From Marginal to Simultaneous Prediction Intervals of Wind Power

Abstract

The current literature in wind power forecast is focused in generating accurate uncertainty forecasts and communicating this information to the end-user. Multi-temporal decision-making problems require information about the temporal trajectory of wind power for the next hours. Presently, this information is provided through a set of temporal trajectories (or scenarios). This paper aims at contributing with an alternative approach for communicating this information through simultaneous prediction intervals. These intervals include the temporal dependency of forecast errors since they provide information about the probability of having the observed wind power trajectory fully inside the quantiles forming the interval. First, a learning sample of temporal trajectories are generated with the Gaussian copula method and using the marginal prediction intervals. Then, two methods proposed in the literature are used to construct the simultaneous intervals. The quality of these intervals is evaluated for three real wind farms.