基于强化学习的多Agent系统规划规则抽取方法  

Planning Rule Extracting Algorithm of Multi-Agent System Based on Reinforcement Learning

作　　者：赵志宏[1] 黄蕾[1] 刘峰[1] 骆斌[1] 

机构地区：[1]南京大学软件学院,江苏南京210093

出　　处：《广西师范大学学报（自然科学版）》2008年第1期174-177,共4页Journal of Guangxi Normal University:Natural Science Edition

基　　金：国家自然科学基金资助项目(60503021);江苏省高新技术资助项目(BG2006027)

摘　　要：强化学习和规划技术在目标上有着很高的相似性,而在技术上又具有互补性,因此,基于强化学习的Agent规划规则抽取问题长期以来一直是研究的热点。针对基于强化学习的多Agent系统在规划规则抽取方面存在的问题,提出了一种从多Agent Q学习中抽取满足规划条件的规划规则的RL-MAPRE算法,并给出了理论分析。As two important aspects of artificial intelligence research, the reinforcement learning and planning method are very similar in retrieving goals, while they are also inter-complementary in the technique. Therefore, taking planning rules from reinforcement learning based system has been a hot spot for a long time in the field. This paper proposes a RL-MAPRE algorithm, which extracts the planning required rules in the multi-agent Q-learning based on Nash equilibrium.

关 键 词：强化学习 多AGENT系统 规划 规则抽取 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]