Ecological driving strategy for dual-motor electric vehicles based on sequential hierarchical deep reinforcement learning

Ecological driving (eco-driving) means optimizing the driving method to reduce energy consumption. This paper proposes a sequential hierarchical deep reinforcement learning-based eco-driving (SHDRL-ECO) strategy for dual-motor electric vehicles. In this framework, the agent of energy management strategy (EMS) is first trained, and a mechanism of multi-Actor network (MAN) with expectation-oriented (EO) optimization is introduced to address the hybrid action space issue in the EMS training process. Then the agent of the adaptive cruise control (ACC) is trained under the guidance of the trained model of Agent EMS. DDPG is used as the basic algorithm in the experiments, and the results have shown that the proposed method can greatly improve performance of both EMS and ACC, and that the energy economy of the proposed method can reach more than 97% of SUMO-ACC+DP on both the training and testing cycles while keeping excellent performance in car-following, and the method has high adaptability across various ACC scenarios.