Estimating Wind Power Uncertainty using Quantile Smoothing Splines Regression

Forecast errors in wind power forecasting are unavoidable due to the complex nature of weather systems and other influences. As a result, quantifying wind power uncertainty is essential for optimally operating grids with a high share of wind energy. This paper uses the quantile smoothing splines (QSS) regression to estimate conditional quantiles of wind power forecast error for a given wind power forecast. This approach is tested using data from eight wind farms in South Africa and evaluated using reliability, sharpness, resolution, and skill score. The results are compared to that of two commonly used approaches: linear regression and fitting beta distributions in different bins. Despite the slight superiority of QSS regression, this paper finds that the results of QSS regression and fitting beta distributions in different bins are comparable. The benefit of using QSS regression, however, is that it is a nonparametric approach that produces smooth results with no discontinuities, and no need for parameter estimations for each bin, making it easily applicable. System operators can use the estimated quantiles to allocate operating reserves and hence ensure the efficient integration of wind farms into the power grid.