Vapor-liquid equilibria modeling using gray-box neural networks as binary interaction parameters predictor

Simulations of vapor-liquid equilibrium (VLE) are widely used given their impact on the scale, design, and extrapolation of different operational units. However, due to a number of factors, it is almost impossible to experimentally study each of the VLE systems. VLE simulations can be developed using representations that are strongly dependent on the nature and interactions of the compounds forming mixtures. A model that helps in predicting these interactions would facilitate simulation processes. A Gray Box Neural Network Model (GNM) was created as Binary Interaction Parameters predictors (BIP), which are estimated using state variables and information from pure components. This information was used to predict VLE behavior in mixtures and ranges not used in the mathematical formulation. The GNM prediction capabilities (including temperature dependency) showed an error level lower than 5% and 20% for mixtures considered and not considered in the training data, respectively.