Prediction of Carbon Dioxide Reduction Catalyst Using Machine Learning with a Few-Feature Model: WLEDZ
Miaojuan Xing, Yunjiang Zhang, Shuyuan Li, Hong He, Shaorui Sun
Abstract
Cu-based alloy catalysts are widely used in the field of carbon dioxide reduction reaction (CO2RR), due to the good selectivity and low overpotential. In order to achieve efficient exploration of alloy catalysts for CO2RR, a machine learning (ML) model, based on a gradient boosting regression (GBR) algorithm, is developed. By implementing a rigorous feature selection process, the dimensionality of feature space is reduced from thirteen to five, including work function (W), local electronegativity (Loc_EN), electronegativity (EN), interplanar spacing (D), and atomic number (Z), which is referred to as the WLEDZ model. The few-feature model has a high performance as that with many features, and the ML model successfully and rapidly predicts the adsorption energy of the key intermediates (HCOO, CO, and COOH) in the CO2RR process. In addition, eight Cu-based bimetallic catalysts are predicted with highly promising alternatives. This demonstrates that the WLEDZ few-feature ML model can screen highly promising bimetallic alloy for CO2RR and can also be used for the design of other types of catalysts.