Toward artificial intelligence-based modeling of vapor liquid equilibria of carbon dioxide and refrigerant binary systems

Abstract

The objective of this study is to design and validate a highly accurate approach based on an artificial neural network (ANN) to predict both bubble and dew point pressures of various CO2?refrigerant binary systems in the temperature range of 263.15?367.30 K and pressure of 0.18?9.09 MPa. 503 Experimental vapour?liquid equilibria (VLE) data of nine different CO2-refrigerant binary mixtures were used for preparation, validation and testing of ANN model. The developed ANN model correlates bubble and dew point pressure to reduced temperature, critical pressure, acentric factor of refrigerant, and distibution of CO2 between the vapour and liquid phases. Trial and error procedure reveals that a three-layer neural network with fourteen neurons in the hidden layer is able to predict the pressure with mean square error (MSE), average absolute relative deviation (AARD), root mean square error (RMSE), and correlation coefficient (R2) of 0.0133, 2.79 %, 0.1153 and 0.99836, respectively. The results confirmed that the ANN model can accurately apply for predicting the VLE data of different binary CO2?refrigerant systems.