Electric Load Forecasting Based on Model Sample Re-Weighting Method
To address the adverse generalization effects caused by assigning equal weights to data samples in the electric load forecasting algorithm,a novel load prediction method based on sample weighting is proposed.In this approach,gradients and impact functions of samples are utilized during the training process to determine their respective weights.This weighting method aids in adjusting the model's fitting,allowing it to better adapt to those samples that are weighted,thereby enhancing the model's generalization performance.In deterministic prediction scenarios,through case studies involving linear models and artificial neural networks applied to real load datasets,it is observed that allocating different weights to diverse samples can further improve prediction accuracy.
electric load forecastingsample weightingempirical risk minimization
万方数据
维普
