A weighted integrated forecasting approach based on VMD decomposition reconstruction for high energy-consuming loads

At present, high energy-consuming industrial loads such as fused magnesium, industrial manufacturing, and high-tech development have the characteristics of large energy consumption and peak control in the power conversion process. Accurate load forecasting is needed to carry out reasonable production scheduling. In the process of electro motor magnesium smelting, the power load usually has the characteristics of non-stationary, non-periodic, high fluctuation, etc. It makes power load forecasting and scheduling extremely difficult. Therefore, a weighted integrated forecasting method is proposed in this paper for high energy-consuming loads based on variational mode decomposition (VMD) and reconstruction. Firstly, an integrated forecasting model based on a decomposition strategy is established. The original sequence of high energy-consuming loads is decomposed by the VMD method. The multiple models are used to predict and reconstruct the sub-sequences, which improved the forecasting accuracy of the load sequence of fused magnesium by the traditional single model. Secondly, the multi-objective parameter optimization method of the weighted integrated forecasting model is proposed to learn multi-angle sequence implicit information and avoid the influence of the hybridization of the decomposed sequence information. The weight combination is optimized by a multi-objective optimization algorithm, and the accuracy and robustness of the model are optimized simultaneously with MAE and EVS as the optimization objectives. Finally, through the Pareto optimal solution method of adaptive variance risk threshold, the robustness of the model is taken as the constraint condition. The results show that the weighted combination model shows excellent forecasting performance on subsequences with different frequency characteristics. At the same time, the proposed weighted integrated forecasting method scientifically adjusts the combined weights of each model based on multi-objective parameter evolution optimization. It improves the forecasting accuracy and ensures the robustness of the model.