A NARMAX model representation for adaptive control based on local models

Abstract

Here we address the problem of representing NARMAX (nonlinear ARMAX) models with application to adaptive control. We propose a nonlinear model representation where a number of simple local models are combined. The local models are valid in specific operation regimes of the process. Explicitly defined model validity functions make it possible to combine the local models by interpolation. During online identification, only the local models corresponding to the current process operation regime are updated. It is therefore not necessary to relearn the model each time there is a change in the operation regime of the process. The concept is illustrated by a simulation example of a nonlinear pH-neutralization process.