Application Research of Power Load Forecasting and Real-time Scheduling Based on SHAP Interpretable Combination Model

A multi-source fusion framework is developed for power load forecasting, integrating meteorological, pricing, and historical data. Temperature-humidity-load coupling effects are quantified through nonlinear modeling. A spatiotemporal ensemble combining LSTM temporal modeling and XGBoost feature interactions is optimized via dynamic adaptive weighting. SHAP analysis identifies temperature abruptness as primary error contributors. The framework demonstrates 23% robustness improvement under extreme weather conditions with 15% overfitting reduction. The "data-algorithm-decision" paradigm is validated to enhance grid security for renewable energy integration.