Predictor-weighting strategies for probabilistic wind power forecasting with an analog ensemble

Abstract

Unlike deterministic forecasts, probabilistic predictions provide estimates of uncertainty, which is an additional value for decision-making. Previous studies have proposed the analog ensemble (AnEn), which is a technique to generate uncertainty information from a purely deterministic forecast. The objective of this study is to improve the AnEn performance for wind power forecasts by developing static and dynamic weighting strategies, which optimize the predictor combination with a brute-force continuous ranked probability score (CRPS) minimization and a principal component analysis (PCA) of the predictors. Predictors are taken from the high-resolution deterministic forecasts of the European Centre for Medium-Range Weather Forecasts (ECMWF), including forecasts of wind at several heights, geopotential height, pressure, and temperature, among others. The weighting strategies are compared at five wind farms in Europe and the U.S. situated in regions with different terrain complexity, both on and offshore, and significantly improve the deterministic and probabilistic AnEn forecast performance compared to the AnEn with 10-m wind speed and direction as predictors and compared to PCA-based approaches. The AnEn methodology also provides reliable estimation of the forecast uncertainty. The optimized predictor combinations are strongly dependent on terrain complexity, local wind regimes, and atmospheric stratification. Since the proposed predictor-weighting strategies can accomplish both the selection of relevant predictors as well as finding their optimal weights, the AnEn performance is improved by up to 20 % at on and offshore sites.