Phase equilibria modeling of polystyrene/solvent mixtures using an artificial neural network and cubic equations of state

Vapor-liquid equilibria (VLE) of polymer/solvent solutions is a topic of importance because of several areas of applications, including the designing of process equipment. Theoretical and thermodynamic models are reported in the literature for the estimation of VLE. However, up until now, the simultaneous representation of VLE and pressure-volume-temperature data is not satisfactory enough with respect to experimental accuracies. New models are therefore highly required. In the present study, a hybrid model including artificial neural networks (ANN) and genetic algorithm (GA) were applied to estimating the VLE data of seven binary polystyrene (PS)/solvents. The ranges of variables used were 283.15-343.15 K and 0.105-7.46 MPa. The VLE data of these systems were taken from the literature. The net was trained, validated and tested with randomly 65% (108 data points), 10% (17 data points) and the 25% (42 data points), respectively. The mean deviations from the experimental data were determined for the model. The ability of the proposed model was compared with cubic equations of state (CEOS). It was observed that the data found by the ANN model was in excellent agreement with the experimental data, while the CEOS models showed more deviations, particularly at low pressures. In fact, the ANN model can be treated as a powerful technique for VLE data prediction in a fast and reliable way compared with the conventional thermodynamic models.