Operational improvement of hybrid-source microgrid using unified power quality conditioner optimized with reinforcement learning brainstorm algorithm

Abstract Power quality is a critical concern in modern microgrids, particularly those integrating renewable energy sources such as photovoltaic arrays, wind turbines, and batteries. However, variability in renewable generation often introduces problems such as voltage instability, harmonics, and current imbalance, which degrade system efficiency and reliability. To address these challenges, this study proposes an Artificial Neural Network (ANN)-based Reinforcement Learning Brainstorm Optimization (RLBSO) controller enhanced with a reward function for a Unified Power Quality Conditioner (UPQC). The RLBSO algorithm is employed for efficient offline training of the ANN, while the reinforcement learning reward function ensures adaptive control under grid-connected microgrid operations. MATLAB/Simulink simulations demonstrated that the ANN-based RLBSO controller significantly outperformed traditional proportional-integral (PI) controllers in key performance metrics. The proposed method reduced Total Harmonic Distortion from 4.2% to 2.59% and decreased current imbalance from 5% to 2% while maintaining voltage stability within ± 2% of the nominal value. Compared with PI and Fuzzy Logic controller (FLC), the ANN-based RLBSO achieved superior performance under varying load conditions owing to its dynamic control mechanism. These results highlight the potential of the proposed controller as a promising approach to improve power quality and efficiently integrate renewable energy into microgrids, thereby setting a new benchmark for intelligent control in hybrid renewable energy systems.