Research on Key Technologies for Static Weather Forecasting of Xinjiang Power System Based on Machine Learning and Multi source Numerical Model Fusion

Xinjiang’s power system is characterized by a vast service area, complex terrain, long transmission corridors, widely distributed substations, and a high proportion of renewable generation. Static and stable weather, marked by low wind speed, inversion, and weak diffusion conditions, can intensify risks such as pollution flashover, icing accumulation, subsequent line galloping, renewable output fluctuation, and regional load imbalance. To improve power-oriented stagnant weather forecasting, this study integrates machine learning with multi-source numerical model fusion. ECMWF, GRAPES, and WRF data are standardized, bias-corrected, and downscaled to a unified high-resolution grid, then combined with ground observations, reanalysis data, grid equipment records, and fault histories. A power-specific stagnant weather index system is constructed, and machine-learning models, including random forest, gradient boosting, LSTM, and attention networks, are used for fusion forecasting. The resulting products support grid-based and line-level forecasting for dispatching, maintenance, and risk warning. The study directly contributes to electromagnetic insulation safety, high-voltage transmission reliability, and power-system electromagnetic environment management.