Probabilistic solar nowcasting based on all-sky imagers

Abstract

The interest in shortest-term solar irradiance forecasts (nowcasts) increases steadily with the increase share of distributed solar power generation. Such solar irradiance nowcasts are beneficial for different stakeholders, from generation to transmission and demand, in order to tackle challenges caused by the variability of solar irradiance. In space and time highly resolved nowcasts can be obtained by all sky imager (ASI) systems, which analyze the sky conditions from sky images. Deterministic nowcasts from ASI systems are subject to significant uncertainties. Reliable uncertainty information are very helpful for any application, in order to derive practical benefit from nowcasts. Therefore, such nowcasts should be probabilistic in nature, which provide probability distributions. Meaningful indicators for the uncertainties at hand are provided by prediction intervals for distinct confidence levels derived from the probability distributions. Thus, a real time capable nonparametric probabilistic quantile nowcasting method based on deterministic ASI nowcast is developed. The method takes irradiance variabilities as main predictor of nowcast uncertainties into account. A benchmark against three distinct baseline models is conducted over an extensive data set, using a variety of recently recommended scores. Overall average continuous ranked probability skill scores (Clear-Sky Dependent Climatology as baseline) for nowcasts up to 20 min ahead of 0.72 ± 0.08 (direct normal irradiance) and 0.62 ± 0.09 (global horizontal irradiance) are reached. For a better evaluation of the actual performance of the probabilistic nowcasts, a discretization of the validation data set into eight irradiance variability conditions is performed. All scores are determined for each of these distinct conditions.