Validation of GFS day-ahead solar irradiance forecasts in China

Abstract: This paper provides a benchmark to evaluate operational day-ahead solar irradiance forecasts of Global Forecast System (GFS) for solar energy applications in China. First, GFS day-ahead solar irradiance forecasts are validated quantitatively with hourly observations at 17 first-class national solar monitoring stations and 1 Baseline Surface Radiation Network (BSRN) station all over China. Second, a hybrid forecast method based on Gradient Boosting (GB) and GFS product is proposed to improve forecasts accuracy. Both GFS forecasts and GB-based forecasts are compared with persistence forecasts. The results demonstrate persistence model is more accurate than GFS forecasts, and the hybrid method has the best performance. Besides, parameter optimization of direct-diffuse separation fails to reduce the errors of direct normal irradiance (DNI) forecasts.