Application of natural computation techniques to optimal design of flow injection systems

Row injection (FI) systems are widely used for on-line monitoring of chemic al processes. Several approaches have been made in order to achieve the opt imal design of the FI system, mainly based on the approach of deterministic models that describe the process using the mass balances around the system and the corresponding kinetic relations. Although, good results have been obtained with this approach, the complexit y of the system and the effort necessary to calculate the parameters that c haracterize the FI system using a deterministic model, have led to the cons ideration of more empirical approaches to obtain a model of the process. In this paper, the authors present the results obtained in the application of two techniques, known as natural intelligence techniques, in the optimal design of a flow injection sandwich system for glucose and glycerol analys is. The optimization is performed using a genetic algorithm, in which a populat ion evolves combining the genetic code of the most capable individuals of t he previous generation. To evaluate the performance of each individual an a rtificial neural network is used. The results obtained with this approach a re comparable with the one previously developed using a deterministic descr iption of the FT system. (C) 1999 Elsevier Science B.V. All rights reserved .