Abstract: To address the issue of low accuracy in short-term power load forecasting caused by the nonlinearity, volatility, and inadequate feature extraction of historical load data, a short-term power load forecasting model based on VMD-SGWO multi-model optimization was proposed. Firstly, the variational mode decomposition (VMD) method was employed to decompose the historical load data into five modal components and a residual sequence. Then, the Sine chaotic mapping was used to improve the grey wolf optimizer (GWO), resulting in the sine-enhanced grey wolf optimizer (SGWO). This improved algorithm was applied to optimize the parameters of the light gradient boosting machine (LightGBM), support vector regression (SVR), and extreme gradient boosting (XGBoost) models for different modals. Subsequently,the optimized models were used to predict the different modal components separately. Finally, the prediction results were reconstructed and optimized to obtain the final forecast. Experiments demonstrated that VMD, SGWO, and the multi-model collaborative forecasting approach effectively enhanced the accuracy of short-term power load forecasting. The proposed VMD-based multi-model optimization algorithm was compared with LightGBM, SVR, XGBoost, temporal convolutional network (TCN), gated recurrent unit (GRU), and long short-term memory network (LSTM) in terms of short-term power load forecasting results, showing reductions in root mean square error by 86.25%, 82.68%, 87.29%, 86.30%, 89.78%, and 84.79%, respectively, and the prediction accuracy was significantly improved based on VMDSGWO multi-model optimization.

    Key words: short-term power load forecasting; variational mode decomposition; grey wolf optimizer; multi-model optimization