Consensus Forecasting of Global Horizontal Irradiance

As the penetration of solar energy advances, it becomes increasingly important to be able to forecast Global Horizontal Irradiance (GHI) at lead times ranging from 5 min to several days. Although equivalent variables may be available from models, they are seldom tuned to match observations. Typical methods to blend GHI predictions from these models are not straightforward, due to differences in the averaging times and meaning of the variables between models, as well as the difficulty in providing locale-specific predictions that depend on local effects of clouds and aerosols. Thus, there is a need to blend information from multiple models to provide a consensus GHI forecast that has been tuned to match observations from the site, whether they be commercial solar arrays or pyranometers in the vicinity of distributed rooftop solar installations.

Consensus forecasting methods have been shown to reduce forecast errors for a variety of forecast variables. These methods typically use machine learning techniques to improve forecasts over Direct Model Output (DMO). Here we describe the Dynamic Integrated ForeCast system (DICast), which is a consensus forecast system that has been recently extended to predict Global Horizontal Irradiance. Observations from pyranometers are used for system tuning and verification. This talk will present initial results from research funded by the US Department of Energy as part of the Solar Forecast Improvement Project.