基于每阶段平均费用最优的激励学习算法

文中利用求解最优费用函数的方法给出了一种新的激励学习算法，即基于每阶段平均费用最优的激励学习算法。这种学习算法是求解信息不完全Markov决策问题的一种有效激励学习方法，它从求解分阶段最优平均费用函数的方法出发，分析了最优解的存在性，分阶段最优平均费用函数与初始状态的关系以及与之相关的Bellman方程。这种方法的建立，可以使得动态规划（DP）算法中的许多结论直接应用到激励学习的研究中来。

REINFORCEMENT LEARNING ALGORITHM BASED ONAVERAGE COST OPTIMIZATION FOR EACH STAGE

This paper is concerned with the problem of a novel reinforcement learning algorithm for solving optimal average cost function. Q-learning is a reinforcement learning method to solve Markovian decision problems with incomplete information. This paper begins with solving optimal average cost function for each stage, studies the existence of optimal solution, the relationship between the optimal average cost function for each and initial state, and corresponding Bellman equation,proposes the relative value iteration Q-learning algorithm. It can make many results of dynamic programming algorithm for studying Q-learning directly.