Spatial and Spatio-Temporal Modelling of Wind Speed Using IDW, Kriging, and XGBoost with Weibull Deviance

Accurate modelling of wind speed (WS) is crucial for wind energy forecasting. This study used spatial interpolation methods—Inverse Distance Weighting (IDW) and kriging with truncated Weibull data and spatio-temporal modelling approach using Extreme Gradient Boost (XGBoost). While IDW and kriging demonstrated lower root mean square error (RMSE), the spatio-temporal model implemented with XGBoost provided a framework to obtain dependencies across both space and time. To address the skewness characteristic in wind speed data, a custom Weibull deviance loss function was developed and integrated into the XGBoost framework. This research has done some simulations for making a comparison between proposed custom loss and default loss functions and found custom loss in XGBoost outperformed default loss functions in terms of Mean Absolute Error (MAE) and Gamma deviance. These results emphasize the importance of using Weibull deviance as loss functions for Weibull-distributed data and highlight it as a promising tool for enhancing wind speed prediction accuracy within XGBoost.