Abstract: Techniques for operating a wind farm include setting an area of interest, a forecast interval, and a maximum lag time for using mesoscale forecasts. Mesoscale forecasts are collected for a training time interval TT at model grid locations. TT is at least ten times the maximum lag time. Fine-scale wind measurements are collected in the area during TT. Selected parameters of the mesoscale forecasts, and coefficients of an evolving ML forecast model are determined based on the mesoscale forecasts and the fine-scale wind measurements during the TT ending at the current time. Then, the coefficients and the mesoscale forecast for the selected parameters during the lag time produce a forecast wind at the wind turbines during the forecast interval. Operation of the wind farm is based on the forecast wind.

Abstract: Techniques for operating a wind farm include setting an area of interest, a forecast interval, and a maximum lag time for using mesoscale forecasts. Mesoscale forecasts are collected for a training time interval TT at model grid locations. TT is at least ten times the maximum lag time. Fine-scale wind measurements are collected in the area during TT. Selected parameters of the mesoscale forecasts, and coefficients of an evolving ML forecast model are determined based on the mesoscale forecasts and the fine-scale wind measurements during the TT ending at the current time. Then, the coefficients and the mesoscale forecast for the selected parameters during the lag time produce a forecast wind at the wind turbines during the forecast interval. Operation of the wind farm is based on the forecast wind.