Clear Sky Models Assessment for an Operational PV Production Forecasting Solution

Abstract

Photovoltaic production is mostly driven by the solar irradiance received at ground level. Forecasting surface solar irradiance remains to predict the cloudiness and combine it with the value of the irradiance modeled under a clear sky for the same area at the same forecast horizon. Thus, uncertainty of irradiance under clear sky can affect significantly the photovoltaic production forecast. Clear sky irradiance can be accurately computed if concentration of some atmospheric components (aerosol, water vapor and ozone) are sufficiently known above a location. Many clear sky models have been designed allowing a various number of inputs. In this work, we analyzed the performance of four different clear sky models. We compared their outputs against ground measurements located in Reunion Island, Corsica and French Guiana. We used the models with atmospheric parameters provided by two different sources (neighboring ground measurements and reanalysis). Best results lead to a relative root mean square error (rRMSE) of 3 % and an absolute relative mean bias error (rMBE) less than 1 %, for minutely irradiance. Using atmospheric parameters from reanalysis instead of punctual measurements significantly reduces errors in clear sky models.