集成博弈和多智能体的人群工作互动行为研究  

Study on interaction behavior between group and work based on game and multi-agent

作　　者：蒋国银[1,2] 胡斌[1] 

机构地区：[1]华中科技大学管理学院,武汉430074 [2]湖北经济学院信息管理学院,武汉430205

出　　处：《管理科学学报》2011年第2期29-41,共13页Journal of Management Sciences in China

基　　金：国家自然科学基金资助项目(70671048);湖北省教育厅人文社科资助项目(2010q091)

摘　　要：基于进化博弈视角,对人群工作互动行为进行多智能体模拟研究.建立了收益和惩罚共享的群体工作收益博弈模型,考虑工作个体的个性决策特征,设计基于历史信息和个体决策特性的混合学习规则,并用多智能体方法对群体工作场景进行描述.在Repast类库基础上,用Ja-va实现该多智能体模拟系统.模拟结果表明:1)群体规模对宏观工作趋势影响小,但对合作频率影响大;2)工作总收益b越大,越有利于工作人群的工作状态稳定.当工作付出c与惩罚d相当时,群体行为呈针锋相对态,当工作付出大于惩罚时,背叛占优,反之,则合作占优;3)工作难度对群体行为有影响.高难度下,群体行为不稳定,获利低,风险大,而低难度下,群体行为稳定,获利高,风险小;4)由具有不同比例决策特性个体组成的群体工作收益和状态各异,保守型和中立个体较多时,合作比例大,而获利也多.该研究可为电子/移动商务环境下的工作行为管理问题提供决策支持.Agent-based simulation for interaction behavior between group and work was explored from an evolutionary game-theoretical perspective. This paper develops a payoff-shared and punishment-shared game model, design evolution learning rules considering of historical information and decision characteristic of neighbors, and uses multi-agent approach to represent group work. Based on class lib of Repast, we use Java 2 to program the multi-agent simulation system. Simulation results indicate that （ 1 ） the size of work group has a minor effect on cooperation trend and major effect on cooperation frequency of group, （2） total work payoff b has a positive effect on the stability of work state of group. Group behavior is in a state of Tit-for-Tat when work cost c is equivalent to work punishment d, more players want to cooperate when c 〉 d, and vice versa, the more players wants to defect when c 〈 d, （3） group behavior in doing highly difficult work is unstable, players can get lower profit from more highly difficult work with larger risk, or vice versa; and （4） groups made up of individuals with different decision-making characteristics have different effects on the work state of group. The number of conservative and neutral individuals has a positive effect on cooperation and profit. This study contributes some decision support to behavior management under E-commerce or Mobile commerce environment.

关 键 词：进化博弈 互动行为 多智能体 混合学习 

分 类 号：C93[经济管理—管理学]