Artificial Intelligence-driven maximum power point estimation for enhanced adaptive power regulation in rooftop PV systems

This paper presents a novel framework for improving grid stability in rooftop photovoltaic (PV) systems by integrating an AI-based Maximum Power Point (MPP) estimation model with a newly designed adaptive power regulation (APR) strategy. The primary objective is to operate the inverter-based power conditioning unit dynamically based on the real-time load demand of individual buildings, ensuring that power injection aligns with local consumption while avoiding grid disturbances. A second-order polynomial regression model is trained on historical PV voltage and current data to estimate the MPP without environmental sensors. This estimated MPP informs an RHS-based APR controller that curtails surplus power by shifting operation away from the MPP during periods of excess generation. A real-time scenario involving a 2 MW rooftop PV deployment across 20 buildings demonstrates the necessity of such regulation, with simulation studies in MATLAB/Simulink showing up to 21% curtailment, voltage reduction from 250 V to 231 V, and frequency deviation controlled within ±0.2 Hz. The proposed method achieves <2% power tracking error, 0.01 s dynamic response, and steady-state oscillation below 0.5%. Hardware implementation on a 1 kW rooftop setup using an NI myRIO controller confirms reliable performance under irradiance fluctuations, with THD maintained at 4.76% in compliance with IEEE 519 standards. These results validate the effectiveness of combining lightweight AI-based estimation with real-time, load-aware adaptive power regulation to enhance inverter performance and ensure grid compliance in distributed solar applications.