An improved estimation method for unmodeled dynamics based on anfis and its application to controller design

Yajun Zhang, Tianyou Chai, Hong Wang, Xinkai Chen, Chun Yi Su

Research output: Contribution to journalArticle

16 Citations (Scopus)

Abstract

By representing nonlinear systems as a combination of linear part and unmodeled dynamics, in this paper, an improved estimation algorithm using an adaptive neuro-fuzzy inference system (ANFIS) for unmodeled dynamics is presented. At first, the unmodeled dynamics is divided into two parts using the differential expansion of the control input at the last time instant; then, the two parts are estimated by the ANFIS. It has been shown that the proposed algorithm overcomes the problem that the unknown control input is embedded in unmodeled dynamics, which makes the true value of unmodeled dynamics difficult obtain. Moreover, the method improves the precision of the estimation of unmodeled dynamics. Second, under the assumption that the growth rate of unmodeled dynamics does not exceed its input vector, the "oneto- one mapping" and "regularization technique" are adopted to deal with the input and output data and the unmodeled dynamics, respectively. As a result, the data vector can be guaranteed to lie inside a compact set, which ensures the use of the universal approximation property of the ANFIS. On the other hand, it has been shown that datum of a system can be fully used to obtain the parameters (centers, widths) in membership functions and the network connection weights in the ANFIS by offline training. These parameters are tuned online to improve the estimation convergence rate of the unmodeled dynamics. The effectiveness of the proposed estimation method is illustrated by comparing it with the simulation results that are obtained from the other existing methods. Finally, the proposed estimationmethod is applied to the nonlinear switching control design. Both simulation and theoretical analysis have confirmed that the nonlinear switching control which adopts the proposed estimation method cannot only guarantee the stability and convergence of the system but can exhibit a desired dynamic performance for the closed-loop system as well.

Original languageEnglish
Article number6399584
Pages (from-to)989-1005
Number of pages17
JournalIEEE Transactions on Fuzzy Systems
Volume21
Issue number6
DOIs
Publication statusPublished - 2013 Dec

Fingerprint

Controllers
Fuzzy inference
Membership functions
Closed loop systems
Nonlinear systems

Keywords

  • Adaptive neuro-fuzzy inference system (ANFIS)
  • Data
  • Nonlinear systems
  • Switching control
  • Unmodeled dynamics

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Artificial Intelligence
  • Computational Theory and Mathematics
  • Applied Mathematics

Cite this

An improved estimation method for unmodeled dynamics based on anfis and its application to controller design. / Zhang, Yajun; Chai, Tianyou; Wang, Hong; Chen, Xinkai; Su, Chun Yi.

In: IEEE Transactions on Fuzzy Systems, Vol. 21, No. 6, 6399584, 12.2013, p. 989-1005.

Research output: Contribution to journalArticle

@article{418400d008d84aa09b867a56007279ed,
title = "An improved estimation method for unmodeled dynamics based on anfis and its application to controller design",
abstract = "By representing nonlinear systems as a combination of linear part and unmodeled dynamics, in this paper, an improved estimation algorithm using an adaptive neuro-fuzzy inference system (ANFIS) for unmodeled dynamics is presented. At first, the unmodeled dynamics is divided into two parts using the differential expansion of the control input at the last time instant; then, the two parts are estimated by the ANFIS. It has been shown that the proposed algorithm overcomes the problem that the unknown control input is embedded in unmodeled dynamics, which makes the true value of unmodeled dynamics difficult obtain. Moreover, the method improves the precision of the estimation of unmodeled dynamics. Second, under the assumption that the growth rate of unmodeled dynamics does not exceed its input vector, the {"}oneto- one mapping{"} and {"}regularization technique{"} are adopted to deal with the input and output data and the unmodeled dynamics, respectively. As a result, the data vector can be guaranteed to lie inside a compact set, which ensures the use of the universal approximation property of the ANFIS. On the other hand, it has been shown that datum of a system can be fully used to obtain the parameters (centers, widths) in membership functions and the network connection weights in the ANFIS by offline training. These parameters are tuned online to improve the estimation convergence rate of the unmodeled dynamics. The effectiveness of the proposed estimation method is illustrated by comparing it with the simulation results that are obtained from the other existing methods. Finally, the proposed estimationmethod is applied to the nonlinear switching control design. Both simulation and theoretical analysis have confirmed that the nonlinear switching control which adopts the proposed estimation method cannot only guarantee the stability and convergence of the system but can exhibit a desired dynamic performance for the closed-loop system as well.",
keywords = "Adaptive neuro-fuzzy inference system (ANFIS), Data, Nonlinear systems, Switching control, Unmodeled dynamics",
author = "Yajun Zhang and Tianyou Chai and Hong Wang and Xinkai Chen and Su, {Chun Yi}",
year = "2013",
month = "12",
doi = "10.1109/TFUZZ.2012.2236889",
language = "English",
volume = "21",
pages = "989--1005",
journal = "IEEE Transactions on Fuzzy Systems",
issn = "1063-6706",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "6",

}

TY - JOUR

T1 - An improved estimation method for unmodeled dynamics based on anfis and its application to controller design

AU - Zhang, Yajun

AU - Chai, Tianyou

AU - Wang, Hong

AU - Chen, Xinkai

AU - Su, Chun Yi

PY - 2013/12

Y1 - 2013/12

N2 - By representing nonlinear systems as a combination of linear part and unmodeled dynamics, in this paper, an improved estimation algorithm using an adaptive neuro-fuzzy inference system (ANFIS) for unmodeled dynamics is presented. At first, the unmodeled dynamics is divided into two parts using the differential expansion of the control input at the last time instant; then, the two parts are estimated by the ANFIS. It has been shown that the proposed algorithm overcomes the problem that the unknown control input is embedded in unmodeled dynamics, which makes the true value of unmodeled dynamics difficult obtain. Moreover, the method improves the precision of the estimation of unmodeled dynamics. Second, under the assumption that the growth rate of unmodeled dynamics does not exceed its input vector, the "oneto- one mapping" and "regularization technique" are adopted to deal with the input and output data and the unmodeled dynamics, respectively. As a result, the data vector can be guaranteed to lie inside a compact set, which ensures the use of the universal approximation property of the ANFIS. On the other hand, it has been shown that datum of a system can be fully used to obtain the parameters (centers, widths) in membership functions and the network connection weights in the ANFIS by offline training. These parameters are tuned online to improve the estimation convergence rate of the unmodeled dynamics. The effectiveness of the proposed estimation method is illustrated by comparing it with the simulation results that are obtained from the other existing methods. Finally, the proposed estimationmethod is applied to the nonlinear switching control design. Both simulation and theoretical analysis have confirmed that the nonlinear switching control which adopts the proposed estimation method cannot only guarantee the stability and convergence of the system but can exhibit a desired dynamic performance for the closed-loop system as well.

AB - By representing nonlinear systems as a combination of linear part and unmodeled dynamics, in this paper, an improved estimation algorithm using an adaptive neuro-fuzzy inference system (ANFIS) for unmodeled dynamics is presented. At first, the unmodeled dynamics is divided into two parts using the differential expansion of the control input at the last time instant; then, the two parts are estimated by the ANFIS. It has been shown that the proposed algorithm overcomes the problem that the unknown control input is embedded in unmodeled dynamics, which makes the true value of unmodeled dynamics difficult obtain. Moreover, the method improves the precision of the estimation of unmodeled dynamics. Second, under the assumption that the growth rate of unmodeled dynamics does not exceed its input vector, the "oneto- one mapping" and "regularization technique" are adopted to deal with the input and output data and the unmodeled dynamics, respectively. As a result, the data vector can be guaranteed to lie inside a compact set, which ensures the use of the universal approximation property of the ANFIS. On the other hand, it has been shown that datum of a system can be fully used to obtain the parameters (centers, widths) in membership functions and the network connection weights in the ANFIS by offline training. These parameters are tuned online to improve the estimation convergence rate of the unmodeled dynamics. The effectiveness of the proposed estimation method is illustrated by comparing it with the simulation results that are obtained from the other existing methods. Finally, the proposed estimationmethod is applied to the nonlinear switching control design. Both simulation and theoretical analysis have confirmed that the nonlinear switching control which adopts the proposed estimation method cannot only guarantee the stability and convergence of the system but can exhibit a desired dynamic performance for the closed-loop system as well.

KW - Adaptive neuro-fuzzy inference system (ANFIS)

KW - Data

KW - Nonlinear systems

KW - Switching control

KW - Unmodeled dynamics

UR - http://www.scopus.com/inward/record.url?scp=84897695612&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84897695612&partnerID=8YFLogxK

U2 - 10.1109/TFUZZ.2012.2236889

DO - 10.1109/TFUZZ.2012.2236889

M3 - Article

VL - 21

SP - 989

EP - 1005

JO - IEEE Transactions on Fuzzy Systems

JF - IEEE Transactions on Fuzzy Systems

SN - 1063-6706

IS - 6

M1 - 6399584

ER -