强化学习在混合动力汽车能量管理方面的应用  

Research on the Application of Reinforcement Learning in Hybrid Electric Vehicle Energy Management

作　　者：郑春花[1] 李卫 ZHENG Chun-hua;LI Wei(Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China)

机构地区：[1]中国科学院深圳先进技术研究院,广东深圳518055

出　　处：《哈尔滨理工大学学报》2020年第4期1-11,共11页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金(51305437);深圳市海外高层次人才创新创业专项(KQJSCX20180330170047681).

摘　　要：针对混合动力汽车(hybrid electric vehicle,HEV)的能量管理问题,提出一种基于强化学习(reinforcement learning,RL)的能量管理策略。首先建立HEV的动力系统模型和需求功率的马尔科夫概率转移模型,接着设计了基于RL的控制策略,最后与基于动态规划算法(dynamic programming,DP)和基于规则的能量管理策略进行比较分析。在UDDS,NEDC,Japan1015三个循环工况下,基于RL的策略燃油经济性相比基于规则的能量管理策略,分别提高了14.4%,10.22%和7.67%,而且燃油经济性均达到基于动态规划策略的92%以上,表明了基于RL能量管理策略的有效性。Aiming at the energy management of hybrid electric vehicles(HEVs),an energy management strategy based on the reinforcement learning(RL)is proposed.First,the HEV′s power system model and the Markov probability transfer model of the required power are established.Then,the RL-based control strategy is designed.Finally,it is compared with the dynamic programming(DP)algorithm and the rule-based energy management strategy.In UDDS,NEDC and Japan1015,the fuel economy of the RL-based strategy is 14.4%,10.22%,and 7.67%higher than that of the rule-based energy management strategy respectively.In addition,the fuel economy is over 92%of that of the DP algorithm on the three driving cycles,showing the effectiveness of the RL-based energy management strategy.

关 键 词：混合动力汽车 能量管理策略 强化学习 Q学习 

分 类 号：U469.7[机械工程—车辆工程]