Solar Energy Irradiance and Electric Generation Forecasting

Recent and projected increases in solar energy generation are creating a need for short term forecasts for this variable energy source. The emphasis of this presentation is on hourly and sub-hourly prediction for the day-ahead time period, which is well within the time range when numerical weather prediction (NWP) models must be relied upon but also must be corrected for various biases. Forecast systems for predicting real-time solar energy generation are being developed along similar lines to those of more established wind forecast systems, but the challenges and constraints are different. Clouds and aerosols play a large role, and for tilted photovoltaic panels and solar concentrating plants, the direct beam irradiance, which typically has much larger forecast errors than global horizontal irradiance, must be utilized.

At MDA Information Systems, we are developing a forecast system based on first principles, with the well-validated REST2 clear sky model (Gueymard, 2008) at its backbone. In tuning the model and addressing aerosol scattering and surface albedo, etc., we relied upon the wealth of public data sources including AERONET (aerosol optical depth at different wavelengths), Suominet (GPS integrated water vapor), NREL MIDC solar monitoring stations, SURFRAD (includes upwelling shortwave), and MODIS (albedo in different wavelength bands), among others. The forecast itself utilizes a blend of NWP model output, which must be brought down to finer time resolution based on the diurnal cycle rather than simple interpolation. Many models currently do not output the direct beam irradiance, and one that does appears to have a bias relative to its global horizontal irradiance, with equally good performance attained by utilizing REST2 and the model global radiation to estimate the direct component. We will present a detailed assessment of various NWP solar energy products, evaluating forecast skill at a range of photovoltaic installations.

Most forecasts and forecast users have so far been interested in generation by utility-scale installations, but the cumulative hidden load reductions due to smaller distributed installations such as residential rooftop panels also will soon become important. We will be developing a system for forecasting the generation from these distributed sources as well, and if we have new findings on this endeavor at the time of the conference, we will include them in our presentation.