Power-Load Forecasting Model Based on Informer and Its Application
Hongbin Xu, Qiang Peng, Yuhao Wang, Zengwen Zhan
Abstract
Worldwide, the demand for power load forecasting is increasing. A multi-step power-load forecasting model is established based on Informer, which takes the historical load data as the input to realize the prediction of the power load in the future. The constructed model abandons the common recurrent neural network to deal with time-series problems, and uses the seq2seq structure with sparse self-attention mechanism as the main body, supplemented by specific input and output modules to deal with the long-range relationship in the time series, and makes effective use of the parallel advantages of the self-attention mechanism, so as to improve the prediction accuracy and prediction efficiency. The model is trained, verified and tested by using the power-load dataset of the Taoyuan substation in Nanchang. Compared with RNN, LSTM and LSTM with the attention mechanism and other common models based on a cyclic neural network, the results show that the prediction accuracy and efficiency of the Informer-based power-load forecasting model in 1440 time steps have certain advantages over cyclic neural network models.