Estimating the solubility of different solutes in supercritical CO2 covering a wide range of operating conditions by using neural network models

Abstract

A new neural network-based model is proposed for solute solubility in supercritical carbon dioxide (CO2). The solubility of fifteen solutes in supercritical carbon dioxide at different temperatures and pressures are estimated. Four neural network models have been tested to investigate the generalize-ability of each network. The results from different semi-empirical models, namely Chrastil, Kumar and Johnston, Bartle, Mendez-Santiago and Teja, the Peng–Robinson and Soave–Redlich–Kwong equation of state models are compared to the one estimated by using the proposed neural network model. The average absolute deviation (AAD) of 5.42% compared to those of density-based and equation of state models, representing the reasonable accuracy of the developed model for estimating the solubility of solutes in supercritical fluid extraction process.