Carbon capture using ionic liquids: An explicit data driven model for carbon (IV) Oxide solubility estimation

Abstract

In this study, carbon dioxide (CO2) solubility in 20 different ionic liquids (ILs) belonging to various chemical families was estimated across a broad range of pressures (0.0098–72.24 MPa) and temperatures (292–450 K) utilizing an artificial neural network. The developed model exhibited high accuracy in predicting experimental values, yielding an R2 value of 0.992, a root mean squared error (RMSE) of 0.04, and a mean square error (MSE) of 0.00173. Additionally, the model was presented in an explicit and transparent manner that facilitates its deployment in software applications, which is not commonly seen in other machine learning models in this field. Moreover, the model was made interpretable by utilizing the connection weights algorithm to demonstrate the relative contribution of each model input and the direction in which the input influences the output prediction. In this regard, it was observed that parameters such as pressure, critical pressure, acentric factor, and ionic liquid density enhance CO2 solubility as their values increase, while parameters like temperature and critical temperature exhibit the opposite effect. Another distinguishing aspect of this study is the evaluation of the computational cost of the formulated model and a proposal for its implementation in practical settings. Given that CO2 solubility plays a crucial role in solvent selection for CO2 capture, this research holds relevance for industry professionals in the oil and gas sector who aim to remove CO2 from natural gas streams with the objective of enhancing the calorific value of the product and reducing the damage caused by corrosion to transmission pipelines.