Volume 20: Sustainable Energy Solutions for a Post-COVID Recovery towards a Better Future: Part III

A DRL-based Ecological Driving Strategy for Series Hybrid Energy Vehicle Including Battery Degradation Yi Fan,Hongwen He,Zexing Wang,Jiankun Peng,Hailong Zhang,Weiqi Chen

Abstract

An ecological driving strategy considered battery State-of-Health is proposed based on Deep reinforcement learning. Not only does this strategy try to minimize fuel consumption while maintaining the safe car-following sate, it also seeks to lower the battery aging speed. In order to optimize the car-following and energy management performance, reward functions are developed by combing driving features of car-following, engine and battery characteristics. The agent maximizes the accumulated reward by interacting with the simulation environment to explore the action space. While controlling the SHEV to maintain a safe car-following distance, the proposed method reduces the effective Ah-throughput by 15 -57.6% and only increases the fuel consumption within 5% compared with the case of achieving the best fuel economy. In addition, this method is proven to achieve similar results in different driving cycles.

Keywords Energy management strategy,car-following,battery state-of-health,deep reinforcement learning