Short-Term Power Load Forecasting Based on Clustering and XGBoost Method

Abstract

According to the problems of high computational cost and over-fitting in traditional forecasting methods, a short-term power load forcasting method is put forward based on combining clustering with xgboost (eXtreme Gradient Boosting)algorithm. The method mainly does research on correlation between influence factors and load forecasting results. Firstly, Features extracted from original datum and missing values are filled during preprocessing stage. Secondly, the changing trend of load is divided into four classifications by K-means algorithm. Meanwhile, classification rules are set up between temperature and category. Finally, xgboost regression model is established for different classifications separately. Furthermore, forecasting load is calculated according to scheduled date. Experimental results indicate the method can to some extent predict the daily load accurately.