Data Feature Processing and Improved Informer for Medium and Long-Term Electrical Power Forecasting

Accurate medium and long-term electrical power forecasting can not only effectively reduce energy consumption, but also plays an important role in the scheduling planning and economic operation of power grid. In order to improve the accuracy of medium- and long-term electrical power forecasting, this paper proposes a method of data feature processing and improved Informer for medium and long-term electrical power forecasting. Firstly, polynomial feature construction is used to expand the features of power consumption data. Secondly, factor analysis is used to eliminate the noise of power consumption data and reduce the complexity of medium- and long-term electrical power forecasting data. Thirdly, relative position encoding is introduced instead of absolute position encoding in the Informer model to better capture dependencies in sequence data and avoid information leakage. Finally, using dilated causal convolution instead of regular convolution increases the receptive field and enhances the ability to extract local information, then conduct comparative experiments with mainstream prediction models (GRU, LSTM, Transformer, and Informer) on multiple datasets. In addition, the applicability and accuracy of the informed Informer model were further validated by adding convolution kernels to evaluate the effectiveness of dilated causal convolution.