Predicción del porcentaje de acetonitrilo aplicando redes neuronales artificiales a partir del proceso de simulación en DWSIM

Abstract

The objective of this research was to predict the percentage of acetonitrile recovery applying artificial neural networks from the simulation of the acetonitrile/water azeotrope separation process, using DWSIM software. For this, the operating variables proposed in the work were considered. “Pressure-swing or extraction-distillation for the recovery of pure acetonitrile from ethanol ammoxidation process: A comparison of efficiency and cost” with which this simulation was developed and validated. Based on the results obtained from the simulation of the process, it was possible to generate a database of 125 data values for feeding the Artificial Neural Network (ANN) which has a structure of 4 inputs, which vary according to the ranges established to achieve the 5 outputs of the network in which the percentage of recovery of acetonitrile is included. For the operation of this network, its structure has 11 hidden layers, and its training process was carried out with the Levenberg-Marquardt logarithm To validate the results provided by the RNA, a graphic and statistical analysis was carried out, of which it can be mentioned that the P-value generated in the ANOVA was greater than 0.05, so it was shown that there is no statistically significant difference between the means of the variables compared with a confidence level of 95.0%, and at the same time the graphic analysis showed the presence of over-peaks in some of the analysed fractions, however with the average percentage error analysis it was shown that this does not exceed in any case more than 1%. Therefore, based on these mentioned analyses, it can be concluded that the designed Artificial Neural Network predicts in a satisfactory, effective, and efficient manner the percentage of acetonitrile recovery.