Research On Control System Design And Energy Management Strategy Of Hybrid Electric Vehicle

In recent years,the energy utilization efficiency of energy-saving and newenergy vehicles has been greatly improved with the rapid development of electrical and intelligent technologies related to vehicles.At the same time,it is required by policies and regulations to increasingly improve the vehicles' energy utilization efficiency.How to further improve the energy utilization efficiency by using new intelligent technology is an important issue for energy-saving and new-energy vehicles.The design of Hybrid Electric Vehicle(HEV)control system and the intelligent energy management strategies(EMSs)were studied in this dissertation.The vehicle under research is a parallel hybrid electric vehicle.The simulation model of the vehicle was established.The vehicle control system of hybrid electric vehicle was designed along with the introduction of its hardware design method,model-based control strategy design method and practical application on the real vehicles.In order to ulteriorly enhance the efficiency of energy utilization for HEV,three learning based EMSs were proposed: EMS based on fuzzy Q learning(FQL),EMS based on deep reinforcement learning(DRL)and EMS based on deep deterministic policy gradient(DDPG).FQL based EMS was proposed to address the problem that the fuzzy control method cannot adapt to different driving cycles.Q learning method and fuzzy control were combined to realize the self-tuning of fuzzy control parameters.The proposed method incorporates the Q estimator network(QEN),the fuzzy parameters tuning(FPT)as well as the action exploration modifier(AEM).The FQL based EMS was verified through simulation experiments.Simulation results indicate that the FQL based EMS achieves good performance in fuel economy as well as the fuzzy control parameters can tune online to be adapted to different driving cycles.The fuzzy rules of FQL based EMS need to be designed by experts.Therefore,a data-driven deep reinforcement learning(DRL)based EMS was designed without experts' experience.DRL method combines reinforcement learning and deep learning to form a deep Q network(DQN)that can learn to select actions directly from the states without any prediction or predefined rules.Therefore,it is an end-to-end method.The key concepts of DRL based EMS were introduced,and then the deep neural network was established to estimate the Q function.Finally,the algorithm stepswere described.Furthermore,a DRL based online learning framework was presented to adapt to different driving cycles.Simulation validation has been conducted.Results validate the effectiveness of the proposed DRL based EMS in both offline and online cases in terms of fuel economy.The output actions need to be discretized in DRL based EMS.However,the engine torque is a continuous output in HEV EMS.The discrete actions cannot adequately approximate all continuous actions.Accordingly,DDPG based EMS was proposed in order to output actions directly without action discretization.This algorithm is based on actor-critic framework,mainly comprising actor network and critic network.The framework and algorithm steps of DRL based EMS were presented.Simulation results show good performance in fuel economy in the whole SOC range.Three learning based EMSs were proposed in this dissertation to improve the energy utilization efficiency of HEV.Meanwhile,the research of this dissertation is meaningful for applying artificial intelligence to new energy vehicle control problems.