数学物理学报, 2019, 39(4): 823-831 doi:

论文

具有变时滞的神经型Hopfield神经网络的全局吸引子研究

周庆华1, 万立,2, 刘杰2

Global Attracting Set for Neutral Type Hopfield Neural Networks with Time-Varying Delays

Zhou Qinghua1, Wan Li,2, Liu Jie2

通讯作者: 万立, E-mail: wanlinju@aliyun.com

收稿日期: 2018-04-1  

基金资助: 国家自然科学基金.  11501499
国家自然科学基金.  61573011
国家自然科学基金.  11301403
国家自然科学基金.  61304022
国家自然科学基金.  11271295
广东省自然科学基金.  2015A030313707

Received: 2018-04-1  

Fund supported: the NSFC.  11501499
the NSFC.  61573011
the NSFC.  11301403
the NSFC.  61304022
the NSFC.  11271295
the Natural Science Foundation of Guangdong Province.  2015A030313707

摘要

该文探讨了一类具有变时滞的非线性及非自治的神经型Hopfield神经网络的渐近性质.利用非负矩阵的性质和矩阵不等式,得到了保证该系统全局吸引集存在和Lagrange稳定性的充分条件.最后,给出一个例子说明理论的有效性.

关键词: 神经型Hopfield神经网络 ; 变时滞 ; Lagrange稳定 ; 全局吸引集 ; 积分不等式

Abstract

This paper deals with the asymptotic properties of a class of nonlinear and nonautonomous neutral type Hopfield neural networks with time-varying delays. By applying the property of nonnegative matrix and an integral inequality, some sufficient conditions are derived to ensure the existence of the global attracting set and the stability in a Lagrange sense for the considered system. Finally, an example is given to demonstrate the effectiveness of our theoretical result.

Keywords: Neutral-type Hopfield neural networks ; Time-varying delays ; Lagrange stability ; Global attracting set ; Integral inequality

PDF (364KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

周庆华, 万立, 刘杰. 具有变时滞的神经型Hopfield神经网络的全局吸引子研究. 数学物理学报[J], 2019, 39(4): 823-831 doi:

Zhou Qinghua, Wan Li, Liu Jie. Global Attracting Set for Neutral Type Hopfield Neural Networks with Time-Varying Delays. Acta Mathematica Scientia[J], 2019, 39(4): 823-831 doi:

1 引言

自1982年,越来越多的学者关注到Hopfield神经网络,其在分类、联想记忆、并行计算和优化中的应用日益广泛.这类应用强烈依赖于网络的动力行为.大量的学者们研究了Hopfield神经的一些动力学行为并取得了可观的成果.例如,文献[1-9]研究了分支和混沌现象, [10-12]研究了周期解, [13-19]讨论了平衡点的稳定性.

值得一提的是, Lyapunov稳定性[13-19]指的是平衡点的稳定性,此稳定性的前提条件是平衡点存在.但有时在许多真实的物理系统中,平衡点并不存在.因此,一些学者开始研究具有时滞的神经网络的吸引集[20-27].在文献[20]中验证了,在全局吸引集之外,不存在平衡点、周期性状态和几乎周期性的混沌状态.在文献[27]中针对不同时滞范围的系统给出了原点吸引范围的估计.

众所周知,不等式是一种研究微分方程的重要的技术工具,见文献[28-37].但是,对于一类具有变时滞的非线性和非自治的神经型Hopfield神经网络,研究其渐近行为时,上述参考文献中的不等式是无效的.我们将给出一个积分不等式,用它来研究上述模型的全局吸引集的存在性.

本文的结构如下.首先是预备知识,包括一些必要的定义、假设以及引理.第3部分讨论模型的全局吸引集.第4部分列举例验证主要结果.第5部分给出结论.

2 预备知识

本文中, $E_n$表示$n\times n$维单位矩阵. ${\Bbb R} $是实数集, ${\Bbb R} _+=[0, +\infty)$,符号${\Bbb R} ^n$表示$n$维Euclidean空间. $A$是一个方阵, $A^{-1}$为其逆矩阵, $\rho(A)$是其谱半径. $A \leq B (A<B)$表示矩阵$A$$B$中对应元素均满足$\leq (<)$的不等式关系.如果$ A \geq 0$,则$A$称为一个非负矩阵.所有从拓扑空间$X$到拓扑空间$Y$的连续映射构成空间$C(X, Y)$.对于$x\in {\Bbb R} ^n, \varphi \in C [(-\infty, t_0], {\Bbb R} ^n]$$ (-\infty, t_0]$上有界, $\tau(t) \in C({\Bbb R} , {\Bbb R} _+)$,我们定义

我们考虑如下具有变时滞的神经型Hopfield神经网络模型

$ \begin{eqnarray} &&{\dot{x}_{i}(t)}=-a_{i}(t)x_{i}(t)+\sum^{n}_{j=1}b_{ij}(t)f_j(x_{j}(t)) +\sum^{n}_{j=1}c_{ij}(t)f_j(\dot{x}_{j}(t-\tau(t))) +I_i(t), \end{eqnarray} $

模型(2.1)由一元神经域$F_x$构成,其中$x_i(t)$$F_x$中第$i$个神经元的激活, $a_{i}(t)\geq 0$表示被动衰变率, $b_{ij}(t)$$c_{ij}(t)$分别代表各自的神经元权重系数. $f_j$是激活函数, $I_i$是外部输入. $\tau(t) \in C({\Bbb R} , {\Bbb R} _+)$是传输延迟, $ \lim\limits_{t\rightarrow +\infty }(t-\tau(t))=+\infty$, ${\dot{\tau}(t)}\leq \tau <1$,其中$\tau$是常数.

系统(2.1)的初始条件是$x_i(s)=\varphi_i(s)$, $s \in (-\infty, t_0]$, $i=1, 2, \cdots, n, $其中$\varphi_i(\cdot)$定义在$(-\infty, t_0]$上的实值连续函数.

我们定义如下符号

为了方便进一步讨论,我们给出如下假设,定义及重要引理.

(A1)对任意$x_{j}\in {\Bbb R} $, $j \in {\Bbb N}$,存在常数$l_j\geq 0$,使得成立

(A2)对$ t\geq t_0$, $\forall s \geq t_0$,存在常数矩阵$\Sigma \geq 0$,使得成立

(A3)对$ t\geq t_0$,存在非负常数矩阵$\Pi$和常数向量$I \geq 0$,使得成立

(A4) $\bar{\Pi}\triangleq \Sigma/(1-\tau)+\Pi \geq 0$, $\rho(\bar{\Pi}) \leq 1$.

(A5)对某些$\theta >0$,有$k_i \triangleq \inf\limits_{t_0 \leq s \leq t}\int_{s}^{s+\theta}a_i(v){\rm d}v >0$, $i=1, 2, \cdots, n.$

定义2.1  如果对任意常数$\varepsilon>0$, $t_0 \geq 0$,存在常数$\delta(\varepsilon)>0$,使得对所有$t \geq t_0$, $\parallel x(t, t_0, \varphi) \parallel \leq \varepsilon$$\sup\limits_{t_0 -\tau \leq s \leq t_0}\parallel \varphi(s) \parallel<\delta(\varepsilon)$成立,则称系统(2.1)关于局部状态$x(t)$一致有界.

定义2.2  如果存在一个紧集$\Omega \subset {\Bbb R} ^n$,对任意初始值$\varphi \in C [(-\infty, t_0], {\Bbb R} ^n] $,成立

其中$x(t)=x(t, t_0, \varphi)$, $d(x(t), \Omega)$表示在${\Bbb R} ^n$$x(t)$$\Omega$的距离,则称$\Omega$为系统(2.1)的全局吸引集.

定义2.3  如果系统(2.1)是一致有界并且有一个全局吸引集$\Omega$,则称其为Lagrange稳定.

定义2.4[28]  设$f \in C({\Bbb R} _+, {\Bbb R} _+)$,对任意给定的$\eta$和任意$\varepsilon >0$,存在常数$\beta$, $T$$\alpha$,使得,对任意$ t \geq \alpha, $下式成立

则称$f(t, s) \in UC_t$.

引理2.1[37]  对任意非负矩阵$A \geq 0$,如果$\rho(A)<1$,则$(E-A)^{-1}\geq 0$.

引理2.2  令$G(t, t_0) \in C({\Bbb R} \times {\Bbb R} , {{\Bbb R} _+^n})$, $B \in {{\Bbb R} _+^{n \times n}}$, $Q(t, s) \in C({\Bbb R} \times {\Bbb R} , {{\Bbb R} _+^{n \times n}})$, $I=(i_1, i_2, \cdots, i_n)^T \geq 0$, $\varphi(t) \in C((-\infty, t_0], {{\Bbb R} _+^n})$, $\alpha_1$是常数, $x(t) \in C({\Bbb R} , {{\Bbb R} _+^n})$是如下时滞积分不等式的一个解

$\begin{eqnarray} && x(t)\leq G(t, t_0)+B{{\parallel x(t)\parallel}_ {\tau(t)}} +\int_{\alpha_1}^t {Q(t, s){\parallel x(s)\parallel}_{\tau(s)}{\rm d}s}+I, \nonumber\\ && x(t)\leq \varphi(t), \;\;\;\;\;\;\;\;\;\;\;\; t\neq t_0. \end{eqnarray} $

若满足如下条件

(ⅰ) $G \triangleq \sup_{t_0 \leq s \leq +\infty}G(s, t_0)$,存在常数非负矩阵$P$满足对$t \geq t_0, $

$\begin{eqnarray} \int_{\alpha_1}^t Q(t, s){\rm d}s \leq P .\end{eqnarray}$

(ⅱ) $\bar P=P+B$, $\rho(\bar P)<1$.

则存在常数向量$M>0$,使得对于$t \geq t_0, $

$\begin{eqnarray}\;\;x(t)<(E_n-\bar P)^{-1}(M+I).\end{eqnarray}$

  由条件(ⅱ)和引理2.1, $(E_n-\bar P)^{-1}$存在并且$(E_n-\bar P)\geq 0.$由此存在一个常数向量$\bar G >G$,使得

假设(2.3)式不成立,则一定存在一个常数$t_1>t_0$和一些整数$\alpha \in \{1, 2, \cdots, n\}$使得

$x_{\alpha}(t_1)=\{(E_n-\bar P)^{-1}(M+I)\}_{\alpha}, $

$x(t)\leq (E_n-\bar P)^{-1}(M+I), t \leq t_1, $

其中$\{\cdot \}_i$表示向量$\{\cdot \}$的第$i$个元素.

不失一般性,我们假设存在一个常数$t_1$, $t_0<t_1$,及一些整数$\alpha$满足(2.5)和(2.6)式.

考虑到表达式(2.2), (2.4)和$G(t_1, t_0)<M$,可得

$\begin{eqnarray}x_{\alpha}(t_1) & \leq & \{G(t_1, t_0)+B \parallel x(t_1)\parallel_{\tau(t_1)} +\int_{\alpha_1}^{t_1}Q(t, s)\parallel x(s)\parallel_{\tau(s)}{\rm d}s+I\}_{\alpha}\nonumber\\ \nonumber & < &\{M+[B+\int_{\alpha_1}^{t_1}Q(t, s){\rm d}s](E_n-\barP)^{-1}(M+I)+I\}_{\alpha}\\ \nonumber & \leq & \{M+[B+P](E_n-\barP)^{-1}(M+I)+I\}_{\alpha}\\ \nonumber & = & \{M+\bar P(E_n-\barP)^{-1}(M+I)+I\}_{\alpha}\\ \nonumber & = & \{(\bar P(E_n-\barP)^{-1}+E_n)(M+I)\}_{\alpha}\\ \nonumber & = & \{(E_n-\barP)^{-1}(\bar P+(E_n-\bar P)E_n)(M+I)\}_{\alpha}\\ & = &\{(E_n-\bar P)^{-1}(M+I)\}_{\alpha}, \end{eqnarray}$

这与等式(2.5)矛盾,因此不等式(2.3)成立.证明完毕.

注2.1  与文献[28-33]中的积分不等式相比较,引理2.2中的时滞积分不等式更适用于神经型Hopfield神经网络模型.

3 主要结论

定理3.1  假设(A1)-(A4)成立,则系统(2.1)一致有界.

  由条件(A1),对任意$ t \geq t_0$,有

从而

$ \begin{eqnarray} |x(t)| & \leq & {\rm e}^{-\int_{t_0}^t a(v){\rm d}v}|\varphi(t_0)|+\int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}B(s)L \parallel x(s)\parallel _{\tau(s)}{\rm d}s \nonumber \\&&+\int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}C(s)L | \dot{x}(s-\tau(s))|{\rm d}s +\int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}\bar{I}(s){\rm d}s. \end{eqnarray} $

由(A2)可得

$ \begin{eqnarray} \int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}C(s)L| \dot{x}(s-\tau(s))|{\rm d}s &\leq &\Sigma \int_{t_0}^t| \dot{x}(s-\tau(s))|{\rm d}s \nonumber\\ &\leq& \frac{\Sigma}{1-\tau}(\parallel x(t)\parallel _{\tau(t)}+|\varphi(t_0-\tau(t_0))|). \end{eqnarray} $

由(3.1)和(3.2)式,可得

$ \begin{eqnarray} |x(t)| & \leq & {\rm e}^{-\int_{t_0}^t a(v){\rm d}v}|\varphi(t_0)|+\frac{\Sigma}{1-\tau}|\varphi(t_0-\tau(t_0))|+\frac{\Sigma}{1-\tau}\parallel x(t)\parallel _{\tau(t)} \nonumber \\&& +\int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}LB(s) \parallel x(s)\parallel _{\tau(s)}{\rm d}s+\int_{t_0}^t {\rm e}^{-\int_s^t a(v){\rm d}v}\bar{I}(s){\rm d}s. \end{eqnarray} $

结合(3.3)式,条件(A3), (A4)和引理2.2,可得存在一个常数向量$K>0$,使得

$\begin{eqnarray}|x(t)|<(E_n-\bar{\Pi})^{-1}(\bar{\Pi}+K+I), \;\;\;\; \forall t \geq t_0, \end{eqnarray} $

即系统(2.1)的解关于局部状态$x(t)$一致有界.

注3.1  由(3.4)式易得每个解的分量的有界性.

定理3.2  假设满足(A1)-(A5),则集合

为系统(2.1)的全局吸引集,即系统(2.1)为全局Lagrange稳定.

  由定理3.1,存在一个非负常数向量$\delta \in {{\Bbb R} }^n $,使得

$\begin{eqnarray}\lim\limits_{t \rightarrow +\infty}\sup |x(t)|=\delta \leq (E_n-\bar\Pi)^{-1}(\Pi+K+I).\end{eqnarray}$

下面我们将证明$\delta \in \Omega$.

由条件(A5)和定义2.4, $ {\rm e}^{-\int_s^t a_i(v){\rm d}v} \in UC_t, \; i=1, 2, \cdots, n$.事实上,取$T=m\theta$,存在整数$n$满足$n\theta \leq t-T-t_0 \leq (n+1) \theta$,可得

$\begin{eqnarray}\int_{t_0}^{t-T}{\rm e}^{-\int_s^t a_i(v){\rm d}v}{\rm d}s &\leq&\nonumber \sum\limits_{j=0}^n \int_{t_0+j\theta} ^{t_0+(j+1)\theta}{\rm e}^{-\int_s^t a_i(v){\rm d}v}{\rm d}s \\ &\leq&\nonumber \sum\limits_{j=0}^n \theta {\rm e}^{-\int_{t_0+(j+1)\theta} ^{t_0+(n+m)\theta} a_i(v){\rm d}v}\\ &\leq&\nonumber \theta {\rm e}^{-mk_i}({\rm e}^{k_i}+1+{\rm e}^{-k_i}+\cdots+{\rm e}^{-(n-1)k_i})\\ & \leq &\frac{\theta {\rm e}^{-(m-1){k_i}}}{1-{\rm e}^{-k_i}}. \end{eqnarray} $

$m=0$,可得

$\begin{eqnarray} \int_{t_0} ^{t}{\rm e}^{-\int_s^t a_i(v){\rm d}v}{\rm d}s &=&\nonumber\int_{t_0} ^{t_0+n \theta}{\rm e}^{-\int_s^t a_i(v){\rm d}v}{\rm d}s+ \int_{t_0+n \theta} ^{t }{\rm e}^{-\int_s^t a_i(v){\rm d}v}{\rm d}s \\&\leq&\nonumber \theta(1+{\rm e}^{-k_i}+\cdots+{\rm e}^{-(n-1)k_i})+\theta \\&\leq&\theta+\frac{\theta}{1-{\rm e}^{-k_i}}. \end{eqnarray}$

由(3.6)和(3.7)式,可推得对于$i=1, 2, \cdots, n$,有${\rm e}^{-\int_s^t a_i(v){\rm d}v}\in UC_t$.易得${\rm e}^{-\int_s^t a_i(v){\rm d}v}a_i(s)\in UC_t$, $i=1, 2, \cdots, n$.因此,对于任意$\epsilon>0$$\varepsilon=(1, 1, \cdots, 1)\in {\Bbb R} ^n$,有正数$A$和常数矩阵$R$,使得对任意$ t>t_0+A$,有

$ \begin{equation} {\rm e}^{-\int_{t_0}^t a(v){\rm d}v}|\varphi(t_0)|<\frac{\epsilon\varepsilon}{4}, \;\;\;\; \int_{t-A} ^{t}{\rm e}^{-\int_{s}^t a(v){\rm d}v}{\rm d}s\leq R, \end{equation} $

$\begin{equation} \int_{t_0} ^{t-A}{\rm e}^{-\int_{s}^t a(v){\rm d}v}B(s)L(E_n-\bar{\Pi})^{-1}(\Pi+K+I){\rm d}s<\frac{\epsilon\varepsilon}{4}. \end{equation} $

根据上极限的定义和$ \lim\limits_{t \rightarrow +\infty}(t-\tau(t))=+\infty$,存在充分大$t_2\geq {t_0+2A}$,满足

$\begin{equation} \parallel x(t) \parallel_{\tau(t)}<\delta+\epsilon\varepsilon, \;\;\;\;t \geq t_2, \end{equation} $

其中$ \tau(t)=2A+\sup\limits_{t-2A\leq s\leq t}\tau(s)$.

因此,结合条件(A3), (3.3)和(3.8)-(3.10)式,可得对任意$ t \geq t_2$,有

再由(3.5)式和上极限定义,存在$t_3\geq t_2$满足$|x(t_3)|>\delta-\epsilon\varepsilon$.

$\epsilon \rightarrow 0$,可得

$\delta \in \Omega.$因此集合$\Omega $为系统(2.1)的全局吸引集,即系统(2.1)为全局Lagrange稳定.

注3.2  显然,如果系统(2.1)有一个全局吸引集,则系统(2.1)的解一致有界.

注3.3  若把平衡点作为吸引集的特殊情况,则Lyapunov稳定为Lagrange稳定的特例.因此,如果系统存在平衡点,定理3.2的结论课用于讨论Lyapunov稳定性.

定理3.3  假设定理3.2的所有条件成立及$I_i(t)=0, i=1, \cdots, n$.$x(t)=0$为系统(2.1)的平衡点,并且全局渐近稳定.

4 数值举例

例4.1  考虑如下模型

$\begin{eqnarray}\left \{\begin{array}{lll}{\dot{x}_{1}(t)}&=&-x_{1}(t)+0.2165\sin(t) x_{1}(t) +0.2165\cos(t) x_{2}(t)+0.0301\dot{x}_{1}(t-\tau)\\&&+0.1624\dot{x}_{2}(t-\tau)+\sin(t), \\{\dot{x}_{2}(t)}&=&-2x_{2}(t)+0.0005 x_{1}(t)+0.3\dot{x}_{1}(t-\tau)+0.5\dot{x}_{2}(t-\tau)+\arctan(t), \end{array}\right.\end{eqnarray}$

其中$ t \geq 0, $$\tau=0.05$,初始值为$\varphi_1(t)=\varphi_2(t)=40.$显然

$\bar{I}(t)=(|\sin(t)|, |\arctan(t)|)^T.$经过计算,可得

$\rho(\bar{\Pi})=0.8575<1$.

定理3.2中所有的条件都满足,所以存在$k=\bigg(\begin{array}{c} 0.2857\\ 0.4 \end{array}\bigg), $使得

为系统(4.1)的全局吸引集.

图 1

图 1   系统解的两个分量的轨迹


5 结论

基于积分不等式和非负矩阵的性质,考虑了变时滞神经型Hopfield神经网络模型,得到了保证系统Lagrange稳定和全局吸引集的充分条件.并且,本文的方法可以用来研究平衡点的全局渐近稳定性.

参考文献

Huang Y , Yang X S .

Hyperchaos and bifurcation in a new class of four-dimensional Hopfield neural networks

Neurocomputing, 2006, 69: 1787- 1795

DOI:10.1016/j.neucom.2005.11.001      [本文引用: 1]

He W , Cao J .

Stability and bifurcation of a class of discrete-time neural networks

Appl Math Model, 2007, 31: 2111- 2122

DOI:10.1016/j.apm.2006.08.006     

Huang W , Huang Y .

Chaos of a new class of Hopfield neural networks

Appl Math Comput, 2008, 206: 1- 11

URL    

Kaslik E , Balint St .

Bifurcation analysis for a discrete-time Hopfield neural network of two neurons with two delays and self-connections

Chaos Soliton Fract, 2009, 39: 83- 91

DOI:10.1016/j.chaos.2007.01.126     

Zheng P S , Tang W S , Zhang J X .

Some novel double-scroll chaotic attractors in Hopfield networks

Neurocomputing, 2010, 73: 2280- 2285

DOI:10.1016/j.neucom.2010.02.015     

Marichal R L , Gonzalez E J , Marichal G N .

Hopf bifurcation stability in Hopfield neural networks

Neural Netw, 2012, 36: 51- 58

DOI:10.1016/j.neunet.2012.09.007     

Mazrooei-Sebdani R , Farjami S .

Bifurcations and chaos in a discrete-time-delayed Hopfield neural network with rings tructures and different internal decays

Neurocomputing, 2013, 99: 154- 162

DOI:10.1016/j.neucom.2012.06.007     

Akhmet M , Fen M .

Generation of cyclic/toroidal chaos by Hopfield neural networks

Neurocomputing, 2014, 145: 230- 239

DOI:10.1016/j.neucom.2014.05.038     

Mazrooei-Sebdani R , Farjami S .

On a discrete-time-delayed Hopfield neural network with ring structures and different internal decays: Bifurcations analysis and chaotic behavior

Neurocomputing, 2015, 151: 188- 195

DOI:10.1016/j.neucom.2014.06.079      [本文引用: 1]

Wang Q , Fang Y Y , Li H , Su L J , Dai B X .

Anti-periodic solutions for high-order Hopfield neural networks with impulses

Neurocomputing, 2014, 138: 339- 346

DOI:10.1016/j.neucom.2014.01.028      [本文引用: 1]

Yang L , Li Y K .

Existence and exponential stability of periodic solution for stochastic Hopfield neural networks on time scales

Neurocomputing, 2015, 167: 543- 550

DOI:10.1016/j.neucom.2015.04.038     

Wang C .

Piecewise pseudo-almost periodic solution for impulsive non-autonomous high-order Hopfield neural networks with variable delays

Neurocomputing, 2016, 171: 1291- 1301

DOI:10.1016/j.neucom.2015.07.054      [本文引用: 1]

Kaslik E , Sivasundaram S .

Multistability in impulsive hybrid Hopfield neural networks with distributed delays

Nonlinear Anal-Real, 2011, 12: 1640- 1649

DOI:10.1016/j.nonrwa.2010.10.018      [本文引用: 2]

Park M J , Kwon O M , Park J H , Lee S M .

Simplified stability criteria for fuzzy Markovian jumping Hopfield neural networks of neutral type with interval time-varying delays

Expert Syst Appl, 2012, 39: 5625- 5633

DOI:10.1016/j.eswa.2011.11.055     

Wang H , Yu Y G , Wen G G .

Stability analysis of fractional-order Hopfield neural networks with time delays

Neural Netw, 2014, 55: 98- 109

DOI:10.1016/j.neunet.2014.03.012     

Wang H , Yu Y G , Wen G G , et al.

Global stability analysis of fractional-order Hopfield neural networks with time delay

Neurocomputing, 2015, 154: 15- 23

DOI:10.1016/j.neucom.2014.12.031     

Dharani S , Rakkiyappan R , Cao J D .

New delay-dependent stability criteria for switched Hopfield neural networks of neutral type with additive time-varying delay components

Neurocomputing, 2015, 151: 827- 834

DOI:10.1016/j.neucom.2014.10.014     

Song B , Zhang Y , Shu Z , Hu F N .

Stability analysis of Hopfield neural networks perturbed by Poisson noises

Neurocomputing, 2016, 196: 53- 58

DOI:10.1016/j.neucom.2016.02.034     

Zhang S , Yu Y G , Wang Q .

Stability analysis of fractional-order Hopfield neural networks with discontinuous activation functions

Neurocomputing, 2016, 171: 1075- 1084

DOI:10.1016/j.neucom.2015.07.077      [本文引用: 2]

Liao X X , Luo Q , Zeng Z G .

Positive invariant and global exponential attractive sets of neural networks with time-varying delays

Neurovomputing, 2008, 71: 513- 518

DOI:10.1016/j.neucom.2007.07.017      [本文引用: 2]

Crauel H , Debussche A , Flandoli F .

Random attractors

J Dynam Differ Equ, 1995, 9: 307- 341

URL    

Kloeden P , Stonier D J .

Cocycle attractors in nonautonomously perturbed dif- ferential equations

Dynam Continuous Discrete Impulsive Systems, 1998, 4: 211- 226

URL    

Cheban D N .

Dissipative functional differential equations

Bull Acad Sci Rep Moldova Mat, 1991, 2 (5): 3- 12

URL    

Caraballo T , Marn-Rubio P , Valero J .

Autonomous and non-autonomous attractors for differential equations with delays

J Differ Equ, 2005, 208: 9- 41

DOI:10.1016/j.jde.2003.09.008     

Caraballo T , Kloeden P E , Real J .

Pullback and forward attractors for a damped wave equation with delays

Stoch Dyn, 2004, 4: 405- 423

DOI:10.1142/S0219493704001139     

Caraballo T , Real J .

Attractors for 2D-Navier-Stokes models with delays

J Differ Equ, 2004, 205: 271- 297

DOI:10.1016/j.jde.2004.04.012     

Zhang L X , Boukas E K , Haidar A .

Delay-range-dependent control synthesis for time-delay systems with actuator saturation

Automatica, 2008, 44: 2691- 2695

DOI:10.1016/j.automatica.2008.03.009      [本文引用: 2]

Xu D Y .

Integro-differential equations and delay integral inequalities

Tohoku Math J, 1992, 44: 365- 378

DOI:10.2748/tmj/1178227303      [本文引用: 3]

Zhao H Y .

Invariant set and attractor of nonautonomous functional differential systems

J Math Anal Appl, 2003, 282: 437- 443

DOI:10.1016/S0022-247X(02)00370-0     

Xu D Y , Yang Z C .

Attracting and invariant sets for a class of implusive functional defferential equations

J Math Anal Appl, 2007, 329: 1036- 1044

DOI:10.1016/j.jmaa.2006.05.072     

Tian H .

The exponential asymtotic stability of singularly perturbed delay differential equations with a bounded lag

J Math Anal Appl, 2002, 270: 143- 149

DOI:10.1016/S0022-247X(02)00056-2     

Xu D Y , Yang Z G .

Exponential stability of nonlinear implusive neutral differential equations with delays

Nonlinear Anal Theory Methods Appl, 2007, 67: 1426- 1439

DOI:10.1016/j.na.2006.07.043     

Zhu W .

Global exponential stability of impulisive reaction-diffusion equation with variable delays

Appl Math Comput, 2008, 205: 362- 369

[本文引用: 1]

Teng L Y , Xu D Y .

Global attracting set for non-autonomous neutral networks with distributed delays

Neurocomputing, 2012, 94: 64- 67

DOI:10.1016/j.neucom.2012.04.020     

Liu L , Zhu Q X , Feng L C .

Lagrange stability for delayed recurrent neural networks with Markovian switching based on stochastic vector Halandy inequalities

Neurocomputing, 2018, 275: 1614- 1621

DOI:10.1016/j.neucom.2017.10.006     

Popa A C .

Global exponential stability of neutral-type octonion-valued neural networks with time-varying delays

Neurocomputing, 2018, 309: 117- 133

DOI:10.1016/j.neucom.2018.05.004     

LaSalle J P. The Stability of Dynamical System. Philadelphia: SIAM, 1976

[本文引用: 2]

/