Statistisches Post-Processing von Windleistungsprognosen für Übertragungsnetzgebiete

The objective of this survey is the improvement of the wind power forecasts generated at the Fraunhofer Institute for Wind and Energy Systems Technology in terms of their accuracy. It considers day-ahead and short-term forecasts that have been generated on the basis of various weather models for the EEG-marketed feed-in to the control areas and throughout Germany. The forecasts are modified with univariate post-processing models of time series analysis. These models receive additional modifications to the characteristics of the wind power forecasts. Therefore methods of exponential smoothing are tested, where forecasts are modified on the basis of all values recorded up to the time the forecast was generated by comparing the previous forecasts with observations. The forecast is done by weighting and summing up the previous values. In this work the error pattern is directly corrected and applied to the forecasts. It is a pure consideration of recorded errors and not a machine learning procedure. For this reason, the advantage of this application is its simplicity and robustness