COOPERATIVE NEGOTIATION FOR THE PROVISIONS

BALANCING IN A MULTI-AGENT SUPPLY CHAIN SYSTEM

FOR THE CRISIS MANAGEMENT

Hayfa Zgaya, David Tang and Slim Hammadi

LAGIS UMR CNRS 8146, Ecole Centrale de Lille, Cité Scientifique - BP 48 - 59851, Villeneuve D’Ascq Cedex, France

Keywords: Distributed System, Supply Chain Management, Multi-Agent System, Interaction, Negotiation Protocol,

Bullwhip Effect.

Abstract: Since a few years, logistics has become a performance criterion for the organizations success. So the Supply

Chain (SC) study is adopted more and more for the competitiveness of companies development. In previous

works we proposed an approach, which aims to reduce an emerging phenomenon of the demand

amplification, called the Bullwhip Effect. In this paper, we present a model, based on the proposed

approach, for a Cooperative Negotiation for the Provision Balancing in a SC system. The studied SC is a

hierarchical system dedicated to the Crisis Management. A Multi-Agent architecture is then proposed to

design this distributed chain through interactive software agents. The results of simulation, presented in this

paper, prove the importance of the interaction between the SC entities for the Provisions Balancing.

1 INTRODUCTION

A Supply Chain (SC) represents the whole of the

links starting from the final customer to the first

level supplier. The main objective of such a structure

is the final customer satisfaction; it is thus necessary

to progress the SC management by optimizing flows

going from the supplier to the customer and also

from the customer to the supplier (e.g. information

and goods flows). In our work, we focus on a special

kind of SC: a distributed Crisis Management SC

(CMSC) based on a hierarchical structure. In a

previous work, we proved that a minimal

communication between the different SC links

reduce considerably Bullwhip effect. Basing on

previous these further verifications, we are interested

here to develop a cooperative Multi-Agent System

(MAS) negotiation for the ammunition balancing in

our CMSC system, in order to balance ammunition

in a disturbed mode. The remainder of this paper is

organized as follows: initially the CMSC will be

described in next paragraph, followed in paragraph 3

by the proposed multi-agent architecture

characterized by the communication and the

information sharing between the different distributed

entities. The negotiation protocol for the provision

balancing is presented and detailed in paragraph 4.

Finally, experimentations in paragraph 5 show the

contribution of the proposed protocol and its limit.

2 THE CRISIS MANAGEMENT

SUPPLY CHAIN

The CMSC is an L-levels SC links; from the

provisions warehouse for routing Z

(exclusive first

level) to several disaster zones Z

. All other zones

are of level i with 1<i<L. So for a given zone Z

, a

downstream zone is of Level i+1: Z

i+1

and its

upstream zone is of level i-1: Z

i-1

.The retro logistic

is not allowed within our CMSC, so the matter flow

goes from the upstream to the downstream nodes.

However, the data flow can take place in the two

directions according to the interaction protocol

expressed later. When a crisis takes place (e.g. a

natural disaster), the manager affected to a disaster

victim zone (Z

), orders the necessary products from

an upstream zone Z

L-1

, which, in its turn, addresses

its request to the zone Z

L-2

and so on. Each zone has

a partial sight of the environment, which results in

incomplete data and limited capacities. Thus, we

propose to model the CMSC by communicating

agents within a distributed MAS which should

follow a formal mathematical model (e.g. Least

280

Zgaya H., Tang D. and Hammadi S. (2008).

COOPERATIVE NEGOTIATION FOR THE PROVISIONS BALANCING IN A MULTI-AGENT SUPPLY CHAIN SYSTEM FOR THE CRISIS MANAGE-

MENT.

In Proceedings of the Third International Conference on Software and Data Technologies - ISDM/ABF, pages 280-283

DOI: 10.5220/0001883102800283

 SciTePress

Square Method and Gaussian Distribution) and

pumps data from real case studies. The idea is to

prepare a mathematical model package and to

instantiate the decided model which can be a single

form or a combination of several ones. The decision

is done thanks to a strategic level within the

reasoning layer of the interaction model presented

afterwards in this paper. This feature of the studied

CMSC is not detailed in this paper.

3 THE PROPOSED

MULTI-AGENT

ARCHITECTURE

3.1 Model Representation

As it was previously mentioned, the hierarchical

feature between the various entities characterizes the

multi-zone logistic system. So there is an agent

responsible of each zone representing it, we call this

agent: an agent-zone. Each agent-zone can

communicate only with another agent-zone that is

hierarchically higher to him (an upstream agent-

zone) or with another agent of the same hierarchical

level. For example, if N, M and P correspond

respectively to the zones numbers Z

, Z

and Z

in a

4-levels CMSC, then :

 Ag

: the Z

agent-zone,

 Ag

Z2i

: the Z

agent-zone (1≤i≤N) who can

interact with the Ag

or with another agent-

zone Ag

Z2i’

(1≤i’≤N and i’≠i ),

 Ag

Z3i,j

: the Z

3i,j

agent-zone (1≤i≤N and

1≤j≤M) who can interact with an agent-zone

or with another agent-zone Ag

Z3i,j’

(1≤j’≤M

and j’≠j),

 Ag

Z4i,j,k

: the Z

4i,j,k

agent-zone (1≤i≤N , 1≤j≤M

and 1≤k≤P), who can interact with an agent-

zone Z

or with another agent-zone Ag

Z4i,j,k’

(1≤k’≤P and k’≠k ).

3.2 Interaction Mode

We adopt the “with agreement” mode, which

expresses the collaboration between the agent-zones

thanks to an effective communication to make better

decisions to the demands. The goal is to find

ammunition balancing in our CMSC system thanks

to a cooperative negotiation between the disaster

sectors and their upstream zones in a disturbed

mode.

4 THE NEGOTIATION

PROTOCOL MODEL

The cooperative negotiation aims to provide urgent

ammunitions to the zones, in case of need, while

waiting for the help. We propose a negotiation

architecture based on the abstract one presented in

(Wooldridge and al., 1995). This architecture is

composed of three layers:

1- Communication Layer: corresponds to the

interaction layer of the architecture, it is responsible

for receiving and sending messages between agents;

2- Control Layer: corresponds to the negotiating

agent behaviours, which will be specified by UML

activities diagrams in further works;

3- Reasoning Layer: corresponds to the decision-

making part of the negotiating agent and interacts

with his Knowledge base module. Through this

layer, an agent (identified by Ag_Id) can evaluate

his own emergency degree for a given resource r

according to his mental statements. This emergency

degree is called Emergency Index and noted by

index

,Ag_Id). The measurement of this emergency

index exceeds the topic of this paper. More details

will given in future publications. A negotiation

process is decomposed of:

 Initiators of the negotiation who start the

process. We focus on the case of a single

initiator for hierarchical reasons. This Initiator

is noted by Init,

 Participants who contribute to this negotiation.

An upstream node can command one or several

downstream nodes noted by Part

(1≤j≤P),

 Objects of the negotiation: limited resources on

which the negotiation members (Initiators end

Participants) negotiate. A resource is noted by r

(1≤i≤R).

The decision of which protocol will be used

(Communication Layer) depends on the agent-zone

Reasoning Layer. In this paper, we focus on an

agent-zone Communication Layer; instance of the

Help-One-To-Many (HOTM) protocol. In future

work, we will compare this proposition to another

kind of negotiation protocol: Help-Many-To-Many

(HMTM) protocol. The proposed HOTM protocol is

described as follows (Figure 1):

 Modification Request: If the Initiator (upstream

zone) realizes that he cannot satisfy all his

subordinate zones demands before some

period of time Δt corresponding to the new

supply delay. So, he informs all the

subordinate agent-zones about the situation

COOPERATIVE NEGOTIATION FOR THE PROVISIONS BALANCING IN A MULTI-AGENT SUPPLY CHAIN

SYSTEM FOR THE CRISIS MANAGEMENT

281

proposing them to renounce to their demands

if they can wait for an additional period of

time. In other words, as soon as an upstream

agent-zone is not able to response to some

resources demands (Reasoning Layer), the

Control Layer is activated by a modification

demand and an “output event” starts the

HOTM protocol,

Figure 1: HOTM protocol.

 Modification Proposition: each Participant

agent-zone sends his emergency degree to the

initiator. For example, if an agent-zone

intends to desist, he should send a weak

emergency degree. This corresponds to an

“input event” within the Initiator negotiating

Agent Architecture,

 Propose (contract): The initiator sends new

contract expressing the new provisions

quantities balancing evaluated within the

Reasoning layer,

 Accept/Refuse: After estimation of remaining

stocks of all the provisions (water, medicines,

clothes, etc.), a participant agent-zone Z

i+1

realizes that he can accept:

 All the Initiator propositions (Total

Accept),

 A sub-set of the Initiator propositions

(Partial Accept). For example, he can

accept the given Initiator proposition for

clothes but not for water and medicines,

 None proposition (Refuse).

 Confirm: Since the Initiator receives enough

desisting Participant responses for a kind of

provisions, he confirms that he can now

satisfy:

 All the demands (Total Confirm): there

is enough quantities for all kinds of

provisions,

 A sub-set of demands (Partial Confirm):

there is enough quantities for some

kinds of provisions.

Further to a confirmation, the Initiator sends

the provisions. In that case, if there are still

some downstream agent-zones (Z

i+1

) who

need some provisions in real time and the

correspondent upstream agent-zone (Z

)

can’t satisfy all the demands, the negotiation

process loops (go to Modification Request

demand) and so on. Otherwise, the protocol

ends.

 Cancel: the negotiation process can be

cancelled (e.g. at the end of authorized

negotiating time).

5 SIMULATION RESULTS

We simulate here a cargo loss between days 35 and

36, checking the HOTM Protocol and we focus on a

single resource satisfaction. The experimentation

aims to find an effective provisions balancing within

the different 4-Levels CMSC zones. The bullwhip

effect is not considered here.

5.1 Case 1: Without HOTM

In this case, the upstream agent-zone Z

sends all his

resources to his subordinates. The problem here is

that this agent resides with an out-of-stock condition

during 3 days (Figure 2-a). This is a serious

problem, because there are no more provisions for

his own consumption. We notice here that security

stocks of subordinate zones are slightly picked

(Figure 2-b). In this context, the principle of the

negotiation is to demand to subordinates zones if

they agree to pick in their own security stocks in

order to avoid emptying totally the upstream zone

stock.

5.2 Case 2: With HOTM

In this case, when the upstream agent-zone Z

receives subordinate agents demands, he realizes

that he cannot satisfy all his subordinate zones

demands before the new supply delay. Thus, he

informs his subordinates (Z

4,1

and Z

4,2

) about his

situation and proposes them to renounce to their

demands if they can wait for an additional period of

time. So, each Z

agent-zone will estimate if his

(Initiator)

i+1

(Participants)

Propose (contract)

Acce

Partial (parameters)

(èt)

Total

Refuse

Confirm

Total

Partial (parameters)

Modification Request (parameters)

Modification Proposition (parameters)

Cancel

ICSOFT 2008 - International Conference on Software and Data Technologies

282

Figure 2-a: Z3.

Figure 2-b: Z4,1.

Figure 2: Case without HOTM.

remaining security stock is sufficient to wait the

required time. Here, the agent-zone Z

4,1

was agree to

desist giving a weak emergency index in order to

avoid the out-of-stock condition of the upstream

agent-zone Z

(decision through the Reasoning

Layer). Thus, Z

gives some amount of goods only

to Z

4,2

. Consequently, the negotiation allowed to all

the zones to survive (Figure 3).

6 CONCLUSIONS

We are interested in our work to a special kind a

distributed SC of which the different interactive

entities are hierarchically related. We proposed for

this SC a multi-agent architecture characterized with

independent agent-zones sharing information. In

this paper, we focus on the provision balancing

thanks to a One-To-Many negotiation protocol

(HOTM). We showed within the experimentation

results that this protocol allows avoiding the

Figure 3: Z4,2.

out-of-stock condition in different cases for a special

kind of provisions. The problem is that this protocol

is not enough robust when a new disaster crisis

overlaps with the current one. So, we propose

another variant of the proposed protocol: the Many-

To-Many protocol (HMTM) which gives the

possibility to an upstream agent-zone to serve a

downstream agent-zone who is not necessarily one

of his subordinates. This protocol and other kind of

interactions models will be proposed, studied and

compared to the HOTM in future work in order to

find the best near optimal robust solution for the

provision balancing and bullwhip effect reduction

through an L-levels CMSC.

REFERENCES

Beer, M. and al., 1999. “Negotiation in Multi-Agent

Systems”. Knowledge Engineering review 14(3), 285-

289.

Forrester, J.W., 1961. Industrial Dynamics, the M.I.T

press-Massachusetts, Institute of Technology-

Cambridge, Massachussetts.

Lee and al.,1997a. “Information Distortion in a supply

chain: The Bullwhip effect”, Management Science,

Vol 43, No.4, pp. 546-558.

Lee and al., 1997b. “The Bullwhip Effect in supply

Chains”, Sloan Management Review, Vol 38, No.3

(Spring), pp. 93-102.

Moyaux, T. and al., 2003. “Multi-Agent Coordination

Based on Tokens : reduction of Bullwhip Effect in a

Forest Supply Chain”, In Proceedings of the 2nd

International Joint Conference on Autonomous Agents

and Multi-Agent Systems (AAMAS’03), July 14-18,

Melbourne, Australia.

Moyaux, T., 2004. Design, Simulation and analysis of

collaborative strategies in multi-agent systems: The

case of supply chain management, PhD, Faculté des

sciences et de Génie Laval university Québec,

November.

Smith, J., 1998. The book, The publishing company.

London, 2

edition.

Wooldridge, M. J., and Jennings, N. R. 1995. Intelligent

agents: Theory and practice. Knowledge Engineering

Review 10(2).

COOPERATIVE NEGOTIATION FOR THE PROVISIONS BALANCING IN A MULTI-AGENT SUPPLY CHAIN

SYSTEM FOR THE CRISIS MANAGEMENT

283