Data-driven regression controller-based MPPT with image encryption inspired solar PV array reconfiguration under partial shading conditions

Abstract

Partial shading and environmental variations significantly reduce the power output and efficiency of photovoltaic (PV) systems, posing challenges for conventional maximum power point tracking (MPPT) methods that suffer from slow convergence, local maxima trapping, and high computational cost. To address these limitations, this paper proposes an image encryption-inspired PV array static reconfiguration technique based on the Kolakoski sequence transform (KST), combined with data-driven regression-based MPPT controllers. The proposed KST method minimizes current mismatches by intelligently redistributing shaded modules, while decision tree (DT), support vector machine (SVM), neural network (NN), and machine learning (ML) regression methods are employed to determine the optimal duty cycle for a SEPIC converter under varying irradiance conditions. The system is evaluated on both symmetrical 5 × 5 arrays and unsymmetrical 4 × 6 arrays, including experimental validation using a 250 Wp standalone PV setup. In MPPT performance, the regression-based controllers attain GMP enhancements of 47.09%, 45.14%, 27.27%, 13.62%, and 10.73% for 5 × 5 arrays and 74.96%, 44.11%, 40.14%, 18.29%, and 7.15% for 4 × 6 arrays under diverse environmental conditions. The reconfiguration technique achieves global maximum power (GMP) improvements of 32.79%, 14.98%, and 10.15% across various shading scenarios using 9 × 9 arrays. Notably, the proposed KST integrated with SVM regression-based MPPT delivers up to 68% GMPP enhancement, with >98.5% efficiency, convergence <0.35 s, and ripple ≤1.5%, validated across dynamic shading, temperature variation, rapid irradiance changes, and hotspot conditions. These results confirm the robustness, adaptability, and real-time suitability of the proposed KST integrated with ML-based Regression MPPT approach for practical PV optimization.