Insights into CO2 capture in porous carbons from machine learning, experiments and molecular simulation

Abstract

Porous carbon is one of the important CO2 adsorbents being developed at present. However, interpreting the potential mechanism of CO2 adsorption by porous carbon is still challenging due to their various functional groups, different structural characteristics and different adsorption conditions (temperature and pressure) during CO2 adsorption. Here, this study firstly applied machine learning to study the effects of pore structure, chemical properties, and adsorption conditions on CO2 adsorption performance based on 1594 CO2 adsorption datasets, and to predict CO2 adsorption capacity. The results show that the R2 of the random forest (RF) model is above 0.97 on the training and test data, which has good prediction performance. According to RF analysis results, the nitrogen groups of porous carbon have the greatest impact on CO2 capture at 0–0.15 bar, while ultra-micropores have the greatest impact on CO2 capture at 0.15–1 bar. Subsequently, we prepared three kinds of porous carbons with different pore structures and functional groups, and carried out CO2 adsorption isotherm tests. The results were consistent with the results of machine learning. However, the above results hardly reveal the effect of functional group type and pore size on CO2 capture. Finally, the relative importance of pore size and functional group on CO2 adsorption under different pressures was calculated by molecular simulation, and the mechanism of CO2 adsorption by a single pore size and functional group species was revealed. The results based on the aforementioned machine learning, experimental data and molecular simulation are of great significance for predicting gas adsorption and guiding the development of the carbon-based adsorbents.