Research On Coalition Mechamism In Distributied Intelligent System

The coordination and cooperation among intelligent subsystems in large-scale complicated system is a key problem. There is a lack of effective method in the classical Large System Theory. Basing on the theory of Coalition in Multi-Agent System, did some theoretical research in the problem, and presented some effective collaborative algorithms, strategies, so as to provide theoretical and methodological guidance for the manufacture of practical application systems. The main research contents and innovations in this dissertation are as follows:(1) Coalition generation algorithm in computation resource bounded environment. Given certain task, proper Agents will be chosen and activated to form the task-solving coalition, consequently the task can be accomplished with the optimal configuration and highest efficiency. This is the most important manner for MAS to run. It is the chief problem for the system to generate the optimal task oriented Agent coalition duly in the computation resource bounded practical system. An ant colony system based Agent coalition generation algorithm was presented for the problem. During the process of solution, ants incline to select those Agents who cooperated well before to form coalitions, which realizes the acquaintance mechanism perfectly. A novel "inner pheromone" was proposed to improve the ACA so as not to get in the premature convergence easily. Furthmore, for the task alignment, the algorithm can generate the optimal coalitions one after another, and the learning ability of ant colony system can reduce the searching time and computing works effectively! The results of experiment show that this algorithm is successful to solve the self-organizing, optimal-cooperating problem among mutli-intelligent subsystems in the complicated distributed intelligent system.(2) Utility allocation strategy for coalition formation. In MAS, agents can get utility after completing a task, and by forming coalition, agents can improve their task solving ability and obtain more utility. So an agent will choose some other agents to form such a coalition to enlarge its utility. In order to encourage agents to form coalitions, the MAS designer must specify an appropriate rule for utility allocation. A novel utility allocation strategy based on benefit equilibrium was proposed. It remains the principle of non-decreasing utility allocation, improves the justice of division for accrued utility and does well to the formation of global optimal coalition. In task oriented domains this strategy can achieve a global optimal solution, which is stable, speedy and distributed. This strategy provides the basis for designing the evolving mechanism of the distributed intelligent system.(3) Communication model for the distributed intelligent system. Fluent communication is the base of cooperation among mutli-intelligent subsystems. A hierachical naming and locating mechanism of agents was designed for Agent to communicate with each other directly. Several KQML performatives were extended to form the describing language on content layer, and they can guarantee the exact semantic understanding of the transmitted information. Thereby, the communication model for distributed intelligent system was built.(4) For the two key problems in manufacture system: supply chain partner selection and profit allocation, applying the Agent coalition generation algorithm and utility allocation strategy, devised corresponding methods, and illustrated their validity. Finally, using the Visual C++ language developed an agile supply chain decision-making supporting system "ASCDSS".