Feature Selection using Select-k-Best Approach with Support Vector Regression to Estimate Ultraviolet Irradiance from Solar Radiation Data

Terrestrial ultraviolet (UV) radiation is a significant factor that impacts the long-term reliability and durability of photovoltaic (PV) modules. This study introduces a two-step model designed to estimate the UV irradiance in six different wavebands based on broadband global horizontal irradiance and various atmospheric quantities. The model utilizes a large look-up table of spectral data generated by the SMARTS radiation model under diverse atmospheric conditions and solar positions. The first step involves selecting the most important features for optimal prediction performance using the k-Best selection approach. The selected optimal subset is then used as input to the Support Vector Regression (SVR) prediction technique. The results demonstrate strong prediction accuracy for all UV wavebands, particularly in the UV-A band.