Abstract: A Social Cooperation based Multi-Agent Evolutionary Algorithm (SCMAEA) which integrates the social cooperation mechanism and multi-agent evolution for numerical optimization is proposed. Using the social cooperation mechanism, trust degree, which denotes the historical information for agents, is defined to control the interaction between agents. At the same time, the ‘acquaintance net model is imported to construct and update the local environment of the agent. It improves the convergence rate by the cooperation characteristic of agents. Furthermore, adopting the non-uniform mutation operation improves the searching for optimal solutions in the local region and assures the diversity of the solution. Simulation results show that compared with the multi-agent genetic algorithm, the social cooperation based multi-agent evolutionary algorithm can find the optima by a smaller number of function evaluations.

Key words: function optimization, multi-agent evolution, social cooperation mechanism, acquaintance net, convergence