Carbon capture via aqueous ionic liquids intelligent modelling

Abstract

With conventional thermodynamic models, it is challenging to estimate the solubility of a gas in the presence of impurities such as water (H2O). Intelligent models can be utilised for this goal in a computationally efficient manner. In this paper, the carbon dioxide (CO2) solubility in ionic liquids (ILs) containing water is predicted using three intelligence models: artificial neural network (ANN), support vector machines (SVM), and least square support vector machine (LSSVM). The shuffled complex evolution (SCE) is used to optimise the intelligent models SVM and LSSVM hyperparameters (σ2 and γ), whereas trial and error are used to determine the optimum numbers of neurons and layers for the ANN. To identify the most efficient model, the capabilities of applied intelligent models for determining solubility were compared. The findings show agreement between the experimental values and model estimations. Given that the coefficient-of-determination (R2) and root-mean-squared-error (RMSE) were found to be, respectively, 0.9965 and 0.0104 for the test data points, ANN is shown to be moderately more accurate than SVMs or LSSVM at predicting solubility. It can also be inferred that from a statistical point of view, when fed with parameters such as R2, RMSE, standard deviation (STD), and average-absolute-percentage-deviation (AARD), the ANN model demonstrated superior precision in predicting gas solubilities compared to the SVM and LSSVM models.