Simulación de una planta industrial para la producción de Dimetil Éter por deshidratación de metanol mediante el uso de DWSIM

Abstract

The objective of this study was to develop an artificial neural network capable of predicting the production of dimethyl ether by dehydration of methanol. For the design of the network, the process was modelled in DWSIM using operating parameters described by Dimer and Luyben. 115 data collected by simulation were used to train and validate the network. The network design was carried out in Matlab with six input variables, 465 neurons in the hidden layer, two output variables and the Levenberg-Marquardt, Regularization-Bayesian, and Scaled-Conjugate-Gradient training algorithms. The temperature and mole fraction of methanol in feed, conversion in the reactor, reflux ratio and reboiler temperature in the first column and the reflux ratio in the second column have been selected as input variables; while the molar flow and the molar fraction of dimethyl ether in distillate from the first column as output variables. A correlation coefficient of 0.96941 and a mean square error of 1.56E-02 show the good performance of the network during its training with 465 neurons and the Bayesian regularization algorithm. A p-value greater than 0.05 allows the prediction model to be validated with 95% confidence, confirming that there is no statistically significant difference between the real and predicted data by the network. The results indicate that the prediction model using artificial neural networks is efficient in predicting the production of dimethyl ether, demonstrating that neural networks can be a viable option in predicting the results of a plant compared to traditional simulations that obtain the results based on previously established equations, achieving to be able to predict results of more complex systems in less time. It is recommended to simulate and predict the production of dimethyl ether using synthesis gas.