Abstract
Accurate 24-hour solar irradiance forecasting is essential for the safe andeconomic operation of solar photovoltaic systems. Traditional numerical weatherprediction (NWP) models represent the state-of-the-art in forecastingperformance but rely on computationally costly data assimilation and solvingcomplicated partial differential equations (PDEs) that simulate atmosphericphysics. Here, we introduce SolarSeer, an end-to-end large artificialintelligence (AI) model for solar irradiance forecasting across the ContiguousUnited States (CONUS). SolarSeer is designed to directly map the historicalsatellite observations to future forecasts, eliminating the computationaloverhead of data assimilation and PDEs solving. This efficiency allowsSolarSeer to operate over 1,500 times faster than traditional NWP, generating24-hour cloud cover and solar irradiance forecasts for the CONUS at 5-kilometerresolution in under 3 seconds. Compared with the state-of-the-art NWP in theCONUS, i.e., High-Resolution Rapid Refresh (HRRR), SolarSeer significantlyreduces the root mean squared error of solar irradiance forecasting by 27.28%in reanalysis data and 15.35% across 1,800 stations. SolarSeer also effectivelycaptures solar irradiance fluctuations and significantly enhances thefirst-order irradiance difference forecasting accuracy. SolarSeer's ultrafast,accurate 24-hour solar irradiance forecasts provide strong support for thetransition to sustainable, net-zero energy systems.