Neural network modeling of ternary solubilities of 2-naphthol in supercritical CO2: A comparative study

A back-propagation multilayer artificial neural network (ANN) has been constructed for prediction of the solubility of 2-naphthol in ternary systems. Different networks were trained and tested with different network parameters using training and testing data sets. Using a validating data set the network having the highest regression coefficient and the lowest mean square error was selected. The comparison with the Peng–Robinson (PR) equation of state (EoS) was investigated. The binary interaction parameters were calculated by fitting the solubility data of the constituent binary systems. However, the predicted average relative deviation (ARD) and the root mean squared error (RMSD) for the trained ANNs data points were 3.15 and 0.81%, respectively. For the PR EoS, the overall average predicted ARD and RMSD for all systems were as high as 11.82 and 8.44%, respectively. The present work demonstrates that the ANN method is a powerful approach with better accuracy compared with the classical thermodynamic methods.