Diurnal Changes and Machine Learning Analysis of Perovskite Modules Based on Two Years of Outdoor Monitoring
Vasiliki Paraskeva, Matthew Norton, Andreas Livera, Andreas Kyprianou, Maria Hadjipanayi, Elias Peraticos, Aránzazu Aguirre, Santhosh Ramesh, Tamara Merckx, Rita Ebner, Tom Aernouts, Anurag Krishna, George E. Georghiou
Abstract
Long-term stability is the primary challenge for the commercialization of perovskite photovoltaics, exacerbated by limited outdoor data and unclear correlations between indoor and outdoor tests. In this study, we report on the outdoor stability testing of perovskite mini-modules conducted over a two-year period. We conducted a detailed analysis of the changes in performance across the day, quantifying both the diurnal degradation and the overnight recovery. Additionally, we employed the XGBoost regression model to forecast the power output. Our statistical analysis of extensive aging data showed that all perovskite configurations tested exhibited diurnal degradation and recovery, maintaining a linear relationship between these phases across all environmental conditions. Our predictive model, focusing on essential environmental parameters, accurately forecasted the power output of mini-modules with a 6.76% nRMSE, indicating its potential to predict the lifetime of perovskite-based devices.