match ratio

Service Quality d g MR

3 3 2.236 0.254 0.7756924 5 2.000 0.227 0.7969214 5 0.000 0.000 1.0000003 5 1.414 0.160 0.8521444 2 3.162 0.359 0.698374

Supplier 2(Cost,Service,Quality) = (G,VG,E) ? S2(3,4,5)Supplier 4(Cost,Service,Quality) = (B,G,E) ? S4(2,3,5)Supplier’s attributes

arch optimal supplier.


Heuristic_Decision_Rule: this procedure computes the heuristicinformation gij and match ratio, where MRCriteria is the match ratiofor policy criteria, and MRWeight is the match ratio for the attributeweight.

Update_Pheromone_Trails: this procedure computes the generalpheromone update sij after evaporation, where Dsij is the amountof pheromone ant k deposits on the arcs it has visited.

Daemon_Actions: the procedure which computes the globalpheromone update value Dsij, where DsWeight

ij ðtÞ and Ds Criteriaij ðtÞ

are the variable amounts of pheromone deposited in the arc(i, j),and this represents the policies which ‘‘MATCH” the supplier’sattributes.

Get_Optimal_Supplier: the procedure which gets the optimalsupplier candidates from solution nodes list sL, and then returnsit to the system.

In order to further explain the idea of the ‘‘attribute” ant of AACS,we through an example to illustrate the search rule. As shown inFig. 2, this study uses cost, service, and quality as the rules for sup-plier selection, when the decision group gets policy p1 from the pol-icy pool, AACS searches the n suppliers (S1,S2,S3, . . .,Sn�1,Sn) fromthe source node p1(Cost = VB, Service = VG, Quality = E) to its neigh-bor nodes (the suppliers), the next n node, for example, nodeS1(Cost = VB, Service = G, Quality = G), node S2(Cost = G, Service = E,Quality = VG), node S3(Cost = VB, Service = VG, Quality = E), nodeSn�1(Cost = B, Service = G, Quality = E) and node Sn(Cost = B, Ser-vice = VG, Quality = B), and so on. Before the research, the criteriaand weight are set according to the sequence cost, service, andquality and transformed into mathematical coordinates, as follows:

p1ðCost¼VB;Service¼VG;Quality¼EÞ! p1ðVB;VG;EÞ! p1ð1;4;5ÞS1ðCost¼VB;Service¼G;Quality¼GÞ! S1ðVB;G;GÞ! S1ð1;3;3ÞS2ðCost¼G;Service¼E;Quality¼VGÞ! S2ðG;E;VGÞ! S2ð3;5;4ÞS3ðCost¼VB;Service¼VG;Quality¼ EÞ! S3ðVB;VG;EÞ! S3ð1;4;5ÞSn�1ðCost¼B;Service¼G;Quality¼ EÞ! Sn�1ðB;G;EÞ! Sn�1ð2;3;5ÞSnðCost¼B;Service¼VG;Quality¼BÞ! SnðB;VG;BÞ! Snð2;4;2Þ

Table 2A case for policy p2(Cost, Service,Quality ) = (G,E, B) vs supplier attributes.

Policy pool Policy 2: p2(Cost,Service,Quality) = p2(G,E,B) ? p2(3,5,2)

Item Policy criteria Supplier att



Supplier 1(Cost,Service,Quality) = (VB,G,G) ? S1(1,3,3),Supplier 3(Cost,Service,Quality) = (VB,VG,E) ? S3(1,4,5),Supplier 5(Cost,Service,Quality) = (B,VG,B) ? S5(2,4,2)

Table 3A case for policy p3(Cost, Service,Quality ) = (VG,G, B) vs supplier attributes.

Policy pool Policy 3: p3(Cost,Service,Quality) = p3(VG,G,B) ? p3(4,3,2)

Item Policy criteria Supplier att



Supplier 1(Cost,Service,Quality) = (VB,G,G) ? S1(1,3,3),Supplier 3(Cost,Service,Quality) = (VB,VG,E) ? S3(1,4,5),Supplier 5(Cost,Service,Quality) = (B,VG,B) ? S5(2,4,2)

In this example, the supplier S3’s (Cost = VB, Service = VG,Qual-ity = E) attribute, weight and policy p1(Cost = VB, Service = VG,Qual-ity = E) match. This means the node (p1 and S3) distance d13 = 0, theother one as like p1 and S1, p1 and Sn�1 are partially match. The tworst situation is p1 and S2, p1 and Sn do not match. The AACS isaccording to the decision probability mode to do the optimal selec-tion. In this example, S3 has more opportunity to be selected withinthe optimal supplier candidate list. In addition to this, it is also needto match the other factor to make the final decision, such as q, s, q,a, b can affect the parameter of decision probability model.

4. Case study

The decision makers choose policy p1(Cost = B, Service =VG,Quality = E) from the policy pool (as shown in Table 1) which in-cludes three criteria, Cost, Service, and Quality, and set the weight forthese. AACS will apply an optimal search rule to get an optimal sup-plier relying on policy criteria and supplier’s attributes. Each policymust be transformed into a vector pair before using the AACS searchalgorithm, for example, if the decision group gets the policyp1(Cost,Service,Quality) = p1( VB,VG,E) ? p1(1,4,5), then the AACSsearch rule gets ‘‘MATCH” or ‘‘PARTIALLY MATCH” informationfrom candidate supplier’s attributes, and AACS calculates the heu-ristic (parameters d and g) and match ratio (the MR value), the glo-bal daemon pheromone update will change the s value accordingthe high MR value and reinforce the search path in order to getthe optimal suppliers which accommodate the firm policy. In Table1, the supplier 3 attribute totally ‘‘MATCH” the policy criteria(MR = 1), the MR value is closer to 1, meaning that the supplierattribute and weight more closer to the policy criteria and weight.Otherwise, the MR value more far away from 1 meaning that thesupplier attribute and weight unavailable to the policy criteriaand weight. The supplier 1, supplier 2, supplier 4, and supplier 5are ‘‘PARTIALLY MATCH” to the policy criteria weight. Obviously,AACS will put the supplier 3 into the optimal supplier candidate list.

ribute Heuristic Match ratio


3 3 3.000 0.199 0.8195504 5 3.162 0.210 0.8105844 5 3.742 0.248 0.7803603 5 3.742 0.248 0.7803604 2 1.414 0.094 0.910283


ribute Heuristic Match ratio


3 3 3.162 0.190 0.8269594 5 3.317 0.199 0.8195504 5 4.359 0.261 0.7702813 5 3.606 0.216 0.8057354 2 2.236 0.134 0.874590


Table 4A case for policy p4(Cost, Service,Quality,Financial) = (VG, G, B, G) vs supplier attributes (4 policy criteria and 4 supplier attributes).

Policy pool Policy 4: p4(Cost,Service,Quality,Financial) = (VG,G,B,G) ? p4(4,3,2,3)

Item Policy criteria Supplier attribute Heuristic Match ratio

Criteria/attribute Cost Service Quality Financial Cost Service Quality Financial d g MR

Supplier 1 4 3 2 3 1 3 3 4 4.162 0.113 0.893151Supplier 2 4 3 2 3 3 5 4 2 6.449 0.175 0.839457Supplier 3 4 3 2 3 1 4 5 2 12.317 0.334 0.716054Supplier 4 4 3 2 3 2 3 5 3 11.000 0.298 0.742301Supplier 5 4 3 2 3 2 4 2 5 3.000 0.081 0.922194

Supplier 1(Cost,Service,Quality,Financial) = (VB,G,G,VG) ? S1(1,3,3,4), Supplier 2(Cost,Service,Quality,Financial) = (G,VG,E,B) ? S2(3,4,5,2)Supplier 3(Cost,Service,Quality,Financial) = (VB,VG,E,B) ? S3(1,4,5,2), Supplier 4(Cost,Service,Quality,Financial) = (B,G,E,G) ? S4(2,3,5,3)Supplier 5(Cost,Service,Quality,Financial) = (B,VG,B,E) ? S5(2,4,2,5) Supplier’s attributes

Table 5A case for policy p5(Cost, Service,Quality,Financial) = (B,G,G, E) vs supplier attributes (4 policy criteria and 4 supplier attributes).

Policy pool Policy 5: p5(Cost,Service,Quality,Financial ) = (B,G,G,E) ? p5(2,3,3,5)

Item Policy criteria Supplier attribute Heuristic Match ratio

Criteria/attribute Cost Service Quality Financial Cost Service Quality Financial d g MR

Supplier 1 2 3 3 5 1 3 3 4 1.414 0.066 0.936131Supplier 2 2 3 3 5 3 5 4 2 4.742 0.221 0.801717Supplier 3 2 3 3 5 1 4 5 2 7.317 0.341 0.711059Supplier 4 2 3 3 5 2 3 5 3 6.000 0.279 0.756540Supplier 5 2 3 3 5 2 4 2 5 2.000 0.093 0.911194

Supplier 1(Cost,Service,Quality,Financial) = (VB,G,G,VG) ? S1(1,3,3,4), Supplier 2(Cost,Service,Quality,Financial) = (G,VG,E,B) ? S2(3,4,5,2)Supplier 3(Cost,Service,Quality,Financial) = (VB,VG,E,B) ? S3(1,4,5,2), Supplier 4(Cost,Service,Quality,Financial) = (B,G,E,G) ? S4(2,3,5,3)Supplier 5(Cost,Service,Quality,Financial) = (B,VG,B,E) ? S5(2,4,2,5) Supplier’s attributes


As shown in Table 2, supplier 2 and supplier 5 are ‘‘PARTIALLYMATCH” to the policy p2(Cost,Service,Quality) = p2(G, E,B) ?p2(3,5,2), the others were not match. In this study, the MR valueof supplier 2 and supplier higher than their other suppliers. TheMR value is near one mean the Daemon actions of performs proce-dure increase the amount of pheromone Ds. In other words, it canenhance the pheromone on the search path, and more have thechance to be the list of optimal supplier candidates.

As shown in Table 3, supplier 1 and supplier 5 are ‘‘PARTIALLYMATCH” to the policy p3(Cost,Service,Quality) = p3(VG, G,B) ?p3(4,3,2), the others were not match. In this case, the MR valueof supplier 1 and supplier 5 higher than the other suppliers. Andsupplier 2, supplier 3, and supplier 4 more get more the amountof pheromone and increase the opportunity within the list of opti-mal supplier candidate.

Given as shown in Tables 4 and 5, policy p4(Cost,Service,Qual-ity,Financial) = p4(VG,G, B,G) ? p4(4,3,2,3) and policy p5(Cost,Ser-vice,Quality,Financial) = p5(B,G,G,E) ? p5(2,3,3,5) are four criteria.

The attributes of supplier need to prepare the four attributes.For AACS platform, the criteria of policy and attribute of supplierare opposite so that the number of them do not affect the factorsfor AACS to choose the optimal supplier. The reason is the numberof policy criteria or supplier attribute for AACS is about the differ-ence of math dimension and does not affect the logic of algorithmand only increase the complex of calculation. For example, the sup-plier within Table 4, Table 4 has four attributes but the function ofsearch within AACS also can find the best policy immediately.

From Table 4 can see the MR value of supplier 5 and policyp4(Cost,Service,Quality,Financial) = p4(VG,G,B,G) ? p4(4,3,2,3)higher than the other candidates. Although, supplier 1, supplier 4,and supplier 5 to p4(Cost,Service,Quality,Financial) = p4(VG,G,B,G)? p4(4,3,2,3) are ‘‘PARTIALL MATCH”, and according to AACS pol-icy model, the heuristic factor g and pheromone s has the impor-tant role for the decision model.

The q number of supplier within waiting queue and the factor gof heuristic is an important parameter. Form supplier 5, the valueof waiting queue is smaller than supplier 1 and supplier 4.Although, supplier 1, supplier 4, and supplier 5 have match ‘‘PAR-TIALL MATCH”, but the value of waiting queue q difference so thattheir MR value is different.

5. Conclusion

The supplier selection problem not only to select the right sup-pliers but also to select the optimal suppliers, based on a number ofkey criteria such as costs, quality, and service, etc. For these pur-poses, the decision making procedure based on AACS platformwas developed in this paper. First, we provide the developmentof policy model and can effective and immediately to choose thebest suppliers from the company’s policy and the attribute of sup-pliers. Secondly, this decision system is based on the platform ofAACS and also modifies the new heuristics and adaptive rule to en-hance the search algorithm. Our models have several advantages:

� They can dynamically change the policy criteria and supplierattributes.� The AACS platform can dynamically and efficiently determine

the optimal supplier.� The AACS platform is also variable as the rapidly and vigorously

circumstances from the buyer decision policy in order to helpthe buyer firm to choose their finest supplier.

In the future, we will integration with related database in theAACS platform, such the buyer database, inventory database, andsupplier database. These databases will help the buyer evaluatethe policy criteria and potential suppliers faster, and make moreefficient and effective decision.


References

Bhutta, K. S., & Huq, F. (2002). Supplier selection problem: A comparison of the totalcost of ownership and analytic hierarchy process approaches. Supply ChainManagement: An International Journal, 7(3), 125–135.

Bullnheimer, B., Hart1, R. F., & Strauss, C. (1999). A new rank-based version of theant system: A computational study. Central European Journal for OperationsResearch and Economics, 7(1), 25–38.

Chou, S.-Y., & Chang, Y.-H. (2008). A decision support system for supplier selectionbased on a strategy-aligned fuzzy SMART approach. Expert Systems withApplications, 34(4), 2241–2253.

Cordon, O., Herrera, F., Fernandez de Viana, I., & Moreno, L. (2000). A new ACOmodel integrating evolutionary computation concepts: The best-worst antsystem. In Proceedings of the 2nd international workshop on ant algorithms,Brussels, Belgium (pp. 22–29).

Dorigo, M. (1992). Optimization, learning and natural algorithms. PhD thesis,Dipartimento di Elettronica, Politecnico di Milano, Milan.

Dorigo, M., & Gambardella, L. M. (1997). Ant colony system: A cooperative learningapproach to the traveling salesman problem. IEEE Transaction on EvolutionaryComputation, 1(1), 53–66.

Dorigo, M., Maniezzo, V., & Colorni, A. (1991). Positive feedback as a search strategy.Technical report 91-016 revised, Dipartmento di Elettonica, Politecnico diMilano, Milan.

Dorigo, M., Maniezzo, V., & Colorni, A. (1996). Ant system: Optimization by a colonyof cooperating agents. IEEE Transactions on System, Man, and Cybernetics – Part B,26(1), 29–41.

Garfamy, R. M. (2006). A data envelopment analysis approach based on total cost ofownership for supplier selection. Journal of Enterprise Information Management,19(6), 662–678.

Gill, D., & Ramaseshan, B. (2007). Influences on supplier repurchase selection of UKimporters. Marketing Intelligence and Planning, 25(6), 597–611.

Hsu, C. C., Kannan, V. R., Leong, G. K., & Tan, K. C. (2006). Supplier selectionconstruct: Instrument development and validation. The International Journal ofLogistics Management, 17(2), 213–239.

Huang, S. H., Uppal, M., & Shi, F. (2002). A product driven approach tomanufacturing supply chain selection. Supply Chain Management: AnInternational Journal, 7(4), 189–199.

Jamont, Y., Valigiani, G., Collect, P., Lutton, E., Biojout, R., & Bourgeois Republique, C.(2005). Experimenting with a real-size man-hill to optimize pedagogical paths.In SAC’05 proceedings (pp. 13–17). Santa Fe, NM, USA: Mars.

Kirytopoulos, K., Leopoulos, V., & Voulgaridou, D. (2008). Supplier selection inpharmaceutical industry an analytic network process approach. Benchmarking:An International Journal, 15(4), 494–516.

Lo, V. H. Y., & Yeung, A. (2006). Managing quality effectively in supply chain: Apreliminary study. Supply Chain Management: An International Journal, 11(3),208–215.

Micheli, G. J. L., Cagno, E., & Zorzini, M. (2008). Supply risk management vs supplierselection to mange the supply risk in the EPC supply chain. ManagementResearch News, 31(11), 846–866.

Ndubisi, N. O., Jantan, M., Hing, L. C., & Ayub, M. S. (2005). Supplier selection andmanagement strategies and manufacturing flexibility. The Journal of Enterpriseinformation Management, 18(3), 330–349.

Nwankwo, S., Obidigbo, B., & Ekwulugo, F. (2005). Allying for quality excellence:Scope for expert systems in supplier quality management. International Journalof Quality and Reliability Management, 19(2), 187–205.

Ramanathan, R. (2007). Supplier selection problem: Integration DEA with theapproaches of total cost of ownership and AHP. Supply Chain Management: AnInternational Journal, 12(4), 258–261.

Semet, Y., Jamont, Y., Biojout, R., Lutton, E., & Collet, P. (2003). Artificial ant colonyand e-learning: An optimization of pedagogical paths. HCI.

Semet, Y., Lutton, E., & Collet, P. (2003). Ant colony optimization for elearning:Observing the emergence of the pedagogic suggestions. In IEEE swarmintelligence symposium, April (pp. 46–52).

Sevkli, M., Koh, S. C. L., Zaim, S., Demirbag, M., & Tatoglu, E. (2008). Hybrid analyticalhierarchy process model for supplier selection. Industrial Management andSystems, 108(1), 122–142.

Teng, S. G., & Jaramillo, H. (2005). model for evaluation and selection of suppliersand global textile and apparel supply chains. International Journal of PhysicalDistribution and Logistics Management, 35(7), 503–523.

Ting, S. C., & Cho, D. I. (2008). An integrated approach for supplier selection andpurchasing decisions. Supply Chain Management: An International Journal, 13(2),116–127.

Wang, H. S. (2008). Configuration change assessment: Genetic optimizationapproach with fuzzy multiple criteria for part supplier selection decisions.Expert Systems with Applications, 34(2), 1541–1555.

Yang, C. C., & Chen, B. S. (2006). Supplier selection using combined analyticalhierarchy process and grey relational analysis. Journal of ManufacturingTechnology Management, 17(7), 926–941.

Yang, Y. Y., & Wu, C. (2008). An attribute-based ant colony system for adaptivelearning object recommendation. Expert Systems with Applications, 36(2),3034–3047 (Part 2).

a dynamic decision approach for supplier selection using ant colony system

Documents