Predicting Power Conversion Efficiency of Organic Photovoltaics: Models and Data Analysis

In this paper, the ability of three selected machine learning neural and baseline models in predicting power conversion efficiency (PCE) organic photovoltaics (OPVs) using molecular structure information as an input is assessed. The bidirectional long short-term memory (gFSI/BiLSTM), attentive fingerprints (attentive FP), simple graph networks (simple GNN) well support vector regression (SVR), random forests (RF), high-dimensional model representation (HDMR) methods are trained to both large computational Harvard clean energy project database (CEPDB) much smaller experimental photovoltaic 15 dataset (HOPV15). It was found that neural-based generally performed better on with FP reaching a state-of-the-art performance test set mean squared error 0.071. proved harder fit, all exhibiting rather poor performance. Contrary dataset, were perform than models. To improve predict PCEs for OPVs, either results correlate experiments or more data at well-controlled conditions likely required.