Multi-Factor Multi-step Power Peak Load Forecasting and Grid Scheduling Model Based on Mutual Information and Multiple Wrapper Feature Selection

The escalating complexity and scale of power systems necessitate precise peak load forecasting for grid stability and energy management. Current forecasting methods struggle with nonlinear internal relationships and multifactorial external interactions. This study presents a feature selection technique that combines mutual information with various wrapper algorithms to enhance the precision and efficiency of peak load forecasting. Initially, mutual information is used to screen features, quantifying their interdependence with loads to discard irrelevant and redundant features. The refined features are then processed through wrapper algorithms such as GA, PSO, SA, PCA, RFE, and SFS to optimize feature subsets and improve model generalization. The selected features are employed in an SVR model for both single-step and multi-step forecasting. The model was trained and validated using real-world power load and meteorological data from Pingxiang City, Jiangxi Province, China, and evaluated using statistical metrics like MAE and RMSE. Results indicate that the proposed feature selection method significantly enhances SVR forecasting accuracy, reduces computational complexity and operational costs in electric grids, and exhibits robust adaptability and stability in practical applications.