Abstract: The plug-in hybrid is energy efficient, and the energy management strategy among multiple power sources has a great impact on the energy consumption of hybrid vehicles. Taking a run on a fixed line series plug-in hybrid city bus as an example, based on the previous cycles, according to markov theory, the demand for power transformation along with the change of time for the state transition matrix, the random power sequence is generated by probability matrix. Module Q reinforcement learning and training, by using the random power sequence to realize the energy distribution of the city bus real-time optimization control. Simulation results show that the proposed reinforcement learning energy management algorithm can significantly optimize energy consumption compared with the rule control algorithm, and can greatly improve the time efficiency compared with dynamic programming and other global optimization strategies, and realize real-time control.

Key words: Plug-in hybrid electric vehicle; Energy management; Reinforcement learning; Transition probability matrix

摘要： 插电式混合动力汽车具有节能的特点，而多动力源之间的能量管理策略对混合动力汽 车能耗有很大的影响，故文章以一款在固定线路上运行的串联插电式混合动力城市客车为例， 基于以往的行驶工况，根据马尔科夫理论，把需求功率随时间的变化转换为状态转移矩阵， 用概率矩阵生成符合实际使用条件的随机功率序列。建立 Q 强化学习模块，使用得到的随机 功率序列进行训练，实现城市客车能量分配的实时优化控制。仿真结果表明，文章所提出的 强化学习能量管理算法相比于规则控制算法能明显的优化能耗，对比动态规划全局优化策略 总能耗仅轻微增长，控制策略可实现实时应用。

关键词: 插电式混合动力汽车；能量管理；强化学习；转移概率矩阵