A Method for Daily Electricity Load Forecasting of Urban and Rural Residents Considering Data Discrepancies

To address the challenges of strong periodicity and high volatility in residential load data, which lead to low accuracy in direct forecasting, a load prediction model integrating Seasonal-Trend Decomposition using Regression (STDR) with Extreme Gradient Boosting Trees (XGBoost) was developed. This study first examines how data discretization affects load forecasting, then applies the Maximal Information Coefficient (MIC) to select and extract key load features, and finally evaluates prediction errors using the XGBoost model. The findings indicate that accounting for discreteness enhances prediction accuracy. The results confirm that the proposed approach performs effectively and outperforms alternatives in short-term residential load forecasting.