Control of nonlinear chemical processes using neural models and feedback linearization

Abstract

Black-box modeling techniques based on artificial neural networks are opening new horizons for the modeling and control nonlinear processes in biotechnology and the chemical process industries. The link between dynamic process models and actual process control is provided by the concept of model-based control (MBC), e.g. internal model control (IMC) or model-based predictive control (MBPC). To avoid timeconsuming calculations, feedback-linearization techniques are used to linearize the nonlinear process model. The resulting linear model then is used in a linear MBC scheme, allowing for standard linear control techniques to be applied. Two methods of input–output feedback linearization are described in combination with the use of neural process models: the exact input–output feedback linearization and the approximate input–output feedback linearization. The proposed methods are applied to a MISO (multi-input single-output) laboratory-scale pressure process, which shows good results compared to conventional linear techniques.