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

Area of Science:

• Materials Science
• Computational Chemistry
• Renewable Energy

Background:

• Organic photovoltaics (OPVs) offer a promising avenue for renewable energy generation.
• Accurate prediction of power conversion efficiency (PCE) is crucial for OPV development.
• Machine learning (ML) models can potentially accelerate the discovery of high-performance OPVs.

Purpose of the Study:

• To evaluate the performance of various ML models in predicting OPV PCE.
• To compare the efficacy of neural network and baseline models on both computational and experimental datasets.
• To identify key challenges and requirements for improving ML-based PCE prediction in OPVs.

Main Methods:

• Utilized three neural network models: bidirectional long short-term memory (BiLSTM), attentive fingerprints (attentive FP), and simple graph neural networks (simple GNN).
• Employed three baseline models: support vector regression (SVR), random forests (RF), and high-dimensional model representation (HDMR).
• Trained and tested models on the Harvard Clean Energy Project database (CEPDB) and the Harvard Organic Photovoltaic 15 dataset (HOPV15).

Main Results:

• Neural-based models, particularly attentive FP, achieved state-of-the-art performance on the large computational CEPDB dataset (test MSE of 0.071).
• All models exhibited poor performance on the smaller, experimental HOPV15 dataset.
• Baseline models outperformed neural models on the experimental dataset, contrary to findings on the computational data.

Conclusions:

• Predicting OPV PCE accurately requires either improved computational data that correlates well with experimental outcomes or more extensive, well-controlled experimental data.
• The performance gap between computational and experimental datasets suggests limitations in current predictive models for real-world OPV applications.
• Further research should focus on bridging the gap between theoretical predictions and experimental validation for robust OPV material design.