Simulación y predicción por RNA de la composición de Tetrahidrofurano separado de una mezcla Azeotrópica Tetrahidrofurano - agua utilizando DWSIM

Abstract

The objective of this study was to simulate an extractive distillation system to separate the tetrahydrofuran-water mixture and predict the composition of tetrahydrofuran using artificial neural networks. The simulation of the system was done with the DWSIM software, and the prediction model was developed in Matlab. The experimental data collected from the literature was used to build and validate the simulation and the neural network. The prediction model was designed using 100 data as with 4 input variables, the molar fraction of tetrahydrofuran in the inlet, the reflux ratio, and the temperature of the reboiler in the first column; the reflux ratio in the second column; 2 output variables, the molar fraction and mass flow of tetrahydrofuran recovered in the extractive distillation column. In the hidden layer, 400 neurons and the Bayesian regularization algorithm. During the training and testing of the neural network, a correlation coefficient of 0.97225 was obtained; a root means the square error of 5.17 E-03; a p-value> 0.05, which validates the neural network, showing that there is no statistical difference between the predicted values and the real data. The network simulates a process that is fed with 3000 Kg / h of the azeotropic mixture at 320 K; 1.57 Kg / h of ethylene glycol in a distillation system that operates in two columns maintained 1.1 atm. The results indicate that the artificial neural network can be widely used as an efficient simulation tool when predicting the concentration of tetrahydrofuran that is desired to be separated by extractive distillation of an azeotropic mixture of tetrahydrofuran-water. It is recommended to use the prediction model to develop neural networks that allow calculating the amount of the separated components in other azeotropic mixtures by extractive distillation with the appropriate entrainment agent.