Intelligent physics-aware deep Q-network controller for resilient and battery-sustainable electric vehicle energy management
TL;DR
The proposed physics-guided deep reinforcement learning framework based on a Physics-Guided Deep Q-Network for multi-objective EV energy management under varying grid operating conditions demonstrates the potential to improve energy efficiency, reduce degradation-related battery stress, and support stable grid interaction within the assumptions of the adopted simulation models.
Abstract
The increasing penetration of electric vehicles (EVs) in modern power systems introduces significant challenges in energy efficiency, battery health management, and stable grid interaction. Conventional EV energy management strategies often fail to simultaneously optimize energy utilization, battery degradation-related stress, and power system performance under dynamically changing operating conditions. To address these limitations, this paper proposes a physics-guided deep reinforcement learning framework based on a Physics-Guided Deep Q-Network (PG-DQN) for multi-objective EV energy management under varying grid operating conditions. The proposed approach integrates physical system knowledge, battery dynamic constraints, a model-based degradation index, and grid interaction characteristics within a deep reinforcement learning architecture, enabling the controller to learn adaptive and physically consistent energy-management policies. A comprehensive system model incorporating vehicle dynamics, lithium-ion battery behavior, and bidirectional grid power exchange is developed to evaluate the effectiveness of the proposed method. Extensive simulation studies are performed to compare the proposed PG-DQN framework with rule-based control, optimization-based control, and conventional deep reinforcement learning approaches. The results demonstrate improved overall system performance under the considered simulation conditions. In particular, the PG-DQN strategy reduces EV energy consumption by approximately 11.6%, lowers the model-based battery degradation index by nearly 30%, and improves electrical performance by reducing DC-bus voltage ripple by about 75%. Additionally, grid power fluctuations are mitigated by approximately 66.7%, resulting in smoother EV–grid interaction. These results demonstrate the potential of the proposed framework to improve energy efficiency, reduce degradation-related battery stress, and support stable grid interaction within the assumptions of the adopted simulation models. Not applicable.