Optimizing Large-Scale PV Systems with Machine Learning: A Neuro-Fuzzy MPPT Control for PSCs with Uncertainties

Abstract

The article proposes a new approach to maximum power point tracking (MPPT) for photovoltaic (PV) systems operating under partial shading conditions (PSCs) that improves upon the limitations of traditional methods in identifying the global maximum power (GMP), resulting in reduced system efficiency. The proposed approach uses a two-stage MPPT method that employs machine learning (ML) and terminal sliding mode control (TSMC). In the first stage, a neuro fuzzy network (NFN) is used to improve the accuracy of the reference voltage generation for MPPT, while in the second stage, a TSMC is used to track the MPP voltage using a non-inverting DC—DC buck-boost converter. The proposed method has been validated through numerical simulations and experiments, demonstrating significant enhancements in MPPT performance even under challenging scenarios. A comprehensive comparison study was conducted with two traditional MPPT algorithms, PID and P&O, which demonstrated the superiority of the proposed method in generating higher power and less control time. The proposed method generates the least power loss in both steady and dynamic states and exhibits an 8.2% higher average power and 60% less control time compared to traditional methods, indicating its superior performance. The proposed method was also found to perform well under real-world conditions and load variations, resulting in 56.1% less variability and only 2–3 W standard deviation at the GMPP.