Use of Neural Networks for the Predictive Development of Thermodynamic Properties of Ionic Liquids


Moretó Bravo, Núria


The objective of increasing costs and environmental problems related to CO2 capture is to study new alternatives to the classic amine absorption processes. In this sense, ionic liquids (IL) seem to be a promising alternative, but the selection of an appropriate cation-anion combination with solvents of suitable properties is a heavy task due to the large number of possible combinations. The use of theoretical tools with prediction capacity is necessary to find solutions with a limited amount of experimental work. Despite extensive research on the study of these compounds using Equations of State (EOS), a new approach based on Multiple Linear Regression (MLR), Random Forest Regression (RFR), and Artificial Neural Network (ANN) has been explored in this project for the prediction of CO2 solubility in five ILS based on the imidazolium cation. The results show that ANN has the best predictive capacity. A procedure has also been implemented to adjust the hyperparameters of the ANN, capable of predicting the solubility of eight gases (CO2, SO2, H2S, N2O, CO, N2, O2 and H2) in the five ILS. A comparison on a reduced data set between the global ANN model and the Soft-SAFT, a body type, has shown that the latter offers better solubility predictions (the MSEs have been respectively 0.2174 and 0.0010).



Llovell Ferret, Fèlix
Fernández Esmerats, Joan


IQS SE - Undergraduate Program in Chemical Engineering