Bandgap prediction of ABX3-type perovskites using Broad Learning System

Perovskite materials have attracted significant attention of scholars at home and abroad due to their excellent photoelectric properties, with the bandgap being a crucial factor that directly determines the performance of perovskite solar cells. However, density functional theory (DFT) calculations of bandgap are time-consuming and costly. To address this issue, Broad learning System (BLS), a data-driven machine learning algorithm, is proposed to predict the bandgap of hybrid organic-inorganic halide perovskites using a dataset of 227 ABX3-type perovskites with experimental bandgap values, and the stoichiometric ratio of positions A, B, X of perovskite as input features. The results demonstrate that BLS algorithm provides superior predictions compared to Support Vector Regression (SVR), Multi-Layer Perceptron (MLP) and Random Forest (RF). BLS algorithm has a lower root mean square error (RMSE), and a higher correlation coefficient (R2), with RMSE being 0.0855 eV and R2 reaching 92.24 %. This work indicates that BLS can be a powerful and potential tool for predicting the bandgap of perovskites.