一种有限时段Markov决策过程的强化学习算法

An algorithm of reinforcement learning for finite-horizon Markov decision processes

研究有限时段非平稳的 Markov决策过程的强化学习算法。通过引入一个人工吸收状态 ,把有限时段问题变为无限时段问题 ,从而可利用通常的强化学习方法来求解。在文献 [3]提出的算法思想基础上 ,提出了一种新的有限时段非平稳的 Markov决策过程的强化学习算法 。

In this paper, reinforcement learning algorithms for finite horizon non stationary Markov decision processes are studied By introducing an artificialabsorbingstate, the finite horizon problem transforms to an infinite horizon one, so that a normal method of reinforcement learning algorithm can be used to solve the finite-horizon problem A new reinforcement learning algorithm of the finite horizon non stationary Markov decision process is put forward based on the algorithm thought presented by reference book\ And an experiment in inventory control problem with non complete model has been done