Modelling power output from correlated wind farms

Abstract

Wind data from fifteen potential wind farm sites covering New Zealand were analysed. The data was fitted with a generalised gamma distribution and transformed to a Gaussian distribution. A regression was used to model the temporal correlation and the spatial correlation was modelled with a multivariate Gaussian distribution. This enabled prediction of the distribution of power generated and of the changes in power over ten minute intervals. For predicting up to two hours ahead, prediction of wind speed was moderate, but the steep relation of wind to power meant that power predictions were not good.