Reinventing power quality enhancement: deep reinforcement learning control for PV-UPQC in microgrids

A pioneering technique for optimizing the functionality of a Photovoltaic-Unified Power Quality Conditioner (PV-UPQC) is proposed in this work by replacing conventional synchronous reference frame (SRF)-based control with deep reinforcement learning (DRL). The PV-UPQC is integrated with a microgrid to improve power quality and system efficiency. In this novel approach, a DRL agent is designed to autonomously learn, adapt, and optimize control strategies for the UPQC, without the need for predefined mathematical models. The DRL agent utilizes real-time data from the microgrid to make intelligent control decisions. Additionally, for accurate DC bus voltage management, a gazelle optimization algorithm (GOA)-tuned PI controller is implemented. The proposed DRL-based control system aims to minimize power quality (PQ) issues, enhance the UPQC’s performance, and ensure system stability by dynamically adjusting control actions. The proposed system shows significant improvements in mitigating voltage sags and swells, as evidenced by MATLAB simulations and hardware testing. The DRL-based system excels in managing various fault conditions and adapting to dynamic load changes, outperforming traditional methods. It achieves remarkably low total harmonic distortion (THD) levels, ranging between 0.9 and 1.3%. This performance indicates a significant enhancement in power quality, showcasing the system’s advanced capabilities in ensuring efficient and stable power distribution.