Nonlinear dynamic MISO identification in a complex noisy environment

Abstract

An algorithm for nonlinear dynamic system identification using the generalised total least squares parameter estimation algorithm is presented in this paper. In many practical applications noise in measured input channels results in parameter estimates which are not consistent when conventional least squares parameter estimation methods are used. total least squares methodologies are limited to situations where all inputs and the output respectively are corrupted by noise. A proper extension is the generalised total least squares algorithm which is able to cope with situations where some input channels of a MISO system are noise-free while others are taken from measurements and are thus subject to noise. The GTLS algorithm together with a well-tried model construction algorithm which is based on an hierarchical logistic discriminant tree results in an efficient algorithm for nonlinear dynamic identification.