Modeling of liquid-vapor equilibrium at constant pressure for ethanol-water systems, using artificial neural network

Abstract

The liquid-vapor equilibrium for the binary mixture ethanol-water at constant pressure, using low, moderate and high pressures from the literature was modeled. Artificial neuronal networks with multilayer architecture perceptron and back propagation learning algorithm, implemented in, KNIME 3.1.1 and Matlab 2013, were used. To determine the reliability of the data, the presence of non-systematic errors was examined finding fails in six experimental points. The presence of systematic errors was proven using thermodynamic consistency tests. Herington’s areas test and Wisniak’s point to point test were applied. Experimental data were reliable, except two points at low pressures. Topologies were obtained from two neurons in the hidden layer up to 10. For the selection, Friedman and Wilcoxon non parametric tests were applied. The selected topology was the one obtained in Matlab 2013 with eight neurons in the hidden layer with a mean of the mean square error of 0,005 4, a deviation of the square means error of 0,000 6 and a correlation coefficient of 0,972 9. With this result, a single model can be used to predict the liquid- vapor equilibrium from 6.6 kPa to 1520 kPa.