SPLTRAK Abstract Submission
Using machine learning to predict the complete degradation of accelerated damp heat testing in just 10% of testing time
Zubair Abdullah-Vetter, Priya Dwivedi, Robert Lee Chin, Brendan Wright, Thorsten Trupke, Ziv Hameiri
UNSW, Sydney, Australia

The ability to accurately predict the long-term performance of photovoltaic modules would have substantial benefits for the photovoltaic market. If we can precisely determine how new modules will perform after 25-30 years in the field, the reliability and bankability of photovoltaic systems will significantly increase. Keeping this target in mind, this study presents the first step towards achieving more cost-effective degradation monitoring. We develop machine learning models to predict the performance of photovoltaic modules at the end of 1,000 hours of damp heat tests after modules have only spent less than 10% of that time in damp heat conditions. Hence, we investigate the ability of unsupervised neural ordinary differential networks to model the entire dynamics of the degradation during a damp heat test using only the data that is collected in the first 10% of the process. The developed algorithms can significantly reduce the required time for damp heat tests and pave the way to transform the photovoltaic market.