Communication And Task Allocation Based On Negotiation Among Distributed Multi-Agent In CSCMS

In the last ten years, the rapid development of information technology and computer technology has brought great changes for human life. In commercial areas, the market has taken on several characteristics such as demand personalization, demand diversification, the acceleration of the pace of product updates and so on. The simple form of sale and supply between suppliers and customers has failed to meet the demand, and it has gradually formed two trends: on the one hand, supply and demand structure has developed from a single chain structure to multi-node, multi-project crisscross network structure; on the other hand, the relations between supply and demand described by static "push", has developed into complicated ones of "push" and "pull" combination, competition and cooperation coexistence, resulting in the modern supply chain and supply chain management. The introduction of concepts of Dynamic Ally and Collaborative E-business more visually reflects the dynamic, distributed, parallel and collaborative features of the modern supply chain, and information technology as a supply chain management platform will no doubt be faced with tremendous challenges. The information management system established in the traditional context can no longer meet the needs of modern supply chain management. While the Multi-Agent thinking and technology sourced from distributed artificial intelligence has gradually attracted more and more researchers for supply chain management and the supply chain management platform, and it becomes the most potential development direction in supply chain management system.The project of Auto Parts Networked Manufacture System (APNM) is developed in this very background, and the Collaborative Supply Chain Management System (CSCMS), as the subsystem of APNM, is mainly based on the Distributed Multi-Agent thinking and technology.This thesis firstly describes concepts and difficulties in developing supply chain management system. According to the needs of supply chain management, we construct the architecture of collaborative supply chain management system, and put forward an easily ignored problem that autonomous Agent is independence. Independence is the root of many problems which can not be solved in the research and development of supply chain management, and it results in the motives of applying Multi-Agent thinking and technology. Secondly, based on the theoretical background of Agent and Multi-Agent systems, we analysis the serviceability of Multi-Agent system, and point out the feasibility and effectiveness that bring the Multi-Agent theory and technology into the collaborative supply chain management.Agents simulate the subjects in collaborative supply chain management system, so they are distributed in the system and need to cooperate with others through coordination and negotiation. While the communication and interaction among distributed Multi-Agents is the foundation of achieving collaborative work.Communicating among distributed Multi-Agents intrinsically differs from objects'communication in object-oriented programming. Because Agents can control their own behavior and state, the concept of"call some object's methods"does not exist among them. Agents adopt appropriate manners to affect others directly or indirectly through"communication acts". Agent communication has two basic contents: the first is communication pattern and the second is the Agent Communication Language (ACL for short). Agent communication patterns in common use are the blackboard pattern, messages passing pattern, and without communications and Plan Passing mode in addition. The blackboard and messages passing pattern are the main patterns of Agent's communication; ACL based on the speech acts (Speech Act) theory believes the communication is considered as an act, that is, Agents do speech acts like a physical action, to promote their (Agents) intents. Nowadays, many Agent Communication Languages based on Speech Act have been developed, in which KQML (Knowledge Query and Manipulation Language) and FIPA ACL are most popular. KQML is released by KSE (Knowledge Sharing Plan) which is supported by DARPA(Defense Advanced Research Projects Agency),and Foundation for Intelligent Physical Agents has developed the FIPA ACL. These two ACL relate to the five aspects of the study and definition as follow: Message Structure, Communication Primitive (or Communicate Act), Content Language, Communication Ontology and Interaction Protocol.In this thesis, based on specification of FIPA, using object-oriented programming language, we design and implement communication and interaction among distributed Multi-Agents in Collaborative Supply Chain Management System. Firstly, the framework of distributed Multi-Agent communications will be constructed. In the system, we use message passing as the main communication pattern, and the main idea of constructing the framework is that: Transform the message objects which is convenient to operate by Agent at the sending end to some format that can be easily transported and analysed, and do the contrary operation at receiving end; Secondly, content reference model will be built, so that the content objects can be converted to the part of ACL's content; Finally, we define the communication act, behavior and interactive role, and thereby implement the Agent interaction protocol as a group of interactive behaviors performed by relevant roles. AUML (Agent Unified modeling Language) protocol diagram will be introduced and illustrated using FIPA-ContractNet-Protocol as an example.Another main task of this thesis is to research the task allocation on the basis of achieving communication and interaction among distributed Multi-Agents in the system. It consists of two parts: the first is utility of the outcome of the task allocation for the participants who choose strategies according to the utility; the second is negotiation mechanisms, and it is the interactive protocol which organizes the participants together to negotiate the strategies to dominate.Firstly, on the basis of theory of Multi-Agent encounters problem, we analyze and formally describe the task allocation problem, and establish the task holder's and potential executor's utility functions for allocation outcome. A new notion–"scheme"will be put forward, and according to this notion, from the holder's perspective, the potential executors can be divided into two categories: executors related to certain scheme and unrelated to certain scheme. Combined with the holder's utility for outcome, schemes can be explained by three states: able, sure and unable. Secondly, we describe the main ideas of common distributed problem solving architecture– contract net, and conclude the researchers'focus on improvements from four stages of its process, that is, promulgating, bidding, awarding and contracting. Based on the theory of negation, the"collusion"which exists in one-to-many and many-to-many negotiated patterns, will be illustrated. And then, the fundamental character of negotiation mechanism will be explained.Combined with formal description of task allocation and basic theory of contract network and negotiation, the task allocation process based on the improved contract network will be developed, and three problems, namely role assignment, negotiation strategies, and rules of ending, will be discussed. At last, we design the algorithm and interaction protocol, and achieve the aim of task allocation among distributed Multi-Agents in Collaborative Supply Chain Management System.Finally, we design simulating instances of purchasing task assignments, to demonstrate the negotiating process and results of the allocation of tasks, through interactive communication between Agent of task holders and Agent of suppliers in Collaborative Supply Chain Management System.