多步截断优先扫描强化学习算法  

A Truncated Multi-step Prioritized Sweeping Reinforcement Learning Algorithm

作　　者：李春贵[1] 

机构地区：[1]广西工学院计算机工程系,柳州545006

出　　处：《计算机工程》2005年第11期13-15,共3页Computer Engineering

基　　金：广西自然科学基金资助项目(桂科自0481016);广西工学院博士基金资助项目

摘　　要：研究了优先扫描的强化学习方法,通过定义新的迹,把多步截断即时差分学习用于集成规划的优先扫描强化学习,用多步截断即时差分来定义扫描优先权,提出一种改进的优先扫描强化学习算法并进行仿真实验,实验结果表明,新算法的学习效率有明显的提高。Prioritized sweeping reinforcement learning method is investigated. Giving a new definition of eligibility traces, truncated multi-step temporal difference learning can be used in integrating planning prioritized sweeping learning. Setting the priority as truncated multi-step temporal errors, an advanced prioritized sweeping RL algorithm is proposed. The results of experiment show that the algorithm has more effective performance.

关 键 词：强化学习 优先扫描 多步截断 MARKOV过程 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]