Machine learning exploration of the critical factors for CO₂ adsorption capacity on porous carbon materials at different pressures

The growing environmental issues caused by CO₂ emission accelerate the development of carbon capture and storage (CCS), especially bio-energy CCS as an environment-friendly and sustainable technique to capture CO₂ using porous carbon materials (PCMs) produced from various biomass wastes. This study developed quantitative structure-property relationship models based on 6244 CO₂ adsorption datasets of 155 PCMs to predict the CO₂ adsorption capacity and analyze the relative significance of physicochemical properties. The results suggested that random forest (RF) models showed good accuracy and predictive performance based on physicochemical parameters of PCMs and adsorption conditions with the test dataset (R² > 0.9). In general, textural properties were more crucial than chemical compositions of porous carbons to the change of CO₂ adsorption capacity. At a low pressure (0.1 bar), the volumes of mesopore and micropore played an important role according to the RF analysis, but had a negative correlation with CO₂ adsorption capacity based on the Pearson correlation coefficient (PCC) analysis. The relative importance of ultra-micropore increased along with the increase of pressure. The PCC value between ultra-micropore volume and CO₂ uptake amount was up to 0.715 (p < 0.01) at 1 bar and 0 °C. The influence of chemical compositions was complex. The N content was confirmed to positively correlate to the CO₂ adsorption capacity but its contribution was much lower than that of ultra-micropores. This study provided a new approach for fostering the rational design of porous carbons for CO₂ capture via statistical analysis and machine learning method, which facilitated adsorbents screening for the cleaner production.