UC San Diego

Sky imagers are used for short-term forecasting of solar irradiance, which can be used to counter ramp errors caused by larger clouds or extensive changes in cloud cover. Sky imager forecast algorithms usually detect cloud classes in the image. However, the assignment of cloud optical depth or surface Global Horizontal Irradiance (GHI) to cloud classes remains a challenge. One method to connect GHI to cloud classes involves the use of a histogram of recently measured clear sky indices to assign a clear sky index/GHI to each cloud class. While this method improves upon choosing static GHI values for each cloud class, this thesis presents a modification which improves the histogram method. Considering data from a significantly lesser time period than the existing method emphasizes more recent cloud conditions. Individual histograms for each cloud class are used to analyze the data instead of a singular histogram. This modification ensured that the algorithm worked better even during sudden cloud formation or dissipation. The new algorithm gave a 30-45% reduction in error against the existing algorithm when tested over a period of time.

Keywords: solar, forecast, GHI, algorithm, whole sky imager