Second‐order optimization methods for time‐delay Autoregressive eXogenous models: Nature gradient descent method and its two modified methods

Summary This article proposes several second‐order optimization methods for time‐delay ARX model. Since the time‐delay in the information vector makes the traditional identification algorithms be inefficient, a redundant rule based method is utilized to transformed the model into a redundant model....

Full description

Saved in:
Bibliographic Details
Published inInternational journal of adaptive control and signal processing Vol. 37; no. 1; pp. 211 - 223
Main Authors Chen, Jing, Pu, Yan, Guo, Liuxiao, Cao, Junfeng, Zhu, Quanmin
Format Journal Article
LanguageEnglish
Published Bognor Regis Wiley Subscription Services, Inc 01.01.2023
Subjects
adaptive gradient descent
Algorithms
ARX model
Autoregressive models
convergence rate
Delay
momentum based method
nature gradient descent
Optimization
Parameter modification
time‐delay
Online AccessGet full text
ISSN0890-6327
1099-1115
DOI10.1002/acs.3519

Cover

More Information
Summary:Summary This article proposes several second‐order optimization methods for time‐delay ARX model. Since the time‐delay in the information vector makes the traditional identification algorithms be inefficient, a redundant rule based method is utilized to transformed the model into a redundant model. Then, the nature gradient descent (NGD) algorithm is developed for such a model. To reduce the computational efforts of the NGD algorithm and to adaptively update each element in the parameter vector, two modified NGD algorithms are also presented. The simulation examples verify the effectiveness of the proposed algorithms.
Bibliography:Funding information
the Fundamental Research Funds for the Central Universities, Grant/Award Number: JUSRP22016; the Funds of the Science and Technology on Near‐Surface Detection Laboratory, Grant/Award Number: TCGZ2019A001; the National Natural Science Foundation of China, Grant/Award Number: 61973137
ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0890-6327
1099-1115
DOI:10.1002/acs.3519