Analysis of confidence intervals for the prediction of regional wind power output

Abstract

According to the increased penetration of wind energy into the utilitygrids, various methods for the short-term (time horizon6h-48h) prediction of the power output of single installations aswell as the ensemble output of groups of turbines withinspecific regions are in application or under development. For methodsthat are based on the application of the outcome ofnumerical weather prediction schemes, the overall quality of the predictionsas measured by long term values of the meanbias and the root mean square errors has been extensively analyzed.For the operational application however, the assignmentof a specific confidence interval for each individual prediction wouldbe desirable. This paper aims in sketching out a way toidentify this information.We present the results of an approach to sort the magnitude of theprediction error in terms of its standard deviationaccording to parameters describing the meteorological situation arisingin course of the forecast procedure (e.g. the predictedpower output itself, the temporal evolution of the forecasted data).Based on the respective values and with respect to the nongaussian nature of the distribution of the forecast errors, a procedureto determine confidence intervals for the expected actualpower is proposed.