Abstract: Based on the evolution game theory,the multi-agent systems is implemented,the system can be used to supporting collaborative management for dynamic coalition.Extended the traditional symmetric game to the discrete evolution game,which includes collaborative coefficient and punishment parameter,then designed evolution learning rules including historical information and neighbor characters,and applied java to programming the group evolution game agent simulation system based on Repast.Simulation analyt ical and numerical results lend insight into how impacts on the profit expectations in the several different size of individuals,different proportion of individual distribution,different communication mode and pun ishment parameters,which contribute some decision support for dynamic alliance under the environment of e-commerce or mobile commerce.

Key words: evolution game, multi-agent simulation system, dynamic coalition, mobile commerce