【作者】 刘伟兵； 王先甲；

【Author】 LIU Wei-bing~1 WANG Xian-jia~(1,2) (1.Institute of Systems Engineering,Wuhan University,Wuhan 430072,China; 2.College of Science,Wuhan University of Science and Technology,Wuhan 430081,China)

【机构】 武汉大学系统工程研究所； 武汉科技大学理学院；

【摘要】 基于Q-强化学习算法,建立了进化博弈中代理人的决策模型.考虑到强化学习算法不需要建立环境模型,可以用于不完全、不确定信息问题中,将Q-强化学习算法引入到进化博弈中,研究了进化博弈中的两种Q一学习决策模型:单代理人Q-学习决策模型和多代理人Q-学习决策模型,并针对不同结构的进化博弈选择不同的决策模型和算法进行了讨论.仿真算例的结果说明基于Q-学习的决策模型能指导代理人学习、选择最优策略.更多还原

【Abstract】 Based on Q-reinforcement learning,decision-models of gents in evolutionary games are established.Considering that reinforcement learning does not need a model of its environment, and it can be used in problems with incomplete and uncertain information.Q-learning is introduced in evolutionary games in this paper,and the single-agent Q-learning decision-model and the multi- agent Q-learning decision-model are presented.In addition,the paper disciples how to choose de- cision-models and algorithms according to the type of games.The results of simulation experiments show that the decision-modeh based on Q-learning can make agents choose the optimal strategy by learning.更多还原