Assessment of Wind Speed Dependent Prediction Error

Abstract

The investigations in this chapter follow the idea that the prediction error quantitatively depends on the meteorological situation that has to be predicted. As a first approach the wind speed as a main indicator of the forecast situation is considered in greater detail. The probability density functions (pdf) of the measured wind speed conditioned on the predicted one are found to be Gaussian in the range of wind speeds that is relevant for wind energy applications. An analysis of the standard deviations of these conditional pdfs reveals no systematic dependence of the accuracy of the wind speed prediction on the magnitude of the wind speed. With the pdfs of the wind speed as basic elements, the strongly non-Gaussian distribution of the power prediction error is explained underlining the central role of the non linear power curve. Moreover, the power error distribution can easily be estimated based on the statistics of the wind speed, the wind speed forecast error and the power curve of the turbine. Thus, it can be reconstructed without knowing the actual measured power output, which is interesting for future sites or sites where no data are available. In addition, a simple formula based on linearising the standard deviation of error is derived. This model illustrates the dominating effect of small relative errors in the wind speed prediction being amplified by the local derivative of the non-linear power curve.