Refinement of irradiance models in PV simulators can improve accuracy of system-level yield predictions

As the market emphasis on grid-tied PV systems shifts from peak power rating to total energy production, the accuracy of system-level production simulators becomes increasingly important. SunPower Corporation uses a proprietary in-house software package (PVGrid) that allows a high level of flexibility for modeling the mechanisms specific to our line of PV mounting products, including flat, fixed-tilt roof- and ground-mounts, and single NS-axis trackers. We have data available from a large fleet of installed fixed-tilt and tracking systems for validating the PVGrid models, and capitalized on this data to make improvements to the irradiance models in PVGrid. PVGrid previously used the DISC beam irradiance model, which may become unstable under certain climatic conditions and under-predict beam irradiance. We replaced this model with an updated version of the Perez beam model that includes a Linke turbidity term and an irradiance variability index. The refinement of our POA irradiance models and algorithms in PVGrid resulted in a reduction in MBE from - 36.7 to −2.31 and in RMSE from 87.18 to 74.07 for calculated POA irradiance using measured beam, diffuse, and POA irradiance data furnished by the Sandia National Labs. This reduction in MBE improved the total annual calculated POA from 94.6% to 99.6% of measured POA irradiance. The irradiance model refinements improved calculated POA irradiance accuracy for both high-beam climates (e.g., Southwest US) as well as for lower-beam climates (e.g., Northeast US, Germany).