基于演化博弈的多智能体任务转发机制  

FORWARDING TASK MECHANISM OF MULTI-AGENT BASED ON EVOLUTIONARY GAME THEORY

作　　者：叶超[1] 许洪云[2] 李紫川 曹奇英[1] 

机构地区：[1]东华大学计算机科学与技术学院,上海201620 [2]东华大学信息科学与技术学院,上海201620

出　　处：《计算机应用与软件》2015年第4期15-18,103,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61272034)

摘　　要：在多智能体系统中,智能体执行任务时往往会碰到无法单独完成的任务,这个时候或者放弃任务,或者向其他智能体寻求帮助。在寻求帮助的过程中,假定某智能体不能全部广播任务到所有其他智能体(比如由于网络因素等只能局部广播到邻近的智能体),这种情况下,智能体更希望邻近的智能体尽可能多地接力转发自己委派的任务,最终选出能胜任该任务的智能体。针对这种问题,提出一种通用的面向多智能体的任务转发机制,同时考虑智能体自身的反思,并以演化博弈为理论基础进行分析。实验结果验证了多智能体系统中转发任务演化稳定的条件以及相关因素对稳定状态的影响。When performing a task in multi-agent systems,agents sometimes encounter such one which they cannot complete it alone.In this case,they either give up the mission or ask other agents for help.In the process of seeking help,assuming that an agent cannot broadcast tasks to all other agents （e.g.can only broadcast locally to adjacent agents due to the network factors）,then the agents will expect more their neighbours can relay one by one to forward the tasks they delegated as much as possible,and finally select the best agent who is competent. In response to such problem,we propose a common multi-agent-oriented task forwarding mechanism.Meanwhile the reflection of agent＆#39;s own is taken into consideration,and is analysed as well based on the evolutionary game theory.Experimental results validate the conclusion,the evolution stability conditions of forwarding the task and the effect of related factors on stable states in multi-agent system are derived.

关 键 词：多智能体 演化博弈 转发任务 奖励 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]