Research And Application On Trust Model In Multi-agent System

Compraring with security techniques and methods in the traditional software system, the security techniques and methods in the open multi-agent systems meet new challenges. Thus the traditional security techniques and method is unsuitable for solving these problems. Trust plays an important role in solving the security problem in the mult-agent system. Our research focuses on experience-based and credential-based trust management.In the experience-based trust management, the trust value is deduced from the previous interactions. However, there are still some problems: how to deal with attachment from the malicious agent with strategy behaviors; how to handle the inaccurate reports from witness, especially who reports strategically; how to get feedback from the witness; how to compute the trust in less time cost and how to store local information in the less space etc.In credential-based trust management, trust is deduced from another agent's credential and the local policy. In such trust management, credential can deduce the fact such as role, attribute and organization, yet can not deduced some implicit information. Further, such trust management may limit because of great time cost in deducing trust using credential.To solve these problems above, we present dynamic metric dealing with strategy behavior; give the approach to handle the inaccurate reports; presents trust-based security model etc. The main research contents of this thesis are as follows:(1)Dynamic trust metric based on time frameIn order to deal with the strategic falling behaviors strategically altering, oscillating behaviors and changing 'name' behavior, we given a dynamic trust metric, which combines recent behaviors factor, history behaviors factor, recent changed behavior factor, history changed behavior factor and exiting factor. Theoretic analysis and simulation show that this dynamic trust metric can deal with the strategic behavior. Further, the agent using such trust metric needs less space storing local interaction information, and reduces computing cost. Lastly, we present the way to collect the witness reports, which is suited for the multi-agent systems.(2)Approach to deal with the inaccurate reportsDue to existence of malicious witness, we present two approaches to compute the witness credibility: previous behavior-based witness credibility and opinion state -based witness credibility, and different approach is suited for different situation. Further, we give algorithm to reduce the effect of witness behaviors.(3)Security model combing different trustBased on the credential-based trust management, this thesis takes experience-based trust into account, and discusses how to combing these two types of trust. Further, this thesis divide the verification module into credential-based verification model, ontology-based reasoning module and policy-based verification module, thus make it the security model to be able to deduce 'implicit' knowledge.(4)Trust-based web services selectionThis thesis gives the approach to compute the trust-based on quality of services, and agent can select the web services according to the trust. In addition, the design of a new type of UDDI, which divides their traditional functions match verification into modules based on the discovery of ontology and the module of user preferences-based. In the module based on preferences, the users are divided into different cluster, which is based on the user preferences about quality of services. Due to the user's preference clustering, the establishment of relations of various types of users and the corresponding services reduce customer the amount of computation when selects the suited services.