Two-Stage Short-Term Power Load Forecasting Based on SSA–VMD and Feature Selection

Short-term power load forecasting is of great significance for the reliable and safe operation systems. In order to improve accuracy short-term forecasting, problems random fluctuation in complexity load-influencing factors, this paper proposes a two-stage method, SSA–VMD-LSTM-MLR-FE (SVLM–FE) based on sparrow search algorithm (SSA), optimize variational mode decomposition (VMD) feature engineering (FE). Firstly, an evaluation criterion loss VMD proposed, SSA used find optimal combination parameters under criterion. Secondly, first stage carried out, different components obtained from SSA–VMD are predicted separately, with high-frequency input long memory network (LSTM) low-frequency multiple linear regression model (MLR) forecasting. Finally, values second error correction; factors high degree influence selected using Pearson correlation coefficient (PCC) maximal information (MIC), value at moment that has time be autocorrelation function (ACF). The fused construct vector, which fed into fully connected layer paper, performance SVLM–FE evaluated experimentally two datasets places China. Place 1, RMSE, MAE, MAPE 128.169 MW, 102.525 1.562%, respectively; 2, 111.636 92.291 1.426%, respectively. experimental results show stability.