Research On Autonomic Computing Oriented Policy-based Multi-agent Cooperation System

Nowadays, how to form multi-institutional virtual organizations, establish efficient cooperation system for resource sharing and cooperative problem solving and make these systems having self-management capability in distributed, dynamic and shared network computing environment, having been one of hot research area. Multi-agent systems have become an important technique of artificial intelligence and intelligentized software, software agent, which is having capability of semantics inter-operation and coordination for cooperation behavior, not only provide stronger support for collaborative work, but also create the foundation of open, flexible and scalable new computing environment. So, an autonomic computing oriented and policy-based multi-agent cooperation system is presented in this thesis, which uses software agent as autonomic element. The system is driven by policy and carried out resource sharing and collaborative work in network computing environment. The theory, principle and method of this cooperation system are discussed in four aspects: establishment, maintenance, evolvement and application case of cooperation system. The main creative outcomes are composed of follows:1. A policy-based multi-agent cooperation logic model is proposed. Reform the mental model of the traditional agent, make policy became thecore part of rational reasoning process of software agent. Based on it, a policy based multi-agent cooperation logic model is present, which described the whole cooperation process of multi-agent, including two stages: cooperation establishment and cooperation maintenance. This model became the theory foundation of agent cooperation discussed in follows parts.2. The architecture of policy-based multi-agent cooperation is explained. Ontology Based Knowledge Representation Language(OKRL) and Policy Representation and Assignment Language (PRAL) are also definedThis multi-agent cooperation architecture is composed of software agents, base on policy-driven distributed self-management pattern and support virtual organizations and its members dynamic to establish cooperation system on-demand. Based on description logic and concept instance mode, Ontology Based Knowledge Representation Language (OKRL), Policy Representation andAssignment Language (PRAL) are defined, which became the foundation of agent cooperation and semantic interoperation.3. The running mechanism of policy-based multi-agent system is established.In cooperation establishment stage, Ontology-based service compatible matchmaking mechanism and policy-based rational negotiation mechanism are applied. Among them, the former selected the candidate agents according to conditions and the latter realized the cooperation establishment process with candidate agents. In cooperation maintenance stage, a contract-based cooperation coordination mechanism based on Joint-obey-policy mode is applied, which maintain the normal state of multi-agent cooperation federation.4. The experiment platform integrating with ebXML is realizated.By bring ebXML into ontology obeyed by agent organization and establish map relation between agent mental model and ebXML cooperation process specification, the experiment platform integrating with ebXML is realizated. The theory and run mechanism of policy-based multi-agent cooperation system are verified by case of meeting arrangement running in this platform.Comparing with traditional multi-agent system, policy-based multi-agent system described in this thesis has three merits: (1) Adaptable, according to application demand, the system can became cooperation federation automatically; (2) Robustly, once the cooperation federation is established, with guidance of policy and contract, the system can handling variety exception automatically and maintain the cooperation federation of collaboration work; (3) Openly, integrate with ebXML specification of this system makes it possible to cooperate with other system supporting ebXMl specification, which highlights openly feature of system and makes this system have practical value.