Advances in Predicting Solar Power for Utilities

NCAR is leading a team to advance solar power forecasting by focusing on a series of technologies that span the continuum of solar forecasting from 15 min to 36 hours. Several approaches are being taken for the short-term nowcasting: 1) using statistical learning methods tuned for specific use-dependent regimes, 2) near-plant prediction based on multiple sky imagers, 3) advecting satellite-derived clouds, including parallax and shadowing effects, and 4) assimilating observations from both satellites and ground-based sensors such as total sky imagers into the dynamical core of the Weather Research and Forecasting (WRF) model to be run in rapid-update (15 min cycle) mode. Beyond a few hours, numerical weather prediction is the focus. NCAR is developing WRF-Solar, that incorporates new algorithms for shallow convection, cloud physical processes and parameterizations, radiative transfer algorithms, and data assimilation advances that allow full use of high resolution satellite imaging. The team is implementing the advanced aerosol estimation schemes that are used both in the radiative transfer schemes and in the cloud physics parameterization, NWP modeling capabilities are tuned for solar radiation at specific locales, and preparation is being made for very high resolution runs (up to 1 km resolution around solar power plants) to distinguish cloud development, movement, and dissipation. The data from these models are being blended with publically available forecasts using computational learning methods that optimize weighting of disparate forecast models. Finally irradiance-to-power conversion methods, including data mining, provide an estimate of power production from each solar array on several time scales. Finally, economic analysis is implemented to assess the value of the forecast improvements. The project incorporates a prototype solar forecasting system into operations and tests it in collaboration with solar plant developers, utilities, and independent system operators (ISOs) in geographically diverse areas, such Long Island, Colorado, coastal California, Florida, and Hawaii. The system technologies leverage solar radiation and cloud measurements, including images from total sky imagers; satellite observations; local meteorological observations; publically available NWP modeling; a customized version of the WRF model (WRF-Solar) tuned for cloud prediction and assimilating specialized data; radiative transfer modeling; statistical blending of forecast technologies tuned to each prediction time; irradiance to power conversion models; and built-in assessment metrics. This is an iterative process with user feedback leading to a forecast capability that adds value to the commercial partners. This talk will provide an update on progress as well as lessons learned.