Mid Term Power Load Forecasting Using Blending Integrated Model Based on Prophet, XGBoost, and LSTM

Accurate forecasting of the electrical load provides strong data support for formulating power generation plans. To improve the accuracy of mid-term power load forecasting, this paper combines the power system load data of a certain region's power grid in the past three years to construct three types of power load forecasting models: time series Prophet, deep learning LSTM, and machine learning XGBoost. The goal is to predict the maximum daily load in the region for the next three months. To balance the unique characteristics of each of the three models and their excellent performance in power load forecasting, a blending integrated model based on Prophet, LSTM, and XGBoost was ultimately established for mid-term power load forecasting in the region. The results show that the integrated model has better prediction effect than the single model. Evaluate the accuracy of each model based on four evaluation metrics: MSE, MAE, MAPE, and R-square. The blending ensemble model proposed in this article achieved the best results with all four indicators, with a model accuracy of 97% and an R of 0.93, demonstrating a high degree of model fitting.