Development of deep learning framework to predict physicochemical properties for Ionic liquids

Abstract

In this paper, a deep learning-based group contribution approach has been developed to identify the optimum structure for ionic liquids (ILs) and to maximize the CO2 absorption capacity. The suggested methodology demonstrates the steps required to build two deep-learning-based group contribution models, DNN-GC and ANN-GC, separately for IL viscosity and CO2 solubility in the programming language 'Python' by using the two widely used modules, 'Scikit-Learn' and 'TensorFlow'. The CO2 solubility of IL candidates may not result in the best CO2 absorption process performance because many other properties of IL, such as viscosity, may affect process performance. To resolve this issue, the two developed models were merged to increase the model accuracy and to predict CO2 solubility and IL viscosity at the same operating pressure and temperature. Based on the latter steps, an illustrative post-combustion CO2 capture case study was performed to assess the applicability of the model to select the best IL candidate. With this method, a trade-off between CO2 solubility and viscosity of ILs can be explicitly studied.