An effective dimensionality reduction approach for short-term load forecasting

Accurate power load forecasting has a significant effect on smart grid by ensuring effective supply and dispatching of power. However, electric data generally possesses the characteristics nonlinearity, periodicity, seasonality. For complex systems, presence redundant information potentially reduces real pattern extraction for forecasting. Bearing in mind these issues, we propose an model which feature module is introduced that combined with variational mode decomposition (VMD) autoencoder (VAE). In this combination, VMD utilized decomposing series VAE used to filter (noises) from each decomposed series. With two sets China, demonstrate proposed can achieve highly accurate predictions, as find mean absolute percentage error (MAPE) values one-step-ahead prediction be 1% (Nanjing) 0.8% (Taixing), respectively.