Elsevier

Information Sciences

Volume 506, January 2020, Pages 148-160
Information Sciences

Event-triggered adaptive neural network controller for uncertain nonlinear system

https://doi.org/10.1016/j.ins.2019.08.015Get rights and content

Abstract

In this paper, an event-triggered adaptive controller, consisting of a basic adaptive neural network controller and an event-triggered mechanism, is developed for a class of single-input and single-output high-order nonlinear systems with neural network approximation. Both the static and the dynamic event-triggered mechanisms are proposed in our design, without the input-state stability (ISS) assumption which is needed in most existing results. It is shown that the proposed methods can ensure that the closed loop system is globally stable. The minimal inter-event time internal is lower bounded by a positive number so that no Zeno behavior occurs. Finally, the numerical simulations are presented to illustrate our theory.

Introduction

Event-triggered control has gained considerable attention in the control community, especially in network control field. Traditional control method is based on continuous control input to guarantee the system stability which need much communication resource to transfer the control signal. Event-triggered strategy can be seen as a supervised control which can decide if control signal needs to be sent to the plant based on some predefined performance condition. In recent years, many results have been reported with different event-triggered conditions, for example see [1], [6], [11], [12], [13], [19], [20], [21].

Most of the existing event-triggered results are developed for known and linear systems, e.g. [12], [20], [24], [27], and the references therein. In [20], a very simple but useful event-triggered scheduling strategy is proposed to ensure system stability under the condition that the measurement error caused by the event-triggered mechanism. Furthermore, the event-triggered mechanism has been studied for robust control of linear uncertain system [23]. Nonlinear systems, which are more practical, also attract attentions from many researchers [1], [11], [19], [21]. Event-triggered schemes are developed in [15] based on hybrid systems tools for the stabilization of nonlinear system with ISS condition. Based on the decreasing property of Lyapunov function, results with static event-triggered conditions are established in [7], [11], [20]. A dynamic event-triggered condition is considered in [3] to reduce the communication burden which allows the Lyapunov function to increase at some points but decrease in the whole execution times. As a more realistic scenario, event-triggered design for unknown nonlinear systems is more meanful. However, because of the complicated system structure and unknown parameters, only limited results on event-triggered control of unknown nonlinear systems are available. Three different mechanisms are proposed in [29] to address the adaptive event-triggered problem of nonlinear systems. On the other hand, neural networks have been studied for decades and the universal property of neural network has been used in adaptive control process, see for example [4], [14], [25], [26], [30]. The application of on-line neural network for control and learning system need response rapidly, but the neural network need a lot of calculation and learning processes which cost too much time, so it is necessary to find a practical method to reduce the calculation burden of the neural network application. Considering the event-triggered mechanism, the adaptive neural controllers with event-triggered condition have been proposed in [17] and [16], in order to reduce the computational and communication burden of the closed-loop system for the continuous-time and discrete-time domains, respectively.

The objective of this paper is on designing event-triggered adaptive neural controllers for unknown nonlinear systems. We begin with designing a basic adaptive controller based on the backstepping technique [9] and neural network with low-frequency learning [31]. Then we propose static mechanism and dynamic mechanism for the obtained adaptive neural controller. In the static mechanism, we design an event-triggered condition based on a given fixed bound of the measurement error, which is the error between the control signals generated by the event-triggered controller and the designed controller. In order to ensure the stability of the system, the basic adaptive neural network controller with the addition of introducing an extra term. In the dynamic mechanism, a new variable which can be considered as a filtered value of the signal in the static mechanism is designed. The stability of the closed-loop system is then established. Both event-triggered mechanisms can guarantee that the lower bound of inter-event time is larger than zero, i.e., Zeno behavior is avoided. Compared with the event-triggered mechanisms in existing neural controllers such as those in [17] and [16], our event-triggered conditions are designed without the input-to-state linearizable condition while with simpler threshold. Furthermore, the established results are to solve the nonlinear tracking problems which are more practical than the stabilization problem in [17]. It is noted that ISS condition is also required in [17] where continuous-time systems are considered.

The remainder of the paper is organized as follows. Section 2  formulates the control problem. Static and dynamic event-triggered controllers are given in Section 3, followed by some simulation examples which illustrate our theories in Section 4. The paper is concluded in Section 5.

Section snippets

System description and problem formulation

In this section, we formulate the design problem and introduce the RBF neural network which will be used in next section.

Event-triggered neural network with adaptive backstepping

In this section, we present a basic adaptive backstepping control with low-frequency leanrning neural network firstly. Then static event-triggered mechanism and dynamic event-triggered mechanism are introduced based on the proposed adaptive neural network control method.

Simulation example

In this section, we consider the following system from a rigid robot of actuator dynamics [10], [22] to illustrate the established theoretical results.Define F2=D1x3D1Nsinx1 and F3=M1Km*x2M1Hx3, which are unknown functions to designers. The initial states of the system are chosen to be x(0)=[0.4,10,0]. The parameters of the system are D=1, N=10, B=1, M=0.05, H=0.5, Km=10 for simulation purpose. We chooseγ2,1=10, γ2,2=10, γ3,1=1 and γ3,2=100. The event-triggered parameters are chosen as B

Conclusions

In this paper, we consider the event-triggered adaptive control for unknown nonlinear system. Two different event-triggered mechanisms are proposed to ensure the stability of the closed-loop system with low communication burden. The static event-triggered method is designed based on the event-triggered error of the designed control signal and event-triggered control signal. And in dynamic mechanism, a filtered variable of the event-triggered condition in static mechanism is introduced to build

Declaration of Competing Interest

We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript.

References (31)

  • M. Velasco et al.

    The self triggered task model for real-time control systems

    24th IEEE Real-Time Systems Symposium (work in progress)

    (2003)
  • T. Yucelen et al.

    Low-frequency learning and fast adaptation in model reference adaptive control

    IEEE Trans. Autom. Control

    (2013)
  • A. Anta et al.

    To sample or not to sample: self-triggered control for nonlinear systems

    IEEE Trans. Autom. Control

    (2010)
  • C.M. Bishop

    Neural networks for pattern recognition

    Agric. Eng. Int.

    (1995)
  • A. Girard

    Dynamic triggering mechanisms for event-triggered control

    IEEE Trans. Autom. Control

    (2014)
  • R. Goebel et al.

    Hybrid Dynamical Systems: Modeling, Stability, and Robustness

    (2012)
  • W.M. Haddad et al.

    Impulsive and Hybrid Dynamical Systems:Stability, Dissipativity, and Control

    (2014)
  • W.P.M.H. Heemels et al.

    Periodic event-triggered control for linear systems

    IEEE Trans. Autom. Control

    (2013)
  • W.P.M.H. Heemels et al.

    An introduction to event-triggered and self-triggered control

    Decision and Control

    (2012)
  • H.K. Khalil

    Nonlinear Systems

    (2002)
  • M. Krstic et al.

    Nonlinear and adaptive control

    Lect. Notes Control Inf. Sci.

    (1995)
  • F.L. Lewis et al.

    Control of Robot Manipulators

    (1993)
  • T. Liu et al.

    A small-gain approach to robust event-triggered control of nonlinear systems

    IEEE Trans. Autom. Control

    (2015)
  • J. Manuel Mazo et al.

    Brief paper: an ISS self-triggered implementation of linear controllers

    Automatica

    (2009)
  • M. Mazo et al.

    Input-to-state stability of self-triggered control systems

    Decision and Control, 2009 Held Jointly with the 2009 Chinese Control Conference. CDC/CCC 2009. Proceedings of the IEEE Conference on

    (2009)
  • Cited by (42)

    View all citing articles on Scopus
    View full text