Chemical formula input relied intelligent identification of an inorganic perovskite for solar thermochemical hydrogen production

Abstract

An efficient prediction procedure is designed for the quick screening of solar thermochemical (STC) perovskites for H₂ production. The core classifier is based on the random forest method, and the input information is easily accessible from chemical formula and periodic table. For ABO₃-type perovskites, the prediction accuracy is high even within a small number of training samples. The prominent feature of the program is the fast and accurate identification of doped perovskites, which is almost impossible when studied individually by experimental and density functional theory methods. By using more than 380 ABO₃ as train samples, a stable accuracy of more than 90% is obtained, which is much larger than the results of probabilistic neutral network and exact radial basis network methods. All the results demonstrated the effectiveness of prediction procedure and provided a valuable reference for high throughput exploration in fields other than STC H₂ production.