信用制度与合作的Multi-Agent仿真模型  

A Multi-Agent Simulation Model of Credit System and Cooperation

作　　者：张发[1] 赵巧霞[2] 

机构地区：[1]西安交通大学管理学院,西安710049 [2]空军工程大学电讯学院,西安710077

出　　处：《计算机工程与应用》2004年第24期97-99,217,共4页Computer Engineering and Applications

摘　　要：信用体系中的各项制度对合作的作用尚缺乏充分的研究。文章用Multi-Agent仿真方法建立了一个人工社会,其中的主体进行重复囚徒博弈,并通过模仿学习不断进化。引入一类信誉机制和惩罚机制,该信誉机制计算主体信誉水平,主体交互前根据对方信誉水平调整策略,但不能选择伙伴或拒绝交互;惩罚机制对反信用行为进行强制惩罚。信誉机制、惩罚机制的有无构成四种信用体系,分别进行多次仿真。结果显示:惩罚机制能显著促进合作,而该信誉机制对促进合作无显著效果。The effect of different institutions in credit system is not very clear.An artificial society is built based on multi-agent simulation theory.There are multiple agents playing iterated prisoner's dilemma and adapting by imitation learning.A credit system composed of reputation and punishment is introduced.Within the reputation institution agents change their strategy temporarily based on rival's reputation but without partner choice and refusal.Within the punish-ment institution cheater will be fined.Four different credit systems are combined.The result of multiple runs of each credit system shows that punishment can promote cooperation significantly,but this kind of reputation institution not.

关 键 词：信用制度 重复囚徒博弈 信誉 惩罚 仿真 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]