Adaptive sliding-mode observer for second order discrete-time MIMO nonlinear systems based on recurrent neural-networks

Ivan Salgado, Hafiz Ahmed, Oscar Camacho, Issac Chairez

Research output: Contribution to journalArticle

Abstract

This manuscript introduces a novel methodology to solve the state estimation of discrete-time multi-input multi-output (MIMO) nonlinear systems with uncertain dynamics. The mathematical model of the nonlinear systems considered in this paper satisfies the usual Lagrangian structure that characterizes many mechanical, electrical or electromechanical systems. A recurrent neural network (RNN) estimates the uncertain dynamics of the MIMO system with an updating law based on a particular variant of the discrete-time version of the super-twisting algorithm (DSTA). A Lyapunov stability analysis defines the convergence zone for the state estimation error throughout the solution of a matrix inequality. The convergence zone for the estimation is smaller when the DSTA and the RNN work together in an observer. Numerical examples demonstrate how the adaptive observer reduces the zone of convergence and the oscillations in the steady state compared with a discrete version of the STA with additional linear correcting terms. An experimental implementation shows how the observer estimates the unknown states of a Van Der Pol Oscillator. A comparison against some variations of the DSTA justifies the advantages of the mixed DSTA-RNN observer.
Original languageEnglish
Pages (from-to)(In-Press)
Number of pages16
JournalInternational Journal of Machine Learning and Cybernetics
Volume(In-Press)
Early online date7 Jan 2019
DOIs
Publication statusE-pub ahead of print - 7 Jan 2019

Fingerprint

Recurrent neural networks
Nonlinear systems
State estimation
Mathematical models

Keywords

  • State estimation
  • Lyapunov theory
  • Sliding modes
  • Recurrent neural networks
  • Discrete-time super twisting algorithm
  • Second order systems

Cite this

Adaptive sliding-mode observer for second order discrete-time MIMO nonlinear systems based on recurrent neural-networks. / Salgado, Ivan ; Ahmed, Hafiz; Camacho, Oscar; Chairez, Issac .

In: International Journal of Machine Learning and Cybernetics, Vol. (In-Press), 07.01.2019, p. (In-Press).

Research output: Contribution to journalArticle

@article{a2f535a48b2f410b885f7231cdc1459e,
title = "Adaptive sliding-mode observer for second order discrete-time MIMO nonlinear systems based on recurrent neural-networks",
abstract = "This manuscript introduces a novel methodology to solve the state estimation of discrete-time multi-input multi-output (MIMO) nonlinear systems with uncertain dynamics. The mathematical model of the nonlinear systems considered in this paper satisfies the usual Lagrangian structure that characterizes many mechanical, electrical or electromechanical systems. A recurrent neural network (RNN) estimates the uncertain dynamics of the MIMO system with an updating law based on a particular variant of the discrete-time version of the super-twisting algorithm (DSTA). A Lyapunov stability analysis defines the convergence zone for the state estimation error throughout the solution of a matrix inequality. The convergence zone for the estimation is smaller when the DSTA and the RNN work together in an observer. Numerical examples demonstrate how the adaptive observer reduces the zone of convergence and the oscillations in the steady state compared with a discrete version of the STA with additional linear correcting terms. An experimental implementation shows how the observer estimates the unknown states of a Van Der Pol Oscillator. A comparison against some variations of the DSTA justifies the advantages of the mixed DSTA-RNN observer.",
keywords = "State estimation, Lyapunov theory, Sliding modes, Recurrent neural networks, Discrete-time super twisting algorithm, Second order systems",
author = "Ivan Salgado and Hafiz Ahmed and Oscar Camacho and Issac Chairez",
year = "2019",
month = "1",
day = "7",
doi = "10.1007/s13042-018-00908-z",
language = "English",
volume = "(In-Press)",
pages = "(In--Press)",
journal = "International Journal of Machine Learning and Cybernetics",
issn = "1868-8071",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - Adaptive sliding-mode observer for second order discrete-time MIMO nonlinear systems based on recurrent neural-networks

AU - Salgado, Ivan

AU - Ahmed, Hafiz

AU - Camacho, Oscar

AU - Chairez, Issac

PY - 2019/1/7

Y1 - 2019/1/7

N2 - This manuscript introduces a novel methodology to solve the state estimation of discrete-time multi-input multi-output (MIMO) nonlinear systems with uncertain dynamics. The mathematical model of the nonlinear systems considered in this paper satisfies the usual Lagrangian structure that characterizes many mechanical, electrical or electromechanical systems. A recurrent neural network (RNN) estimates the uncertain dynamics of the MIMO system with an updating law based on a particular variant of the discrete-time version of the super-twisting algorithm (DSTA). A Lyapunov stability analysis defines the convergence zone for the state estimation error throughout the solution of a matrix inequality. The convergence zone for the estimation is smaller when the DSTA and the RNN work together in an observer. Numerical examples demonstrate how the adaptive observer reduces the zone of convergence and the oscillations in the steady state compared with a discrete version of the STA with additional linear correcting terms. An experimental implementation shows how the observer estimates the unknown states of a Van Der Pol Oscillator. A comparison against some variations of the DSTA justifies the advantages of the mixed DSTA-RNN observer.

AB - This manuscript introduces a novel methodology to solve the state estimation of discrete-time multi-input multi-output (MIMO) nonlinear systems with uncertain dynamics. The mathematical model of the nonlinear systems considered in this paper satisfies the usual Lagrangian structure that characterizes many mechanical, electrical or electromechanical systems. A recurrent neural network (RNN) estimates the uncertain dynamics of the MIMO system with an updating law based on a particular variant of the discrete-time version of the super-twisting algorithm (DSTA). A Lyapunov stability analysis defines the convergence zone for the state estimation error throughout the solution of a matrix inequality. The convergence zone for the estimation is smaller when the DSTA and the RNN work together in an observer. Numerical examples demonstrate how the adaptive observer reduces the zone of convergence and the oscillations in the steady state compared with a discrete version of the STA with additional linear correcting terms. An experimental implementation shows how the observer estimates the unknown states of a Van Der Pol Oscillator. A comparison against some variations of the DSTA justifies the advantages of the mixed DSTA-RNN observer.

KW - State estimation

KW - Lyapunov theory

KW - Sliding modes

KW - Recurrent neural networks

KW - Discrete-time super twisting algorithm

KW - Second order systems

U2 - 10.1007/s13042-018-00908-z

DO - 10.1007/s13042-018-00908-z

M3 - Article

VL - (In-Press)

SP - (In-Press)

JO - International Journal of Machine Learning and Cybernetics

JF - International Journal of Machine Learning and Cybernetics

SN - 1868-8071

ER -