毛凱,時寶
(海軍航空工程學院,a.系統科學與數學研究所;b.基礎部,山東煙臺264001)
基于自由權矩陣的時滯神經網絡系統時滯相依全局穩定性分析
毛凱a,時寶b
(海軍航空工程學院,a.系統科學與數學研究所;b.基礎部,山東煙臺264001)
在激勵函數有界且滿足扇區條件的情形下研究了一類同時具有時變時滯和無窮分布時滯的神經網絡系統全局穩定性.通過構造一個包含更多信息的新的Lyapunov泛函,利用自由權矩陣更好地描述Newton-Leibniz公式中各項以及系統各項之間的關系,充分利用狀態變量、時滯函數、狀態變量以及Lyapunov泛函導數中出現的負項中隱藏的信息,并結合S-Procedure等方法以線性矩陣不等式(LMI)的形式給出了系統時滯相依全局指數穩定充分條件.文中結果去掉了以往文獻對時變時滯函數的導數具有不超過1的上界的限制條件,具有更低的保守性,且利用Matlab工具箱極易驗證條件的正確性,并推廣改進了相關文獻的結果.
時變時滯;無窮分布時滯;自由權矩陣;S-Procedure;Lyapunov泛函
眾所周知,時滯通常是引起系統不穩定或振蕩的重要原因,對于時滯神經網絡系統穩定性的研究無論在理論上還是在實踐中都具有重要意義.文獻[1-4]研究了離散定常時滯神經網絡的穩定性.文獻[5-9,13-14,17-18]研究了具有時變時滯的神經網絡的穩定性.文獻[10-11,18]研究了具有時變時滯和有限分布時滯的神經網絡系統的穩定性.文獻[12,19]研究了含有形gj(uj(s))ds的無窮分布時滯項的神經網絡系統的穩定性.但對于含有形的無窮分布時滯項的神經網絡系統的穩定性研究卻很少見.
一方面已有文獻[5-9,18]中對于時變時滯函數通常都要求其導數具有不超過1的上界的限制條件.另一方面如何降低穩定性判據的保守性已成為研究熱點.時滯相依的穩定性判據利用了時滯長度的信息,通常比時滯不相依的穩定性判據具有更低的保守性.而文獻[13-16]引入了自由權矩陣法,用以描述Newton-Leibniz公式中各項之間的關系.它充分利用狀態變量、狀態變量的導數、時滯函數以及Lyapunov泛函導數中出現的負項中隱藏的信息,能夠極大降低保守性.
基于以上分析,文中將通過構造一個新的恰當的Lyapunov泛函,利用自由權矩陣法以及SProcedure[15,21-22]研究同時具有時變時滯和形無窮分布時滯系統的全局漸近穩定性和全局指數穩定性,其中不再要求時變時滯函數的導數具有不超過1的上界,并以線性矩陣不等式(LMI)給出系統時滯相依全局指數穩定充分判據.
考慮如下具有時變時滯和無窮分布時滯的系統:對于系統(1),我們作如下假設:

H1:時變時滯函滿足0≤τ1≤τ(t)≤τ2,τ觶(t)≤μ.
H2:激勵函數有界且滿
H3:核函數kij(s)非負,
注1:通常文獻都假設0≤τ(t)≤τ2,τ觶(t)≤μ<1.文中無論對于時變時滯函數的下界還是其導數的上界的要求都更具一般性.
注2:關于激勵函數的假設H2最早出現在文獻[23]中,現在已經被廣泛采用.常以為正數,負數或零,相對于Sigmoid型和Lipschitz型激勵函數而言,顯然H2更具一般性,這樣的假設被廣泛應用于相關文獻.
由假設H2知系統存在平衡點,記為y*.作變換yj(t)=xj(t)-,則系統(1)變為

寫成矩陣向量形式即為

為方便,記Ij為第j行元素全部為1而其余元素全部為零的n×n矩陣,Ijj是第j行j列元素為1,其余元素全部為零的n×n矩陣L=diag(l1,l2,…,ln),

引理1[21]對于任意的正定常數矩陣M,常數r〉0和向量值函數x:[0,r]→Rn有

引理2(S-procedure[15,21-22])設Ti(i=0,1,2,…,k)為對稱矩陣,若存在τi≥0(i=1,2,…,p)使立,則對于任意滿足ξTTiξ≥0(i=1,2,…,k)的ξ≠0都
定理假設H1~H3成立.對于給定的τ1,τ2,μ,若存在正定矩陣P,Qi(i=1,2,3,4),Ri(i=1,2,3),正對角陣E=diag(e1,e2,…,en),H=diag(h1,h2,…,hn),半正定對角矩陣Λi=diag(λi1,λi2,…,λin),λij≥0,i=1,2;j=1,2…,n以及任意適當維數的矩陣NT=(,,…,),M1,M2使得矩陣Π<0,則系統(2)的零解全局指數穩定.其中:

證明:作Lyapunov泛函V(t)=V1(t)+V2(t)+V3(t)+V4(t),其中:

對V(t)沿系統(2)的軌跡對t求導數,并利用引理1有:



對于任意適當維數的矩陣M1,M2顯然還有下式成立.


另外,由假設H2知有:

則由引理2知,若存在Λi=diag(λi1,λi2,…,λin),λij≥0,i=1,2;j=1,2,…,n使得對任意的η(t)≠0都有下式成立,則系統(2)的零解全局漸近穩定.

由定理條件Π<0知,上式成立,從而系統(2)的零解全局漸近穩定性得證.
下面進一步證明系統(2)的零解全局指數穩定.由式(10)以及Π<0有:

再由V(t)的構造容易得:

其中α1=λmax(P)+λmax(LE),L=diag(l1,l2,…,ln),


考察函數e2λtV(t),其中λ〉0待定.有

將上式從0到t積分,得

對于上式中的二重積分和三重積分,只要交換積分次序并適當放大積分區域,不難得到:

同理,

于是,

只要取λ滿足下列條件

則由式(15)知:

由式(12)有:

利用Cauchy-Schwartz不等式,有


另外,顯然還有:

由V(t)的表達式以及式(16),式(17),式(18)得到:


通過構造一個新的Lyapunov泛函,并利用自由權重矩陣法和S-Procedure研究了一類同時具有時變時滯和無窮分布時滯的細胞神經網絡的全局指數穩定性,以LMI形式給出了系統全局漸近穩定以及指數穩定的充分判據,由Matlab工具箱能非常方便地對判據進行驗證.文章去掉了時變時滯函數的導數具有不超過1的上界的嚴格要求.自由權重矩陣的引入有利于更好地描述Newton-Leibnitz公式中各項之間、以及系統各項之間的關系,能更好地利用狀態變量、各時滯狀態變量以及狀態變量的導數之間隱藏的信息,從而極大地減少了穩定性判據的保守性,改進了相關文獻的結果.
[1]Liao X F,Chen G R,Sanchez E N.Delay-dependent exponential stability analysis of delayed neural networks:an LMI approach[J].Neural Networks,2002,5(7):855-866.
[2]Li C D,Liao X F,Zhang R.Delay-dependent exponential stability analysis of bi-directional associative memory neural networks with time delay:an LMI approach[J].Chaos,Solitons&Fractals,2005,24(4):1119-1134.
[3]Xu S Y,Lam J,Ho D W C,et al.Novel global asymptotic stability criteria for delayed cellular neural networks[J].IEEE Trans Circ Syst II,2005,52(6):349-353.
[4]Xu S Y,Lam J,Ho D W C.A new LMI condition for delaydependentasymptoticstabilityofdelayedHopfieldneural networks[J].IEEE Trans Circ Syst,2006,53(3):230-234.
[5]Arik S.An analysis of exponential stability of delayed neural networks with time-varying delays[J].Neural Networks,2004,17(7):1027-1031.
[6]Zhang Q,Wei X P,Xu J.Delay-dependent exponential stability of cellular neural networks with time-varying delays[J].Chaos,Solitons&Fractals,2005,23(4):1363-1369.
[7]Zhang H B,Li C D,Liao X F.A note on the robust stability of neural networks with time delay[J].Chaos,Solitons&Fractals,2005,25(2):357-360.
[8]Liu Jiang.Global exponential stability of Cohen-Grossberg neural networks with time-varying delays[J].Chaos,Solitons&Fractals,2005,26(3):935-945.
[9]He Y,Wang Q G,Zheng W X.Global robust stability for delayed neural networks with polytopic type uncertainties[J].Chaos,Solitons&Fractals,2005,26(5):1349-1354.
[10]Li H Y,Chen B,Zhou Q,et al.Robust exponential stability for
uncertain stochastic neural networks with discrete and distributed
time-varying delays[J].Physics Letters A,2008,372(19):3385-3394.[11]RRakkiyappan,PBalasubramaniam,SLakshmanan.Robuststability results for uncertain stochastic neural networks with discrete interval and distributed time-varying delays[J].Physics Letters A,2008,372(32):5290-5298.
[12]Ju H Park.On gobal stability criterion of neural networks with continuously distributed delays[J].Chaos,Solitons&Fractals,2008,37(2):444-449.
[13]Song Q K,Wang Z D.A delay-dependent LMI approach to dynamics analysis of discrete-time recurrent neural networks with time-varying delays[J].Physics Letters A,2007,368(1-2):134-145.
[14]Liu H L,Chen G H.Delay-dependent stability for neural networks with time-varying delay[J].Chaos,Solitons&Fractals,2007,33(1):171-177.
[15]Jiang Y H,Yang B,Wang J C,et al.Delay-dependent stability criterion for delayed Hopfield neural networks[J].Chaos,Solitons&Fractals,2009,39(5):2133-2137.
[16]Park P.A delay-dependent stability criterion for systems with uncertain time-invariant delays[J].IEEE Transactions on Automatic Control,1999,44(4):876-877.
[17]Poogyeon Park,Jeong Wan Ko.Stability and robust stability for systems with a time-varying delay[J].Automatica,2007,43(10):1855-1858.
[18]Li T,Luo Q,Sun C Y,et al.Exponential stability of recurrent neural networks with time-varying discrete and distributed delays[J].Nonlinear Analysis:Real World Applications,2009,10(4):2581-2589.
[19]Su Huan,Li Wenxuw,Wang Ke,et al.Stability for stochastic neural network with infinite delay[J].Neurecomputing 2011,74(10):1535-1540.
[20]宮大為,王占山,黃博南.一類帶有混合時滯的神經網絡全局漸近穩定分析[J].東北大學學報:自然科學版,2011,32(6):773-776.
[21]S Boyd,L EI Ghaoui,E Feron,et al.Linear matrix inequalities in system and control theory[M].SIAM,Philadelphia,PA,1994.
[22]S Udpin,P Niamsup.Robust stability of discrete-time LPD neural networks with time-varying delay[J].Commun Nonlinear Sci Numer Simulat,2009,14(11):3914-3924.
[23]Liu Y R,Wang Z D,Liu X H.Global exponential stability of generalizedrecurrentneuralnetworkswithdiscreteand distributed delays[J].Neural Networks,2006,19(5):667-675.
On the delay-dependent global stability analysis of delayed neural networks based on the free-weighting matrix method
MAO Kaia,SHI Baob
(a.Institute of Systems Science and Mathematics;b.Department of Basic Science,Naval Aeronautical and Astronautically University,Yantai 264001,China)
The global stability of the neural networks with both time-varying and infinite distributed delays is studied based on the condition that the activation functions satisfy the sector condition.By constructing a new Lyapunov functional which contains more information,utilizing the free-weight matrices which can better describe the relations among the terms of Newton-Leibnitz formula and the system,taking advantage of the information hidden in the state variable,the time delay functions,the derivative of the state variable and the Lyapunov functional,combining with the S-procedure,a delay-dependent sufficient condition for guaranteeing the globally exponential stability of the system is derived in form of LMI,which can be easily checked by the Matlab toolbox.The result obtained in this article improves the previous ones on which it threw off the constraint that the derivative of the time-varying function has an upper bound no larger than 1 and is less conservative than the relevant ones.
time-varying delays;infinite distributed delays;free-weighting matrix;S-Procedure;Lyapunov functional
O175.13
A
2012-06-05
海軍航空工程學院專業技術拔尖人才基金(名師工程)
毛凱(1972-),男,博士,副教授,主要從事神經網絡動力學等方面的研究,E-mail:maokaif@hotamil.com.
2095-3046(2012)05-0082-06