The Research Of Using Cooperative Co-Evolution Algorithm In Multi-Agent Cooperation

Multi-agent system (MAS) is a new field of AI. There are many interesting subjects in this field, such as distributed AI, sociology, psychology, etc. MAS has caused widely attention and been applied in many fields. However, there is no mature theory about the cooperation of MAS, which causes the problem that MAS can not be used effectively in practise. Dynamic programming and Markov decision, which are common methods in the cooperation mechanism of MAS, face the challenges such as uncertain environmental situation and the lack of integrated information got by agent. The cooperative co-evolution algorithm simulating the co-evolution of the nature species provides us a new way of cooperation mechanism of MAS. Cooperative co-evolution algorithm is applied to the cooperation of MAS in the paper. The main work is listed below:(1) A self-adaptive selection method of representation is proposed. The validity of this method is proved by the simulation of function optimization.(2) A cooperative co-evolution neuron method is proposed, which applies the cooperative mechanism of the cooperation co-evolution algorithm to evolving network.(3) A layered architecture is constructed based on cooperative co-evolution algorithm. One network is used to map environment to the action of one agent. And the network is evolved by the co-evolution neuron method. The cooperation of MAS is realized by combining the evolution of networks with cooperative co-evolution algorithm.