Photovoltaic Power Prediction and SHAP Interpretability Analysis Based on Multi-Model Comparative Learning

In recent years, photovoltaic (PV) power generation has been developing rapidly, but its power instability poses a challenge to the stable operation of power systems. In this study, three machine learning models, CatBoost, Random Forest, and Support Vector Regression (SVR), were used for PV power prediction, and it was found that the SVR model performed the best, with the highest fit between predicted and true values (R² ≈ 0. 95) and the smallest fluctuation in residual error (-50% to 50%). Through SHAP analysis, the study reveals that illumination intensity is the most critical variable affecting PV power generation with a contribution of more than 70%. Additional machine learning models and optimization algorithms can be further explored in the future to improve prediction performance. In summary, this study provides an effective combinatorial machine learning method for high-precision PV power prediction, revealing the dominant role of key factors such as illumination intensity on the prediction results. Future research directions include optimizing the model structure and integrating multi-source data to further improve the accuracy and reliability of PV power prediction.