Prediction of H2S solubility in aqueous solutions using a feed forward multilayer perceptron neural network model

Favorable properties of aqueous solutions are improved with addition of different materials for separation of hydrogen sulfide (H2S). Also, equilibrium data and available equations for solubility estimation of this gas are only valid for specific solutions and limited ranges of temperature and pressure. In this regard, a model based on artificial neural network (ANN) is proposed and developed with mixtures containing monoethanolamine (MEA), diethanolamine (DEA), methyldiethanolamine (MDEA), 2-amino-2-methyl-1-propanol (AMP), triethylene glycol (TEG), piperazine (PZ), 2-Piperidineethanol (PDEA), ammonium nitrate (NH4NO3), sodium nitrate (NaNO3), n-methylpyrrolidone (NMP), sodium hydroxide (NaOH), 1-Amino-2-propanol (MIPA), and ionic liquids containing 1-hexyl-3-methylilmidazolium hexafluorophosphate ([hmim][PF6]), 1-hexyl-3-methylimidazolium tetrafluoroborate ([hmim][BF4]), 1-hexyl-3-methylimidazolium bis(trifluoromethanesulfonyl)imide ([hmim][Tf2N]), 1-(2-hydroxyethyl)-3-methylimidazolium hexafluorophosphate ([HOemim][PF6]), 1-(2-hydroxyethyl)-3-methylimidazolium trifluoromethanesulfonate ([HOemim][OTf]), 1-(2-hydroxyethyl)-3-methylimidazolium bis-(trifluoromethyl) sulfonylimide ([HOemim][Tf2N]), 1-ethyl-3-methylimidazolium ethylsulfate ([emim][EtSO4]), 1-ethyl-3-methylimidazolium tris(pentafluoroethyl)trifluorophosphate ([C2mim][eFAP]) and 1-octyl-3-methylimidazolium hexafluorophosphate ([C8mim][PF6]) to predict H2S solubility in mixed aqueous solution (especially in binary and ternary mixtures) over wide ranges of temperature (298 to 434.5 K), pressure (13 to 9319 kPa), overall mass concentration (3.82 to 100 percent) and mixture’s apparent molecular weight (18.39 to 556.17 g/mol). Accuracy of performance of this network was evaluated by regression analysis on calculated and experimental data, which had not been used in network training. The optimal neural network that was trained by the Levenberg-Marquardt back-propagation algorithm and the Gauss-Newton method with combination of a Bayesian regularization technique contains two hidden layers, having 8 and 4 neurons, respectively. Tan-sigmoid function was used as transfer function of hidden and output layers of this network.

Keywords:

H2S solubility , Multi-layer neural network , H2S loading , blended aqueous solutions