Direct Load-Aware MPPT Control in Photovoltaic Systems: A Comprehensive Evaluation of Ensemble Learning Architectures

The global transition toward sustainable energy highlights the importance of photovoltaic (PV) systems. However, the non-linear characteristics of PV cells, which are heavily influenced by environmental fluctuations, make optimal energy extraction a complex task. To maximize efficiency, Maximum Power Point Tracking (MPPT) controllers are vital in power converters. Traditional MPPT methods, such as Perturb and Observe (P&O), are widely used but suffer from low efficiency, slow tracking under dynamic weather, and steady-state oscillations. To address these limitations, this paper investigates advanced Ensemble Learning (EL) architectures to ensure robust and precise control. Distinctively, load resistance is integrated as a dynamic input alongside irradiance and temperature to enable direct duty cycle generation. This load-aware strategy completely eliminates auxiliary controllers and intermediate calculations. A comprehensive ablation study comprising nine predictive models is conducted. Simulation results confirm that the Stacking (MLR Meta) framework outperforms all traditional and standalone machine learning models. It achieves an exceptional coefficient of determination (R۲) of ۰.۹۹۸۹۶, a Root Mean Square Error (RMSE) of ۰.۰۰۵۸۷, and an outstanding total dynamic tracking efficiency of ۹۹.۲۰%. Consequently, the evaluated architecture provides a highly reliable and ripple-free power output under severe dynamic conditions.

کلیدواژه ها:

Maximum Power Point Tracking (MPPT) ، Ensemble Learning (EL) ، Photovoltaic (PV) Systems ، Direct Duty Cycle Control ، Stacking Architecture ، Ablation Study