General machine learning models for interpreting and predicting efficiency degradation in organic solar cells

Abstract

Photovoltaic (PV) energy plays a key role in addressing the growing global energy demand. Organic solar cells (OSCs) represent a promising alternative to silicon-based PVs due to their low cost, lightweight, and sustainable production. Despite achieving power conversion efficiencies (PCEs) over 20 %, OSCs still face challenges in stability and efficiency. Recent advances in manufacturing, artificial intelligence and machine learning (ML) achieve optimized and screened OSCs for greater sustainability and commercial viability, thus potentially reducing costs while ensuring stable and long term performance. This work presents optimal ML models to represent the temporal degradation on the PCE of polymeric OSCs with structure ITO/PEDOT:PSS/P3HT:PCBM/Al. First, we generated a database with 166 entries with measurements of 5 OSCs, and up to 7 variables regarding the manufacturing and environmental conditions for more than 180 days. Then, we relied on a software framework that provides a conglomeration of automated ML protocols that execute sequentially against our database by simply command-line interface. This easily permits hyper-optimizing the ML models through exhaustive benchmarking so that optimal models are obtained. The accuracy for predicting PCE over time reaches values of the coefficient determination widely exceeding 0.90, whereas the root mean squared error, sum of squared error, and mean absolute error are significantly low. Additionally, we assessed the predictive ability of the models using an unseen OSC as an external set. For comparative purposes, classical Bayesian regression fitting are also presented, which only perform sufficiently for univariate cases of single OSCs.