A two-layer SSA-XGBoost-MLR continuous multi-day peak load forecasting method based on hybrid aggregated two-phase decomposition

Abstract

To further improve the prediction accuracy under the realization of rolling peak load prediction scenarios for the coming month, an aggregated hybrid two-phase decomposition with two-layer prediction model architecture (ICEEMDAN-SE-VMD-SSA-XBoost-MLR) is proposed in this paper. First, a fully integrated empirical modal decomposition with improved adaptive noise (ICEEMDAN) is used to initially decompose the historical peak load time series, which leads to several eigenmodal functions (IMFs), and by calculating the sample entropy (SE) of each IMF, the IMFs with similar SEs are aggregated and reconstructed into load components representing different time scales. Then, the variational modal decomposition (VMD) is introduced to quadratically decompose the less regular parts of the load components to fully weaken their non-smooth characteristics. Then, extreme gradient boosting (XGBoost) is used to assist in establishing the feature engineering, combining each load component to form the data set required for the prediction model, and using the sparrow search algorithm (SSA) optimized XGBoost with multiple linear regression (MLR) to construct the first layer prediction model for each component, and the prediction results of each component are superimposed to obtain the preliminary prediction results. Finally, inspired by the idea of integrated learning, an error correction link is proposed in this paper to make a secondary prediction of the preliminary prediction results by XGBoost, to better track the fluctuation of load details, realize the error correction of the preliminary prediction results and improve the prediction accuracy.

Keywords

• Aggregate mixed binomial decomposition
• Two-layer prediction model
• Extreme gradient boosting
• Rolling prediction
• Error correction