iBusiness, 2010, 2, 333-341
doi:10.4236/ib.2010.24043 Published Online December 2010 (http://www.scirp.org/journal/ib)
A Study of Multi-Agent Based Supply Chain Modeling and
Management
WanSup Um1, Huitian Lu2, Teresa J. K. Hall2
1Gangneung-Wonju National University, Gangneung, Korea; 2South Dakota State University, Brookings, USA.
Received June 15th, 2010; revised October 13th, 2010; accepted November 20th, 2010.
ABSTRACT
Supply Chain Management (SCM) is a management paradigm to understand and analyze the flow of goods, services
and the accompanying values reaching to the consumers followed by the processes of purchasing, production and dis-
tribution with combining and connecting the whole system. Today, SCM is regarded as an essential strategic factor
which has a great deal of influence on earning competitiveness in the abruptly changing global business environment.
Multi-agent technology becomes the best candidate for problem solver under these circumstances. An agent performs
given tasks automatically using inter-collaboration or negotiation with other agents on behalf of a human on the basis
of real-time connectivity.
There will be the conflict among the pursuit of the profit of all members of the SCM. In order to maximize the total
profit of the SCM, negotiation among all members is necessary. In this research, we propose to find the best negotiation
strategy that makes all members of the SCM satisfied in a simple SCM. We suggest a new negotiation algorithm in the
SCM environment with using multi-agent technology. The ideas behind the suggested model are negotiation algorithm
with a trading agent and we consider multiple factors that are price, review point and delivery time. We created agents
with Java Agent Development Framework (JADE) and performed the simulation under JADE and Eclipse environment.
The case study denotes that our algorithm gives a better result than the Kasbah system that is a typically well known
system where users create autonomous agents that buy and sell goods on their behalf. We’ve used benefit/cost ratio as a
performance measure in order to compare our system with the Kasbah system.
Keywords: Supply Chain Management, Multi-Agent, Trading Agent, JADE, Eclipse
1. Introduction
The supply chain is a worldwide network of suppliers,
factories, warehouses, distribution centers, and retailers
through which raw materials are acquired, transformed,
and delivered to customers. In order to optimize per-
formance of a supply chain, its functions must operate in
a coordinated manner. But the dynamics of the enterprise
and the market make this difficult: materials are delayed
in shipment, production facilities experience downtime,
workers call in sick, customers change orders or cancel,
and other issues cause deviations from the plan. In the
global marketplace with shortening product life cycles
and fast changing trends, the need for real time supply
chain coordination is vital. Information technology and
information sharing make coordination possible. The
major contribution of information technologies such as
the Internet is to enable many companies to make contact
with customers directly without time zone or distance
intervention. Collaborations in supply chains cannot be-
governed by any single company in a one-directional
way, but need to be coordinated by autonomous partici-
pation of companies. For these reasons, agent technology
is regarded as one of the best candidates for supply chain
management.
To optimize supply chain decisions, an agent cannot
by itself make a locally optimal decision rather it must
determine the effect its decisions will have on other
agents and coordinate with others to choose and execute
an alternative that is optimal over the entire supply chain.
In dealing with stochastic events, the agents must make
optimal decisions based on complex global criteria that
are not completely known by any one agent and may be
contradictory, therefore require trade-offs. Internet tech-
nologies have contributed significantly to e-commerce
by increasing the mutual visibility of consumers and
suppliers, and by raising the possibility that some of their
trading processes may be automated. Despite these ad-
vances, most procurement activities within supply chains

A Study of Multi-Agent Based Supply Chain Modeling and Management 334
are still based on static long-term contracts and relation-
ships. In many cases, such contracts are detrimental be-
cause they fail to handle the dynamic nature of these
environments.
To rectify this, we believe agent-based solutions are
needed. This research suggests a multi-agent system for
distributed and collaborative supply chain management.
Multi-agent technology has many beneficial features for
autonomous, collaborative, and intelligent systems in
distributed environments, which makes it one of the best
candidates for complex supply chain management. A
new negotiation algorithm for SCM system with a trad-
ing agent is proposed to make the most appropriate deci-
sion using multiple attributes for buyer demand. We cre-
ated agents with Java Agent Development Framework
(JADE) and performed the simulation with JADE and
Eclipse environments. The results indicate the simulated
environment had a higher rate of return than a traditional
negotiated exchange.
To date, however, the use of agents within e-com-
merce has generally focused on simple auctions. But the
supply chain domain typically requires handling a much
more complex setting where decisions must be made in
the presence of greater degrees of uncertainty. To this end,
the International Trading Agents Competition for Supply
Chain Management (TAC SCM) represents an ideal envi-
ronment in which to test the autonomous agents.
The remainder of the paper is organized as follows:
Section 2 describes a review of the literature; Section 3
presents the framework for agent development; Section 4
suggests the modeling of the problem under study; Sec-
tion 5 provides a case study; and finally, Section 6 con-
cludes this paper and outlines the areas of future work.
2. Literature Review
In supply chain management, improving the efficiency of
the overall supply chain is of key interest. Because of
market globalization and the advancement of e-com-
merce the importance of supply chain network is in-
creasing [1]. It is very difficult for different companies in
supply chains to share information. A supply chain can
produce products for multiple markets. Also, an individ-
ual company is likely to have only limited visibility of
the supply chain structure, which makes it difficult to
make future demand estimations, because the pattern of
demand propagation through the supply chain depends
on the capabilities and strategies of companies along the
path from the markets to the company.
These problems are further amplified if the supply
chain changes over time dynamically. As a result of
these problems, individual companies are likely to make
inaccurate demand estimations and the supply chain can
suffer from the well-known Bullwhip effect [2]. The
Bullwhip effect refers to the problem where the fluctua-
tions of productions and inventory levels are amplified in
the upstream parts of supply chains than in the down-
stream parts. The bullwhip effect was first observed by
Forrester [2], and has been studied further by Lee et al.
[3]. One of the solutions proposed to deal with the bull-
whip effect is to have information sharing across the
companies in the supply chain.
There are unique characteristics required for informa-
tion systems that support supply chain management. First,
they should be able to support distributed collaboration
among companies. Second, a single company cannot
govern collaborations in supply chains in a one-direc-
tional way, but need to be coordinated by autonomous
participation of companies. Third, they need a high level
of intelligence for planning, scheduling, and change ad-
aptation. For these reasons, agent technology is regarded
as one of the best candidates for supply chain manage-
ment [4].
Since the mid 1990s, the agent concept has emerged in
the literature relevant to computer applications. Agents
may have many other properties. Agents can exhibit au-
tonomy, social ability and responsiveness, in addition to
adaptability, mobility, and rationality [5]. Studies on
agent-based supply chain management can be classified
into three categories. The first type of research is con-
cerned with the coordination aspect. In this type of re-
search, various types of companies and their capabilities
are modeled into individual agents and their interactions
are designed for efficient collaboration [6]. The second
type of research focuses on simulation of supply chains
using agent-based models. This type of research tries to
discover the performance of agent-based supply chain
architectures under various strategies and constraints [7].
The third type of research studies how virtual supply
chains can be organized flexibly by multi-agent systems
[8]. For example, research by Chen et al. [8] showed how
virtual supply chains can be formed by solving distributed
constraint satisfaction problems by agents.
Multi-agent technology has many beneficial features for
autonomous, collaborative, and intelligent systems in dis-
tributed environments, which makes it one of the best can-
didates for complex supply chain management [7]. Recent
research literature acknowledges intelligent agents as the
most appropriate technology for trading and auctioning in
electronic markets [9]. A software agent is viewed as an
encapsulated computer system that is capable of flexible
autonomous action in that environment in order to meet its
design objective [10]. In automated negotiations, the agents
prepare bids and evaluate offers in order to obtain the
maximum return for the parties they represent [11]. Such
automated negotiation leads to dynamic pricing which en-
sures that goods and services are allocated to the entity that

A Study of Multi-Agent Based Supply Chain Modeling and Management 335
values them most highly [12]. Optimal control deals with
the problem of finding a control law for a given system
such that a certain optimality criterion is achieved. A con-
trol problem includes a cost functional that is a function of
system state and control variables.
3. Agent Development Framework
3.1. Agent
Since the mid 1990s, the agent concept is increasingly
emerging in study relevant to computer applications.
Agents may have goals and an ability to plan based on
their goals. They may be able to execute actions based on
their goal-directed plans, monitor their environment to
determine the effects of their actions, analyze the extent
to which their actions brought about the desired changes
in the environment, and replan their actions when neces-
sary to reach their goals. Agents may have many other
properties. Agents can exhibit social ability, responsive-
ness in addition to adaptability, mobility, and rationality
[5]. Furthermore, agents may have both high-level and
low-level reasoning capabilities [13] and the actions that
result from these capabilities may be influenced by in-
tentions and beliefs. Also, agents provide a modular or
object-oriented modeling framework from the point of
system development’s view.
In recent years, a new software architecture for man-
aging the supply chain at the tactical and operational
levels has emerged. It views the supply chain as com-
posed of a set of intelligent software agents, each re-
sponsible for one or more activities in the supply chain
and each interacting with other agents in planning and
executing their responsibilities. In order to carry out their
common project such as the supply chain planning in a
decentralized supply chain environment, agents are de-
signed to undertake several different tasks by means of
cooperation or negotiation among themselves.
3.2. Multi-Agent System (MAS)
The dynamics of the supply chain makes coordinated be-
havior an important factor in its integration. To optimize
supply chain decisions, an agent cannot just make a locally
optimal decision by itself, but must determine the effect of
its decisions on other agents and coordinate with others to
choose and execute an alternative that is optimal over the
entire supply chain. The problem is exacerbated by the
stochastic events generated by the flow of new objects into
the supply chain. These include modifications to customer
orders at the customer’s request, resource unavailability
from suppliers, and machine breakdown all drive the sys-
tem away from any existing predictive schedule. Agents
operate within organizations where humans must be rec-
ognized as privileged members. This requires knowledge
of organization roles and respecting the obligations and
authority incurred by the roles.
Recent works on the dynamics of industrial systems
and supply chains have attempted to describe the net-
works of relationship that characterize contemporary
businesses’ trading situations and internal functional
structures. In modeling these relations, research has in-
creasingly turned to frameworks derived from multi-
agent system (MAS). The concept of agents and of
MASs emerged from a number of research disciplines
including artificial intelligence, systems design and
analysis using object-oriented methodology and human
interfaces.
Agents send and receive messages concerning their
current situations to agents in other related or same sys-
tem, and display evolutionary behavior in response to
changes. Within the MAS, different types of agents have
different degrees of problem solving capabilities within
different problem domains. MAS architectures vary ac-
cording to the complexity of problem domains, i.e., in
number of agents, system design, and the number of va-
riables determining agents’ decision making behavior.
Effective coordination mechanisms regulating agents’
interaction are particularly needed in these circumstances.
There are many multi-agent development tools.
4. Multi-Agent Modeling with Trading Agent
4.1. Java Agent Development Framework (JADE)
JADE is a software framework fully implemented in
Java language. It simplifies the implementation of mul-
ti-agent systems through a middle-ware that complies
with the Foundation for Intelligent Physical Agents (FI-
PA) specifications and through a set of graphical tools
that supports the debugging and deployment phases. The
agent platform can be distributed across machines that
not even need to share the same OS and the configura-
tion can be controlled via a remote graphical user inter-
face (GUI). The configuration can be even changed at
run-time by moving agents from one machine to another
one, as and when required. Eclipse is an open source
community, whose projects are focused on building an
open development platform comprised of extensible
frameworks, tools and runtimes for building, deploying
and managing software across the lifecycle. Eclipse is a
software development platform helping the software de-
veloper to rapidly build new JAVA applications: Eclipse
is a JAVA-based developing platform and service set
that makes developing environment with plug-in com-
ponents easier.
4.2. Negotiation System
Kasbah is a Web-based system where users create auto-
nomous agents that buy and sell goods on their behalf
where the original idea was to reinvent the classified ads.

A Study of Multi-Agent Based Supply Chain Modeling and Management 336
We observed that there are many sites on the Web that
post adv. These classified ad sites all provide tools to
help the user find adv. of interest. Certainly, such tools
are useful yet they only assist with one step in the mul-
ti-step process of buying and selling that of finding ads
which match what one is looking for. The idea behind
Kasbah is to help users with the other step in the process,
the negotiations between buyer and seller, by providing
agents which can autonomously negotiate and make the
best possible deal on the user’s behalf. Kasbah is avail-
able through a Web site where users go to buy and sell
things. They do this by creating buying and selling
agents, which then interact in the marketplace, thus,
Kasbah is a multi-agent system.
The marketplace is designed to handle any type of
agent that supports the appropriate protocol, though the
current prototype only has a single kind of relatively
simple buying and selling agents. It is these agents that
will be described in the remainder of the paper. The
agents themselves are not tremendously smart. They do
not use any AI or Machine Learning techniques, which
usually exist in current agent developer. Despite the
growth in the number of online auction sites, there is still
a need for a more dynamic, personalized bidding experi-
ence. Existing bidding and auction sites overemphasize
bid price as the sole parameter determining the match of
a buyer and a seller. We believe that for dynamic pricing
systems to truly benefit the buyer and seller, the negotia-
tion interaction needs to extend further than a simplistic
exchange of bid and ask prices. On-line auction systems,
such as eBay, Amazon Auctions, and Priceline’s airline
bidding system violate several principles we believe are
necessary for a bidding system to benefit both the buyer
and the seller.
These principles are: 1) offers should be evaluated and
selected on multiple criteria, not just price; 2) the nego-
tiation should be a non-binding arrangement, allowing
the buyer to make multiple bids on multiple offers, in-
creasing the chance of a successful match; and, 3) sellers
should have the tools to evaluate bids based on complex
criteria, not just immediate revenue.
4.3. Mathematical Modeling with Trading Agent
4.3.1. Design Element and Notation
We propose multi-agent modeling for a SCM system
using adaptive trading agent to make the most appropri-
ate decision using multi attribute for demand of buyer.
We had created agents with JADE and performed the
simulation under JADE and Eclipse environments.
Communication among agents is performed by a set of
messages that follow predefined protocols. In our model,
FIPA’s two protocols, Query and Request, were used to
model the conversations among agents. Each supply
chain entities exhibit a tendency to have independent
authority with conflicting requirement, and may possess
local information relevant to its interests. In order to
maximize the total profit of the SCM, negotiation among
all members is necessary. In this research, we want to
find the best negotiation strategy that makes all members
of the SCM satisfy in the simple SCM. But the interest of
supplier, producer and distributor is not in keeping with,
there will be the conflict among the pursuit of the profit
of all members of the SCM.
All supply chain entities pursue their own profit. The
objective function for supply chain is to maximize the
total profit of the self-interested supply chain entities. So
the trading agent performs a negotiation among supply
chain entities. We consider a simple supply chain that
only consists of suppliers, producers, distributors, and
customers. We assume that we restrict our research scope
to coordination of the unit price that producer paid to
supplier, the unit price that distributor paid to producer,
and trading quantities. We assume that the demand of
end products is determined by the price that the distribu-
tor has set up. And we assume that trading agent tries to
communicate and adjust to supply chain entities. We
want to maximize the total profit of a simple supply
chain. Distributor is supplied with finished products by
producer and sells end products to the customer. Pro-
ducer is supplied with parts by supplier and makes a fin-
ished product. Supplier makes parts and supplies the
producer with parts. The process of a simple supply
chain is depicted as Figure 1.
where

the total profit of supplier is

, producer
j
m

, and
distributor kd

D the demand for customer or production quantities
x the number of supplier
y the number of producer
z the number of distributor
is
P the unit price that producer paid to supplier i, i =
1,2,…,x
j
m
P the unit price that distributor paid to producer j, j =
1,2,…,y
kd
P the unit price that customer paid to distributor k, k =
1,2,…,z
is
C the unit production cost of supplier i, i = 1,2,…,x
j
m
C the unit production cost of producer j, j = 1,2,…,y
kd
C the unit delivery cost of distributor k, k = 1,2,…,z
i, j, k the indices for supplier, producer, and distributor
respectively
4.3.2. Mathematical Modeling
In our model we assume that
1) Each entity in supply chain makes decision inde-
pendently but they are ready to negotiate in order to op-
timize all supply chain;

A Study of Multi-Agent Based Supply Chain Modeling and Management 337
Figure 1. The process of a simple supply chain.
2) Distributors have exclusive selling rights in the sup-
ply chain;
3) The price and demand are determined by negotia-
tion between distributor and customer, and
4) After negotiation between distributor and customer,
the demand for producer and supplier is determined as
constant D.
In SCM modeling study, we look for max

from the
supply chain. The total profit of the supply chain is the
sum of the profit of distributor, producer, and supplier.
111
 



y zx
kdjm is
kji
Maximize (1)
subject to


11 1
,, () 
 

 
y
zz
kdkd jmkdkd
kj k
DP PP DDP
   
11


y z
kdkd jmjm
kj
CDDP PDDP (2)
Distributor is supplied with finished products by pro-
ducer and sells end products to the customer. The profit
of distributor is total price that customer paid to dis-
tributor minus both total delivery cost and total price
paid to producer.


111 11
,,
yyy xx
jmjm isjmjmis
jij ji
DPPPDCD DPD 
 
 
 
(3)
Producer is supplied with parts by supplier and makes
a finished product. The profit of producer is total price
that distributor paid to producer minus both total produc-
tion cost and total price paid to supplier.
 
111
,
xxx
is isisis
iii
DPPDCD D 



(4)
Supplier makes parts and supplies the producer with
parts. The profit of supplier is total price that producer
paid to supplier minus total production cost.
In order to maximize supplier’s profit, the optimality
condition is found as follows,
 
111
,0
xx
ii
x
is isisis
i
dDPdDPdCDDdD 

 
 
 
 

let

 
1
x
i
i
issDd CDDCdD





the condition for optimization is



1 1 




 
xx
is isis
i i
Pd CDDdDCD (5)
The same procedure to maximize producer’s profit,

11 1
,,
yy
x
j
mjmis jm
ji j
dDPPdDP 
 







11
0
yx
jm is
ji
dCDDdD P







(6)
Let

 
1
dC
y
jm jm
j
CD DDdD





the condition for optimization is


11
yx
j
mjmis
ji
CD PP



 (7)
The same procedure for distributor,


11 1
,,
y zz
kdkd jmkd
kj k
dD PPdDdPDDdD 
 









11
0
yy
jm jm
jj
dCDDdD P







(8)
let

 
1
dP
z
kd kd
k
RD DDdD





the condition for optimization is found as




1
y
kd kdjm
j
RD CDP



(9)
Trading agent try to coordinate their different views
and lead to optimal negotiation. So the negotiation of
trading agent depends on following optimality conditions.
The optimal condition for supply chain must satisfy the
optimal condition for distributor, producer and supplier
simultaneously. We’ve found the condition for optimiza-
tion from Equations (5), (7), and (9)




1
y
jm jmis
j
PCDCD


 (10)




1
y
jm kdkd
j
PRDCD


 (11)
These equations are interpreted to mean that optimal
conditions will be obtained if sum of the marginal pro-
duction cost of producer and the marginal production
cost of supplier equal to the unit production cost of pro-
ducer or the difference between marginal profit of dis-
tributor and marginal cost of distributor equal to the unit

A Study of Multi-Agent Based Supply Chain Modeling and Management 338
production cost of producer.
4.3.3. Trading Agent Algorithm
The coordination procedure for the trading agent to make
a negotiation is as follows: First, for given demand for
customer Dt distributor j tries to find
j
m
P that satisfy
Equation (11). Second, for given
j
m
P after making a
negotiation between producer and supplier we find new
'
t
D that satisfy Equation (10). Third, if new '
t
D equal
to D then that is optimal D = new '
t
D, otherwise we re-
define Dt+1 as the middle point of Dt and new '
t
D. We
repeat the same procedures until we find solution. We
can summarize coordination step as follows:
Initialization step
1) Trading agent: let t = 1.
2) Trading agent: report the forecasted demand D1
to distributor, and go to Step t.
Step t
t1) distributor: for given Dt find the optimal Pjmt that
satisfy Equation (11) and report it to the trading
agent.
t2) trading agent: let '
t
D = Dt and present that to the
producer and supplier.
t3) producer: calculate

'
()
j
mt
CD with '
t
D of trading
agent and report it to the trading agent; supplier:
calculate

'
()
is t
CD with '
t
D of trading agent and
let Pist =

'
()
is t
CD, and report Pist and

'
()
is t
CD to
the trading agent.
t4) trading agent:
t4-1) investigate whether Equation (10) based on

'
()
j
mt
CD provided by producer and

'
()
is t
CD pro-
vided by supplier is to be satisfied or not;
t4-2) if Equation (10) is to be satisfied then go to
step t5, otherwise
t4-3) redefine '
t
D and report this '
t
D to the pro-
ducer and supplier and go to step t3.
t5) trading agent:
t5-1) if |Dt-'
t
D|<  , 0,  is a very small value,
then we find solution and terminate, otherwise
t5-2) let Dt+1 = (Dt+'
t
D)/2 and report Dt+1 to dis-
tributor and let t = t+1 and go to step t1.
In the e-commerce the important factors that have
considerable influences upon the buying are price, re-
view point and delivery time. Our system can be de-
picted in Figure 2.
In this example, we consider not only price but also
other factors that affect trading in the e-marketplace. We
consider a system that consists of a seller agent, a buyer
agent, and a trading agent. Agent takes part in the trading
on one’s behalf. Seller or buyer create agents and give
them some parameters for trading. Information that seller
agent or buyer agent is engaged is delivered to the trad-
ing agent. Figure 3 shows the seller agent process.
Figure 2. System of trading agent.
Figure 3. Seller agent process.
4.3.4. Negotiation Function
When the customer is going to purchase some goods they
want to choose the best one from among goods on the in-
ternet shopping mall. They first review the comments post-
ed by other customers who gave the review point on the
goods they purchased. The customer checks the delivery
time and then make a decision with considering price,
review point and delivery time altogether. In order for
the trading agent to calculate trading point that has
weighted average characteristics we consider price, re-
view point and delivery time simultaneously. For the
seller agent, because we use weighted average value in-
stead of price as negotiation point, we first calculate the
weight.
In the kth negotiation the weight for price is calculated
as follows.

maxmax min
1
k
iw
i
P
PP tTC




 
 


, 1, 2,,kn (12)
In the kth negotiation the weight for review point is
calculated as follows.

A Study of Multi-Agent Based Supply Chain Modeling and Management 339

maxmax min
1
k
iw
i
SSS tTS




 


, 1, 2,,kn (13)
In the kth negotiation the weight for delivery time is
calculated as follows.

maxmax min
1
k
iw
i
DDD tTD




 


,1, 2,,kn (14)
where
T the date to sell the item by
w
C the weight for price
w
S the weight for review point
w
D the weight for delivery time
i
t th inegotiation time
i current number index of negotiation
n total number of negotiation
max
P the first suggested highest acceptable price
min
P the first suggested lowest acceptable price
max
S the first suggested highest acceptable review point
min
S the first suggested lowest acceptable review point
max
D the first suggested longest acceptable deliver time
min
D the first suggested shortest acceptable delivery
time
For the buyer agent, in the kth negotiation the weight
for price is calculated as follows.
max
1
k
iw
i
PtTC




 


, 1, 2,,kn

 (15)
In the kth negotiation the weight for review point is
calculated as follows.
max
1
k
iw
i
StTS




 


, 1, 2,,kn

 (16)
In the kth negotiation the weight for delivery time is
calculated as follows.
max
1
k
iw
i
DtTD




 


, 1, 2,,kn

 (17)
where T is the date to buy the item by.
Trading agent compares points suggested by seller agent
and buyer agent and determines how to make progress the
negotiation. Trading agent determine the negotiation price
with Eclipse. We first run the JADE and make selling
agent and buyer agent. And both agents negotiate in the
marketplace that is controlled by JADE GUI tools.
5. Case Study
After trading agent modeling of SCM is introduced, a
fictitious bookstore case will be investigated. We have
considered the simple bookstore case that consists of 2
distributors, 2 producers and 2 suppliers. Distributors
have made a negotiation with customer. Distributor’s
maximum price is $ 15,000 and minimum acceptable
price is $ 13,200 and customer’s maximum acceptable
price is $ 15,000.
Trading agents seek the best purchasing point with
buyer and seller agent through the use of JADE tools.
We developed trading agent based negotiation system
with following developing environment in Table 1.
We compare the benefit/cost (b/c) ratio of trading
agent with that of the Kasbah system.
In this paper, we’ve defined buyer b/c ratio and seller
b/c ratio as following:
Buyer b/c Ratio
= negotiation price suggested by seller agent/negotiation
price suggested by buyer agent
Seller b/c Ratio
= negotiation price suggested by buyer agent/highest
acceptable price suggested by seller agent
In this case we assume that we’ll make a 10 round ne-
gotiation and give 30 minutes to each negotiation time for
seller and buyer agent. We want to find negotiation strat-
egy that gives us the high benefit/cost ratio. Table 2 gives
the attribute table of input data for seller and buyer agents.
We can compare the number of trading negotiation
and total negotiation b/c ratio for Kasbah and our trading
agent. The number of trading negotiation is similar be-
tween two methods. In order to compare the efficiency of
the negotiation algorithm, we use total negotiation b/c
ratio. The results are shown in Tables 3, 4, 5. As shown
in Figure 4, trading agent gives larger benefit/cost ratio
than the Kasbah system.
6. Discussion
The importance of supply chain management is increas-
ing with globalization and the widespread adoption of
Table 1. Developing platform.
item System
Operating System Windows XP
JAVA compiler JDK 1.6.0
JAVA developing tool Eclipse SDK 3.4.1
Eclipse Plug-in EJADE RMA
Agent Protocol FIPA ACL
JADE JADE 3.6
Figure 4. Negotiation b/c ratio.

A Study of Multi-Agent Based Supply Chain Modeling and Management 340
Table 2. Attribute table for seller and buyer agent.
price Review point Delivery time
Maximum value 15,000 100 3 days
Minimum value 13,200 1 1 day
Book seller agent
Weight 0.6 0.2 0.2
Maximum value 15,000 100 1 day
Book buyer agentweight 0.6 0.2 0.2
Table 3. Negotiation result of kasbah.
Seller Agent 1 2 3 4 5 6 7 8 9 10
InitPrice 14946 14972 14941 14945 14943 14973 1497314972 14960 14965
NgtPrice 13371 13364 13359 1334613351 13378 13383 13376 13363 13291
Buyer Agent 1 2 3 4 5 6 7 8 9 10
InitPrice 452 462 491 459 473 443 458 475 661 541
NgtPrice 13571 13861 13745 1377513703 13741 13750 13765 13883 13523
Table 4. Negotiation result of trading agent.
Seller Agent 1 2 3 4 5 6 7 8 9 10
InitPrice 8981 9004 9001 8988 8974 9000 9002 9008 9001 8973
NgtPrice 8020 8022 8045 8038 8045 8050 8047 8044 8043 8026
Buyer Agent 1 2 3 4 5 6 7 8 9 10
InitPrice 328 266 327 275 383 338 313 208 327 267
NgtPrice 8207 8259 8163 8249 8052 8120 8130 8111 8182 8263
Table 5. Negotiation b/c ratio.
Kasbah Trading Agent
Seller b/c Buyer b/c total Seller b/c Buyer b/c total
1 0.9080 0.9852 1.8932 0.9138 0.9772 1.8910
2 0.9257 0.9641 1.8898 0.9172 0.9713 1.8885
3 0.9199 0.9719 1.8918 0.9068 0.9855 1.8923
4 0.9217 0.9688 1.8905 0.9177 0.9744 1.8921
5 0.9170 0.9743 1.8913 0.8972 0.9991 1.8963
6 0.9177 0.9735 1.8912 0.9022 0.9913 1.8935
7 0.9183 0.9733 1.8916 0.9031 0.9897 1.8928
8 0.9193 0.9717 1.8910 0.9004 0.9917 1.8921
9 0.9280 0.9625 1.8905 0.9090 0.9830 1.8920
10 0.9036 0.9828 1.8864 0.9208 0.9712 1.8920
electronic commerce. However, supply chains can
change over time and companies in supply chains can
have only limited visibility of the supply chains. This
paper suggested a multi-agent system that has trading
agent and make negotiation through suggested model.
The ideas behind the suggested model are making a new
negotiation algorithm, making an agent by using JADE
and Eclipse environment. It is multi-agent technology
and information sharing among neighboring agents that
is very important to the SCM. When we make negotia-
tion we consider not only price but also review point and
delivery time. We suggest a new negotiation in the SCM
environment with using multi-agent technology. And in
order to verify the performance of our algorithm we’ve
made the simulation on the simple supply chain and
compared benefit/cost ratio with Kasbah system.
The performance of the suggested model was analyzed
by a simulation experiment with JADE and Eclipse. The
result of the simulation revealed that the simulated model
had better performance than existing negotiation model.
After 10 rounds of negotiation we investigated the nego-
tiation benefit/cost ratio between the proposed model and
the Kasbah system. Our trading agent model gave us a
slightly higher rate of return than the Kasbah system. We
used weighted average of multi-attributes of negotiation.
There are some limitations in this research: First, de-
termining the weight is entirely subjective and may be
not adequately represent true marketplace conditions.
Second, there are few evaluation tools to compare the
performance of the proposed system with others. Lastly,

A Study of Multi-Agent Based Supply Chain Modeling and Management 341
the simulation was limited to a single simple system.
However, we believe this model and system simulation
have provided an alternative application of agent-based
trading that shows potential for future research on mul-
ti-agent trading environments.
7. Acknowledgements
This paper is the work supported by the research fund of
2009 long-term oversea visiting scholar program from
Gangneung-Wonju National University, Korea.
REFERENCES
[1] M. E. Nissen, “Agent-Based Supply Chain Dis-Interme-
diation vs. Re-Intermediation: Economic and Techno-
logical Perspectives,” Intelligent Systems in Accounting,
Finance, & Management, Vol. 9, No. 4, 2000, pp. 237-
256.
[2] J. W. Forrester, “Industrial Dynamics,” MIT Press, Cam-
bridge, 1961.
[3] H. L. Lee, P. Padmanabhan and S. Whang, “Information
Distortion in a Supply Chain: the Bullwhip Effect,”
Management Science, Vol. 43, No. 4, 1997, pp. 546-558.
[4] M. E. Nissen, “Agent-Based Supply Chain Integration,”
Information Technology & Management, Vol. 2, No. 3,
2001, pp. 289-312.
[5] N. R. Jennings and M. Wooldridge, “Applying Agent
Technology,” Applied Artificial Intelligence, Vol. 9, 1995,
pp. 357-369.
[6] M. S. Fox, M. Barbuceanu and R. Teigen, “Agent-Ori-
ented Supply-Chain Management,” Flexible Manufac-
turing Systems, Vol. 12, No. 2, 2000, pp. 165-188.
[7] J. M. Swaminathan, “Modeling Supply Chain Dynamics:
A Multiagent Approach,” Decision Sciences, Vol. 29, No.
3, 1997, pp. 607-632.
[8] Y. Chen, Y. Peng, Y. Labrou, S. Cost, B. Chu, J. Yao, R.
Sun and B. Willhelm, “A Negotiation-Based Multi-Agent
System for Supply Chain Management,” Workshop on
Agents for Electronic Commerce and Managing the In-
ternet-Enabled Supply Chain, Seattle, 1999, pp. 15-20.
[9] M. He, H. F, Leung and N. R. Jennings, “A Fuzzy Logic
Based Bidding Strategy for Autonomous Agents in Con-
tinuous Double Auctions,” IEEE Transactions on Know-
ledge and Data Engineering, Vol. 15, No. 6, 2003, pp.
1345-1363.
[10] Y. Yuan, T. P. Liang and J. J. Zhang, “Using Agent
Technology to Support Supply Chain Management: Po-
tentials and Challenges,” Michael G. DeGroote School of
Business Working Paper Series, 2001, p. 453.
[11] N. R. Jennings, “An Agent-Based Approach for Building
Complex Software Systems,” Communications of the
ACM, Vol. 44, No. 4, 2001, pp. 35-41.
[12] R. H. Guttman, A. G. Moukas and P. Maes, “Agent-Me-
diated Electronic Commerce: A Survey,” Knowledge En-
gineering Review, Vol. 13, No. 2, 1998, pp. 147-159.
[13] P. Davidsson, E. Astor and B. Ekdahl, “A Framework for
Autonomous Agents Based on the Concept of Anticipatory
Systems,” In: Proceedings of Cybernetics and Systems,
World Scientific, Singapore, Vol. 2, 1994, pp. 1427-1434.