Control of Nonlinear Systems Using Input Output Information

Abstract

Most chemical processes are nonlinear and difficult to model. Usually detailed first principle dynamic models are either not available or they are not amenable to practical controller designs. On the other hand, traditional linear controllers often fail to control nonlinear process units over a wide range of operating conditions and external process disturbances. With tighter constraints on product quality, environmental regulations and energy utilization, there is now a growing need for reliable predictive models and new on-line control algorithms for nonlinear processes.

This paper addresses the control of nonlinear systems using available plant data (input-output information). In particular the use of polynomial ARX models for nonlinear process control purposes is emphasized. After discussing theoretical justification for the use of such models and their practical advantages, the first part of the paper presents an identification algorithm for the construction of polynomial ARX models where we discuss data generation, model structure selection, parameter estimation and model validation. Next, for these classes of models, we show how to design nonlinear predictive controllers. Implementation issues such as disturbance modeling, state and parameter estimation are discussed. By using polynomial models, it is shown that the nonlinear optimization problem can be solved globally at each sampling time to compute the control moves. One detailed reactor example is given to demonstrate the utility of the polynomial ARX based MPC algorithms.