基于博弈论及惩罚机制的多Agent协作控制算法  

The Algorithm for Mulit-agents Cooperation Controling Based on Game Theory and Punishment Mechanism

作　　者：郑延斌[1,2] 陶雪丽[1,2] 段领玉 李波[1] 

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453007 [2]河南师范大学智慧商务与物联网技术河南省工程实验室,河南新乡453007

出　　处：《河南师范大学学报（自然科学版）》2015年第6期146-151,共6页Journal of Henan Normal University(Natural Science Edition)

基　　金：河南省重点科技攻关项目(132102210537;13210221053)

摘　　要：针对协作过程中自利的Agent选择回报更高的任务去执行,导致当前任务不能正常执行的问题,提出一种基于博弈论及惩罚机制的协作控制方法.在形成协作时优先选择信誉值较高的Agent,在协作执行过程中利用惩罚机制来约束退出协作的Agent的行为.仿真结果表明,该算法能有效地避免Agent在协作中随意的退出,确保协作任务的顺利执行,提高协作成功率.For the problem of currents task cannot be normally performed when selfish agents would select higher utility tasks during collaborative process,a cooperative controlling method based on game theory and punishment mechanism was proposed.In phase of cooperative teamforming,selected agents which have higher credibility,during taskexecuting,used punishment mechanism to constrain the behaviors of agent for exiting.Simulation results showed that the algorithm can effectively avoid agent arbitrary exiting collaboration,ensured the smooth implementation of collaborative tasks and improve the success rate of collaboration.

关 键 词：博弈论 协作 惩罚机制 信誉值 MULTI-AGENT 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]