MODWT-Based Wavelet Neuro-Fuzzy MPPT for Photovoltaic Systems with Comparative Analysis to Machine Learning Controllers

Abstract

Accurate maximum power point tracking (MPPT) is critical for photovoltaic (PV) systems because the optimal operating point shifts continuously with irradiance and cell temperature. This paper studies a predictive MPPT strategy that combines (i) shift-invariant, multi-resolution features extracted from irradiance/temperature time series using the maximal overlap discrete wavelet transform (MODWT) with a Daubechies-4 (db4) mother wavelet and a three-level decomposition and (ii) a first-order Takagi-Sugeno neuro-fuzzy model that maps those features to a reference MPP voltage. The proposed wavelet neuro-fuzzy (WNF) controller is benchmarked against four common regression baselines (decision tree, random forest, support vector regression, and a multilayer perceptron ANN) using 48 h of per-minute weather data for Lahore, Pakistan (22--23 June 2025). In a 100 kW PV boost-converter simulation, the WNF controller achieves 99.988% average tracking efficiency and yields 1533.7 kWh, which is approximately 2% higher energy than the compared learning baselines under the same test conditions. We also discuss computational implications of MODWT-based feature extraction for real-time embedded MPPT.