Accurate prediction of carbon dioxide capture by deep eutectic solvents using quantum chemistry and a neural network

Abstract

Carbon dioxide (CO₂) emissions from fossil fuel combustion are a significant source of greenhouse gas, contributing in a major way to global warming and climate change. Carbon dioxide capture and sequestration is gaining much attention as a potential method for controlling these greenhouse gas emissions. Among the environmentally friendly solvents, deep eutectic solvents (DESs) have demonstrated the potential capability for carbon capture. To establish a theoretical framework for DES activity, thermodynamics modeling and solubility predictions are significant factors to anticipate and understand the system behavior. Here, we combine the COSMO-RS model with machine learning techniques to predict the solubility of CO₂ in various deep eutectic solvents. A comprehensive data set was established comprising 1973 CO₂ solubility data points in 132 different DESs at a variety of temperatures, pressures, and DES molar ratios. This data set was then utilized for the further verification and development of the COSMO-RS model. The CO₂ solubility (ln(x_CO2)) in DESs calculated with the COSMO-RS model differs significantly from the experiment with an average absolute relative deviation (AARD) of 23.4%. A multilinear regression model was developed using the COSMO-RS predicted solubility and a temperature-pressure dependent parameter, which improved the AARD to 12%. Finally, a machine learning model using COSMO-RS-derived features was developed based on an artificial neural network algorithm. The results are in excellent agreement with the experimental CO₂ solubilities, with an AARD of only 2.72%. The ML model will be a potentially useful tool for the design and selection of DESs for CO₂ capture and utilization.