Abstract:Electricity load forecasting has become an important task in power dispatching and one of the important indicators to measure whether the power enterprise is modernized or not. Accurate power load forecasting is of great significance in reasonably arranging the start and stop of generating units of power enterprises, reducing excess power loss, improving economic efficiency and guaranteeing the social demand for electricity. Short-term load forecasting is aimed at short-term load change or even real-time load change, but it is difficult to forecast because short-term load change is more abrupt. In this paper, we propose a short-term power load forecasting method based on the broad learning system(BLS), by using the neighborhood rough set classification algorithm to extract features from the input parameters. Then, the trained model is used to achieve 24h short-term load forecasting by using the offline training of historical data of electric load using BLS. The results show this method using the broad learning system effectively reduce MAPE and RMSE, also reduce the training time, improve the model training speed, and have excellent forecasting capability.