An Ensemble Electrometallurgical Load Forecasting Method Based on Hierarchical Parameter Optimization

Accurate load forecasting is essential for the electric metallurgy industry. The complex nature of electric metallurgy load complicates forecasting. To address this, we propose an algorithm combining decomposition, reconstruction, and hierarchical parameter optimization. We use Variational Mode Decomposition (VMD) to handle non-stationarity and varying time scales. Both homogeneous and heterogeneous models forecast the decomposed subsequences. Traditional methods often treat decomposition and model parameter selection independently, leading to poor coupling and errors. Our hierarchical optimization approach addresses decomposition and model parameters simultaneously. At the lower level, the goal is to optimize the decomposition parameters by fixing the model parameters and evaluating the fit between the decomposed subsequences and model. At the upper level, the objective is to achieve the highest ensemble prediction accuracy. The optimized decomposition parameters were then incorporated to optimize the model parameters. This iterative process continues until the predetermined termination criteria are fulfilled. To establish the generality of hierarchical optimization, it was applied to single-model predictions under the premise of VMD decomposition. The results consistently demonstrate that the algorithm displays impressive accuracy and versatility.