Machine learning for accelerating the discovery of high-performance donor/acceptor pairs in non-fullerene organic solar cells

Integrating artificial intelligence (AI) and computer science together with current approaches in material synthesis and optimization will act as an effective approach for speeding up the discovery of high-performance photoactive materials in organic solar cells (OSCs). Yet, like model selection in statistics, the choice of appropriate machine learning (ML) algorithms plays a vital role in the process of new material discovery in databases. In this study, we constructed five common algorithms, and introduced 565 donor/acceptor (D/A) combinations as training data sets to evaluate the practicalities of these ML algorithms and their application potential when guiding material design and D/A pairs screening. Thus, the best predictive capabilities are provided by using the random forest (RF) and boosted regression trees (BRT) approaches beyond other ML algorithms in the data set. Furthermore, >32 million D/A pairs were screened and calculated by RF and BRT models, respectively. Among them, six photovoltaic D/A pairs are selected and synthesized to compare their predicted and experimental power conversion efficiencies. The outcome of ML and experiment verification demonstrates that the RF approach can be effectively applied to high-throughput virtual screening for opening new perspectives to design of materials and D/A pairs, thereby accelerating the development of OSCs.