• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2021, Vol. 57 ›› Issue (22): 237-246.doi: 10.3901/JME.2021.22.237

• 运载工程 • 上一篇    下一篇

扫码分享

基于深度强化学习的混合动力汽车智能跟车控制与能量管理策略研究

唐小林1, 陈佳信1, 刘腾2, 李佳承1, 胡晓松1   

  1. 1. 重庆大学汽车工程学院 重庆 400044;
    2. 滑铁卢大学机械与机电学院 滑铁卢 ON N2L 3G1 加拿大
  • 收稿日期:2020-11-02 修回日期:2021-05-20 出版日期:2021-11-20 发布日期:2022-02-28
  • 通讯作者: 胡晓松(通信作者),男,1983年出生,博士,教授。主要研究方向为电动汽车电池管理技术、机电复合动力传动系统优化与控制。E-mail:xiaosonghu@ieee.org
  • 作者简介:唐小林,男,1984年出生,博士,副教授,博士研究生导师。主要研究方向为混合动力汽车NVH与能量管理。E-mail:tangxl0923@cqu.edu.cn;陈佳信,男,1996年出生,硕士研究生。主要研究方向为混合动力汽车能量管理与深度强化学习。E-mail:201932132050@cqu.edu.cn
  • 基金资助:
    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)和重庆市自然科学基金(cstc2020jcyj-msxmX0956)资助项目。

Research on Deep Reinforcement Learning-based Intelligent Car-following Control and Energy Management Strategy for Hybrid Electric Vehicles

TANG Xiaolin1, CHEN Jiaxin1, LIU Teng2, LI Jiacheng1, HU Xiaosong1   

  1. 1. School of Automotive Engineering, Chongqing University, Chongqing 400044;
    2. Mechanical and Mechatronics Engineering, University of Waterloo, Waterloo ON N2L 3G1, Canada
  • Received:2020-11-02 Revised:2021-05-20 Online:2021-11-20 Published:2022-02-28

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。

关键词: 混合动力汽车, 深度强化学习, 跟车控制, 能量管理

Abstract: Limited by the cost, high-speed on-off valves are often used in the electro-hydraulic braking system. However, the knocking noise of the solenoid valve, hydraulic impact noise and pressure fluctuation in the switch control of the hydraulic valve lead to the quality and precision of brake control to deteriorate. Therefore, the key to the design of the high-speed on-off valve is to realize the linearization by pulse width modulation (PWM) control. The dynamic characteristics of high-speed on-off valve are affected by transient flow force, nonlinear electromagnetic force, mechanical inertia and spring force. Therefore, the dynamic model and joint simulation model of high-speed on-off valve and hydraulic control unit (HCU) are established. Through simulation and experimental verification, the influence of structural parameters such as poppet angle, orifice diameter, air gap size on the linear characteristics of solenoid valve is analyzed. The reasonable structure parameters of the solenoid valve are designed and applied to a hydraulic control circuit to widen the linear working range and meet the requirements of automobile braking safety control.

Key words: hybrid electric vehicle, deep reinforcement learning, car-following control, energy management

中图分类号: