Globally linearising control using artificial neural networks

In previous publications the advantage of the globally linearising control (GLC) technique has been demonstrated. It can lead to improved control performances over conventional (linear) methodologies when applied to ´highly´ non-linear processes. However, it should be noted that when synthesising the GLC law, a mechanistic model of the process must be available. Unfortunately, the development of an accurate mechanistic model of a chemical process can often be a costly, time consuming exercise. As an alternative to the use of a mechanistic model within the GLC framework this paper proposes the use of a generic cost effective modelling philosophy: artificial neural networks. Wherever possible, a priori knowledge is incorporated within the network architecture. The paper is organised as follows. First, the fundamental concepts behind the GLC are presented. Artificial neural networks are then briefly discussed, and a neural network architecture suitable for incorporation within the GLC framework is proposed. Finally, the performance of the resulting control strategy is illustrated by application to a simulated batch chemical reactor system.