Reinforcement Learning Optimized Energy Control of Electric Vehicles Considering Powertrain Thermal Dynamics

In view of the contradiction between the best engine monomer performance and the poor vehicle performance existing energy management strategies, the objective of this study is to leverage deep reinforcement learning to incorporate the thermal characteristics of the engine into the optimization process of energy management strategies, thereby enhancing fuel economy under real-world vehicle operating conditions. Combining the real-time road condition information provided by the vehicle network system, the state space and action space are formulated based on the Soft Actor-Critic (SAC) reinforcement learning algorithm, taking into account energy power and engine cooling constraints, while a generalized reward function design methodology is proposed. Based on bench test data, this paper establishes a series hybrid electric vehicle model with integrated engine thermal characteristics, and validates the effectiveness of the algorithm under actual road conditions by using the engine bench test. The results show that the proposed control strategy can improve fuel economy by more than 3.4% under -10°C environment compared with the traditional SAC enhanced learning energy management strategy without considering engine thermal characteristics. Fuel economy can be improved by more than 4.0% compared to rule-based energy management strategies.