Machine Learning-based Classification of Spectral Conditions for High-Throughput Indoor Testing of Photovoltaic Modules

Abstract

High-throughput testing of solar modules to accurately predict energy yield (EY) is increasingly important as more of the power grid runs on photovoltaics (PV). Modules are sold based on power ratings measured under standard testing conditions, not fully considering environmental conditions of the real world. In this work, we use the k-means algorithm to extract the best representative conditions of the environment that minimizes error in EY. The work presented here is a fully scoped proof-of-concept demonstrated on a year of spectral data clustered and analyzed for every month of 2017 in Boulder, Colorado. Preliminary results demonstrate a decrease in 5 percent relative error in energy yield predictions between one standard testing condition and up to seven clusters found with this method. This can be generalized to more locations around the world as a powerful tool for EY estimation. These results demonstrate the capacity for high throughput, accurate EY prediction using clustered conditions.