A novel smart maximum power point tracking approach using an M5-pruned algorithm for photovoltaic systems

Introduction. Maximizing power extraction from photovoltaic (PV) systems under varying environmental conditions requires robust and efficient control strategies. DC-DC boost converter plays a critical role in this process, where reducing output ripples is essential for power quality and system longevity. Problem. Conventional maximum power point tracking (MPPT) methods such as incremental conductance (INC) suffer from high steady-state oscillations (ripples) and power loss. While intelligent methods like fuzzy logic controllers (FLC) offer better stability, they introduce significant computational complexity and design difficulties, hindering their implementation in low-cost converters. Goal. To develop and validate a novel hybrid MPPT control approach that combines the M5-pruned (M5P) algorithm with the modified INC (MINC) method to achieve the high accuracy of AI-based methods while retaining the low computational complexity of conventional techniques, specifically targeting ripple reduction and efficiency maximization under dynamic environmental variations. Methodology. A mathematical model of the PV system was simulated using MATLAB/Simulink, and the M5P model was trained using Weka software. A comparative study was conducted against MINC and FLC under rapid changes in irradiation and temperature levels. Results. The simulation results confirm the superiority of the proposed M5P-MINC method. It achieved the highest overall system efficiency of 88.31 %, outperforming FLC (87.87 %) and MINC method (85.5 %). In terms of stability, M5P-MINC demonstrated the lowest oscillations, reducing PV power ripples to 0.32 % and boost converter output power ripples to 4.35 %, compared with 0.57 % and 6.22 % for FLC and 3.6 % and 12.23 % for MINC, respectively. Although the FLC method showed a marginally higher tracking efficiency (99.88 %) than M5P-MINC (99.83 %), the difference is negligible compared to the significant gain in simplicity. Scientific novelty. This study proves that the regression-tree-based M5P algorithm provides an optimal trade-off: it delivers superior overall efficiency and stability comparable to complex AI controllers but maintains a low computational burden suitable for real-time applications. Practical value. The proposed algorithm offers a cost-effective solution for commercial PV systems, ensuring high-quality power delivery with minimal stress on electrical components. References 31, tables 2, figures 13.