Research On Energy Management Strategy Of Hybrid Electric Vehicle Based On Data Driven Method

With the increase in the number of vehicles in China,the consumption of fossil fuels is also constantly rising.In the context of striving to achieve carbon peak and carbon neutrality,the development of hybrid electric vehicles(HEVs)has become an important solution because they combine the high energy efficiency of electric drive with the convenience of traditional petrochemical energy.HEVs can significantly reduce fuel consumption,increase driving range,and help achieve energy conservation and emissions reduction.The energy management strategy(EMS)of HEVs determines the energy distribution and is a key technology for improving vehicle fuel economy and overall performance,and is currently a research focus in the field of HEVs.This thesis focuses on a parallel HEV,establishes models of its major components of the powertrain system,and designs energy management strategies based on deterministic rules and dynamic programming(DP).Meanwhile,a supervised learning strategy based on the energy strategy generated by the deterministic rulebased approach in the model simulation data set is proposed.In addition,a data-driven offline reinforcement learning energy management strategy is further designed.The work included in this thesis is as follows:(1)In this thesis,a system model including engine,motor,battery,driveline,and vehicle longitudinal dynamics is developed for a parallel HEV.Based on this model,an energy management strategy based on deterministic rules and an energy management strategy based on dynamic planning are built.(2)By using the data obtained from the simulation of the deterministic rule-based energy management strategy in the built model,this thesis designs a supervised learningbased energy management strategy that takes the deviation of the current moment of the battery’s State of Charge(SOC)from the initial SOC,the demand torque,gear position,and vehicle speed information as the states of the strategy.With the goal of improving fuel economy while avoiding battery over-charge or over-discharge,the motor torque is used as the control output.Simulations were also conducted under four different driving cycles,WLTC,FTP-75,Artemis Urban,and Artemis Motorway.The simulation results show that the SOC stability may not be maintained under the working conditions with high-speed demand.(3)In this thesis an offline reinforcement learning energy management strategy based on a data-driven model is proposed.The construction process of the strategy,the interaction process between the reinforcement learning intelligences and the data-driven model,and the optimization method of the data-driven model are introduced.Finally,simulations are conducted on the parallel HEV model,and the results show that the energy management policy shows good adaptability and fuel economy under four different driving cycles.Compared with the logging policy(deterministic rule-based energy management policy),the relative fuel consumption reduction is 3%,5%,5.5% and 0.3% for WLTC,FTP-75,Artemis Urban and Artemis Motorway cycles,respectively.Finally,a hardware-in-the-loop experimental platform is built to validate the effectiveness of the offline reinforcement learning energy management strategy based on the data-driven model proposed in this thesis in a real-time environment.