A Study on Wind Power Output Prediction Using XGBoost and Spatial Interpolation of Meteorological Data

Accurate wind power prediction is essential to ensure grid stability as renewable energy integration increases. However, obtaining precise meteorological data at wind turbine locations is challenging due to technical and economic constraints. This study introduces a prediction model combining spatial interpolation and machine learning techniques using meteorological observations and reanalysis data. By leveraging data from a wind farm in Jeju Island, the model constructs a comprehensive meteorological database using kriging with elevation corrections adapted to seasonal variability, thereby enhancing the spatial coverage of meteorological inputs. Additionally, machine learning-based bias correction is applied to forecast data to improve predictive accuracy and enhance practical applicability. The proposed approach provides a practical solution for system operators, mitigating grid management risks and supporting the energy transition. Future work will focus on incorporating high-resolution regional forecasts and optimizing the integration of multiple reanalysis datasets to further enhance prediction performance.