MAS's Structure And The Application Of Genetic Algorithm In Its Cooperation Mechanism

The theory and technology of agent, especially the Multi-agent, provide a new way for analysis, design and realization of distributed open system, and has been hailed as " the next breakthrough in software development". Theory and technology of agent originally come from Distributed Artificial Intelligence (DAI), with the development of the research of DAI, some new research fields appear such as coordination theory, distributed reasoning, the learning and communication language between the agents. And these researches make the research of DAI gradually turn to the direction of intelligent and independent agents. Since the 1980's, theory and application of agent has extended from DAI and syncretized and referenced with other fields. Agent has been a hot term in the field of distributed computation environment, and agent technology has been one hotspot in the research area of artificial intelligence.Based on the theory and method of the Multi-agent System, this thesis analyzes the dynamic trait of MAS's structure. In order to avoid the superfluous communication cost in forming MAS, th: dynamic evolvement of MAS's structure must be considered. This thesis presents the restriction of the dynamic consistency of MAS's structure: integer dependence and object-extendable, and gives the manipulative formalization description of MAS's structure. And this thesis demonstrates the dynamic consistency of MAS's structure, and provides the theory foundation for the theory of MAS's structure.In addition, this thesis analyzes the present correlated work of the coalition formation problem in MAS, presents the shortcomings of these work, and presents a coalition formation method considering some main factors: task decomposing, attributes of agents, dependent relations among agents. Aimed at the GA's disadvantages: weak local optimization, easiness of falling into local extremum, this thesis presents an improved GA which introduces two operators "local search" and "adaptively avoid entropy decreasing". The improved GA not only increases the ability of local search through the local search operator, but also breaks the monopoly predominance of the genes, and increases the diversity of population. The simulation result shows the effectiveness of the improved algorithm. Then this thesis uses the improved GA to solve coalition formation problem, and the simulation result illustrates the effectiveness of GA and the feasibility of algorithm in this problem.Finally, this thesis discusses the task-allocation problem in MAS and the improved GA to solve the problem is illuminated. And this thesis compares the operation performance of the improved GA to the standard GA and GA&SA hybrid algorithm. The simulation result illustrates the effectiveness of the improved GA and the feasibility of algorithm in this problem. In the end, this thesis discusses the meaning of this algorithm in MAS, and points out its application importance.