Development of machine learning models to enhance element-doped g-C3N4 photocatalyst for hydrogen production through splitting water

Abstract

Elemental doping has been widely adopted to enhance the photoactivity of graphitic carbon nitride (g-C3N4). Correlating photocatalytic performance with experimental conditions could improve upon the current trial-and-error paradigm, but it remains a formidable challenge. In this study, we have developed machine learning (ML) models to link experimental parameters with hydrogen (H2) production rate over element-doped graphitic carbon nitride (D-g-C3N4). Material synthesis parameters, material properties, and H2 production conditions are fed to the ML models, and the H2 production rate is derived as the output. The trained ML models are effective in predicting the H2 production rate using experimental data, as demonstrated by a satisfactory correlation coefficient for the test data. Sensitivity analysis is performed on input features to elucidate the ambiguous relationship between H2 production rate and experimental conditions. The ML model can not only identify important features that are well-recognized and widely investigated in the literature, which supports the efficacy of the developed models but also reveals insights on less explored parameters that might also demonstrate significant impacts on photocatalytic performance. The method described in the present study provides valuable insights for the design of elemental doping strategies for g-C3N4 to improve the H2 production rate without conducting time-consuming and expensive experiments. Our models may be used to revolutionize future catalyst design.