Chemical Engineering Research & Design, Vol.72, No.1, 3-19, 1994
Artificial Neural Networks - Studies in-Process Modeling and Control
In the mid 1980s wide spread interest in Artificial Neural Network research re-emerged following a period of reduced research funding. The much wider availability and power of computing systems, together with new research studies, resulted in a far greater market for the technology. The seeds were sown for claims by some that the technique provided much sought after pragmatic solutions, and by others that it provided a panacea to all complex modelling problems. Unlike ARMA, NARMA and multivariate statistical modelling approaches the methodology has been attributed the potential of accurately describing the behaviour of extremely complex systems. But is the approach so different? Should we not consider the concept of Neural Networks as being an integral part of system representation, modelling and identification? In this respect perhaps we really do have an established, but still developing, theory and technology represented within a new framework. Indeed, selling the technology in anything but this manner might discredit what could well prove to be a valuable engineering tool. This paper examines the contribution that various networks methodologies can make to the process modelling and control toolbox. Feedforward networks with sigmoidal activation functions, radial basis function networks and autoassociative networks are reviewed and studied using data from industrial processes. Finally the concept of dynamic networks is introduced with an example of nonlinear predictive control.
Keywords:PRINCIPAL COMPONENT ANALYSIS;SELF-TUNING CONTROL;NON-LINEAR SYSTEMS;DISTILLATION COLUMN;NONLINEAR-SYSTEMS;VALIDITY TESTS;IDENTIFICATION