Adaptive Frequency Control of an EV-Integrated Microgrid Using Deep Reinforcement Learning

High penetration of renewable energy sources and electric vehicles (EVs) reduces system inertia and degrades frequency stability in microgrids. This paper proposes an adaptive frequency control strategy for an EV-integrated low-inertia microgrid using reinforcement learning (RL). The RL agent dynamically tunes the PI gains of the voltage and current control loops of a grid-forming inverter to enhance virtual inertia support. An AC-DC microgrid with renewable generation, EV charging infrastructure, and a low-inertia synchronous generator is modeled in MATLAB/Simulink. The proposed controller is evaluated under varying load levels, load disturbances, grid disconnection, and fault conditions, and compared with a fixed-gain PI controller. Simulation results show that the RL-based approach reduces minimum frequency overshoot by up to 11.6 % and consistently improves transient response and disturbance rejection.