Prediction of acid gas solubility in amine, ionic liquid and amino acid salt solutions using artificial neural network and evaluating with new experimental measurements

Abstract

In this work presented here attempt is prediction of acid gases (carbon dioxide and hydrogen sulfide) loading capacities by employing artificial neural network (ANN) model in 51 single and blended alkanolamine, ionic liquid and amino acid salt solutions as commonly and new industrial absorbents in large domain of operational conditions. Also for evaluating extrapolation capability of ANN, new experimental data on CO2 solubility in aqueous solutions of Potassium Glycinate blended with Piperazine (PZ) and 2-amino-2-methyl-1-propanol (AMP) at different temperatures and pressures are measured. It should be mention that CO2 solubility data for these two solutions are not available in literature. For developing ANN, solution pH, total mass concentration, partial pressure of CO2 and H2S, apparent molecular weight, critical temperature, critical pressure and temperature are assumed as inputs. A band of 2982 experimental data points for CO2 and H2S loading capacities have been collected from literature to create the suggested ANN. The best structure of the suggested network is achieved by employing these literature data points. The network is trained by algorithm of Levenberg–Marquardt back-propagation, consists of 9 and 6 neurons in first and second hidden layers, respectively. For the hidden and output layers, Tan-sigmoid transfer function is utilized. The output results of developed network show that suggested network that is created with solubility data of single and blended alkanolamine, ionic liquid and amino acid salt solutions has capability to predict accurately CO2 and H2S loading capacities in dissimilar commonly and new industrial solutions with Average Relative Deviation (ARD %) equal to 2.7992, Mean Square Error (MSE) value of 3.7468 × 10−5 and correlation coefficient (R2) equal to 0.9984.