Vapor-liquid phase equilibria behavior prediction of binary mixtures using machine learning

Abstract

Basic thermodynamic data plays an important role in chemical applications. However, the traditional acquisition of thermodynamic data through experiments is laborious. Thermodynamic data prediction is considered as an alternative to the experiments, especially when qualitative analysis is needed prior to experimental studies. In this work, we report a successful machine-learning based approach to predict the fundamental thermodynamics characteristics of vapor–liquid equilibrium (VLE). A new dataset of the VLE experimental data of 210 binary mixtures with screened descriptors was constructed. The obtained results show that the VLE characteristics of the target system can be fully revealed by machine learning methods and random forest has more excellent predictive ability on the VLE behavior than the neural network. This work provides a new approach to the prediction of VLE data and useful information for the mechanistic study on the VLE phenomenon.