Study On Coalition Formation In Multi-Agent System

The coordination and cooperation among agents in large-scale complicated Multi-Agent System is very important and coalition mechanism is a kind of cooperation for agents. Coalition generation is a key topic in Multi-Agent System. Using intelligent optimization algorithms to solve the agent coalition generation is an important research direction. The coalition utility allocation is another important component of Multi-Agent System and using the reasonable allocation strategy is of great significance. This thesis mainly studies the coalition generation under the condition of constrained computation resources and the allocation of coalition utility.The main research contents and contributions in this thesis are as follows:Coalition generation mainly focuses on how to generate the optimal task-oriented coalition in a dynamic manner. Compared with GA and ACA, the features of PSO algorithm are parallel, distributed and robust. This paper improves on above algorithms, presents a coalition generation strategy based on improved task-matching calculation method which can give full play to the agent learning ability, calculate the similarity more accurately between tasks in MAS, and decide more quickly how to learn from experience. By using this strategy the value of the coalition and the speed of the optimal solution convergence are improved. The experiments show that this strategy can decrease the retrieving time and computation effectively.How to allocate utility is a key problem in order to make the agents more willing to form coalition. Based on credit-degree, this paper proposes a coalition utility allocation scheme and the method. Through examples and the results of analysis, we can see that this approach better meets the request of the reasonableness, effectiveness and timeliness, and is conducive to improve the stability of coalition.