一类基于启发式搜索的激励学习算法  

A Class of Reinforcement Learning Algorithm Based on Heuristic Search

作　　者：唐中勇[1] 付强[1] 卓佳[1] 陈焕文[1] 

机构地区：[1]长沙理工大学计算机通讯工程学院,湖南长沙410076

出　　处：《计算机技术与发展》2006年第8期41-43,共3页Computer Technology and Development

摘　　要：激励学习已被证明是在控制领域中一种可行的新方法。相比其他的方法,它能较好地处理未知环境问题,但它仍然不是一种有效的方法。幸运的是,在现实世界中,智能体总是会有一些环境的先验知识,这些能形成启发式信息。启发式搜索是一种常用的搜索方法,有很快的搜索速度,但需要精确的启发式信息,这在有些时候难以得到。文中分析比较了启发式搜索和激励学习的各自特点,提出一类新的基于启发式搜索的激励学习算法,初步的实验结果显示了较好的性能。The reinforcement learning has been proved to be a new applicable method in control field. It can solve the problems of unknown environment better than the others. But it isn＇t a very effective method yet. Fortunately in real world,the agent often has some knowledge of the environment,which can be used as heuristic information. The heuristic search is a very effective search method,which can search very quickly. But it need very precise heuristic information, which may be hard to get in complex environment. The characteristics of heuristic search and reinforcement learning are compared and a class of reinforcement learning algorithm on heuristic search is introduced. The preliminary empirical result shows better than the previous.

关 键 词：启发式搜索 激励学习 启发式SARSA 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]