Novel ML-based load frequency controller for microgrid stability

Abstract

Microgrids have become an essential component of modern electrical systems due to their economic, technical, and environmental benefits. Despite their advantages, they face significant challenges in maintaining frequency stability, primarily due to the inherent variability of renewable energy sources (RESs), fluctuating electric loads, and the dynamic behaviour of energy storage systems. The intermittent and non-inertial nature of RESs such as solar and wind exacerbates these frequency deviations, particularly in islanded microgrids. To address these issues, this study introduces a novel Load Frequency Control (LFC) strategy based on machine learning (ML) technique. An islanded microgrid configuration is modelled for Madrid, Spain, incorporating a diesel generator, solar photovoltaic (PV) system, wind turbine, and battery energy storage system (BESS). The proposed control approach employs an advanced machine learning-based optimization algorithm Proximal Policy Optimization (PPO) to dynamically regulate system frequency under varying load and generation conditions. To improve PPO’s performance, its hyperparameters are tuned using a Genetic Algorithm (GA). Photovoltaic and wind profiles are obtained from the System Advisor Model (SAM), while the diesel generator and battery systems are sized based on empirical demand data. For comparison, GA is also applied independently as a benchmark heuristic method. While GA offers an evolutionary optimization approach, PPO utilizes reinforcement learning to adapt control strategies through real-time environmental interaction. Both algorithms are rigorously tested through simulations under diverse operating conditions, including load perturbations and renewable variability. Simulation results highlight the superior performance of the PPO-based controller, demonstrating enhanced frequency stability, faster response, and reduced settling times compared to GA. The findings establish PPO as a robust and efficient solution for intelligent LFC, making it well-suited for real-time control applications in isolated, RES-dominated microgrids. This work contributes to the development of resilient and adaptive energy management strategies, supporting the broader transition to sustainable power systems.