Identification of bilinear models for chemical processes using canonical variate analysis

Abstract

The identification of nonlinear models for chemical processes solely from experimental data is described in this paper. The canonical variate analysis (CVA) technique that has served well in the identification of empirical linear process models is extended to construct data-based bilinear models in an iterative fashion. Numerous examples involving engineering systems are included to illustrate the practicality of the suggested approach for bilinear model identification. Finally, the use of the identified nonlinear models for control is demonstrated using the example of a simulated paper machine headbox system.