Multi-Agent System Architecture And Multi-Agent Collaboration Method Research Based On Co-evolution Mechanism

The architecture and collaboration mechanism of multi-agent system is the core and live problem of the multi-agent theory. As a kind of new research idea, the method of multi-agent collaboration based on co-evolution mechanism has wide research foreground. In this paper, we outspreaded the contents mainly surrounded by multi-agent theory and realization method based on co-evolution mechanism, and mainly discussed the modeling method of multi-agent system model based on co-evolution, the architecture and collaboration mechanism, the co-evolution of collaborative behavior among multi-agents, the system simulation framework and the experiment research, and so on.Firstly, we introduced the main idea, the formalized description, and the basic flow of co-evolution algorithm. Then, from the point of pattern analyzation, we established the mathematics model of the multi-population co-evolution algorithm based on Pattern Replicator Equation of the single population genetic algorithm, and made the theoretical analysis and compare for the method of best choice and the method of random choice of the co-evolution algorithm. We put forward a new method for the individual fitness evaluation, and validated the performance of the new method by the simulation experiment.We established the symbol deduction theory model of the multi-agent system based on co-evolution. Combining with co-evolution mechanism, we renewedly defined the several aspect of the co-evolutionary agent, such as the belief, action, communication, decision maker, and so on. We made the co-evolutionary agent model and actual world model become a uniform closed loop system, and established the symbol deduction theory model of multi-agent system based on co-evolution mechanism. The mathematics model of the multi-population co-evolution algorithm and the symbol deduction theory model of multi-agent system based on co-evolution mechanism are theory bases of whole paper.We established the architecture of the multi-agent system based on co-evolution, and analyzed its inner operation mechanism deeply. After reviewing the main design idea of single agent architecture, such as the agent architecture in PTS domain and several typical interactive agent architecture, we defined the model of Multi-Agent System Based on Co-Evolution (MASBCE), and made a deep analysis of the components of MASBCE and the interactions among them. By the means of analysis for the layered design method of co-evolutionary agent architecture, the interaction process of the agent and the world model, the cycling process at deferent level of MASBCE, the periodic step mechanism of MASBCE, etc., we radicated the architecture of MASBCE and its inner operation mechanism.We brought forward the multi-agent co-evolution model based on evolutionary rule and strategy system, and deeply researched the method of multi-agent collaboration by the meansof combining co-evolution mechanism and evolutionary rule and strategy system. Firstly we made an introduce for the typical evolutionary rule and strategy system, SAMUEL. Then, we made a deep analysis for the multi-agent co-evolution model based on evolutionary rule and strategy system from the point of behavior engineering, and analyzed the evaluation problem of the behavior strategy in the collaboration process of multi-agent using co-evolution mechanism.Basing on the main idea of the architecture and collaboration mechanism of multi-agent system based on co-evolution method, we established the simulation framework of the multi-agent system based on co-evolution method lastly, and made an object oriented design for the simulation framework. We also built the basic class base of the multi-agent system based on co-evolution, and created an opening simulation experiment environment for the research of the multi-agent system based on co-evolution mechanism. At last, we made the experiment research on predator-prey problem, the typical research platform of multi-agent collaborative behavior, and drew the conclusion that by combining the co-evolution mec...