The Research On Cooperation Of Multi-Agent System Based On Petri Nets

As their capabilities to imitate the social intelligence of human being flexibly and adap-tively, Multi-Agent Systems (MAS) can widely be applied to construct open and dynamic software systems, and is one of the major research areas in artificial intelligence. How to form,describe, decompose problem and synthesize results among multiple agents using a formal method is one of urgent issues to be solved in multi-agent systems. This thesis regards multiple agents cooperating to accomplish a whole goal as the topic studied, describes the Petri net models of valid task decomposition in detail, and validates the properties which models behave, and on the basis mentioned above proposes the task dynamic allocate Petri net models corresponding to the ideas of contact protocol and Petri net models for plan execution, furthermore related properties are analyzed and validated.The major contributions of this dissertation are as follows:1) Describing the semantics of logic and/or tree at length to assure no misunderstanding occurs, and giving the algorithm to construct the logic and/or tree, proposing the rules that convert the and/or tree of task decomposition to Petri net models.2) By eliminating the unreasonable structures of task decomposition, we get the Petri net systems of valid task decomposition, detailed analyze the relationship between the liveness of the Petri net systems and the valid task decomposition; through the reachability analyze for the valid task decomposition Petri net systems, the all feasible plans solving the whole problem are obtained and the correctness of plans are validated by the ideas of net enlogy, futhermore the determinant theorem for the concurrency among different sub-tasks are given. In the end, an illustrative example is given.3) Constructing task allocate Petri net models by employing the idea of gradually refinement. Task allocate Petri net models are considered from two level points: macroscopical models from the social level and microcosmic models from the agent level. Macroscopical models consist of two kinds of bidding models which are possessing fixed worker agents models and possessing variable worker agents models respectively, and the cooperations among multi-agents are studied through Petri nets. Microcosmic models mainly introduce the Petri net models for agent based on BDI architecture, which regards the inner mental changes of each agent as the subject to study. And this chapter also describes the dynamic semantics of the two level Petri...