Energy Management of Intelligent FCHEVs Under Adaptive Cruise Control Based on Multi-Objective Deep Reinforcement Learning

The integration of advanced adaptive cruise control (ACC) with energy management strategy (EMS) is crucial for improving the fuel efficiency of fuel cell hybrid electric vehicles (FCHEVs). The nonlinearity inherent in ACC and EMS arises from complex interactions between vehicle dynamics system and composite energy sources, necessitating intricate real-time adjustments and resulting in high-dimensional dynamic optimization problems. This paper proposes a data-driven, model-free deep reinforcement learning (DRL) integrated optimization framework that emphasizes balancing multiple objectives—safety, comfort, energy saving, and energy source durability while addressing the challenges of policy convergence exacerbated by potentially sparse rewards in high-dimensional action-state spaces. Specifically, an ACC based on course learning DRL is proposed, with a featuring a staged reward mechanism for driving safety, comfort, and speed consistency to model human-like progressive learning. Then, adaptive equivalent consumption minimization is combined to design the DRL-based EMS that balances fuel economy and power source durability. Finally, the decision sets of the integration optimization framework are generated and verified. The simulation results show that the proposed method can ensure driving safety and traffic throughput, the ego vehicle improves driving comfort by 73.4% and energy saving by 11.5% compared to the preceding vehicle.