Novel Strategy of Machine Learning for Predicting Henry’s Law Constants of CO₂ in Ionic Liquids

Abstract

The Henry’s law constant (HLC) of CO2 is an important parameter to characterize its absorption by solvents. However, the HLC data is incomplete in the NIST ionic liquid (IL) Thermo database. In this work, molecular dynamics (MD) simulations were used to accurately calculate the HLC of CO2 in ILs. Then, machine learning (ML) and artificial neural networks were combined to learn from limited data to rapidly expand the database. Three rapid HLC prediction models were established by cross-validation and grid search based on rigorous data sets of IL’s structure, temperature, and HLC. The determination coefficient, mean absolute error, and mean square error of the optimal multilayer perceptron model were 0.9817, 0.3023, and 0.2104, respectively. Compared with the reported models, the prediction model established in this work has better versatility and higher prediction accuracy. The HLC matrix of CO2 in 306 ILs was completed, which proves the great potential and significance of the MD–ML method in the expansion of green solvent database. Finally, the structure–activity relationship of CO2 absorption by the binary IL mixtures was studied.

Physics-Based Machine Learning Models Predict Carbon Dioxide Solubility in Chemically Reactive Deep Eutectic Solvents

“…For example, Kartal and Özveren (178 data points), Kardani et al (81 data points), and Gao et al , (125–150 data points) developed ML/DL models with 81–178 data points for the prediction of lignocellulosic biomass composition and conversion of biomass during the hydrothermal carbonization. Recently, Zhang et al developed ML models with 132 experimental data points for the prediction of the Henry’s law constant for CO 2 solubility in ionic liquids (ILs). However, recognizing the fact that our data set was small, we took steps to avoid overfitting in training the ML model.…”