Short-term load forecasting for holidays based on similar days selecting and XGBoost model

Daily load curves of legal holidays differs greatly from those of normal days due to the influence of holiday policies and local customs. Since holiday load curves are complex and irregular, the daytime load forecast errors for holidays are comparatively high. To address the issue, this paper proposes a combined method based on similar days matching and XGBoost for short-term load forecasting on holidays. Firstly, the holiday loads are split into two parts: trend curves and daily load extremes. Trend curves are predicted by code-matching similar historical days, while daily load extremes are predicted by the XGBoost model. Finally, the predictions are combined to produce daily load curves. Through experimental verification, compared with single model predictions, the proposed method has better performance.