Work

Machine learning-driven design of high-performance activated carbons for enhanced CO2 capture

Year: 2026

Type: article

Abstract: Rising atmospheric CO 2 levels demand efficient carbon capture technologies. Activated carbons (ACs), known for their high surface area and adjustable porosity, offer a promising pathway for CO 2 adsorption. However, optimizing their structural and chemical properties often requires time-intensive experimental trials. This study addresses this challenge by leveraging machine learning (ML) to predict CO 2 adsorption capacity and guide the design of high-performance ACs. A comprehensive dataset, including pore volume, mean pore diameter, adsorption temperature and pressure, and BET surface area, was utilized to train four ML models using K-Fold approach with five folds, with CO 2 uptake as the target parameter. Among the tested models, Gradient Boosting achieved the best performance (R 2 = 0.9989) with high predictive precision. Experimental validation using asphaltene-derived AC confirmed excellent agreement between predicted and measured CO 2 uptake values. Adsorption experiments at 1 bar and 298 K measured CO 2 uptake at 4.72 mmol/g, closely aligning with the Gradient Boosting model's predicted value of 4.81 mmol/g. The study highlights the model’s capability to simulate CO 2 -AC interactions and its potential to reduce the need for extensive experimental trials. 3D plots generated by the Gradient Boosting model provided valuable insights into optimizing textural and adsorption properties to maximize CO 2 uptake. This integrated approach accelerates the discovery and design of high-performance adsorbents, offering a roadmap for sustainable carbon capture solutions. • Machine learning used to predict CO 2 uptake of activated carbons. • Gradient Boosting model achieved high accuracy (R 2 = 0.9989). • Asphaltene-derived activated carbon synthesized and characterized. • Experimental CO 2 uptake (4.72 mmol/g) matched model prediction. • 3D ML visualizations guided optimization of AC textural properties.

Source: Journal of CO2 Utilization

DOI: 10.1016/j.jcou.2026.103336