Dynamic modelling using a multiple neural network architecture

Abstract

A multiple neural network architecture has been introduced. The methodology makes use of the 'K' means clustering algorithm in order to differentiate between different process operating regions thus allowing an improved utilisation of the training data set. Indeed the ability of the technique to focus on specific operating regions appears to enhance characterisation of process behaviour when compared to a single network trained over a large region. An additional advantage of the technique, when compared to the standard feedforward ANN, is that the reliability of the network prediction may be monitored. Finally the advantages of the proposed technique are highlighted by application to two simulated nonlinear systems: a binary distillation column; and a continuous stirred tank reactor. >