Short-term Load Forecasting Based on Multi-model Fusion of CNN-LSTM-LGBM

Abstract

Inter-regional energy dispatch and regional peak cutting and valley filling require accurate load forecasting as support. In order to improve the forecasting accuracy, this paper proposes a multi-model fusion forecasting method based on CNN (convolutional neural network)-LSTM (long short-term memory)-LGBM (Light Gradient Boosting Machine) considering demand response. The CNN's ability is exploited to effectively extract local features, and LSTM’s ability to grasp time series information is used to build a serial CNN-LSTM model. Meanwhile, LGBM's regression analysis capabilities for nonlinear influencing factors is utilized to build an LGBM prediction model, and then an optimal combination method is used for model fusion. In addition, the impact of demand response, that is, electricity price factors, on regional loads is also considered. Through testing on the load data set, the results show that the fusion model has better load forecasting performance than individual model, and the MAPE (Mean Absolute Percentage Error) of the test set is 1.597%.