Enhanced carbon dioxide adsorption using lignin-derived and nitrogen-doped porous carbons: A machine learning approaches, RSM and isotherm modeling

Abstract

The experimental data obtained from the CO2 adsorption experiments conducted by Saha et al. (2017) were utilized. The Langmuir, Dubinin-Radushkevich (D-R), Hill, and Freundlich models were fitted and compared to determine the best-fitting isotherm model. The models were also used to predict the adsorption behavior of lignin under varying pressure and temperature conditions. Five regression models, including Random Forest (RF), Gradient Boosting (GB), XGBoost (XGB), Extra Trees (ET), and Adaptive Neuro-Fuzzy Inference System (ANFIS), were compared using statistical analysis. The Random Forest model showed the highest prediction accuracy with an R2 value of 0.9996 and a low MSE value of 0.00021991. The optimal hyperparameter settings for the Random Forest model were found to be 50 for n_estimators, 2 for the minimum number of samples for node split, and 1 for min_samples_leaf. Response surface analysis and ANOVA revealed that pressure had the greatest impact on CO2 adsorption effectiveness. The optimal parameter combinations identified through response surface analysis were 2280 m2/g for surface area (BET), 1.1 cm3/g for total pore volume, 293 K for temperature, and 380.0 torr for pressure.

Keywords

• ANNs
• Random forest
• RSM
• Isotherm modeling
• CO2 capture
• Lignin adsorbent