Prediction intervals for wind power forecasting: Using sparse warped Gaussian process

Abstract

The accuracy of short-term wind power forecast is highly variable due to the stochastic nature of wind, so providing prediction intervals for such forecast is important for assessing the risk of relying on the forecast results. This paper focuses on building prediction intervals for the short-term wind power forecasts. A sparse Bayesian model is formulated to provide non-Gaussian predictive distributions for the future wind power, thus yields the prediction intervals. This model based on the warped Gaussian process (WGP), it handles the non-Gaussian uncertainties of the wind power series by automatically converting it to a latent series. The converted series is well-modeled by a Gaussian process (GP), then the non-Gaussian uncertainty of the wind power can be predicted in a standard GP framework. Since the high computational costs of WGP hinder its practical application on large-scale problems such as wind power forecast, we also give a method to sparsify the WGP. The simulation on actual data validates the effectiveness of the proposed model.