Day-Ahead Electric Load Forecasting Based on T2VC-Informer

Accurate day-ahead load forecasting is of great significance to arrange day-ahead schedule plan, maintain the stability of power system and prolong the service life of equipment. This paper proposes a novel day-ahead electric load forecasting model based on time2 vec embedding layer, 1D convolutional layer and Informer network (T2VC-Informer). Firstly, meteorological factors highly correlated with electric load are selected based on Spearman correlation coefficient to determine the input features of the forecast model. Then, the T2VC-Informer forecast model is established, where the time 2 vec embedding layer is used to learn periodic and non-periodic patterns in the original inputs, the 1 D convolutional layer is utilized to extract high-dimensional spatiotemporal features, and the Informer is adopted to efficiently capture long-range dependency between input and output sequences. Finally, in the comparison experiment, the mean absolute percentage error of T2VC-Informer is 35.283% and 11.609 % lower than that of LSTM and Transformer, respectively, fully proving its superiority.