張賀 徐勝元
摘要? ?針對(duì)延遲中立型神經(jīng)網(wǎng)絡(luò)系統(tǒng),研究了反饋控制問(wèn)題.考慮了帶有Levy噪聲的中立型神經(jīng)網(wǎng)絡(luò),建立了一個(gè)適當(dāng)?shù)腖yapunov函數(shù).通過(guò)Lyapunov分析方法并使用一般性It公式和LMI技術(shù),得到了閉環(huán)系統(tǒng)的均方穩(wěn)定性準(zhǔn)則.最后,通過(guò)數(shù)值例子證實(shí)本文方法的有效性.
關(guān)鍵詞
中立型神經(jīng)網(wǎng)絡(luò);Levy噪聲;Lyapunov分析方法;均方穩(wěn)定性
中圖分類(lèi)號(hào)? O429
文獻(xiàn)標(biāo)志碼? A
0 引言
近來(lái),中立型神經(jīng)網(wǎng)絡(luò)得到了廣泛的研究[1-2] .中立型神經(jīng)網(wǎng)絡(luò)是可以被中立型泛函微分方程描述的一類(lèi)神經(jīng)網(wǎng)絡(luò).由于系統(tǒng)模型涉及到當(dāng)前的狀態(tài)導(dǎo)數(shù)和過(guò)去的狀態(tài)導(dǎo)數(shù),所以這類(lèi)系統(tǒng)更適合描述神經(jīng)反應(yīng)過(guò)程和神經(jīng)細(xì)胞特性的動(dòng)力學(xué)[3] .在探索隨機(jī)中立型神經(jīng)網(wǎng)絡(luò)的同步和穩(wěn)定性問(wèn)題中,一般性的中立型模型得到了廣泛的關(guān)注.
需要注意的是,很多工作主要集中在由高斯白噪聲驅(qū)動(dòng)的中立型神經(jīng)網(wǎng)絡(luò)中的隨機(jī)同步[4-5] ,當(dāng)隨機(jī)脈沖輸入量大、振幅小的時(shí)候,可以認(rèn)為是合理的近似.然而,對(duì)于真實(shí)的生物邏輯神經(jīng)元是不合理的.例如,突觸后神經(jīng)元的觸發(fā)區(qū)附近突觸的輸入減少,在真實(shí)神經(jīng)系統(tǒng)的噪聲振幅中產(chǎn)生較大的脈沖[6] ,因此,在這種情況下,Levy噪聲比布朗運(yùn)動(dòng)更合適.實(shí)際上,與純擴(kuò)散布朗運(yùn)動(dòng)相比,跳躍擴(kuò)散的Levy過(guò)程不僅允許個(gè)體的數(shù)量在大部分時(shí)間內(nèi)持續(xù)變化,而且還允許在任意時(shí)刻任意大小的隨機(jī)跳躍是不連續(xù)的.這些證據(jù)表明,雖然隨機(jī)神經(jīng)網(wǎng)絡(luò)增加了復(fù)雜的數(shù)學(xué)描述,但是隨機(jī)神經(jīng)網(wǎng)絡(luò)系統(tǒng)的研究更有意義、更具挑戰(zhàn)性.
本文主要分析中立型神經(jīng)網(wǎng)絡(luò)的反饋控制問(wèn)題,所研究的系統(tǒng)包含固定常數(shù)延遲以及Levy噪聲和布朗運(yùn)動(dòng)噪聲.針對(duì)延遲中立型神經(jīng)網(wǎng)絡(luò)系統(tǒng),建立了一個(gè)適當(dāng)?shù)腖yapunov函數(shù).通過(guò)Lyapunov分析方法并使用一般性It公式和LMI技術(shù),得到了閉環(huán)系統(tǒng)的均方穩(wěn)定性準(zhǔn)則.最后,通過(guò)數(shù)值例子證實(shí)了本文方法的有效性.
1 基礎(chǔ)知識(shí)和數(shù)學(xué)模型
在本文中,定義(Ω,F(xiàn),{F t} t≥0 ,P)是一個(gè)完備概率空間,其中{F t} t≥0 是右連續(xù)遞增的且F 0包含所有概率P=0的集合.如果 A 是一個(gè)向量或者矩陣,則 A? ?T 代表矩陣 A 的轉(zhuǎn)置.對(duì)于任意的t>0,C([-t,0]; R? n)表示一個(gè)從[-t,0]到 R? n的連續(xù)函數(shù)的族,其中,‖ A ‖是矩陣 A 的模長(zhǎng)并且‖ A ‖= sup {| Ax | ∶ | x |=1}.L 2 F 0 ([-t,0]; R? n)是所有有界的、F 0-可測(cè)的、值域?yàn)镃([-t,0]; R? n)的隨機(jī)變量族. R? n +為正的集合,即 R? n +={ x ∈ R? n:x i>0,1≤i≤n}.N( d t, d x)是 Poisson 隨機(jī)可測(cè)的,則 ( d t, d x)被定義為N( d t, d x)-λ( d x) d t是補(bǔ)償?shù)?Poisson 隨機(jī)可測(cè)的.定義λ(s)=EN(1,s)是跳躍測(cè)度.定義B t是一個(gè)獨(dú)立的n-維 Brownian 運(yùn)動(dòng).
4 結(jié)束語(yǔ)
對(duì)延遲中立型神經(jīng)網(wǎng)絡(luò)系統(tǒng),均方穩(wěn)定性問(wèn)題被研究.本文中的中立型神經(jīng)網(wǎng)絡(luò)被考慮帶有Levy噪聲.本文考慮的控制器是一個(gè)連續(xù)的模型,這無(wú)疑對(duì)硬件的要求較高.如何在確保系統(tǒng)性能可以達(dá)到的同時(shí),減少數(shù)據(jù)的傳輸量以適應(yīng)有限的交流帶寬是一個(gè)有趣的課題.在未來(lái)的研究工作中,將進(jìn)一步研究帶有離散輸入的中立型系統(tǒng).
參考文獻(xiàn)
References
[ 1 ]?Huang H,Du Q,Kang X.Global exponential stability of neutral high-order stochastic Hopfield neural networks with Markovian jump parameters and mixed time delays[J].ISA transactions,2013,52(6):759-767
[ 2 ] Kiumarsi B,Lewis F L,Levine D S.Optimal control of nonlinear discrete time-varying systems using a new neural network approximation structure[J].Neurocomputing,2015,156:157-165
[ 3 ] Bai C.Existence and stability of almost periodic solutions of Hopfield neural networks with continuously distributed delays[J].Nonlinear Analysis:Theory,Methods & Applications,2009,71(11):5850-5859
[ 4 ] Park J H,Kwon O M.Synchronization of neural networks of neutral type with stochastic perturbation[J].Modern Physics Letters B,2009,23(14):1743-1751
[ 5 ] Zheng C D,Wei Z,Wang Z.Robustly adaptive synchronization for stochastic Markovian neural networks of neutral type with mixed mode-dependent delays[J].Neurocomputing,2016,171:1254-1264
[ 6 ] Patel A,Kosko B.Stochastic resonance in continuous and spiking neuron models with Levy noise[J].IEEE Transactions on Neural Networks,2008,19(12):1993-2008
[ 7 ] Mao ?W,Mao X.On the approximations of solutions to neutral SDEs with Markovian switching and jumps under non-Lipschitz conditions[J].Applied Mathematics and Computation,2014,230:104-119
Feedback control of neutral-type neural networks with Levy noise
ZHANG He 1 XU Shengyuan1
1 School of Automation,Nanjing University of Science and Technology,Nanjing 210094
Abstract? The feedback control problem in delayed neutral-type neural networks with Levy noise is addressed in this paper.Delay is considered as constant in this study.An appropriate Lyapunov function is used to analyze the mean square stability of the closed-loop system.Using the Lyapunov method,the general It formula,and the linear matrix inequality technique,the sufficient condition to guarantee stability in mean square sense for the closed-loop system is derived.Finally,a numerical example is given to illustrate the effectiveness of the obtained results.
Key words? neutral-type neural networks;Levy noise;Lyapunov method;mean square stability
南京信息工程大學(xué)學(xué)報(bào)2018年6期