Machine learning approaches for prediction of phase equilibria in poly (ethylene glycol) + sodium phosphate aqueous two-phase systems

Document Type: Research Article


1 Faculty of Chemical Engineering, Babol University of Technology, PO Box, 484, Babol, Iran

2 Chemical Engineering Department, Faculty of Engineering, Shomal University, PO Box 731,Amol, Mazandaran , Iran

3 Faculty of Chemical Engineering and Environmental Protection "Cristofor Simionescu", "Gh. Asachi" Technical University, Bld Mangeron 73, 700050, Iasi, Romania


Liquid-liquid equilibrium (LLE) data for aqueous two-phase systems (ATPSs) containing sodium di hydrogen phosphate and poly (ethylene glycols) (PEG) (1500, 2000, 3000, 4000 or 8000 g/mol) have been determined experimentally at different temperatures (298.15 and 308.15 K). The effects of molecular weights of PEG and temperature on the binodal curve were studied. It was found that an increase in PEG molecular weight and temperature led to the binodal curve displacement towards the origin and increasing two-phase region. In order to generate predictions and to optimize the system, Artificial Neural Networks and Support Vector Machines, simple, or in combination with Differential Evolution were employed. The simulation results indicated that the combination that best fits to the experimental data is Support Vector Machines optimized with Differential Evolution, with an average absolute error of 5.97%.