夏 曄, 鐘守銘, 鐘福利
(電子科技大學 數(shù) 學科學學院,四川 成 都611731)
近幾年神經(jīng)網(wǎng)絡(luò)廣泛應(yīng)用于模式識別、信號處理、圖像處理等方面.發(fā)現(xiàn)網(wǎng)絡(luò)的動態(tài)性是影響其應(yīng)用的主要原因,因此要想更好地應(yīng)用神經(jīng)網(wǎng)絡(luò)就必須對其網(wǎng)絡(luò)的動態(tài)特性尤其是穩(wěn)定性進行全面而深入分析和研究[1-3].而時滯的存在往往會對系統(tǒng)的穩(wěn)定性有所影響,因此研究有時滯的神經(jīng)網(wǎng)絡(luò)更有實際意義[4-10].被動分析是研究非線性系統(tǒng)的一個重要工具,被動理論經(jīng)常應(yīng)用于控制系統(tǒng)來更好地研究系統(tǒng)內(nèi)部的穩(wěn)定性,所以很多學者致力于研究這方面的問題[11-18].其中文獻[11]通過用分段時滯構(gòu)造新的Lyapunov泛函來判定混合時滯模型的被動性,文獻[12-14]研究帶有一個時滯的神經(jīng)網(wǎng)絡(luò)系統(tǒng)的被動性,文獻[17-18]研究神經(jīng)網(wǎng)絡(luò)的指數(shù)被動性.
本文在已有的基礎(chǔ)上進行了補充,研究帶有離散時滯和分布時滯的神經(jīng)網(wǎng)絡(luò),通過構(gòu)造新的Lyapunov泛函,建立了判定帶有混合時滯的模型是被動的新標準.本文采用了Lyapunov穩(wěn)定理論、線性矩陣不等式(LMI)及自由權(quán)矩陣等方法,實例很好地說明了本文結(jié)論是正確有效的,且在一定程度上降低了保守性.
考慮如下帶有混合時滯的神經(jīng)網(wǎng)絡(luò)系統(tǒng)
其中,x(t)=[x1(t),…,xn(t)]T∈Rn是神經(jīng)元狀態(tài)向量,g(x(t))=[g1(x1(t)),…,gn(xn(t))]T∈Rn代表激活函數(shù)向量,u(t)=[u1(t),…,un(t)]T是輸入向量,y(t)是輸出函數(shù),變量h(t)和r(t)代表模型中混合時滯且滿足
A=diag{a1,…,an}是正定的對角矩陣,W=(wij)n×n、W1=(w1ij)n×n、W2=(w2ij)n×n是代表權(quán)重系數(shù)的互連矩陣,而且激活函數(shù)gi(xi)(i=1,2,…,n)假定滿足如下條件:
表1 取不同值時,u可取到的上界Table 1 when gets different values,the upper bounds of u can get
表1 取不同值時,u可取到的上界Table 1 when gets different values,the upper bounds of u can get
ˉr 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 u 0.895 5 0.885 5 0.873 5 0.859 1 0.839 8 0.812 4 0.769 7 0.689 7 0.4235
表2 u取不同的值時,能取到的上界Table 2 when u gets different values,the upper bounds of can get
表2 u取不同的值時,能取到的上界Table 2 when u gets different values,the upper bounds of can get
u 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9ˉr 0.931 1 0.924 6 0.915 8 0.903 6 0.885 0 0.853 4 0.791 1 0.634 4 0.0503
由表1可知,當分布時滯的上界在0.1~0.9范圍內(nèi)由小至大變化時,u能取得的最大值也發(fā)生相應(yīng)的變化,u的上界隨的增大而減小,u最大值所允許的變化范圍是在0.423 5~0.895 5.當?shù)淖兓秶?.6以下時,u所能取到的最大值均在0.8以上.其中當取0.1時,u最大可以取到0.895 5.本例的實驗結(jié)果說明當分布時滯上界的大小處在一定的范圍(0.1~0.9)時,u處于一個相對較寬的范圍內(nèi)變化,能夠很好地保證模型穩(wěn)定.
由表2可知,當u在0.1~0.9范圍由小至大變化時,能取得的最大值也發(fā)生相應(yīng)的變化,的上界隨u的增大而減小,最大值的變化范圍為0.050 3~0.931 1,取值范圍較廣.當u的變化范圍在0.6以下時,的所能取到的最大值均在到0.8以上且變化的幅度不大.
綜合表1和表2中的數(shù)值實驗結(jié)果,可知在本文中u所允許的取值范圍較廣,當u處在一定范圍時,允許的分布時滯的上界也相對較寬,說明所提的方法在一定程度上降低了保守性,具有更好的實際應(yīng)用性,同時驗證了本文結(jié)論是正確有效的.
本文主要研究帶有混合時滯的神經(jīng)網(wǎng)絡(luò)的被動性,建立了判定的新標準,實例驗證了本文的結(jié)論是正確有效的,同時在一定程度上降低了保守性,具有一定的優(yōu)越性,因而有更好的實際應(yīng)用.
[1] Shao H.New delay-dependent stability criteria for systems with interval delay[J].Automatica,2009,45:744-749.
[2] Xu X,Lam J,Ho D W C.Novel global robust stability criteria for interval neural networks with multiple time-varying delays[J].Phys Lett,2005,A342:322-330.
[3] Cao J,Yuan K,Li H X.Global asymptotical stability of recurrent neural networks with multiple discrete delays and distributed delays[J].IEEE Trans Neural Networks,2006,17:1646-1651.
[4] Liu X,Zhong S.T-S fuzzy model-based impulsive control of chaotic systems with exponential decay rate [J].Phys Lett,2007,A370(3/4):260-264.
[5] Liu X.Delay-dependent H control for uncertain fuzzy systems with time-varying delays[J].Nonlinear Anal:TMA,2008,68(5):1352-1361.
[6] Liu X,Yu W,Wang L.Necessary and sufficient asymptotic stability criterion for 2-D positive systems with time-varying state delays described by Roesser model[J].IET Cont Theory Appl,2011,5(5):663-668.
[7]陳光淦,蒲志林,張健.變時滯Hopfield神經(jīng)網(wǎng)絡(luò)模型的全局指數(shù)穩(wěn)定性和全局吸引性[J].工程數(shù)學學報,2005,22(5):821-826.
[8]龍述君,向麗.一類具有分布時滯的Hopfield神經(jīng)網(wǎng)絡(luò)的穩(wěn)定性[J].四川師范大學學報:自然科學版,2006,29(5):566-569.
[9]趙亮,李樹勇.含時滯和脈沖的雙向聯(lián)想記憶的神經(jīng)網(wǎng)絡(luò)模型的全局魯棒一致漸近穩(wěn)定性分析[J].四川師范大學學報:自然科學版,2013,36(2):177-124.
[10] Kwon O M,Park J H.New delay-dependent robust stability criterion for uncertain neural networks with time-varying delays[J].Appl Math Comput,2008,205:417-427.
[11] Li H Y,Lam J,Cheung K C.Passivity criteria for continuous time neural networks with mixed time-varying delays[J].Appl Math Comput,2012,218:11062-11074.
[12] Xu S,Zheng W,Zou Y.Passivity analysis of neural networks with time-varying delays[J].IEEE Transcations on Circuits and Systems II,2009,56(4):325-329.
[13] Zhang Z,Mou S,Lam J,et al.New passivity criteria for neural networks with time-varying delay[J].Neural Networks,2009,22(7):864-868.
[14] Park M J,Kwon O M,Park J H,et al.A new augmented Lyapunov-Krasovskii Functional approach for stability of linear systems with time-varying delays[J].Appl Math Comput,2011,217:7197-7209.
[15]Kwon O M,Park J H.Exponential stability for uncertain cellular neural networks with discrete and distributed time-varying delays[J].Appl Math Comput,2008,A203:813-823.
[16] Park J H.Further results on passivity analysis of delayed cellular neural networks[J].Chaos,Solitons& Fractals,2007,34:1546-1551.
[17] Kwon O M,Park J H.A new augmented Lyapunov-Krasovskii functional approach to exponential passivity for neural networks with time-varying delays[J].Appl Math Comput,2011,217:10231-10238.
[18] Zhu S,Shen Y,Chen C.Exponential passivity of neural networks with time-varying delay and uncertainty[J].Phys Lett,2010,A375:136-142.