Industrial reactor modelling with artificial neural networks

Abstract

This paper describes the development of artificial neural network based models of a large-scale industrial reaction process. The process operates continuously but is subject to changes in product demand and variations in raw material quality. The models are required for the purpose of process optimisation which poses additional modelling demands as predictions of process constraints are also required. A blend of artificial neural network and standard linear regression models was found to be most appropriate. The models have subsequently been incorporated within an advisory system to provide the operators with advice on process set-points to maximise process profitability within the constraints. Results from process operation serve to demonstrate the predictive ability of the models and serve to highlight important application lessons. An overview of the advisory system concept is included to place the models in context.