Screening of ionic liquids for CO2 capture using data analytics techniques

Abstract

Carbon dioxide (CO2) is the most prominent greenhouse gas (GHG) present in the atmosphere, making it the most accountable for global warming. CO2 capture is capable of greatly reducing carbon emissions. The current method of CO2 capture by amine-based solvent has drawbacks, such as high demand for energy and intense corrosion, making it a less reliable method. More attention is given to ionic liquids (ILs) for their negligible vapour pressure, low melting point, and high chemical and thermal stability advantage. This study uses data analytics techniques to develop a predictive model for screening ILs for CO2 capture, moving away from the experimental approach, which is burdensome, costly, and less environmental-friendly. Data on the properties and parameters of ILs are collected from COSMO-RS software. CO2 solubility is the function of collected data and developed into 15 models of three different methods: Support Vector Machine (SVM), Neural Networks (NN), and Gaussian Process Regression (GPR). The use of data analytics in this field is new and can provide valuable insight towards CO2 solubility in ILs. The dataset is distributed randomly at 80/20% for training and testing. Each model is evaluated using R-squared and root mean square error (RMSE). The rational Quadratic GPR model shows the lowest RMSE of 0.0002 for training and testing, with R-squared the closest to one. Rational Quadratic GPR is the best model to be used for screening IL for CO2 capture.