Modeling and Optimizing N/O-Enriched Bio-Derived Adsorbents for CO2 Capture: Machine Learning and DFT Calculation Approaches

The CO2 emission issue has triggered the promotion of carbon capture and storage (CCS), particularly bio-route CCS as a sustainable procedure to capture CO2 using biomass-based activated carbon (BAC). The well-known multi-nitrogen functional groups and microstructure features of N-doped BAC adsorbents can synergistically promote CO2 physisorption. Here, machine learning (ML) modeling was applied to the various physicochemical features of N-doped BAC as a challenge to figure out the unrevealed mechanism of CO2 capture. A radial basis function neural network (RBF-NN) was employed to estimate the in operando efficiency of microstructural and N-functionality groups at six conditions of pressures ranging from 0.15 to 1 bar at room and cryogenic temperatures. A diverse training algorithm was applied, in which trainbr illustrated the lowest mean absolute percent error (MAPE) of <3.5%. RBF-NN estimates the CO2 capture with an R2 range of 0.97–0.99 of BACs as solid adsorbents. Also, the generalization assessment of RBF-NN observed errors, tolerating 0.5–6% of MAPE in 50–80% of total data sets. An alternative survey sensitivity analysis discloses the importance of multiple features such as specific surface area (SSA), micropore volume (%Vmic), average pore diameter (AVD), and nitrogen content (N%), oxidized-N, and graphitic-N as nitrogen functional groups. A genetic algorithm (GA) optimized the physiochemical properties of N-doped ACs. It proposed the optimal CO2 capture with a value of 9.2 mmol g–1 at 1 bar and 273 K. The GA coupled with density functional theory (DFT) to optimize the geometries of exemplified BACs and adsorption energies with CO2 molecules.