Stacked Ensemble Machine Learning for MPPT Under Partial Shading: A High-Accuracy Data-Driven Approach

Abstract

In this paper, a Stacked Ensemble Machine Learning (SEML) based MPPT framework is introduced for variable irradiances and temperature. Based on a regression-based group of base learners (Random Forest, Gradient Boosting, Support Vector Regression) and linear meta-model, the SEML-MPPT algorithm works well in estimating the V mp of PV modules. The model, trained on environmentally sourced data provided by NASA, has excellent fitting results with a coefficient of determination R 2 = 0.993 and a mean squared error (MSE) made of 0.0935, evidencing a high generalization ability despite the presence of non-linear operating conditions.Implementation in MATLAB/Simulink is conducted for experimental verification, and the robustness of the proposed model is confirmed with realistic PV and IV characteristics, as well as voltage behavior about environmental parameters. In addition, an efficiency analysis over varying irradiance levels in the range from 200 to 1000 W/m2 has shown a high conversion efficiency of over 94% up to the optimum point of 97.7% under standard test conditions. These results highlight the potential of the SEML-MPPT model to achieve accurate, adaptive, and reliable tracking for the next photovoltaic energy systems.