基于先验知识的强化学习系统

针对强化学习算法收敛速度慢的主要原因为强化学习算法所用模型通常都假设系统参数未知、先验知识未知，由此造成该算法从没有任何基础开始搜索最优策略，搜索范围大的问题，提出将强化学习系统建立在先验知识的基础上，既有效利用了前人的工作成果，又加快了算法的收敛速度．通过解决电梯群控问题验证了所提出系统的合理性和有效性．

Prior Knowledge Based Reinforcement Learning System

The slow rate of convergence is the main disadvantage of the reinforcement learning (RL) algerithm, which results from that the reinforcement learning model is based on model-free and always assuming any prior knowledge unknown. This makes the algorithm must devise a new policy from scratch. This paper presented an RL system based on prior knowledge relevant to the system, which avoids the duplication of these efforts and accelerates the convergence. An elevator domain knowledge based RL system was developed and it has a good performance.