Enhancing Day-Ahead Load Forecasting: Insights from Weather-Integrated Models and SHAP Analysis

The integration of meteorological data into machine learning models offers a promising approach for enhancing the accuracy and interpretability of day-ahead load forecasting. This study contributes to the field by (1) developing a novel framework that integrates ERA5 and CAMS datasets into machine learning-based forecasting, (2) training hourly-specific Gradient Boosting models to improve forecasting performance, and (3) applying SHAP analysis to dynamically evaluate feature importance at different hours of the day. Our approach significantly improves accuracy metrics (MAE, MAPE, RMSE) while revealing critical temporal dynamics in feature importance. Air temperature consistently emerges as the most influential factor, while solar radiation shows limited impact on system-level forecasts. These findings offer practical value for transmission system operators in improving operational reliability.