Publication detail

A Levenberg-Marquardt algorithm with instrumental variables and its application on the identification of dynamic system

DOKOUPIL, J. PIVOŇKA, P.

Original Title

A Levenberg-Marquardt algorithm with instrumental variables and its application on the identification of dynamic system

Type

conference paper

Language

English

Original Abstract

This article deals with process identification, using nonlinear ARX model via feed-forward multilayer neural network. Estimation of network parameters is achieved using the Levenberg-Marquardt (LM) method in iterative batch mode adaptation. In order to obtain consistent estimate, original implementations of LM algorithm - which include instrumental variables (IV) technique - are suggested. Basic and extended IV methods are presented as some of the IV methods. Advantages of the proposed approach are illustrated in the example simulations on the real process, using B&R PLC.

Keywords

Neural networks, Levenberg-Marquardt, NARX, Instrumental variables, System identification

Authors

DOKOUPIL, J.; PIVOŇKA, P.

RIV year

2010

Released

20. 10. 2010

Publisher

DAAAM International Vienna

Location

TU Wien Karlsplatz 13/311 A-1040 Vienna Austria

ISBN

978-3-901509-73-5

Book

Annals of DAAAM for 2010 & Proceedings of the 21st International DAAAM Symposium, No1

Edition

2010

Edition number

1

Pages from

1359

Pages to

1360

Pages count

2

BibTex

@inproceedings{BUT35747,
  author="Jakub {Dokoupil} and Petr {Pivoňka}",
  title="A Levenberg-Marquardt algorithm with instrumental variables and its application on the identification of dynamic system",
  booktitle="Annals of DAAAM for 2010 & Proceedings of the 21st International DAAAM Symposium, No1",
  year="2010",
  series="2010",
  number="1",
  pages="1359--1360",
  publisher="DAAAM International Vienna",
  address="TU Wien
Karlsplatz 13/311
A-1040 Vienna Austria",
  isbn="978-3-901509-73-5"
}