Reinforcement Learning-Driven Sliding Mode Control for Nonlinear Bidirectional Energy Exchange in Vehicle-to-Grid Systems

This paper proposes a reinforcement learning–enhanced sliding mode control (RLSMC) framework for efficient and robust bidirectional energy exchange in vehicle-to-grid (V2G) and grid-to-vehicle (G2V) systems. The proposed approach integrates a Deep Q-Network (DQN) with classical sliding mode control to enable adaptive gain tuning, thereby improving robustness against nonlinearities, parameter variations, and external disturbances. A comprehensive nonlinear model of a hybrid energy storage system, comprising a lithium-ion battery, an ultracapacitor, quadratic buck–boost converters, and a three-phase grid-connected inverter, is developed to evaluate the performance of the proposed controller under both charging and discharging modes. The RLSMC achieves improved dynamic tracking, reduced chattering, and enhanced stability compared to conventional SMC, particularly under sudden load variations and grid disturbances. The controller ensures smoother current transitions, reduced steady-state error, and improved regulation under nonlinear converter dynamics. Extensive simulation results demonstrate the controller’s ability to maintain stable bidirectional power flow and accurate current regulation across a wide range of operating conditions. To validate real-time feasibility, controller-in-the-loop (CIL) experiments are conducted using the Delfino F28369D microcontroller. The experimental results closely align with simulation outcomes, confirming the practical applicability and reliability of the proposed control strategy.