Real-time Energy Management Strategy for Hybrid Electric Vehicles Based on Deep Reinforcement Learning and Model Predictive Control

To address the challenge of balancing real-time performance and adaptability in hybrid electric vehicle (HEV) energy management, this paper proposes a real-time hierarchical energy management strategy (EMS) that integrates deep reinforcement learning (DRL) with model predictive control (MPC). At the upper layer, a deep Q-network (DQN) is employed to construct an EMS controller that rapidly plans a reference trajectory for the state of charge (SOC) prior to vehicle departure. At the lower level, a Long Short-Term Memory (LSTM) network is first employed to construct a velocity predictor, forecasting the velocity sequence over a future time domain. Subsequently, an MPC controller is designed to achieve optimal power flow allocation by tracking the SOC reference trajectory. The proposed strategy is then comprehensively compared with dynamic programming (DP) and rule-based strategies across different test conditions. Simulation results demonstrate that the proposed strategy achieves over 90 of the fuel economy attained by the DP strategy while exhibiting strong real-time application potential. Finally, hardware-in-the-loop (HIL) experiments validate the practical applicability of the proposed strategy.