【摘要】 提出了一种自适应状态集结因子化 SARSA(λ)强化学习算法 ,在学习的过程中利用Bellman余留数进行状态集结 ,通过集结 ,大大减少了状态空间搜索与计算的复杂度 ,有利于求解大状态空间的 MDPs问题 ,而且 ,本算法不需要有关状态特征的先验知识 ,有很好的通用性更多还原

【Abstract】 We propose an new state space aggregation SARSA(λ) algorithm. The main principle of the algorithm is based on Bellman residual, state space is aggregated, significantly reducing the searching and computing complexity of the state space.The algorithm is a promise solving for large scale MDPs problem., and more,it needn′t the prior knowledge of state space feature.更多还原