Computerized Prediction of Perovskite Performance Using Deep Learning

Abstract

Perovskites have exceptional physical and chemical features in different fields. Perovskites have an ABO3 formula with similar sizes of A-site and B-site cations. This research explores the challenges of developing new perovskite solar cells with high performance. Therefore, this article proposes a deep learning model for the prediction of perovskites performance measures. The measures are: energy conversion performance, ABO3 stability, ion volume, and induced oxygen vacancy dimension. These performance measures are very crucial electrochemical reactions in energy conversion in fuel crystals. The challenges in any deep learning model are the lack of the presence of sufficient data and training time. Consequently, in this research, we propose a transfer learning perovskites model. Perovskite performance detection is critical to offer operative energy resources. In the proposed model, the constructed detection model uses a perovskites feature set. The transfer learning model utilizes other materials with large-sized datasets to predict the four performance measures with high accuracy. The output of the transfer learning is then utilized for the proposed deep learning model to predict perovskites performance measures with a small-sized dataset. A dataset of 8500 perovskite samples is utilized in the research. The results prove that a deep learning F2-Score with transfer learning attains high accuracy of 98.95%, recall of 96.91% and F2-score of 97.05%.