Modeling and predicting solubility of n-alkanes in water

Accurate knowledge on the phase equilibrium of hydrocarbons and water is needed to design and operate gas, petroleum and petrochemical plants. Usually famous simulation softwares are used to simulate and design industrial plants during the basic design stage. In most famous softwares, well known cubic equations of state (e.g. Peng–Robinson) are employed. These models, in most cases, are reliable for estimating phase behavior of hydrocarbons but they cannot model and predict solubility of hydrocarbons in water accurately. s communication, solubility of n-alkane hydrocarbons (methane, ethane, propane, n-butane, n-pentane and n-hexane) in water has been studied. Equations of state, simple correlations and artificial neural network have been used to model and predict solubility of selected hydrocarbons in water and their performances have been compared. Soave–Redlich–Kwong (SRK), Peng–Robinson, Kabbadi–Danner and Lee–Kesler–Plocker equations of state have been used in this study. Kabbadi–Danner equation of state has shown the best performance among investigated equations, however, none of the tested equations of state can be totally trusted in this regard. Simple correlations, as another option, have been used to model and predict solubility of n-alkanes in water. Simple correlations are easy to apply and reliable during modeling but they have not shown acceptable predictive capabilities especially when no experimental data is available. Artificial neural network has been used as an alternative to other techniques. A new approach, based on application of a simple structural property as a network input, has been used to estimate solubility of light n-alkanes in water. The overall average relative deviation of this method has been found to be 9.34% which is better than other studied techniques. Also, the designed neural network has shown acceptable predictive capability which confirms its superiority over famous equations of state, simple correlation and traditional neural network techniques. Finally, all investigated techniques have been compared, comprehensively.