Identification of quasi-ARMAX models of nonlinear stochastic sampled-data systems

Authors:	Akesson Bernt, Abo Akademi University, Finland Toivonen Hannu, Abo Akademi University, Finland
Topic:	1.1 Modelling, Identification & Signal Processing
Session:	Nonlinear System Identification II
Keywords:	neural network models, nonlinear models, sampled-data systems,stochastic systems, parameter identification

Abstract

State-dependent parameter representations of nonlinear stochasticsampled-data systems are studied. Velocity-based linearization isused characterize sampled-data systems using nominally linearmodels whose parameters can be represented as functions of pastoutputs and inputs. For stochastic systems the approach leads tostate-dependent ARMAX (quasi-ARMAX) representations. The modelsand their parameters are identified from input-output data usingfeedforward neural networks to represent the model parameters asfunctions of past inputs and outputs.