Dynamic modelling of a falling-film evaporator for model predictive control : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Technology at Massey University

Abstract

Fundamentally this thesis provides a study into dynamic modelling of a pilot-scale falling-film evaporator. Its main aim, however, is to research and demonstrate artificial neural network (ANN) modelling for model predictive control as applied to an evaporator system. As a prerequisite to testing an advanced control strategy such as model predictive control one must have available a suitable dynamic simulation of the process. To this end the thesis initially presents the formulation of a dynamic model of the evaporator system developed from first-principles. A novel approach to developing a dynamic ANN representation of a process is presented and applied to the evaporator. This approach incorporates prior knowledge of the system into the network topology to attain a model with a flexible, modular structure. A dynamic, recurrent training methodology is devised to enable the ANN model to predict over a future horizon of arbitrary length. The performance of the modular ANN model is compared with a linear model of a similar form identified through conventional linear regression methods. It was found that the nonlinear ANN model and subsequent nonlinear MPC scheme exhibited no improvement in performance to that of their linear counterparts.