Identification and Control of a Simulated Distillation Plant using Connectionist and Evolutionary Techniques

Abstract

This paper deals with the application of neural networks and evolutionary techniques to the area of process identification and control. A distillation process is simulated with the dynamic flow-sheet simulator DIVA which employs a first-principle-based model. Several neural paradigms were implemented to adaptively model the concentration dynamics. A combined PI-Neural Net controller for concentration control is presented. Using genetic algorithms it was possible to optimize the network structure and reduce the size of the training set. Finally, some parallelization methods for neural and evolutionary algorithms, implemented on Connection Machine architectures, are briefly explained.

An Adaptive Neuro-Fuzzy Approach to Control a Distillation Column

“…The results obtained show the adequate performance of the neuro-fuzzy composition controller in contrast to the ones obtained using other neuralbased schemes such as in Gariglio et al [8], where the neural controller must operate in conjunction with a PI controller to get an efficient performance, or as in Baratti and Servida [11], where the existence of the plant's inverse is a precondition to calculate the neural controller.…”

“…Neural and fuzzy applications have been successfully applied to the chemical engineering processes [3][4][5], and several control strategies have been reported in the literature for the distillation plant modelling [6][7][8][9] and control tasks [10,11].…”