Very short-term spatio-temporal wind power prediction using a censored Gaussian field

Wind power is a renewable energy resource, that has relatively cheap installation costs and it is highly possible that will become the main energy resource in the near future. Wind power needs to be integrated efficiently into electricity grids, and to optimize the power dispatch, techniques to predict the level of wind power and the associated variability are critical. Ideally, one would like to obtain reliable probability density forecasts for the wind power distributions. We aim at contributing to the literature of wind power prediction by developing and analysing a spatio-temporal methodology for wind power production, that is tested on wind power data from Denmark. We use anisotropic spatio-temporal correlation models to account for the propagation of weather fronts, and a transformed latent Gaussian field model to accommodate the probability masses that occur in wind power distribution due to chains of zeros. We apply the model to generate multi-step ahead probability predictions for wind power generated at both locations where wind farms already exist but also to nearby locations.