Battery Life-Aware Predictive Deep Reinforcement Learning Energy Management for Hybrid Electric Vehicles

Hybrid energy storage system (HESS) is the preferred energy source for hybrid electric vehicles (EVs). Extending system lifespan and improving energy management efficiency are critical factors in enhancing the availability and sustainability of EVs. This study develops a predictive deep reinforcement learning energy management strategy using vehicle historical data and considering the battery life effect during the power optimization process. First, the Autoregressive Integrated Moving Average (ARIMA) model processes the vehicle’s historical data to predict short-term future speed and road gradient changes. Second, a battery life-aware predictive deep Q-Network (LAP-DQN) energy management strategy (EMS) is introduced, and the battery aging effect is incorporated during training to achieve a synergistic optimization of energy consumption and battery lifespan. Finally, the effectiveness of the proposed method is validated via comparative simulations against CD-CS and PMP via three cycles. The results demonstrated that LAP-DQN significantly extended battery life by 8.76% while improving UC utilization ratio by 17.91% in overall performance. This study offers new insight into EMS for EVs and shows promising prospects for engineering sustainability applications and the circular economy.