Enabling Invariant Models to Describe Time-Varying Dynamics: A Case Study ?

摘要

This work addresses the problem of identifying nonlinear and time-varying behaviors using Nonlinear polynomial AutoRegressive models with eXogenous inputs (NARX). Two approaches are investigated. The first one is based on a recursive algorithm with an adaptive forgetting factor to update the parameters of black-box models. The second approach aims to include a certain class of regressors to identify invariant models, i.e. with constant parameters, to describe such behaviors. An experimental case study of a pH neutralization process is carried out, which indicated the presence of behaviors that resemble varying dynamics features. Time-invariant models with a specific class of regressors are proposed and seem to be a promising way to deal with such dynamics.