Short-term load forecasting based on DenseNet-LSTM fusion model

Abstract

Short-term load forecasting is of great significance to the smooth operation and safe dispatch of the power grid, and the accuracy and real-time performance of the existing models need to be further improved. Aiming at the shortcomings of existing models that cannot balance prediction accuracy and computational complexity, a fusion model based on dense block network (DenseNet) and long short-term memory network (LSTM) is proposed. First, the improved DenseNet is introduced to mine the potential characteristics of historical load data, and then the data characteristics are dynamically trained through LSTM to reduce the loss of time series characteristics and realize short-term load forecasting of power data. Finally, the public data set is used to analyze the calculation examples. The experimental results show that the fusion model based on DenseNet-LSTM has higher prediction accuracy and generalization ability, while reducing the amount of calculation, and has a good application prospect.