On Improving TLS Identication Results Using Nuisance Variables with Application on PMSM

Abstract

This article presents a novel total least-squares based method for errors-in-variables model identication with a known structure. This method considers the errors of both input and output variables and thus achieves more accurate estimates compared to conventional ordinary least-squares based methods. The introduced method consists of two recursive total least-squares algorithms connected in a hierarchical structure, which allows for exploitation of nuisance variables and a priori known structure of the identied model. The total least-squares (TLS) method is introduced, and a new “nuisance improved hierarchical total least-squares” (nHTLS) method is derived. Its properties are discussed and proved by simulations. Furthermore, the method is applied in a practical experiment consisting of the state-space identication of the permanent magnet synchronous motor (PMSM). The introduced method is compared with TLS and proven to provide measurably superior dynamical behavior and smaller estimation error of results.This article presents a novel total least-squares based method for errors-in-variables model identication with a known structure. This method considers the errors of both input and output variables and thus achieves more accurate estimates compared to conventional ordinary least-squares based methods. The introduced method consists of two recursive total least-squares algorithms connected in a hierarchical structure, which allows for exploitation of nuisance variables and a priori known structure of the identied model. The total least-squares (TLS) method is introduced, and a new “nuisance improved hierarchical total least-squares” (nHTLS) method is derived. Its properties are discussed and proved by simulations. Furthermore, the method is applied in a practical experiment consisting of the state-space identication of the permanent magnet synchronous motor (PMSM). The introduced method is compared with TLS and proven to provide measurably superior dynamical behavior and smaller estimation error of results.