ANN-based estimation of pure-component parameters of PC-SAFT equation of state using quantum chemical data

The perturbed chain-statistical associating fluid theory equation of state (PC-SAFT EoS) is a physical property estimation tool that can be used to calculate a wide range of substance types, temperatures, and pressures. To perform calculations using the PC-SAFT EoS, substance-specific pure-component parameters are required, which are generally determined from liquid density and saturated vapor pressure. Few studies have reported on these parameters, and methods that can obtain pure-component parameters without relying on measured physical properties remain elusive. In this study, an artificial neural network (ANN) is introduced to predict the pure-component parameters of the PC-SAFT EoS. The molecular information estimated from a Gaussian software was used as the input. In addition, we optimized the structure of the ANN by varying the transfer function, number of neurons, and number of hidden layers. The optimized ANN comprises a hard sigmoid transfer function composed of two hidden layers, with 20 and 10 neurons in the first and second layer, respectively. This model can determine the pure-component parameters of the PC-SAFT EoS for a wide range of substance types. Furthermore, SHapley Additive exPlanations analysis on the optimized ANN demonstrates that the contributions of the polarizability and dipole moment are large. However, the feature values related to the shape of the substance are lacking. These results contribute to expanding the range of applications for property estimation using EoS.