Prediction of the enhanced performance of Cu(I)-modified porous materials towards CO adsorption by using tree-based machine learning models

Abstract

Listen

Translate article

Cu(I)-modified adsorbents have exhibited broad application prospects in the field of CO adsorption by virtue of their high adsorption selectivity and capacity. Machine learning (ML) is expected to explain the structure-activity relationship between Cu(I)-modified adsorbents and CO adsorption, giving valuable suggestions for designing adsorbents with excellent adsorption performance obviating high cost of time and labor via experiments. Herein, a series of tree-based ML models on the Python platform were constructed to precisely predict the enhanced performance of Cu(I)-modified porous materials towards CO adsorption. A dataset was built covering the CO adsorption performance of various porous materials, and principal component analysis was proposed to classify support types, with excellent CO adsorption selectivity as a prerequisite. The extreme gradient boosting (XGB) model exhibited a satisfactory accuracy (R2 > 0.97), achieving accurate predictions of CO adsorption capacity. SHapley Additive exPlanation and Partial Dependence Plots demonstrated that there is a trade-off between the loading amount of Cu(I) and the surface textural properties, which would remarkably affect CO adsorption performance. According to feature importance analysis, the preferred loading amounts of Cu(I) precursor were determined to be 15–23 wt% relative to the weight of the support. After loading Cu(I), the optimized adsorbent possessed 52–70 % specific surface area as that of the support. Meanwhile, the micro-porous volume exceeding 0.43 cm3/g was desirable to achieve excellent performance towards CO adsorption. Under low partial pressure (0–100 kPa) of CO, Linde type A molecular sieves were found to be the most suitable support of Cu(I)-modified adsorbents. This study provided insights for designing Cu(I)-modified adsorbents to guide the synthesis towards enhanced CO adsorption performance.