Deep Reinforcement Learning-Based Energy Management and Fault-Tolerant Control for Vehicle-to-Grid Enabled Electric Vehicle Drive Systems

Abstract

Vehicle-to-Grid (V2G) technology has emerged as an effective approach for supporting bidirectional energy exchange between electric vehicles and smart grids. However, uncertainties associated with renewable energy generation, electricity price fluctuations, varying driving conditions, and component faults make real-time energy management and reliable system operation challenging. This study presents a Deep Reinforcement Learning (DRL)-based framework that integrates intelligent energy management with adaptive faulttolerant control for V2G-enabled electric vehicle drive systems. A Deep Q-Network (DQN) agent continuously evaluates battery state-of-charge, renewable power availability, load demand, electricity pricing, and fault severity to determine optimal charging, discharging, and grid-support actions while maintaining stable system operation under abnormal conditions. The proposed framework is validated using MATLAB/Simulink under diverse driving cycles, renewable energy scenarios, and component fault conditions. Simulation results demonstrate an energy efficiency of 96.1%, battery utilization of 94.2%, grid support capability of 91.3%, and a 34.7% reduction in operating cost compared with a conventional rule-based strategy. The integrated fault diagnosis and adaptive recovery mechanism achieves 98.8% fault detection accuracy, identifies faults within 24 ms, and restores normal operation in 0.14 s. These results demonstrate the effectiveness of the proposed framework in improving the efficiency, reliability, and operational resilience of nextgeneration V2G-enabled electric vehicle drive systems.