Toward reliable prediction of CO2 uptake capacity of metal–organic frameworks (MOFs): implementation of white-box machine learning

Abstract

The burning of fossil fuels is the major cause of the surge in atmospheric CO₂ concentration. The unique properties of Metal–organic frameworks (MOFs) have made them a highly promising and efficient class of materials for gas adsorption projects. In this piece of research, white-box machine learning algorithms, including gene expression programming (GEP), group method of data handling (GMDH), and genetic programming (GP), are implemented to generate reliable and efficient explicit correlations for estimating CO₂ uptake capacity of MOFs based on the most extensive databank gathered up-to-date containing 6530 data points from 88 different MOFs. The CO₂ uptake capacity is considered a strong function of pressure, temperature, surface area, and pore volume. The results indicated that the GMDH correlation could provide more reliable results by showing total root mean square error (RMSE) and correlation coefficient (R²) of 2.77 mmol/g and 0.8496, respectively. Also, the trend analysis reflected that this correlation could precisely detect the physical trend of CO₂ uptake capacity with pressure variations. Moreover, the sensitivity analysis showed the high impact of pressure on the estimated CO₂ uptake capacity values. Based on the sensitivity analysis of the GMDH correlation’s estimations, it can be expected that the CO₂ adsorption capacity of MOFs increases by raising MOFs’ surface area and pore volume and designing the adsorption process at elevated pressures and lower temperatures. The proposed correlation can be simply employed to estimate MOFs’ CO₂ uptake capacity with an acceptable level of confidence using a simple calculator.