Estimation of nonlinear systems using linear multiple models

Authors


Abstract

This article addresses the problem of estimating the states of a nonlinear plant that operates in multiple regimes and makes transitions between them. It is often difficult to obtain a single nonlinear model that accurately describes the plant in all regimes. Even if a global model is available, it sometimes cannot be used conveniently in an estimator. An alternative approach is presented where local linear models are identified at each different operating point, and estimation is performed by tracking the transitions from one regime to another. This is done by first estimating the validity of the local models on-line and then constructing a time-varying global model by interpolating between the local linear models. State and parameter estimation is then performed using this global model in a moving horizon estimator. To demonstrate the effectiveness of this method, it is applied to three test systems: a simple numerical example, a CSTR, and a copolymerization reactor.

Ancillary