摘要
以研究智能混合动力汽车控制技术与深度强化学习算法为目标,首先,在两辆混合动力汽车的跟驰环境中,针对领航车提出一种基于深度值网络算法的能量管理策略,实现深度强化学习对发动机与机械式无级变速器的多目标协同控制;其次,针对跟随车建立基于深度强化学习的分层控制模型,实现面向智能混合动力汽车的上层跟车控制与下层能量管理;最后,仿真验证分层控制模型的有效性。结果表明,基于深度强化学习的跟车控制策略具有理想的跟踪性能;同时,基于深度强化学习的能量管理策略在领航车与跟随车中均实现了较好的燃油经济性;此外,基于深度强化学习的能量管理策略输出每组控制动作的平均时间为1.66 ms,保证了实时应用的潜力。
The aims is to study the control technology of intelligent hybrid electric vehicles(HEVs)and deep reinforcement learning(DRL)algorithms.Firstly,under the car-following model of two HEVs,a deep q-network(DQN)-based energy management strategy(EMS)for the leading car is proposed,which realizes the multi-objective collaborative control of the engine and the continuous variable transmission(CVT)by DRL.Secondly,a hierarchical control model based on DRL is established for the following car,which realizes the upper-level car-following control and lower-level energy management facing to an intelligent HEV.Finally,a simulation verifies the effectiveness of the hierarchical control model.The results show that the DRL-based car-following control strategy has ideal tracking performance.Meanwhile,the DRL-based EMS achieves good fuel economy in both the leading car and the following car.Moreover,the average time of outputting each set of actions is 1.66 ms for the DRL-based EMS,which ensuring the potential for real-time applications.
作者
唐小林
陈佳信
刘腾
李佳承
胡晓松
TANG Xiaolin;CHEN Jiaxin;LIU Teng;LI Jiacheng;HU Xiaosong(School of Automotive Engineering,Chongqing University,Chongqing 400044;Mechanical and Mechatronics Engineering,University of Waterloo,Waterloo ON N2L 3G1,Canada)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2021年第22期237-246,共10页
Journal of Mechanical Engineering
基金
国家自然科学基金(52072051)
汽车测控与安全四川省重点实验室(QCCK2020-006)
重庆市自然科学基金(cstc2020jcyj-msxm X0956)资助项目
关键词
混合动力汽车
深度强化学习
跟车控制
能量管理
hybrid electric vehicle
deep reinforcement learning
car-following control
energy management
