Realtime Decomposition of Site-Measured Solar Irradiance Using Machine Learning for Bifacial System Performance Characterization

This work presents a method for decomposing realtime on-site measured plane of array (POA) and global horizontal (GHI) irradiance from pyranometers into beam and diffuse components. A machine learning method is applied in conjunction with typical metereological year (TMY) irradiance data to facilitate reliable irradiance decomposition of five minute measured data, a regime in which conventional decomposition methods like GTI-DIRINT sometimes struggle. The approach is combined with view factor models of rear side irradiance for bifacial systems to reliably calculate performance ratio and other metrics. Validation of the method on bifacial utility-scale solar power plant data shows credible results.