【作者】 王勇； 何洪文； 彭剑坤； 谭华春；

【Author】 Wang Yong;He Hongwen;Peng Jiankun;Tan Huachun;School of Mechanical Engineering;National Laboratory for Electric Vehicles,Beijing Institute of Technology;School of Transportation,Southeast University;

【机构】 北京理工大学机械与车辆学院； 北京理工大学电动车辆国家工程实验室； 东南大学交通学院；

【摘要】 插电式混合动力汽车作为我国新能源汽车"三纵"整车技术创新链之一,是新能源汽车中的一种重要类型。以能量管理为核心的汽车混合动力技术具有低油耗、低排放、长续驶的优势,使得插电式混合动力车型在市场上广受欢迎,以丰田THS为代表的全耦合架构混合动力系统已经衍生出混合动力版本与插电混合动力版本。本文对丰田THS混合动力系统的控制策略进行研究,提出一种基于数据驱动的智能能量管理策略。首先基于Python构建混合动力整车模型,在开源深度学习平台TensorFlow搭建深度强化学习算法。然后对算法的动作、状态、奖励进行设定,以构建马尔科夫决策模型。为验证该策略的性能,构建了大规模汽车工况数据来训练基于数据驱动的能量管理策略,在NEDC标准工况与GCDC本地实车运行工况对燃油经济性仿真。仿真结果表明提出的控制策略可达到动态规划95%的燃油经济性,比基于规则式的能量管理提升13%～15%;提出的控制策略计算时间与规则式的策略相近,证明其可实时在线应用。最后,数据驱动的控制策略在HEV与PHEV车型实验,验证了PHEV在纯油行驶模式下与HEV油耗差异在5%之内。更多还原

【Abstract】 As one of the "3-vertical" vehicle technology of new energy vehicles in China,plug-in hybrid vehicle(PHEV) is an important type of new energy vehicles.The hybrid technology based on energy management has the advantages of low fuel consumption,low emission and long driving distance,that the PHEV market grows strongly.The full coupling structure hybrid system represented by Toyota Hybrid System(THS) has derived the hybrid version and plug-in hybrid version.In this paper,the control strategy of the THS hybrid system is studied,and a data-driven intelligent energy management strategy is developed.Firstly,the whole vehicle model is built based on python,and the deep reinforcement learning algorithm is built in TensorFlow,an open-source deep learning platform.Then,the action,state and reward of the algorithm are set to build Markov decision model.In order to verify the performance of the strategy,a large-scale vehicle driving cycle dataset is constructed to train the data-driven energy management strategy,and the fuel economy is simulated under the NEDC standard driving cycle and GCDC real driving condition.The simulation results show that the proposed control strategy can achieves 95% fuel economy of the dynamic planning(DP),the average fuel economy is improved by 13% ～15%.The calculation time of the proposed control strategy is close to that of the rule-based strategy,which proves that it is liable to real-time and on-line control.Finally,the data-driven control strategy is simulated on HEV and PHEV models,and it is verified that the fuel consumption difference between PHEV and HEV in charge-sustaining mode is within 5%.更多还原