Application of artificial neural networks for estimation of solubility of acid gases (H2S and CO2) in 32 commonly ionic liquid and amine solutions

Abstract

In the present study, the potential use of a model based on an artificial neural network (ANN) was investigated to predict the solubility of acid gases (H2S and CO2) in 32 commonly single and mixed amine and ionic liquid (IL) solutions over wide ranges of operating conditions. Temperature, partial pressure of acid gas (H2S or CO2), overall mass concentration, apparent molecular weight, critical temperature and critical pressure of solution were chosen as input variables of the proposed network. A collection of 733 experimental data points for H2S solubility (including train, test and validation data points) have been gathered from the literature to develop the network. The best parameters of the developed ANN containing the number of neurons, numbers of hidden layer and transfer function were acquired by using these data points. To evaluate the network accuracy, regression analysis with a data set including 169 data points for H2S solubility which were not considered in the training, testing and validation stages was applied. Furthermore, the extrapolation capability of the network was investigated by an extra data set (114 data points for CO2 solubility). The optimized network was trained by the Levenberg–Marquardt back-propagation algorithm with two hidden layers including 8 and 4 neurons and Tan-sigmoid transfer function for the hidden and output layers. The model results show that developed ANN model has ability to estimate accurately the solubility of acid gases in different solutions with Mean Relative Error (MRE) value of 3.104 and correlation coefficient (R2) of 0.997.