Short-Term Wind-Forecast Calibration for Energy Management Using Numerical Modeling and In Situ Measurements

The increasing integration of renewable energy into transmission grids requires accurate short-term wind forecasting to optimize Dynamic Line Rating systems, yet numerical weather prediction (NWP) models alone show insufficient accuracy for energy management applications. This study presents a machine learning framework that combines numerical weather prediction outputs with real-time observations from 18 meteorological stations in northern Iberian Peninsula to calibrate short-term wind forecasts. A Light Gradient Boosting Machine (LGBM) algorithm was trained on one year of data to generate 24 independent models for prediction horizons from 1 to 24 h, incorporating lagged observations, WRF forecasts, and station metadata as predictors. During a 48 h evaluation test, the LGBM-corrected forecasts reduced mean absolute error from 2.095 to 0.94 m/s for wind speed and from 90.2° to 47.3° for wind direction compared to raw WRF outputs, with improvements exceeding 62% and 56%, respectively, at 5 h lead time. The methodology demonstrates that integrating sensor network observations with numerical modeling through machine learning post-processing substantially enhances short-term wind prediction accuracy, providing a practical solution for real-time energy grid management and renewable integration.