Development of short-term forecast quality for new offshore wind farms

Abstract

As the rapid wind power build-out continues, a large number of new wind farms will come online but forecasters and forecasting algorithms have little experience with them. This is a problem for statistical short term forecasts, which must be trained on a long record of historical power production – exactly what is missing for a new farm. Focus of the study was to analyse development of the offshore wind power forecast (WPF) quality from beginning of operation up to one year of operational experience. This paper represents a case study using data of the first German offshore wind farm "alpha ventus" and first German commercial offshore wind farm "Baltic1". The work was carried out with measured data from meteorological measurement mast FINO1, measured power from wind farms and numerical weather prediction (NWP) from the German Weather Service (DWD). This study facilitates to decide the length of needed time series and selection of forecast method to get a reliable WPF on a weekly time axis. Weekly development of WPF quality for day-ahead WPF via different models is presented. The models are physical model; physical model extended with a statistical correction (MOS) and artificial neural network (ANN) as a pure statistical model. Selforganizing map (SOM) is investigated for a better understanding of uncertainties of forecast error.