Desarrollo de una RNA para la predicción de la concentración de diisobutileno producido en una columna de destilación catalítica

Abstract

The objective of this work was to develop an Artificial Neural Network (ANN) capable of predicting the molar concentration of Diisobutylene (DIB), 3,4,4-trimethyl-1-Pentene (DIM) and Tertiary Butyl Alcohol (TBA) in the distillate and residue streams in the following columns: Reactive (CDC), High Pressure (ADC) and Low Pressure (TDC). The process simulation was carried out in DWSIM; once the validity of the simulation was verified, a sensitivity analysis was performed to identify which operating variables influenced the mole fraction of DIB, DIM and TBA in the outputs of the three columns used. The input variables were: mole fraction of isobutylene (IB) and 2-butene (2-Bu) in the C4 feed; C4 and TBA feed temperature; CDC column operating pressure, ADC column pressure and TDC column pressure. The development, training, validation and testing of the network was performed in MATLAB using the Neural Fitting app. The network structure is based on the Bayesian Regularization (BR) algorithm consisting of 7 inputs and 7 outputs with 30 neurons in the hidden layer. The network achieved a mean square error (MSE) of 0.0008251 with a linear regression coefficient (R) of 0.9946, however, the reliability of the network was evaluated by determining the percentage error between the predicted values and the real values provided by DWSIM, through an internal and external validation, from which it obtained an average error of 5% and 10% respectively. Subsequently, the ANOVA statistic was used to validate the results obtained. Finally, a process optimization was performed by determining the operating conditions that allowed the highest DIB fraction in the CDC column residue, the highest TBA fraction in the ADC column distillate and the highest DIB fraction in the TDC column distillate.