Deep reinforcement learning control scheme for frequency control of microgrids

The growing integration of renewable energy sources into microgrids poses substantial challenges to traditional load-frequency control, primarily due to the intermittent nature of generation and continuously changing operating conditions. To overcome these limitations, this study proposes an adaptive control strategy based on reinforcement learning, in which a deep deterministic policy gradient (DDPG) agent is used to automatically tune a cascaded PIDF controller for frequency regulation. The learning agent interacts directly with the microgrid environment and updates controller parameters online based on observed frequency deviation and its integral and derivative, enabling continuous adaptation without requiring explicit system models or offline optimization. A detailed two-area microgrid incorporating conventional generation, wind, and photovoltaic power profiles derived from real measurement data, as well as energy storage units, is developed for evaluation. Time-domain simulations in MATLAB/Simulink are conducted under diverse disturbance scenarios to examine controller behavior under load variations and renewable intermittency. The proposed learning-based control strategy demonstrates improved dynamic regulation characteristics and stable frequency operation across all tested conditions, contributing to reliable renewable energy integration and the development of sustainable energy systems.