Minimal model dimension/order determination algorithms for recurrent neural networks

Jeen-Shing Wang, Yu Liang Hsu, Hung Yi Lin, Yen Ping Chen

研究成果: Article

3 引文 (Scopus)

摘要

This paper focuses on the development of model dimension/order determination algorithms for determining minimal dimensions/orders of recurrent neural networks using only input-output measurements of unknown systems. We present two types of model dimension/order determination approaches. The first type is named all-in-one strategy that includes the minimum description length (MDL) principle and the eigensystem realization algorithm (ERA). This type is capable of identifying the model dimension/order and model parameters simultaneously. The other type is named divide-and-conquer strategy that includes the Lipschitz quotients and false nearest neighbors (FNN). This type usually requires additional parameter optimization algorithms to estimate the model parameters for closely emulating the dynamic behavior of unknown systems. The effectiveness of these four algorithms has been validated through nonlinear dynamic system identification examples. In addition, we provide performance comparisons and discussion on the characteristics of these four algorithms as method-selection guidelines.

原文English
頁(從 - 到)812-819
頁數8
期刊Pattern Recognition Letters
30
發行號9
DOIs
出版狀態Published - 2009 七月 1

指紋

Recurrent neural networks
Identification (control systems)
Dynamical systems

All Science Journal Classification (ASJC) codes

  • Software
  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Signal Processing

引用此文

Wang, Jeen-Shing ; Hsu, Yu Liang ; Lin, Hung Yi ; Chen, Yen Ping. / Minimal model dimension/order determination algorithms for recurrent neural networks. 於: Pattern Recognition Letters. 2009 ; 卷 30, 編號 9. 頁 812-819.
@article{3f0a616a8816485c9d1c064ded2a2f96,
title = "Minimal model dimension/order determination algorithms for recurrent neural networks",
abstract = "This paper focuses on the development of model dimension/order determination algorithms for determining minimal dimensions/orders of recurrent neural networks using only input-output measurements of unknown systems. We present two types of model dimension/order determination approaches. The first type is named all-in-one strategy that includes the minimum description length (MDL) principle and the eigensystem realization algorithm (ERA). This type is capable of identifying the model dimension/order and model parameters simultaneously. The other type is named divide-and-conquer strategy that includes the Lipschitz quotients and false nearest neighbors (FNN). This type usually requires additional parameter optimization algorithms to estimate the model parameters for closely emulating the dynamic behavior of unknown systems. The effectiveness of these four algorithms has been validated through nonlinear dynamic system identification examples. In addition, we provide performance comparisons and discussion on the characteristics of these four algorithms as method-selection guidelines.",
author = "Jeen-Shing Wang and Hsu, {Yu Liang} and Lin, {Hung Yi} and Chen, {Yen Ping}",
year = "2009",
month = "7",
day = "1",
doi = "10.1016/j.patrec.2008.05.007",
language = "English",
volume = "30",
pages = "812--819",
journal = "Pattern Recognition Letters",
issn = "0167-8655",
publisher = "Elsevier",
number = "9",

}

Minimal model dimension/order determination algorithms for recurrent neural networks. / Wang, Jeen-Shing; Hsu, Yu Liang; Lin, Hung Yi; Chen, Yen Ping.

於: Pattern Recognition Letters, 卷 30, 編號 9, 01.07.2009, p. 812-819.

研究成果: Article

TY - JOUR

T1 - Minimal model dimension/order determination algorithms for recurrent neural networks

AU - Wang, Jeen-Shing

AU - Hsu, Yu Liang

AU - Lin, Hung Yi

AU - Chen, Yen Ping

PY - 2009/7/1

Y1 - 2009/7/1

N2 - This paper focuses on the development of model dimension/order determination algorithms for determining minimal dimensions/orders of recurrent neural networks using only input-output measurements of unknown systems. We present two types of model dimension/order determination approaches. The first type is named all-in-one strategy that includes the minimum description length (MDL) principle and the eigensystem realization algorithm (ERA). This type is capable of identifying the model dimension/order and model parameters simultaneously. The other type is named divide-and-conquer strategy that includes the Lipschitz quotients and false nearest neighbors (FNN). This type usually requires additional parameter optimization algorithms to estimate the model parameters for closely emulating the dynamic behavior of unknown systems. The effectiveness of these four algorithms has been validated through nonlinear dynamic system identification examples. In addition, we provide performance comparisons and discussion on the characteristics of these four algorithms as method-selection guidelines.

AB - This paper focuses on the development of model dimension/order determination algorithms for determining minimal dimensions/orders of recurrent neural networks using only input-output measurements of unknown systems. We present two types of model dimension/order determination approaches. The first type is named all-in-one strategy that includes the minimum description length (MDL) principle and the eigensystem realization algorithm (ERA). This type is capable of identifying the model dimension/order and model parameters simultaneously. The other type is named divide-and-conquer strategy that includes the Lipschitz quotients and false nearest neighbors (FNN). This type usually requires additional parameter optimization algorithms to estimate the model parameters for closely emulating the dynamic behavior of unknown systems. The effectiveness of these four algorithms has been validated through nonlinear dynamic system identification examples. In addition, we provide performance comparisons and discussion on the characteristics of these four algorithms as method-selection guidelines.

UR - http://www.scopus.com/inward/record.url?scp=67349167976&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=67349167976&partnerID=8YFLogxK

U2 - 10.1016/j.patrec.2008.05.007

DO - 10.1016/j.patrec.2008.05.007

M3 - Article

AN - SCOPUS:67349167976

VL - 30

SP - 812

EP - 819

JO - Pattern Recognition Letters

JF - Pattern Recognition Letters

SN - 0167-8655

IS - 9

ER -