This paper describes the collaborative results of a study between The University of Newcastle, Sydney University, ICI Engineering Technology and ICI Australia Pty Ltd into the application of neural networks to Model-Based Predictive Control. The results discussed will describe the methodology of extracting data from a real industrial process, pre-processing the data, selection of key inputs using dynamic correlation and multivariate stcrtistics, process modelling and control. The paper will emphasise the importance of combining engineering knowledge, advanced statistics and neural networks in order to obtain an extremely powerful modelling technique for dealing with non-linear systems. The implementation of the controller was carried out on a validated simulation of the actual process. This Speedup model had been developed by Sydney University over a period of 18 months and had been used previously to design other successful control strategies that are not on-line on the process.
The resultant Model-Based controller was benchmarked against a linear model-based controller and two PID controllers. The neural network- controller not only outperformed the linear MBPC by a 50% reduction in standard deviation but also reduced overshoot and settling time dramatically.
Agamennoni, O. and Romagnoli, J.1993. DICOF: A package for multivariable controller design in the frequency domain'. Cache News, 36, Spring.
2.
Bhat, N., and McAvoy, T.J.1990. 'Use of neural nets for dynamic modeling and control of chemical process systems', Computer Chem Engng, 14(4/5), 573-583.
3.
Bhat, N., Minderman, P., and McAvoy, T.J.1989. 'Use of neurat nets for modelling of chemical process systems', IFAC Symp on Dycord+89, Maastricht, The Netherlands, 21-23 August, 1989. 147-153.
4.
Birky, G.J., and McAvoy, T.J.1989. 'A neural net to learn the design of distillation controls ', IFAC Symp on Dycord+89. Maastricht, The Netherlands, 21-23 August. 205—13.
5.
Cybenko, G.1989. 'Approximation by superposition of sigmoidal function' , Math Control Signal Syst2, 303-314.
6.
Eskinat, E., Johnson, S., and Luyben, W.1991. Use of Hammerstein Models in identification of non-linear systems', AIChE Journal, 37(2), Feb.
7.
Frank, I.E., Feikema, J., Constantine, N., and Kowalski, B.R.1983. 'Prediction of product quality from spectral data using the partial least squares method.' J of Chemical Information and ComputerSci, 24, 20-24.
8.
Hernandez, E., and Arkun, Y.1992. 'Study of the control relevant properties of backpropogation neural network models of non-linear dynamical systems', ComputerChem Engng, 16(4), 227-240.
9.
Hertz, J., Krogh, A., and Palmer, R.1991. Introduction to the Theory of Neural Computation , Addison-Wesley.
10.
Jackson, J.E.1981. 'Principal components and factor analysis: Part 1-Principal components', J Qual Tech13(2).
11.
Marquardt, D.1963. 'An algorithm for least squares estimation of non-linear parameters', SIAMJ Appl Math, 11, 431.
12.
Montague, G.A., and Willis, M.J.1993. 'Artificial neural network model-based control' , Advances in Model Based Predictive Control Conference Proceedings , September, Department of Electrical Engineering, University of Oxford, UK. Volume 2, 65-76.
13.
Ungar, L.H., Powell, B.A., and Kamens, S.N.1990. 'Adaptive networks for fault diagnosis and process control ', Computers Chem Engng, 14(4,5), 561-572.
14.
Williams, R.J., and Zipser, D.1989. 'A learning algorithm for continually running, fully recurrent neural networks', Neural Computation, 1270-1289.
15.
Willis, M.J., Massimi, C.Di., Montague, G.A., Tham, M.T., and Morris, A.J.1991. Artificial neural networks in process engineering' , IEE Proceedings-D, 138(3), May.
