一种基于统计的多智能体Q学习算法  

A Multi-Agent Q-Learning Method Based on Statistic

作　　者：谭天晓[1] 赵辉[1] 赵宗涛[1] 

机构地区：[1]西安高技术研究所,陕西西安710025

出　　处：《微电子学与计算机》2008年第1期17-19,24,共4页Microelectronics & Computer

摘　　要：针对非确定马尔可夫环境下的多智能体系统,提出了一种新的基于统计的多智能体Q学习算法,该算法将统计学习与增强学习有机结合,有效地解决了智能体环境部分感知、信息不确定以及其他智能体行为策略学习的问题,经RoboCup仿真实验证明该算法具有较强的自适应能力和学习效率。This paper proposes a new multi-agent Q -learning method based statistic for MAS in uncertain Markov circumstance. The new algorithm solves problems, including partial information, uncertainty information and interacting with other agent effectly, through jointing statistic and reinforcement learning. The learning efficiency of algorithm is exceeding the conventional Q -learning in RoboCup simulation.

关 键 词：Q学习 多智能体系统 统计 ROBOCUP 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]