Weekly Electricity Forecasting Method Based on Double-Layer lightGBM-GRU-IBES Algorithm

Abstract

This paper presents a power forecasting method based on the two-layer lightGBM-GRU-IBES algorithm, aiming to ensure stable power grid development and control power costs. The proposed method comprises the following steps: First, the processed feature data is fed into the first layer of the lightGBM model to perform feature importance analysis, filtering features with high correlation. Next, the selected features are input into the second layer, which includes both the lightGBM model and the GRU model, for training and prediction. Finally, the GRU model's predictions are weighted and corrected to enhance the lightGBM model's predictions, producing the final power forecasting results. To further enhance prediction accuracy, the improved BES algorithm is used to optimize the model's hyperparameters. The method's performance is evaluated using real electricity data from Yunnan Province through an example analysis. The results demonstrate that the proposed model outperforms traditional classic models in terms of accuracy and fitting, achieving high-quality electricity forecasting. This approach holds promise for accurate and effective power consumption prediction, benefiting power grid stability and cost control.