Título: Process Identification with Artificial Neural Network Applied to Experimental Data from a Continuous Distillation Column
Autores: Vieira, W. G.; Sodré, C. H.; Barcellos, K. B.; Dantas, L. C.
Resumo: This paper describes an identification procedure, based on Artificial Neural Network (ANN) to study the dynamic behavior of the continuous distillation column. The study was focused on an existing 10 trays, 12cm diameter pilot scale distillation column. The column was used to distillate methanol and water mixture. The dynamic model was developed and validated against pilot distillation column data and used to generate a large data set of operational variables. Open-loop responses were used to generate data. A multilayer feed forward network was chosen for the distillation column representation. Based on the theorectical results, the following strategy was adopted for the network architecture: the input layer was composed by four variables and the output layer was formed by the two variables. The number of nodes in the hidden layer was obtained from a trial-and-error procedure. The backpropagation method was used to the process training. It was observed that the network generalization capacity and the training time increased with the number of hidden neurons. This study can be able to develop a Multivariable Predictive Control (MPC) to be implemented in the column control system, using the ANN as internal model. The obtained ANN model agrees very well with the experimental and theorectical data then it could be used to simulate the real process with the control strategy.
Palavras-chave: Process Identification; Artificial Neural Networks; Distillation Column
Código DOI: 10.21528/CBRN2005-156
Artigo em PDF: CBRN2005_156.pdf
Arquivo BibTex: CBRN2005_156.bib