吴海霞
摘要:针对一类具有区间时滞和随机干扰的BAM神经网络的全局渐近稳定性问题,通过构造合适的Lyapunov-Krasovskii泛函,应用随机分析和自由权值矩阵方法,并考虑时滞区间范围,得到了新的稳定性充分条件。该条件能够保证时滞BAM神经网络在均方意义下是全局渐近稳定的,同时适用于快时滞和慢时滞,其适用范围更广。最后,通过一个仿真实例证明了定理的有效性。
关键词:双向联想记忆神经网络;全局渐近稳定性;区间时滞;线性矩阵不等式
中图分类号:TP183 文献标识码:A 文章编号:1009-3044(2014)19-4544-06
1 概述
双向联想记忆神经网络(Bidirectional Associative Memory, BAM)已经成功应用于诸多领域,如模式识别、图像处理、自动控制、模型辨识和优化问题等。所有这些成功的应用都必须依赖于神经网络的稳定性。众所周知,许多实际系统的数学模型中均含有时滞的现象,在BAM神经网络中也不例外。例如,在模拟神经网络电路实现中,由于运放器的开关速度限制会产生时滞,神经网络中的轴突信号传输延迟也会产生时滞。这些都会导致不良的动态网络特性,即系统失稳、产生振荡甚至混沌,造成系统性能指标的下降。目前,时滞BAM神经网络的稳定性分析问题已经取得了大量的研究成果[1-7],其中时滞类型包括常时滞、变时滞与分布时滞等。
事实上,随机因素确实存在于很多的实际系统,比如物理电路、生物系统等。为了抵消这些不确定因素的影响,必须将系统描述为随机系统。一般而言,在生物神经系统中,突触递质的传递会受到随机噪声和其他一些概率事件的影响,这些随机扰动理所当然的会影响神经系统的稳定性。系统建模一个基本原则是尽可能模拟实际情况,因此,在BAM神经网络的建模中考虑随机扰动的也是不可避免的。截止目前,已有许多国内外学者致力于带有随机干扰的时滞BAM神经网络的稳定性有研究工作[8-9]。然而,对于某些系统,在时滞非零时是稳定的,时滞为零时却是不稳定的。因此,研究非零时滞系统的稳定性十分重要,非零时滞可以将时滞定义在一个区间内,应用范围更广。
本文将应用随机分析和自由权值矩阵方法,构造合适的Lyapunov-Krasovskii泛函并考虑时滞区间,研究新的稳定性判定准则,用以保证时滞BAM神经网络在均方意义下是全局渐近稳定的。
2 系统模型及引理
考虑以下带区间时滞的双向联想记忆神经网络模型:
[du1i(t)dt=-a1iu1i(t)+j=1mw1jif1j(u2j(t-τ2j(t)))+Ii, i=1,2,…,n,du2j(t)dt=-a2ju2j(t)+i=1nw2ijf2i(u1i(t-τ1i(t)))+Jj, j=1,2,…,m,] (1)
其中[u1i(t)]和[u2j(t)]分别是第[i]个神经元和第[j]个神经元的状态;[f1j(?)],[f2i(?)]分别表示第[i]个神经元和第[j]个神经元的激活函数;[Ii,Jj]表示在[t]时刻的外部输入;[a1i,a2j]为正数,分别表示第[i]个神经元和第[j]个神经元的被动衰减率;[w1ji,w2ij]表示突触连接权值;[τ1i(t),τ2j(t)]为时变时滞。有关系统(1)的初始条件假设如下:
[u1i(s)=?u1i(s), t∈-τ1,0, i=1,2,…,n,u2j(s)=?u2j(t), t∈-τ2,0, j=1,2,…,m.]
假设1 在系统(1)神经元激活函数[f1j(?)]和[f2i(?)]有界,且存在正数[l(1)j>0]和[l(2)i>0]满足:
[f1j(ξ1)-f1j(ξ2)≤l(1)jξ1-ξ2, f2i(ξ1)-f2i(ξ2)≤l(2)iξ1-ξ2, ?ξ1,ξ2∈R, i=1,2,…,n, j=1,2,…,m.]
按照通常做法,假设[u1?=u?11,u?12,…,u?1nT, u2?=u?21,u?22,…,u?2mT]是系统(1)的平衡点。为了简化证明过程,通过变换[x1i(t)=u1i(t)-u?1i,][x2j(t)=u2j(t)-u?2j,] [f2i(x1i(t))=f2i(x1i(t)+u?1i)-f2i(u?1i),] [f1j(u2j(t))=f1j(u2j(t)+u?2j)-f1j(u?2j), ]转移系统(1)的平衡点到新系统的原点,得到以下系统模型:
[x1i(t)=-a1ix1i(t)+j=1mw1jif1j(x2j(t-τ2j(t))), i=1,2,…,n,x2j(t)=-a2jx2j(t)+i=1nw2ijf2i(x1i(t-τ1i(t))), j=1,2,…,m. ] (2)
将式(2)改写为矩阵形式,则有:
[x1(t)=-A1x1(t)+W1f1(x2(t-τ2(t))),x2(t)=-A2x2(t)+W2f2(x1(t-τ1(t))),] (3)
其中[x1(t)=x11(t),x12(t),…,x1n(t)T, x2(t)=x21(t),x22(t),…,x2m(t)T, ] [A1=diaga11,a12,…,a1n, ][ A2=diaga21,a22,…,a2m],
[W1=w1jim×nT, W2=w2ijn×mT, f1(x2)=f11(x2),f12(x2),…,f1m(x2)T,][f2(x1)=f21(x1),f22(x1),…,f2n(x1)T,]
[τ1(t)=τ11(t),x12(t),…,τ1n(t)T,] [τ2(t)=τ21(t),τ22(t),…,τ2m(t)T.]
显然,神经元激活函数具有如下性质:endprint
[fT1(x2(t))f1(x2(t))≤x2T(t)LT1L1x2(t),fT2(x1(t))f2(x1(t))≤x1T(t)LT2L2x1(t),] (4)
其中[L1=diagl(1)1,l(1)2,…,l(1)m],[L2=diagl(2)1,l(2)2,…,l(2)n.]
接下来,将考虑如下具有区间时滞和随机干扰的BAM神经网络模型:
[dx1(t)=-A1x1(t)+W1f1(x2(t-τ2(t)))dt +C1x1(t)+D1x2(t-τ2(t))dω(t),dx2(t)=-A2x2(t)+W2f2(x1(t-τ1(t)))dt +C2x2(t)+D2x1(t-τ1(t))dω(t),] (5)
其中[ω(t)=ω1(t),ω2(t),…,ωl(t)T]是一个定义在完备概率空间[(Ω,?,?tt≥0,)]上的布朗运动(Brownian motion)。
假设2 时滞[τ1(t)]和[τ2(t)]满足:
[0≤τ_1≤τ1(t)≤τ1, 0≤τ_2≤τ2(t)≤τ2,] (6)
[τ1(t)≤μ1, τ2(t)≤μ2,] (7)
其中[0≤τ_1<τ1, 0≤τ_2<τ2, μ1]和[μ2]为正常量。
引理1 对于任意适当维数常数矩阵[D]和[N],矩阵[F(t)]满足[FT(t)F(t)≤I],有:
(i) 对任意常数[ε>0,][DF(t)N+NTFT(t)DT≤ε-1DDT+εNTN,]
(ii) 对任意常数[P>0],[2aTb≤ aTP-1a+bTPb.]
引理2 [10] 随机微分方程的平凡解
[d(x(t),y(t),t)=G(x(t),y(t),t)dt+H(x(t),y(t),t)dω(t)] [t∈t0,T]
有:
[x(t)=?x(t) t∈-τ,0, y(t)=?y(t) t∈-ρ,0,] [G:R+×Rn×Rn→Rn] 和[H:R+×Rn×Rn→Rn×m]在概率上是全局渐近稳定的,假如存在函数[V(x(t),y(t),t)∈R+×Rn×Rn]在Lyapunov意义上是正定的并且满足
[?V(x(t),y(t),t)=?Vdt+gradVG+12traceHHTHess(V)<0.]
矩阵[Hess(V)]表示Hessian矩阵的二阶偏导数。
3 全局渐近稳定性
首先,定义
[g1(t)=-(A1+ΔA1)x1(t)+(W1+ΔW1)f1(x2(t-τ2(t))),g2(t)=-(A2+ΔA2)x2(t)+(W2+ΔW2)f2(x1(t-τ1(t))),]
[g3(t)=C1+ΔC1x1(t)+D1+ΔD1x2(t-τ2(t)),g4(t)=C2+ΔC2x2(t)+D2+ΔD2x1(t-τ1(t)),]
那么,系统(1) 被记为:
[dx1(t)=g1(t)dt+g3(t)dω(t),dx2(t)=g2(t)dt+g4(t)dω(t).] (8)
定理1 对于给定的正常数[0≤τ_1<τ1, 0≤τ_2<τ2, μ1]和[μ2],系统(5)在均方意义下是全局渐近稳定的,如果存在矩阵, [Pi>0, Qi≥0, Ri≥0, Ti≥0, i=1,2, Zj>0, j=1,2,…,8,][N(i)j,M(i)j,][S(i)j, j=1,2, i=1,2,]和两个正常量[α1, α2,]使得以下LMI(9)成立:
[Ξ=Ξ0Ξ1Ξ2Ξ3Ξ4?-Ξ11000??-Ξ2200???-Ξ330????-Ξ44<0,] (9)
其中
[Ξ0=?1?T2U1?T3U2?T4U3?T5U4?-U1000??-U200???-U30????-U4, Ξi=τiN(i)1hiM(i)1hiS(i)1τiN(i)2hiM(i)2hiS(i)2000???000, Ξ2+i=N(i)1M(i)1S(i)1N(i)2M(i)2S(i)2000???000,]
[?1=Σ10Σ500P1W1M(1)1-S(1)100?Σ20Σ6P2W2000M(2)1-S(2)1??Σ3000M(1)2-S(1)200???Σ40000M(2)2-S(2)2????-α1I00000?????-α2I0000??????-Q1000???????-R100????????-Q20?????????-R2,]
[?2=-A10000W10000, ?3=0-A200W200000,]
[?4=C100D1000000, ?5=0C2D20000000,]
[U1=τ1Z1+h1Z3, U2=τ2Z2+h2Z4, U3=P1+τ1Z5+h1Z7,U4=P2+τ2Z6+h2Z8, Ξ11=diagτ1Z1, h1Z3, h1(Z1+Z3),Ξ22=diagτ2Z2, h2Z4, h2(Z2+Z4),]
[Ξ33=diagZ5, Z7, Z5+Z7, Ξ44=diagZ6, Z8, Z6+Z8,]
[Σi=-PiAi-AiTPi+Qi+Ri+Ti+N(i)1+N(i)1T,Σ2+i=-(1-μi)Ti+S(i)2+S(i)2T-N(i)2-N(i)2T -M(i)2-M(i)2T+αiLT3-iL3-i,]
[Σ4+i=S(i)1-N(i)1-M(i)1+N(i)2T, hi=τi-τ_i, i=1,2.]
证明. 构造如下Lyapunov-Krasovskii泛函:endprint
[V(x1(t),x2(t))=i=12V1(x1(t),x2(t))+V2(x1(t),x2(t))+V3(x1(t),x2(t)),V1(x1(t),x2(t))=xTi(t)Pixi(t),]
[V2(x1(t),x2(t))=t-τ_itxTi(s)Qixi(s)ds+t-τitxTi(s)Rixi(s)ds+t-τi(t)txTi(s)Tixi(s)ds,]
[V3(x1(t),x2(t))=-τi0t+θtgiT(s)Zigi(s)dsdθ+-τi-τ_it+θtgiT(s)Z2+igi(s)dsdθ +-τi0t+θtg2+iT(s)Z4+ig2+i(s)dsdθ+-τi-τ_it+θtg2+iT(s)Z6+ig2+i(s)dsdθ.] (10)
由牛顿-莱布尼茨(Leibniz-Newton)公式可知,对于任意具有适当维数的矩阵[N(i)j,M(i)j,S(i)j, i=1,2,][j=1,2,],以下等式成立:
[0=2xTi(t)N(i)1+xTi(t-τi(t))N(i)2 ×xi(t)-xi(t-τi(t))-t-τi(t)tgi(s)ds-t-τi(t)tg2+i(s)dω(s),] (11)
[0=2xTi(t)M(i)1+xTi(t-τi(t))M(i)2 ×xi(t-τ_i)-xi(t-τi(t))-t-τi(t)t-τ_igi(s)ds-t-τi(t)t-τ_ig2+i(s)dω(s),] (12)
[0=2xTi(t)S(i)1+xTi(t-τi(t))S(i)2 ×xi(t-τi(t))-xi(t-τi)-t-τit-τi(t)gi(s)ds-t-τit-τi(t)g2+i(s)dω(s).] (13)
应用引理1(ii),对任意矩阵[Zj≥0, j=1,2,…,8,],下列不等式成立:
[-2ξT(t)N(i)t-τi(t)tgi(s)ds≤τiξT(t)N(i)Z-1iN(i)Tξ(t)+t-τi(t)tgTi(s)Zigi(s)ds,] (14)
[-2ξT(t)M(i)t-τi(t)t-τ_igi(s)ds≤hiξT(t)M(i)Z-12+iM(i)Tξ(t)+t-τi(t)t-τ_igTi(s)Z2+igi(s)ds,] (15)
[-2ξT(t)S(i)t-τit-τi(t)gi(s)ds≤ hiξT(t)S(i)Zi+Z2+i-1S(i)Tξ(t)+t-τit-τi(t)gTi(s)(Zi+Z2+i)gi(s)ds,] (16)
[-2ξT(t)N(i)t-τi(t)tg2+i(s)dω(s)≤ξT(t)N(i)Z-14+iN(i)Tξ(t)+t-τi(t)tgT2+i(s)dω(s)Z4+it-τi(t)tg2+i(s)dω(s),] (17)
[-2ξT(t)M(i)t-τi(t)t-τ_ig2+i(s)dω(s)ds≤ξT(t)M(i)Z-16+iM(i)Tξ(t)+t-τi(t)t-τ_igT2+i(s)dω(s)Z6+it-τi(t)t-τ_ig2+i(s)dω(s),] (18)
[-2ξT(t)S(i)t-τit-τi(t)g2+i(s)dω(s)≤ ξT(t)S(i)Z4+i+Z6+i-1S(i)Tξ(t) +t-τit-τi(t)gT2+i(s)dω(s)Z4+i+Z6+it-τit-τi(t)g2+i(s)dω(s),] (19)
其中
[N(i)=N(i)1T N(i)2T 0 0 0 0 0 0 0 0T,M(i)=M(i)1T M(i)2T 0 0 0 0 0 0 0 0T,]
[S(i)=S(i)1T S(i)2T 0 0 0 0 0 0 0 0T.]
由式(4)有
[fTi(x3-i(t-τ3-i(t)))fi(x3-i(t-τ3-i(t)))≤x3-iT(t-τ3-i(t))×LTiLix3-i(t-τ3-i(t)), i=1,2.] (20)
沿着系统(5)解的轨迹,对[?V]求时间的导数:
[?V(x1(t),x2(t))=i=12?V1(x1(t),x2(t)) +?V2(x1(t),x2(t))+?V3(x1(t),x2(t)),]
[?V1(x1(t),x2(t))=2xTi(t)Pi-Aixi(t)+Wifi(x3-i(t-τ3-i(t))) +gT2+i(t)Pig2+i(t),] (21)
[?V2(x1(t),x2(t))≤xTi(t)(Qi+Ri)xi(t)-xTi(t-τ_i)Qixi(t-τ_i) -xTi(t-τi)Rixi(t-τi)+xTi(t)Tixi(t) -(1-μi)xTi(t-τi(t))Tixi(t-τi(t)) ,] (22)
[?V3(x1(t),x2(t))=gTi(t)τiZi+hiZ2+igi(t)-t-τitgTi(s)Zigi(s)ds -t-τit-τ_igTi(s)Z2+igi(s)ds+gT2+i(t)τiZ4+i+hiZ6+ig2+i(t) -t-τitgT2+i(s)Z4+ig2+i(s)ds-t-τit-τ_igT2+i(s)Z6+ig2+i(s)ds]
[ =gTi(t)τiZi+hiZ2+igi(t)+gT2+i(t)τiZ4+i+hiZ6+ig2+i(t) -t-τi(t)tgTi(s)Zigi(s)ds-t-τi(t)t-τ_igTi(s)Z2+igi(s)ds -t-τit-τi(t)gTi(s)(Zi+Z2+i)gi(s)ds-t-τi(t)tgT2+i(s)Z4+ig2+i(s)ds -t-τi(t)t-τ_igT2+i(s)Z6+ig2+i(s)ds-t-τit-τi(t)gT2+i(s)(Z4+i+Z6+i)g2+i(s)ds.] (23)endprint
联立式(11)-(23),可得:
[?V(x1(t),x2(t))≤ξT(t)?1+?T2U1?2+?T3U2?3+?T4U3?4+?T5U4?5 +i=12τiN(i)Z-1iN(i)T+hiM(i)Z-12+iM(i)T +hiS(i)Zi+Z2+i-1S(i)T+N(i)Z-14+iN(i)T +M(i)Z-16+iM(i)T+S(i)Z4+i+Z6+i-1S(i)Tξ(t) +t-τi(t)tgT2+i(s)dω(s)Z4+it-τi(t)tg2+i(s)dω(s) +t-τi(t)t-τ_igT2+i(s)dω(s)Z6+it-τi(t)t-τ_ig2+i(s)dω(s)]
[ +t-τit-τi(t)gT2+i(s)dω(s)Z4+i+Z6+it-τit-τi(t)g2+i(s)dω(s) -t-τi(t)tgT2+i(s)Z4+ig2+i(s)ds-t-τi(t)t-τ_igT2+i(s)Z6+ig2+i(s)ds -t-τit-τi(t)gT2+i(s)(Z4+i+Z6+i)g2+i(s)ds,] (24)
其中
[ξ(t)=xT1(t) xT2(t) xT1(t-τ1(t)) xT2(t-τ2(t)) fT2(x1(t-τ1(t))) fT1(x2(t-τ2(t))) xT1(t-τ_1) xT1(t-τ1) xT2(t-τ_2) xT2(t-τ2)T.]
由于
[Εt-τi(t)tgT2+i(s)dω(s)Z4+it-τi(t)tg2+i(s)dω(s)=Εt-τi(t)tgT2+i(s)Z4+ig2+i(s)ds,]
[Εt-τi(t)t-τ_igT2+i(s)dω(s)Z6+it-τi(t)t-τ_ig2+i(s)dω(s)=Εt-τi(t)t-τ_igT2+i(s)Z6+ig2+i(s)ds,]
[Εt-τit-τi(t)gT2+i(s)dω(s)Z4+i+Z6+it-τit-τi(t)g2+i(s)dω(s)= Εt-τit-τi(t)gT2+i(s)(Z4+i+Z6+i)g2+i(s)ds,]
则有
[Ξ=?1+?T2U1?2+?T3U2?3+?T4U3?4+?T5U4?5 +i=12τiN(i)Z-1iN(i)T+hiM(i)Z-12+iM(i)T+hiS(i)Zi+Z2+i-1S(i)T +N(i)Z-14+iN(i)T+M(i)Z-16+iM(i)T+S(i)Z4+i+Z6+i-1S(i)T<0,]
对所有[x1(t),x2(t)]([x1(t)=x2(t)=0]除外),有
[ΕdV(x1(t),x2(t))=Ε?V(x1(t),x2(t))dt<0],
其中[Ε]为数学期望算子。由Schur补充条件,上式等价于[Ξ<0]。那么,由Lyapunov稳定性定理可知,系统(5)均方意义下是全局渐近稳定的。定理1证明完毕。
4 仿真算例
本文将用一个仿真算例说明所得结论的有效性。
例1 考虑具有区间时滞和随机干扰BAM神经网络模型系统(5),其参数为
激活函数为[fi(x)=12x+1-x-1,i=1,2,]
时滞为
那么,显然对任意i和j,有li-lj=1, 即L1=L2=I。同时
运用定理1,通过Matlab求解 (9)式,容易判定系统(5)在均方意义下是全局渐近稳定的。所得的部分可行解如下:
[P1=1.1886-0.0918-0.09182.2512, P2=2.9766-0.3644-0.36442.6715,α1=0.6267, α2=1.3079.]
5 结束语
本文得到了一个BAM神经网络的全局渐近稳定性的新结果,该神经网络是带有区间时滞和随机干扰的。与现有大部分文献相比,该文的稳定性判定准则去掉了导数上界的限制,既可以适用于快时滞的情况,也适用于慢时滞的情况。最后,一个仿真算例验证了结论的有效性。
参考文献:
[1] Senan S,Arik S,Liu D.New robust stability results for bidirectional associative memory neural networks with multiple time delays[J].Applied Mathematics and Computation,2012,218(23):11472-11482.
[2] Zhang Z,Liu K,Yang Y.New LMI-based condition on global asymptotic stability concerning BAM neural networks of neutral type[J].Neurocomputing,2012,81:24-32.
[3] Li X,Jia J.Global robust stability analysis for BAM neural networks with time-varying delays[J].Neurocomputing,2013,120:499-503.
[4] Liu B.Global exponential stability for BAM neural networks with time-varying delays in the leakage terms[J].Nonlinear Analysis: Real World Applications,2013,14(1):559-566.
[5] Zhao Z,Liu F,Xie X,et al.Asymptotic stability of bidirectional associative memory neural networks with time-varying delays via delta operator approach[J].Neurocomputing,2013,117:40-46.
[6] Zhang A,Qiu J,She J.Existence and global exponential stability of periodic solution for high-order discrete-time BAM neural networks[J].Neural Networks,2014,50:98-109.
[7] 潘特铁,时宝,杨树杰,等.具时滞和脉冲的随机BAM型Cohen-Grossberg神经网络的稳定性分析[J].数学物理学报,2013,33(5):937-950.
[8] Du Y,Zhong S,Zhou N,et al.Exponential stability for stochastic Cohen—Grossberg BAM neural networks with discrete and distributed time-varying delays[J].Neurocomputing,2014,127:144-151.
[9] Rao R,Zhong S,Wang X.Stochastic stability criteria with LMI conditions for Markovian jumping impulsive BAM neural networks with mode-dependent time-varying delays and nonlinear reaction-diffusion[J].Communications in Nonlinear Science and Numerical Simulation,2013,19(1):258—273.
[10] Syed Ali M,Balasubramaniam P.Robust stability for uncertain stochastic fuzzy BAM neural networks with time-varying delays[J].Physics Letters A,2008,372(31):5159-5166.endprint
[4] Liu B.Global exponential stability for BAM neural networks with time-varying delays in the leakage terms[J].Nonlinear Analysis: Real World Applications,2013,14(1):559-566.
[5] Zhao Z,Liu F,Xie X,et al.Asymptotic stability of bidirectional associative memory neural networks with time-varying delays via delta operator approach[J].Neurocomputing,2013,117:40-46.
[6] Zhang A,Qiu J,She J.Existence and global exponential stability of periodic solution for high-order discrete-time BAM neural networks[J].Neural Networks,2014,50:98-109.
[7] 潘特铁,时宝,杨树杰,等.具时滞和脉冲的随机BAM型Cohen-Grossberg神经网络的稳定性分析[J].数学物理学报,2013,33(5):937-950.
[8] Du Y,Zhong S,Zhou N,et al.Exponential stability for stochastic Cohen—Grossberg BAM neural networks with discrete and distributed time-varying delays[J].Neurocomputing,2014,127:144-151.
[9] Rao R,Zhong S,Wang X.Stochastic stability criteria with LMI conditions for Markovian jumping impulsive BAM neural networks with mode-dependent time-varying delays and nonlinear reaction-diffusion[J].Communications in Nonlinear Science and Numerical Simulation,2013,19(1):258—273.
[10] Syed Ali M,Balasubramaniam P.Robust stability for uncertain stochastic fuzzy BAM neural networks with time-varying delays[J].Physics Letters A,2008,372(31):5159-5166.endprint
[4] Liu B.Global exponential stability for BAM neural networks with time-varying delays in the leakage terms[J].Nonlinear Analysis: Real World Applications,2013,14(1):559-566.
[5] Zhao Z,Liu F,Xie X,et al.Asymptotic stability of bidirectional associative memory neural networks with time-varying delays via delta operator approach[J].Neurocomputing,2013,117:40-46.
[6] Zhang A,Qiu J,She J.Existence and global exponential stability of periodic solution for high-order discrete-time BAM neural networks[J].Neural Networks,2014,50:98-109.
[7] 潘特铁,时宝,杨树杰,等.具时滞和脉冲的随机BAM型Cohen-Grossberg神经网络的稳定性分析[J].数学物理学报,2013,33(5):937-950.
[8] Du Y,Zhong S,Zhou N,et al.Exponential stability for stochastic Cohen—Grossberg BAM neural networks with discrete and distributed time-varying delays[J].Neurocomputing,2014,127:144-151.
[9] Rao R,Zhong S,Wang X.Stochastic stability criteria with LMI conditions for Markovian jumping impulsive BAM neural networks with mode-dependent time-varying delays and nonlinear reaction-diffusion[J].Communications in Nonlinear Science and Numerical Simulation,2013,19(1):258—273.
[10] Syed Ali M,Balasubramaniam P.Robust stability for uncertain stochastic fuzzy BAM neural networks with time-varying delays[J].Physics Letters A,2008,372(31):5159-5166.endprint