考虑驾驶风格的混合动力汽车强化学习能量管理策略  

Reinforcement Learning-Based Energy Management Strategy Considering Driving Style for Hybrid Electric Vehicle

作　　者：施德华[1,2] 袁超 汪少华 周卫琪[1] 陈龙 SHI Dehua;YUAN Chao;WANG Shaohua;ZHOU Weiqi;CHEN Long(Automotive Engineering Research Institute,Jiangsu University,Zhenjiang,Jiangsu 212013,China;Jiangsu Province Engineering Research Center of Electric Drive System and Intelligent Control for Alternative Vehicles,Zhenjiang,Jiangsu 212013,China)

机构地区：[1]江苏大学汽车工程研究院,江苏镇江212013 [2]江苏省新能源汽车电驱动系统与智能控制工程研究中心,江苏镇江212013

出　　处：《西安交通大学学报》2024年第10期51-62,共12页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(51905219);中国博士后科学基金资助项目(2023M731444);中国科协青年人才托举工程资助项目(2020QNRC001)。

摘　　要：为了提升混合动力汽车能量管理策略对不同风格驾驶员的适应性,基于深度强化学习和等效燃油消耗最小策略(equivalent consumption minimization strategy,ECMS),提出一种考虑驾驶风格的混合动力汽车能量管理策略。通过实车试验采集驾驶员行驶数据,基于采集数据进行驾驶员驾驶风格的聚类分析,建立驾驶风格识别模型;构建基于强化学习和ECMS的能量管理策略,将驾驶风格系数作为强化学习状态变量,利用多种驾驶风格的组合工况训练深度确定性策略梯度智能体,获取不同工况和驾驶风格下ECMS等效因子,采用ECMS求解最优发动机、电机转矩分配以及变速箱挡位;搭建硬件在环测试平台,并基于实际采集的不同驾驶员驾驶数据构建测试工况,验证所提出控制策略的有效性。研究结果表明,相较于基于规则策略、基于等效因子比例修正的自适应ECMS以及DRL-SAC策略,提出的考虑驾驶风格的强化学习能量管理策略使整车能量消耗分别降低16.35%、11.11%和7.56%,所提控制策略的有效性得到了验证。A considering driving style energy management strategy is proposed to enhance the adaptability of energy management strategies for different driving styles in hybrid electric vehicles.The strategy combines deep reinforcement learning with the equivalent consumption minimization strategy(ECMS).Real vehicle experiments are conducted to collect driving data,which is then subjected to clustering analysis to identify distinct driving styles.A driving style recognition model is developed based on this data.The energy management strategy is built using reinforcement learning and ECMS,with driving style coefficients serving as the reinforcement learning state variables.A deep deterministic policy gradient agent is trained using various combinations of driving styles and operating conditions to determine ECMS equivalent factors for different driving styles and conditions.The ECMS is employed to optimize the engine,motor torque allocation,and gearbox gear selection.To validate the effectiveness of the proposed control strategy,a hardware in the loop testing platform is constructed,and test scenarios are generated using real driving data from different drivers.The research findings demonstrate that the reinforcement learning-based energy management strategy considering driving style reduces overall vehicle energy consumption by 16.35%,11.11%,and 7.56%compared with rule-based strategy,equivalent factor ratio correction-based adaptive ECMS,and DRL-SAC strategy,respectively.The effectiveness of the proposed control strategy is successfully validated.

关 键 词：混合动力汽车 能量管理策略 驾驶风格 强化学习 等效燃油消耗最小策略 

分 类 号：U463.2[机械工程—车辆工程]