汽车工程 ›› 2022, Vol. 44 ›› Issue (9): 1400-1409.doi: 10.19562/j.chinasae.qcgc.2022.09.011

所属专题: 新能源汽车技术-电驱动&能量管理2022年

• • 上一篇    下一篇

基于TD3-PER的混合动力履带车辆能量管理

张彬,邹渊(),张旭东,杜国栋,孙文景,孙巍   

  1. 北京理工大学机械与车辆学院,北京  100081
  • 收稿日期:2021-12-30 修回日期:2022-04-09 出版日期:2022-09-25 发布日期:2022-09-21
  • 通讯作者: 邹渊 E-mail:zouyuanbit@vip.163.com
  • 基金资助:
    国家自然科学基金(51775039)

Energy Management Strategy Based on TD3-PER for Hybrid Electric Tracked Vehicle

Bin Zhang,Yuan Zou(),Xudong Zhang,Guodong Du,Wenjing Sun,Wei Sun   

  1. School of Mechanical Engineering,Beijing Institute of Technology,Beijing  100081
  • Received:2021-12-30 Revised:2022-04-09 Online:2022-09-25 Published:2022-09-21
  • Contact: Yuan Zou E-mail:zouyuanbit@vip.163.com

摘要:

为优化串联式混合动力履带车辆(SHETV)的燃油经济性和动力电池性能,提出一种基于优先经验采样的双延迟深度确定性策略梯度(TD3-PER)能量管理策略。TD3算法能实现更精准的连续控制和防止训练陷入过优估计。优先经验采样(PER)算法可加速策略的训练和获得更高的优化性能。在建立包括纵横向动力学的车辆模型的基础上,完成基于TD3-PER的能量管理策略的框架构建和仿真验证。结果表明,与深度确定性策略梯度(DDPG)相比,所提出的策略使SHETV的燃油消耗降低了3.89%,燃油经济性达到了作为基准的动态规划算法的95.05%。同时该策略具有较好的电池SOC保持能力和工况适应性。

关键词: 串联式混合动力履带车辆, 双延迟深度确定性策略梯度, 连续控制, 优先经验采样

Abstract:

To optimize the fuel economy and traction battery performance of series hybrid electric tracked vehicle (SHETV), an energy management strategy (EMS) based on twin delayed deep deterministic policy gradient with prioritized experience replay (TD3-PER) is proposed. The TD3 algorithm can achieve more precise continuous control and prevent training from falling into over-assessment. The PER algorithm can accelerate strategy training and obtain higher optimization performance. Based on the model of the SHETV including longitudinal and lateral dynamics, the framework construction and simulation verification of EMS based on TD3-PER is completed. The results show that compared with deep deterministic policy gradient algorithm, the strategy proposed reduces the fuel consumption of SHETV by 3.89%, making its fuel economy reaching 95.05% of DP algorithm as a benchmark, with a better battery SOC retention ability and working condition adaptability.

Key words: series hybrid electric tracked vehicles, twin delayed deep deterministic policy gradient, continuous control, prioritized experience replay