Short-term energy forecasting based on GBRT and time lag correlation coefficient

Inhalt:
Energy forecasting is a very important research topic in the field of electric power and energy. Accurate energy forecasting can provide support for enterprise production arrangement and energy management, which is conducive to achieving energy saving and maximizing production efficiency. Since energy variation is affected by numerous factors, choosing appropriate input features is the key to improving the accuracy of energy forecasting. In this paper, a GBRT short-term energy forecasting model is established, a feature selection method based on time lag correlation coefficients is introduced to select the sub-equipment loads with high correlation to the total load as input features. Calendar variables, historical load variables, and trend variables are designed to describe cycle characteristics, time series characteristics, and economic impact of loads, respectively. The results based on an industrial energy case show that the GBRT prediction model established in this paper can provide a certain degree of interpretability and has excellent prediction accuracy, and the proposed feature selection method can effectively improve the prediction accuracy of the energy prediction model.