Abstract:

Improving the forecasting accuracy of solar irradiance, as an important safeguard ensuring the stability of grid operations, remains one of the challenging technical bottlenecks. This study constructs a short-time solar irradiance forecasting model (R model) based on the cloud relative radiative forcing ratio R derived by the ground-based solar irradiances from the relationship between clouds and radiation and then evaluates the forecasting accuracy of the R model using 16-year solar irradiance measurements at the Southern Great Plains Central Facility site in the United States. The results show that (1) the R model significantly outperforms the simple persistence model and exhibits a 2%-25% improvement relative to the advanced smart persistence model (also called cloud relative radiative forcing (RCRF) model herein) in the cloudy case. (2) For the overall validation with 2.9×10⁵ individual cloudy cases in 8 categories over 16 years, the forecast accuracy of the R model is significantly better than that of the simple model and the RCRF model at lead times longer than 1 h. Compared with the RCRF model, the forecast performance of the R model is improved by 25% and 19% in global horizontal irradiance and direct normal irradiance, respectively, and the lead time is extended by 1.5 and 1 h. (3) The R model with higher forecast performance sets a higher standard for the benchmark model of short-term solar irradiance forecasting. Meanwhile, the R model only requires short-term ground-based radiation observations for forecasting, facilitating and providing new possibilities for photovoltaic power plants without concurrent meteorological measurements in short-term solar irradiance forecasting.

Key words: persistence model, short-term irradiance forecast, forecasting performance, photovoltaic power plant