Short-Term Electrical Load Forecasting Based on Time Augmented Transformer

Abstract Electrical load forecasting is of vital importance in intelligent power management and has been a hot spot industrial Internet application field. Due to the complex patterns dynamics data, accurate short-term still challenging task. Currently, many tasks use deep neural networks for forecasting, most recurrent network as basic architecture, including Long Short-Term Memory (LSTM), Sequence (Seq2Seq), etc. However, performance these models not good expected due gradient vanishing problem network. Transformer learning model initially designed natural language processing, it calculates input–output representations captures long dependencies entirely on attention mechanisms which great capturing dynamic nonlinear sequence dependence input. In this work, we proposed Time Augmented (TAT) electrical forecasting. A temporal augmented module TAT learn relationships representation between input history series adapt We evaluate our approach real-word dataset extensively compared existed statistical approach, traditional machine methods, experimental results show that higher precision accuracy