Robust training algorithm of multilayered neural networks for identification of nonlinear dynamic systems

Song, Qing

doi:10.1049/ip-cta:19981614

Cited by 26 publications

(30 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The adaptive law consists of a leakage modification of a standard gradient descent algorithm. However, in contrast to commonly leakage modifications [3][4][5][6][7][8][9][10] which aim at stability in the presence of approximation errors and disturbances, we introduce the leakage term here for, in addition to stability, ensuring that the residual prediction error converges to zero. More precisely, it is show by using usual Lyapunov arguments and an adaptive bounding technique [12] that the residual prediction error converges asymptotically to zero, whereas the others error signals remain bounded.…”

Section: Introductionmentioning

confidence: 99%

“…Similarly, also others relevant works, such as [5][6][7][8][9][10], showed that the dead zone, modified −  rule, and 1  -modification and others robust modifications can be used in weight adjustment laws to make the entire identification process stable in the presence of approximation error and disturbances.…”

Section: Introductionmentioning

confidence: 99%

“…Neural identification models usually employed are the dynamic ones, being their weights mainly adjusted using gradient and backpropagation algorithms or their robust modifications [3][4][5][6][7][8][9][10]. Most used robust modifications in neuro-identification are the  , switching- , 1  , parameter projection, and dead zone [3][4][5][6][7][8][9][10] which avoid the parameter drift.…”

Section: Introductionmentioning

confidence: 99%

“…System neuro-identification is important not only to predict the behavior of the system, but also for providing an appealing system parameterization, which can later be used in the synthesis of control algorithms, since mathematical characterization is often a prerequisite to observer and controller design, see for instance [1][2][3][4][5][6][7][8][9][10] and the references therein.…”

Section: Introductionmentioning

confidence: 99%

“…Most used robust modifications in neuro-identification are the  , switching- , 1  , parameter projection, and dead zone [3][4][5][6][7][8][9][10] which avoid the parameter drift. Nevertheless, these modifications, at present, can not ensure that the prediction error converges asymptotically to zero in the presence of approximation error and bounded disturbances.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Stability Analysis of A Neuro-Identification Scheme with Asymptotic Convergence

Vargas¹

2012

IJAIA

View full text Add to dashboard Cite

This paper focuses on the stability and convergence analysis of a neuro-identification scheme for uncertain nonlinear systems. Based on linearly parameterized neural networks and the previous knowledge of upper bounds for the approximation error and disturbances, a robust modification of the descent gradient algorithm is proposed to make the overall identification process stable, and, in addition, the on-line residual prediction error asymptotically null, despite the presence of approximation error and disturbances.A simulation study to show the application and comparative performance of the proposed algorithm is presented.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Stability Analysis of A Neuro-Identification Scheme with Asymptotic Convergence

Vargas¹

2012

IJAIA

View full text Add to dashboard Cite

show abstract

Observer‐based controller design for uncertain disturbed Takagi‐Sugeno fuzzy systems: A fuzzy wavelet neural network approach

Ebrahimi

Asemani

Safavi

2020

Adaptive Control & Signal

View full text Add to dashboard Cite

SummaryIn this article, we develop a novel method to design a controller for nonlinear systems represented by Takagi‐Sugeno (T‐S) fuzzy model in the presence of unknown dynamics, uncertainties in parameters of nonlinear system and external disturbances. The control law is constituted two segments. The first segment derives from parallel distributed compensation (PDC) procedure, in which each control rule is drawn from the respective rule of T‐S fuzzy model. The second segment stems from fuzzy wavelet neural network (FWNN) estimator which is evoked by the hypothesis of multiresolution analysis (MRA) of wavelet transforms and fuzzy notions, so as to approximate the uncertainties and external disturbances in T‐S fuzzy model accurately. In this regard, the Lyapunov stability theorem is applied to acquire the adaptive learning laws for training and tuning online FWNN parameters such as the dilations, translations, and the weights of networks. Moreover, the asymptotic stability of the closed loop system is guaranteed based on the Lyapunov stability theorem. Furthermore, a development of the proposed method to observer‐based controller design for uncertain nonlinear systems descripted by T‐S fuzzy model is provided. The efficiency and robustness of the proposed method are illustrated by simulation outcomes. It is noteworthy that the proposed controller remarkably handles the uncertainties and external disturbances in T‐S fuzzy model without employing traditional conservative lemma and without considering bounds on uncertainties.

show abstract

Modeling of Crude Oil Blending via Discrete-Time Neural Networks

Li¹,

2010

Studies in Computational Intelligence

View full text Add to dashboard Cite

Crude oil blending is an important unit operation in petroleum refining industry. A good model for the blending system is beneficial for supervision operation, prediction of the export petroleum quality and realizing model-based optimal control. Since the blending cannot follow the ideal mixing rule in practice, we propose a static neural network to approximate the blending properties. By the dead-zone approach, we propose a new robust learning algorithm and give theoretical analysis. Real data of crude oil blending is applied to illustrate the neuro modeling approach.

show abstract

Robust training algorithm of multilayered neural networks for identification of nonlinear dynamic systems

Cited by 26 publications

References 1 publication

Stability Analysis of A Neuro-Identification Scheme with Asymptotic Convergence

Stability Analysis of A Neuro-Identification Scheme with Asymptotic Convergence

Observer‐based controller design for uncertain disturbed Takagi‐Sugeno fuzzy systems: A fuzzy wavelet neural network approach

Modeling of Crude Oil Blending via Discrete-Time Neural Networks

Contact Info

Product

Resources

About