Research On Multi-agent Cooperation Based On Co-evolution Algorithm

Theory and applications of multi-agent system involve multiple disciplines, especially controland computer technology, it has become a new hotspot research in the field of distributed artificialintelligence. Most of the theoretical study of multi-agent system is still abstract, much theoryresearch is not mature. The cooperation mechanism of multi-agent system is the key problem of themulti-agent theory in a complex dynamic environment. Developed by imitating the mechanism ofmutual restraint and mutual benefit between species in nature, co-evolution algorithm provides anew idea for multi-agent cooperation research. Co-evolution algorithm is applied to the multi-agentcooperation in this paper. The main research work as follow:(1) For the problems that can not determine the inter-activity between sub-components, thispaper proposes self-adjusting representation selection co-evolution algorithm. The simulationresults show that this algorithm is effective to problems with different inter-activity degree betweensub-components.(2) This paper presents punctuated anytime learning method and sampling method to reducethe computation. Function optimization experiments prove that these two methods can reduce thecalculation of the co-evolution algorithm and ensure the evolution results.(3) Co-evolution algorithm is applied to the multi-agent system. A hierarchical interactivestructure of agent architecture and the cooperation mechanism of multi-agent system based onco-evolution algorithm are designed in this paper.(4) The cooperation mechanism of multi-agent system is applied to the system of multi-robotformation control, the results of simulation experiment demonstrate that the cooperation mechanismof multi-agent system based on co-evolution algorithm is effective.