Elsevier

Neurocomputing

Volume 156, 25 May 2015, Pages 129-133
Neurocomputing

Brief Papers
New mean square exponential stability condition of stochastic fuzzy neural networks

https://doi.org/10.1016/j.neucom.2014.12.076Get rights and content

Abstract

This paper investigates the stability problem for interval type-2 (IT2) stochastic fuzzy neural networks. Firstly, an IT2 stochastic fuzzy neural network is constructed. Secondly, by using stochastic analysis approach and Itô׳s differential formula, a new sufficient condition ensuring mean square exponential stability is obtained. The condition can be expressed in terms of convex optimization problem. The main contribution of this paper is that the IT2 stochastic fuzzy neural network with parameter uncertainties is first proposed. The parameter uncertainties are bounded, and can be effectively expressed by upper and lower membership functions. Two numerical examples are proposed to show the effectiveness of the proposed scheme.

Introduction

Over the past years, neural networks have been widely used in many practical applications such as signal processing, combinatorial optimization and image processing. Many research results have been proposed for neural networks, such as [1], [2], [3], [4], [5], [6], [7], [8], and the references therein. In real nervous systems, it is easy to see that stochastic disturbances are nearly inevitable and affect the stability of the neural networks. Hence, it implies that the stability analysis of stochastic neural networks has primary significance in the research of neural networks. Some related research results have been published in [9], [10], [11], [12], [13].

Fuzzy logic control scheme has been proposed as an effective method to deal with complex nonlinear systems [14], [15], [16]. In recent years, the fuzzy logic control method has been widely used in many practical applications such as chemical processes, automotive systems and robotics systems [17], [18], [19], [20], [21], [22], [23], [24], [25], [26]. Takagi–Sugeno (T–S) fuzzy system [27] is a well-known fuzzy system in model-based fuzzy control. The results on stability analysis and stabilization, control problem and filtering problem for T–S fuzzy systems were reported in [28], [29], [30], [31], [32], [33]. Recently, the T–S fuzzy control approach has been used in the study of stochastic fuzzy neural network [34], [35], [36], [37], [38]. In [35], the authors considered the mean square exponential stability problem of stochastic fuzzy neural networks with time-varying delays. However, it should be mentioned that the above research results are based on type-1 T–S stochastic fuzzy neural networks. By using the type-1 T–S fuzzy model, the stability problem cannot be studied if the nonlinear plants contain parameter uncertainties. Once nonlinear plants contain parameter uncertainties, it will result in the uncertainties of membership functions. To solve this problem, the authors in [39] proposed an IT2 fuzzy logic model, which can be utilized to investigate the nonlinear plants problem. Recently, the authors in [40], [41] have studied the problem of the IT2 fuzzy systems by using the upper and lower membership functions which can be used to deal with parameter uncertainties problem. However, it is mentioned that, so far, there are few results on stability analysis of IT2 stochastic fuzzy neural networks.

Motivated by the above discussion, this paper investigates the stability problem for IT2 stochastic fuzzy neural networks subject to parameter uncertainties. The parameter uncertainties, which can be obtained by the lower and upper membership functions, are bounded. By using stochastic analysis approach and Itô differential formula, a novel sufficient condition ensuring mean square exponential stability for stochastic fuzzy neural networks is achieved. Two examples are given to show the feasibility of the proposed scheme. The remaining parts of this paper are organized as follows. Section 2 introduces IT2 stochastic fuzzy neural networks. Section 3 proposes stability condition for IT2 stochastic fuzzy neural networks and Section 4 provides two illustrative examples to show the merits of the proposed results. Section 5 concludes this paper.

Notation: The superscript “T” denotes matrix transposition, and “−1” stands for matrix inverse. Rn is the n-dimensional Euclidean space and the notation X>0(0) stands for a symmetric and positive definite (semi-definite) matrix. The symbol within a symmetric block matrix represents the symmetric terms, and diag{} denotes a block-diagonal matrix. He(A) is defined as He(A)=A+AT for simplicity. If not explicitly stated, all matrices are assumed to be compatible dimensions for algebraic operations. L2[0,+) stands for the Hilbert space of square integrable functions over [0,+). E{x} means the expectation of the stochastic variable x.

Section snippets

Problem formulation

As discussed in [34], we consider the following IT2 stochastic fuzzy neural network.

Plant Rule i:IF g1(x(t)) is η1i AND AND gp(x(t)) is ηpi,THEN:dx(t)=[Aix(t)+Bif(x(t))]dt+[Cix(t)+Dif(x(t))]dw(t),where ga(x(t)) denotes the premise variable and ηai is an IT2 fuzzy set, i=1, 2,,r,a=1,2,,p. p is a positive integer. x(t)=[x1(t),x2(t),,xn(t)]T stands for the neural state, f(x(t))=[f1(x1(t)),f2(x2(t)),,fn(xn(t))]T denotes the neuron activation function, and w(t) is a one-dimensional Brownian

Main results

In this section, the stability condition for the IT2 stochastic fuzzy neural network (2) is first proposed in Theorem 1. Based on the linear matrix inequality (LMI) approach, we can have the following theorem. For presentation convenience, we denote L1=diag{h1,h2,,hn}, L2=diag{h1,+h2,+,hn+}.

Theorem 1

The IT2 stochastic fuzzy neural network (2) is mean square exponentially stable, if there exist matrices P>0, G1=diag{μ1,μ2,,μn}0, G2=diag{ν1,ν2,,νn}0, K=diag{k1,k2,,kn}0 with appropriate

Simulation results

In this section, two examples are used to show the effectiveness and the merits of the proposed method.

Example 1

Consider the IT2 stochastic fuzzy neural network (2) with the parameters as follows:A1=[2003],A2=[1002],B1=[0.242.210.950.55],B2=[0.370.830.320.7],C1=[1.180.560.420.69],C2=[1.260.590.811],D1=[0.591.010.350.53],D2=[0.70.960.080.63].

Take the activation functions as f1(x1)=tanh(x1) and f2(x2)=tanh(0.3x2). Therefore, it is easy to have L1=diag{1,0}, L2=diag{0,0.3}. Membership

Conclusions

In this paper, the stability problem has been considered for a class of IT2 stochastic fuzzy neural networks. Firstly, the IT2 stochastic fuzzy neural networks have been constructed. Sufficient condition has been given to ensure that the IT2 stochastic fuzzy neural network is mean square exponential stability. The existence condition of the stability can be obtained in terms of LMI. Two numerical examples have been given to show the merits of the proposed approach.

Xing Xing was born in Liaoning Province, China, in 1982. He obtained the M.S. degree in Computer Software and Theory in 2008, and the Ph.D. degree in Computer Applied Technology in 2013, from Dalian Maritime University, China. He was a visiting student at the Department of Computer Science, University of New Mexico, USA, from September 2009 to September 2011. Currently, he is an associate professor at the Department of Information Science and Technology, Bohai University, China. From February

References (49)

  • H. Li et al.

    Mean square exponential stability of stochastic fuzzy Hopfield neural networks with discrete and distributed time-varying delays

    Neurocomputing

    (2009)
  • H. Zhao et al.

    Almost sure exponential stability of stochastic fuzzy cellular neural networks with delays

    Chaos Solitons Fractals

    (2009)
  • L. Chen et al.

    Stability analysis of stochastic fuzzy cellular neural networks with delays

    Neurocomputing

    (2008)
  • H. Li et al.

    Fault-tolerant control of Markovian jump stochastic systems via the augmented sliding mode observer approach

    Automatica

    (2014)
  • S. Yin et al.

    A comparison study of basic data-driven fault diagnosis and process monitoring methods on the benchmark Tennessee eastman process

    J. Process Control

    (2012)
  • J. Cao et al.

    Global asymptotic and robust stability of recurrent neural networks with time delays

    IEEE Trans. Circuits Syst. I: Regul. Pap.

    (2005)
  • J. Cao et al.

    Global asymptotical stability of recurrent neural networks with multiple discrete delays and distributed delays

    IEEE Trans. Neural Netw.

    (2006)
  • Y. Tang et al.

    On controllability of neuronal networks with constraints on the average of control gains

    IEEE Trans. Cybern.

    (2014)
  • Y. Tang et al.

    Multiobjective identification of controlling areas in neuronal networks

    IEEE/ACM Trans. Comput. Biol. Bioinf.

    (2013)
  • Y. Tang et al.

    Distributed robust synchronization of dynamical networks with stochastic coupling

    IEEE Trans. Circuits Syst. I: Regul. Pap.

    (2014)
  • R. Yang et al.

    Exponential stability on stochastic neural networks with discrete interval and distributed delays

    IEEE Trans. Neural Netw.

    (2010)
  • S. Tong et al.

    A combined backstepping and small-gain approach to robust adaptive fuzzy output feedback control

    IEEE Trans. Fuzzy Syst.

    (2009)
  • G. Feng

    A survey on analysis and design of model-based fuzzy control systems

    IEEE Trans. Fuzzy Syst.

    (2006)
  • Y.-Y. Cao et al.

    Analysis and synthesis of nonlinear time-delay systems via fuzzy control approach

    IEEE Trans. Fuzzy Syst.

    (2000)
  • Cited by (16)

    • Cooperative strategy for constructing interval type-2 fuzzy neural network

      2019, Neurocomputing
      Citation Excerpt :

      Although these above algorithms [31,32] and some other efficient learning algorithms [33,34] can improve the accuracy of IT2FNN, the high computational complexity remains an issue in these learning algorithms [35–38]. Furthermore, most of these algorithms have not discussed the convergence [39–42]. Based on the above discussion, a self-constructing interval type-2 fuzzy neural network (SC-IT2FNN) was developed in this study.

    • Synchronization of IT2 stochastic fuzzy complex dynamical networks with time-varying delay via fuzzy pinning control

      2019, Journal of the Franklin Institute
      Citation Excerpt :

      Hence, stochastic systems have received much attention and much efforts have been devoted to extending many fundamental methods for deterministic systems to stochastic systems. Recently, the synchronization issue for various CDNs with stochastic effects has been widely reported in [10–13]. For example, the authors in [12] discussed the mean square exponential stability of stochastic fuzzy neural networks, and in [13], the problem of synchronization for discrete-time stochastic complex networks with randomly occurred nonlinearities and mixed time delays was studied.

    • A biological brain-inspired fuzzy neural network: Fuzzy emotional neural network

      2018, Biologically Inspired Cognitive Architectures
      Citation Excerpt :

      However these models are based on type-1 fuzzy approach, in which the stability of solutions cannot be studied, especially if the problem is uncertain in the nature (Xing, Pan, Lu, & Cui, 2015). To overcome this problem, an interval type-2 fuzzy logic has been proposed (Mendel, John, & Liu, 2006; Xing et al., 2015). Type-2 FNNs minimize the effects of uncertainties in rule-based systems.

    • Observer-based robust consensus control for multi-agent systems with noises

      2016, Neurocomputing
      Citation Excerpt :

      The model of multi-agent systems (MASs) has attracted great interest in the last decades due to its wide applications in the research of sensor networks [1–3], neural networks [4,5], and vehicle formations [6,7].

    • Stable ANFIS2 for Nonlinear System Identification

      2016, Neurocomputing
      Citation Excerpt :

      Expanding research on type-2 fuzzy systems has led to wide applications in, time-series prediction [14]; control of DC motor [15]; Control of nonlinear systems with time-varying delays [16]; sliding mode control [17]; pattern recognition [18] and robot control [19]. Xing et al. discussed the stability of a class of type-2 fuzzy systems [20]. Juang et al. proposed a type-2 neural fuzzy system that learned through its type-1 counterpart by merging and extending the type-1membership function and then implements the built type-2 neural fuzzy system on field-programmable gate array (FPGA) chip [21].

    View all citing articles on Scopus

    Xing Xing was born in Liaoning Province, China, in 1982. He obtained the M.S. degree in Computer Software and Theory in 2008, and the Ph.D. degree in Computer Applied Technology in 2013, from Dalian Maritime University, China. He was a visiting student at the Department of Computer Science, University of New Mexico, USA, from September 2009 to September 2011. Currently, he is an associate professor at the Department of Information Science and Technology, Bohai University, China. From February 2014, he carries out his postdoctoral research in the School of Astronautics, Harbin Institute of Technology. His research interests include data mining, machine learning, and neural networks.

    Yingnan Pan received the B.S. degree in mathematics from Bohai University, Jinzhou, China, in 2012. He is studying for M.S. degree in applied mathematics in Bohai University, Jinzhou, China. His research interests include fuzzy control, robust control and their applications.

    Qing Lu received the B.E. degree in automation from Hebei University of Technology, Tianjin, China in 2013, and is studying for M.E. degree in Control Theory and Control Engineering in Automation Research Institute, Bohai University, Jinzhou, China. She is a master degree candidate in the College of Engineering, Bohai University, Jinzhou, China. Her research interest includes fuzzy control and model predictive control.

    Hongxia Cui received the B.Sc. degree in photogrammetry and remote sensing from Wuhan University, Hubei Province, China, in 1991, the M.E. degree in computer science from Shanxi University, Shanxi, China, in 2002 and the Ph.D. degree in photogrammetry and remote sensing from Wuhan University, Hubei Province, China, in 2006. She is currently a professor at the College of Computer Science and Technology, Bohai University. Her current research interests include image processing, computer vision, neural networks and their applications.

    This work is partially supported by National Natural Science Foundation of China (No. 41371425).

    View full text