A novel molecular structure-based model for prediction of CO2 equilibrium absorption in blended imidazolium-based ionic liquids

Abstract

• A comprehensive dataset including 4397 data points of CO 2 equilibrium solubility in imidazolium-based ionic liquids was acquired. • Various blended of imidazolium-based ILs were characterized based on their molecular frameworks. • Three machine learning predictive models were proposed to predict CO 2 solubility. • The simple but effective FFNN model gives reliable predictions, as compared with RBFNN and SVM algorithms. The present study highlights a comprehensive database including 4397 data points of CO 2 equilibrium solubility measurements in the 43 different imidazolium-based ionic liquids (ILs) over a broad range of pressures and absorption temperatures. The relation between the equilibrium CO 2 solubility and the molecular structure of the imidazolium-based ILs mixed with different kinds of solvents, including Diethanolamine (DEA), Methyl diethanolamine (MDEA), Diisopropylamine (DIPA), Amionomethyl propanol (AMP), and the equilibrium absorption pressure and the temperature has been accurately correlated. According to this database, a novel chemoinformatics-based descriptor model with a large number of 26 input data of structural information of all involved cation and anions and experimental conditions has been extracted. Three different machine learning methods, namely feed-forward neural network (FFNN), radial-based function neural network (RBFNN), and support vector machine (SVM), are employed to develop the derived descriptor-based model. The results of the three machine learning methods demonstrate that the prediction performance of the suggested models is quite reliable. Comparing the results indicate that the FFNN with corresponding values of RMSE = 0.071, R 2 = 0.952, and MAPE = 0.544 is the best paradigm to predict the CO 2 equilibrium solubility in imidazolium-based ILs.