Deep Reinforcement Learning-based Energy Management for Hybrid Electric Vehicles

More about the book

The urgent need for vehicle electrification and improved fuel efficiency has garnered global attention. Hybrid vehicle systems have demonstrated their value in both academic research and industry applications, with energy management playing a crucial role in optimizing hybrid electric vehicles (HEVs). Various energy management approaches, from rules-based strategies to optimization methods, offer diverse options for enhancing fuel economy. However, the research landscape is evolving, particularly with advancements in intelligent transportation systems and onboard sensing and computing capabilities. The rise of machine learning, especially deep learning and deep reinforcement learning (DRL), is propelling research into learning-based energy management strategies (EMSs). These strategies are promising for their ability to handle large datasets and generalize learned rules to new scenarios without extensive manual tuning. This work focuses on learning-based energy management, emphasizing DRL. It begins with an overview of DRL in HEV energy management, identifying the strengths and limitations of typical DRL-based EMSs based on state and action space types. It discusses value-based, policy gradient-based, and hybrid action space-oriented methods. Finally, it presents a general online integration scheme for DRL-based EMS, bridging the gap between strategy learning in simulations and deployment on vehicle controllers.