Short-Term Power Load Forecasting Based onProphet and XGBoost Mixed Model

Accurate and effective load forecasting is of great importance to secure and stable operation of power grid. Through deep analysis on Prophet frame and extreme gradient boosting (abbr. XGBboost) machine learning model, a hybrid load forecasting model based on Prophet-XGBoost was proposed. Based on large amount of historic load data, data information and meteorological data, a Prophet based load forecasting model and an XGBboost based machine learning forecasting model were constructed respectively, and by means of reciprocal error method the Prophet and XGBoost were combined to obtain hybrid forecasting model. Taking historical load data of a certain city located in Southwest China for example, the proposed hybrid model was applied to the load forecasting test, and the results show that using the proposed Prophet-XGBoost hybrid model a higher forecasting accuracy can be obtained than by such models as support vector machine regression (abbr. SVR,support vector regression) model, Prophet model and XGBoost model, and the time wasted for the load forecasting by the proposed method is much shorter than that by SVR model.