Enhancing data-driven energy management strategy via digital expert guidance for electrified vehicles

This study addresses data efficiency and reliability issues in reinforcement learning (RL)-based energy management strategies (EMS) for hybrid electric vehicles (HEVs). A novel expert-guided RL (EGRL) paradigm is proposed, combining deep ensemble methods with a digital expert model (DEM) for real-time EMS intervention across various scenarios. DEM, trained via domain adversarially invariant meta-learning (DAIML), adapts to different driving conditions. An intervention mechanism, based on uncertainty evaluation in the deep ensemble, allows DEM to guide and supervise RL training, ensuring reliability. The EMS optimizes energy consumption, battery health, and electricity maintenance for the range-extended electric bus (REEB) system. Simulation results show the paradigm significantly improves energy management, nearing optimal performance and surpassing traditional RL methods. EGRL achieves an average 15.8% improvement in economic benefit across all test cycles. This research offers an innovative solution for EMS and has broad potential for other automation applications.