Model Reduction and Control of Multistage Flash (MSF) Desalinization Plants
Multistage flash (MSF) desalinization plants are a major means of desalting scawater for human use in several arid regions of the world in the present times. The MSF plants are physically large and their control usually involves more than twenty control loops. According to the present practice, the controllers are of the PI or PID type and their tuning is largely based on experience rather than on systematic modeling of the plant. Plant modeling based on physical principles gives rise to a large and complex set of coupled nonlinear differential equations which has to be linearized about a chosen set of operating conditions. As the operating point changes, the resulting linearized model also changes. This requires retuning of the controllers in certain loops depending on the changing linear plant model in the related loops. The linearized model happens to be enormously large in size requiring reduction for controller design and practical implementation. There exist several model reduction methods and they have to be chosen to meet the objectives of adequate modeling. In a nonlinear plant, the linearized model parameters vary with the operating conditions. To make a controller to simultaneously meet the demands of model reduction and variable operating conditions, the conventional approaches of control are either inadequate or too involved. In this chapter, a technique based on artificial neural networks (ANN) for model reduction under plant parameter perturbations is proposed. The complexity of analysis, reduction, computation and controller design in the variable conditions of operation is avoided by simply training an ANN to give the parameters of a reduced model for use in controller design. An automated decision support may be provided to choose the best ANN configuration for reduced order modeling of a large, complex and variable plant which provides the basis for a robust design of a simple controller. Certain well established model reduction methods are employed in the mainstream of the procedure and the related results are impressive. Based on these results, a scheme based on an added ANN, for direct controller implementation under the discussed conditions is proposed. The results of this modest attempt point out to the strong possibility of more intelligent control of large complex plants under uncertainty and/or variable plant dynamics. The present discussion is centred on SISO loop designs, which does not rule out the possibility of simple extensions to MIMO designs.
KeywordsArtificial Neural Network Model Reduction System Number Step Response Reduce Order Model
Unable to display preview. Download preview PDF.
- Al-Gobaisi D.M.K., A.S. Barakzai, and A.M. El-Nashar, “An overview of modern control strategies for optimizing thermal desalinization plants”. in Desalinization and water Re-use, Proceedings of the 12th International Symposium, (Miriam Balaban, ed.), (Malta), 15–18 April 1991.Google Scholar
- Wilson D.A., “Optimum solution of model reduction problem”, Proceedings of IEE, Vol. 117, pp. 1161, 1165, 1970.Google Scholar
- Luus R. and G.D. Howitt, “Model reduction by optimization”, Hungarian Journal of Industrial Chemistry, Vol. 16, pp. 29, 38, 1988.Google Scholar
- Pal J., “System reduction by a mixed method”, Transactions of IEEE on Automatic Control, Vol. AC-25,no. 5, pp. 973, 976, 1990.Google Scholar
- Chu, S.R., R. Shoureshi, and M. Tenorio, “Neural networks for system identification”, IEEE Control Systems Magazine, pp. 31, 35, 1990.Google Scholar
- Liang Jin, Peter N. Nikiforuk, and Madan M Gupta, “Model matching control of unknown nonlinear systems using recurrent neural networks”, in IFAC 11th Triennial World Congress, (Sydney, Australia), pp. 337, 344, 1993.Google Scholar
- Anderson B.D.O. and J.B. Moore, Optimal Control: Linear Quadratic Methods. Prentice-Hall of India Ltd., 1991.Google Scholar
- Safanov M., “Future directions in the robust control theory”, in IFAC 11th Triennial World Congress, (Tallin, Estoria), pp. 171, 175, 1990.Google Scholar