This paper is concerned with the problem of sampled-data synchronization for complex dynamical networks (CDNs) with time-varying coupling delay and random coupling strengths. The random coupling strengths are described by normal distribution. The sampling period considered here is assumed to be less than a given bound. By taking the characteristic of sampled-data system into account, a discontinuous Lyapunov functional is constructed, and a delay-dependent mean square synchronization criterion is derived. Based on the proposed condition, a set of desired sampled-data controllers are designed in terms of linear matrix inequalities (LMIs) that can be solved effectively by using MATLAB LMI Toolbox. Numerical examples are given to demonstrate the effectiveness of the proposed scheme.
1. Introduction
In the real world, many practical and natural systems can be described by models of complex networks such as internet, food webs, electric power grids, scientific citation networks, and social networks. Therefore, a dynamical network can be regarded as a dynamical system with a special structure. In the last few years, complex dynamical networks (CDNs) have received extensive attention and increasing interest across many fields of science and engineering [1–3]. CDNs are a large set of interconnected nodes, in which each node represents an element with certain dynamical system and edge represents the relationship between them. With the important discovery of the “small-world” and “scale-free” properties, complex dynamical networks have become a focal research topic in the area of complexity science.
It is very common that many natural systems often exhibit collective cooperative behaviors among their constituents. Synchronization, as a kind of typical collective behavior, is one of key issues in the study of complex dynamical networks. The main reason is that network synchronization not only can explain many natural phenomena but also has wide applications in many fields including secure communications, synchronous information exchange in the internet, genetic regulatory process, the synchronous transfer of digital signals in communication networks, and so on. Over the past several decades, the synchronization in CDNs has been intensively investigated from various fields such as sociology, biology, and physics [4–16]. The authors in [5] focused the synchronization stability of general CDNs with coupling delay. In [6], the authors investigated the locally and globally adaptive synchronization of an uncertain complex dynamical network. The problem of globally exponential synchronization of impulsive dynamical networks was investigated in [7]. The pinning synchronization problems in CDNs have been analyzed in [8, 9]. In [10], the authors studied the global exponential synchronization and synchronizability for general dynamical networks. In [11], some sufficient conditions for CDNs with and without coupling delays in the state to be passive were presented. Recently, the guaranteed cost synchronization of a CDN via dynamic feedback control was addressed in [15].
It is well known that the coupling strength of complex dynamical network plays an important role in the realizing synchronization. In general, the coupling strength of the considered CDNs is deterministic [4–16]. If the deterministic coupling strength is large enough, a complex network can realize synchronization by itself. However, according to the discussion in [17, 18], because of the effects of environment and artificial factor, the coupling strength of complex dynamical networks may randomly vary around some constants. If the upper or lower bound of the random coupling strength is only considered, some conservative result will be derived. That is to say, random phenomena in coupling strength should be taken into account when dealing with the synchronization of CDNs. Furthermore, the normal distribution characteristic of random variables can be easily obtained by statistical methods. Therefore, it is interesting to investigate the synchronization of CDNs with random coupling strengths described by normal distribution.
On the other hand, the sampled-data control system, whose control signals are allowed to change only at discrete sampling instants, can drastically reduce the amount of information transmitted and increase the efficiency of bandwidth usage. Therefore, sampled-data control has received notable attention [19–22]. The input delay approach proposed in [19] is very popular in the study of sampled control system, where the system is modeled as a continuous-time system with a time-varying sawtooth delay in the control input induced by sample-and-hold. In [20], by constructing the time-dependent Lyapunov functional, a refined input delay approach was presented. Later, the chaos synchronization problems are investigated by using sampled-data control [23–26]. Recently, in the framework of the input delay approach, the sampled-data synchronization problem has been investigated for a class of general complex networks with time-varying coupling delays in [27]. Furthermore, some improved sampled-data synchronization criterion has been derived to ensure the exponential stability of the closed-loop error system and corresponding sampled-data feedback controllers are designed in [28]. By combining the time-dependent Lyapunov functional approach and convex combination technique, the exponential sampled-data synchronization of CDNs with time-varying coupling delay and uncertain sampling was studied in [29]. However, the Lyapunov functional proposed in [27, 28] ignored the substitutive characteristic of sampled-data system, which leads to some conservatism inevitably. In addition, the results obtained in [29] are sufficient conditions, which imply that there is still room for further improvement. To the best of our knowledge, the sampled-data synchronization problem of complex dynamical networks with time-varying coupling delays and random coupling strengths has not been studied in the literature.
Motivated by the aforementioned discussions, in this paper, the problem of sampled-data synchronization of CDNs with time-varying coupling delay and random coupling strengths is investigated. The sampling period is assumed to be time varying but less than a given bound. The random coupling strengths are described by normal distribution. By capturing the characteristic of sampled-data control system, a new discontinuous Lyapunov functional is constructed. By using the low bound lemma and convex combination approach, a mean square synchronization condition is formulated in terms of LMIs. The corresponding sampled-data controllers are designed that can achieve the synchronization of the considered CDNs. The proposed method can lead to a less conservative result than the existing ones. Finally, numerical examples are given to illustrate the effectiveness and less conservatism of the presented sampled-data control scheme.
Notation. The notations used throughout this paper are fairly standard.RnandRm×ndenote the n-dimensional Euclidean space and the set of allm×nreal matrix, respectively.P>0orP<0means that P is symmetric and positive or negative definite. The superscript “T” represents the transpose, and “I” and “0” denote the identity and zero matrix with appropriate dimensions.diag{l1,l2,…,ln}stands for a block diagonal matrix. The symmetric terms in a symmetric matrix are denoted by *.
2. Preliminaries and Model Description
Consider a class of linearly coupled complex dynamical networks consisting ofNidentical coupled nodes, in which each node is an n-dimensional subsystem
(1)x˙i(t)=f(xi(t))+c1(t)∑j=1NGijDxj(t)+c2(t)∑j=1NGijAxj(t-τ(t))+ui(t),+c2(t)∑j=1NGijAxjiiiiiiiiii=1,2,…,N,
where xi=(xi1,xi2,…,xin)∈Rnandui(t)∈Rnare, respectively, the state variable and the control input of the node i.f:Rn→Rnis a continuous vector-valued function.c1(t)andc2(t)are mutually independent random variables, which denote the random coupling strengths of nondelayed coupling and time-delayed couplings, respectively.τ(t)denotes the time-varying coupling delay satisfying0≤τ(t)≤h,τ˙(t)≤μ, whereh>0andμare known constants.D=(dij)n×n∈Rn×nis the constant inner-coupling matrix andA=(aij)n×n∈Rn×nis the time-delay inner-coupling matrix.G=(gij)∈RN×Nis the coupling configuration matrix, wheregijis defined as follows: if there is a connection between node i and node j (i≠j), thengij>0; otherwise,gij=0, and the diagonal elements of matrix G are defined bygii=-∑j=1,j≠iNGij,i=1,2,…,N.
Remark 1.
The coupling configuration matrix G represents the topological structure of network (1). In this paper, the matrix G is not assumed to be symmetric or irreducible. In [27, 28], the coupling configuration matrix is assumed to be symmetric, which is quite restrictive in practice. In this regards, the network model considered here is more general.
In this paper, similar to [17, 18], we assume that almost all the values ofci(t),i=1,2, are taken on some nonnegative intervals, that is,ci(t)∈(σi,ρi), whereσi,ρi(i=1,2)are nonnegative constants withσi<ρi. Almost all the values ofci(t)satisfyingci(t)∈(σi,ρi)imply thatProb{ci(t)∈(σi,ρi)}=1. It should be noted that the actual minimum and maximum allowable coupling strength bounds are notσiandρi, respectively. It just means thatProb{ci(t)<σi}=0andProb{ci(t)>ρi}=0. The actual lower bounds ofci(t)may be very small and the actual upper bounds of them may be very large. This is very different from synchronization results obtained by traditional method, in which constant coupling strength is always preassumed or deterministic.
Remark 2.
We assume the coupling strengths satisfy the normal distribution, which can randomly vary around some given intervals. This is very different from the considered network models in [27–29], in which constant coupling strengths are always preassumed or deterministic. Therefore, for the random coupling strength, most of existing results with constant coupling strength may not be applicable anymore. In addition, it is worth pointing out that when c1(t)=c10andc2(t)=c20orδ1=δ2=0, system (1) includes the models in [27–29] as a special case.
Assumption 3.
There nonlinear function f satisfies
(2)[f(x)-f(y)-U(x-y)]T[f(x)-f(y)-V(x-y)]≤0,∀x,y∈Rn,
where U and V are constant matrices of appropriate dimensions.
Assumption 4.
The mathematical exception and variance ofci(t)areE{ci(t)}=ci0andE{(ci(t)-ci0)2}=δi2, respectively, whereci0andδiare nonnegative constants.
On the basis of the property of variablesci(t), system (1) can be rewritten in the following form:
(3)x˙i(t)=f(xi(t))+c10∑j=1NGijDxj(t)+(c1(t)-c10)∑j=1NGijDxj(t)+c20∑j=1NGijAxj(t-τ(t))+(c2(t)-c20)∑j=1NGijAxj(t-τ(t))+ui(t),i=1,2,…,N.
Letei(t)=xi(t)-s(t)be the synchronization error, wheres(t)∈Rnis the state trajectory of the unforced isolate nodes˙(t)=f(x(t)). Then, the error dynamics is given by
(4)e˙i(t)=g(ei(t))+c10∑j=1NGijDej(t)+(c1(t)-c10)∑j=1NGijDej(t)+c20∑j=1NGijAej(t-τ(t))+(c2(t)-c20)∑j=1NGijAej(t-τ(t))+ui(t),i=1,2,…,N,
whereg(ei(t))=f(xi(t))-f(s(t))=[f1(ei1(t))f2(ei2(t))⋯fn(ein(t))]T.
The control signal is assumed to be generalized by using a zero-order-hold (ZOH) function with a sequence of hold times0=t0<t1<⋯<tk<⋯. Therefore, the state feedback controller takes the following form:
(5)ui=Kiei(tk)=Ki(xi(tk)-s(tk)),tk≤t<tk+1,Kiei(tk)=Ki(xi(tk)-s(tk))iiiiiiiii,i=1,2,…,N,
whereKiis the feedback gain matrix to be determined andei(tk)is the discrete measurement ofei(t)at sampling instanttk. In this paper, the sampling is not required to be periodic, and the only assumption is that the distance between any two consecutive sampling instants is less than a given bound. It is assumed thattk+1-tk=hk≤pfor any integerk≥0, wherep>0represents the largest sampling interval.
By substituting (5) into (4), we obtain
(6)e˙i(t)=g(ei(t))+c10∑j=1NGijDej(t)+(c1(t)-c10)∑j=1NGijDej(t)+c20∑j=1NGijAej(t-τ(t))+(c2(t)-c20)∑j=1NGijAej(t-τ(t))+Kiei(tk),i=1,2,…,N.
Furthermore, by using the Kronecker product, system (6) can be rewritten as
(7)e˙(t)=g-(e(t))+c10(G⊗D)e(t)+(c1(t)-c10)(G⊗D)e(t)+c20(G⊗A)e(t-τ(t))+(c2(t)-c20)(G⊗A)e(t-τ(t))+Ke(tk),i=1,2,…,N,
wheree(t)=[e1T(t)e2T(t)⋯eNT(t)]T,g-(e(t))=[gT(e1(t))gT(e2(t))⋯gT(en(t))]T, andK=diag{K1,K2,…,KN}.
To proceed further, the following definition and useful lemmas are needed.
Definition 5.
The coupled complex dynamical network (1) is said to be globally synchronized in mean square sense if limt→∞E{∥ei(t)∥2}=0,i=1,2,…,N, holds for any initial values.
Lemma 6 (extended Wirtinger inequality [22]).
Letz(t)∈W[a,b)andz(a)=0. Then for any matrixR>0, the following inequality holds:
(8)∫abzT(α)Rz(α)dα≤4(b-a)2π2∫abz˙T(α)Rz˙(α)dα.
Lemma 7 (reciprocally convex approach [30]).
Let f1,f2,…,fN:Rm↦R have positive values in an open subset D ofRm. Then, the reciprocally convex combination offiover D satisfies
(9)min{αi∣αi>0,∑iαi=1}∑ifi(t)=∑ifi(t)+maxgi,j(t)∑i≠jgi,j(t)
subject to
(10){gi,j:Rm⟼R,gj,i(t)≜gi,j(t),[fi(t)gi,j(t)gj,i(t)fj(t)]≥0}.
The aim of this paper is to design a set of sampled-data controllers (5) with sampling period as big as possible to ensure synchronizing the complex network (1) in mean square sense. By some transformation, the synchronization problem of the delayed complex network (1) can be equivalently converted into the mean square asymptotical stability problem of error system (7). Therefore, we are interested in two main issues in our paper, one is to find some stability conditions for error system (7) in mean square for givenKi, and the other is to derive the design method of sampled-data controllers.
3. Main Results
In this section, by considering the characteristic of sampled-data system, we first give a delay-dependent condition to ensure error system (7) to be globally stable in mean square sense. Then, based on the derived condition, the design method of the sampled-data controllers is proposed. Before presenting the main results, for the sake of presentation simplicity, we denote
(11)U-=(IN⊗U)T(IN⊗V)2+(IN⊗V)T(IN⊗U)2,V-=-(IN⊗U)T+(IN⊗V)T2.
Theorem 8.
Under Assumptions 3-4, for given controller gain matricesKi, the error system (7) is globally asymptotically stable in mean square sense if there exist matricesP>0,Q1>0,Q2>0,Q3>0,Z1>0,Z2>0,U>0,R>0,W>0,S1,S2,N, and a scalarε>0 such that the following LMIs are satisfied:
(12)[Σ-pR-Γ1pN*-X10**-pU]<0,(13)[Σ+pR-Γ2*-X2]<0,(14)[ZiSi*Zi]≥0,i=1,2,
where
(15)Σ=[Σ11Σ12Σ13Σ14Σ15Σ16*Σ22Σ23-N4-N5-N6**Σ33000***Σ44Σ450****Σ550*****-εI],Σ11=c10P(G⊗D)+c10(G⊗D)TP+Q1+Q2+Q3-14π2W-εU--Z1-Z2+N1+N1T,Σ12=PK+14π2W+Z1-S1-N1T+N2,Σ13=S1+N3,Σ14=P(G⊗A)+Z2-S2+N4,Σ15=S2+N5,Σ16=P-εV-+N6,Σ22=-2Z1-14π2W+S1+S1T-N2-N2T,Σ23=Z1-S1-N3,Σ33=-Q1-Z1,Σ44=-(1-μ)Q3-2Z2+S2+S2T,Σ45=Z2-S2,Σ55=-Q2-Z2,Γ1=[Ω1Tδ1Ω2Tδ2Ω3T],Γ2=[Ω^1Tδ1Ω^2Tδ2Ω^3T],Ω1=[c10Z(G⊗D)ZK0c20Z(G⊗A)0Z],Ω2=[Z(G⊗D)00000],Ω2=[000Z(G⊗A)00],Ω^1=[c10(Z+pU)(G⊗D)(Z+pU)K0×c20(Z+pU)(G⊗A)0(Z+pU)],Ω^2=[(Z+pU)(G⊗D)00000],Ω^3=[000(Z+pU)(G⊗A)00],X1=diag{-Z,-Z,-Z},X2=diag{-Z-pU,-Z-pU,-Z-pU},Z=p2Z1+h2Z2+p2W,R-=diag{0,R,0,0,0,0}.
Proof.
Consider the following Lyapunov functional:
(16)V(t)=V1(t)+V2(t)+V3(t)+V4(t),
wheret∈[tk,tk+1)and
(17)V1(t)=eT(t)Pe(t)+∫t-pteT(s)Q1e(s)ds+∫t-hteT(s)Q2e(s)ds+∫t-τ(t)teT(s)Q3e(s)ds,V2(t)=∫-p0∫t+θte˙T(s)Z1e˙(s)dsdθ+h∫-h0∫t+θte˙T(s)Z2e˙(s)dsdθ,V3(t)=(p-(t-tk))∫tkte˙T(s)Ue˙(s)ds+(p-(t-tk))(t-tk)eT(tk)Re(tk),V4(t)=p2∫tkte˙T(s)We˙(s)ds-π24∫tkt[x(s)-x(tk)]TW[x(s)-x(tk)]ds.
It is clear that at anyt>0except the sampling instantstk,V3(t)is continuous and nonnegative, and right after the jump instantstk,V3(t)becomes zero; that is,V3(tk-)≥0,V3(tk+)=0. According to Lemma 7, we can easily find thatV4(t)≥0andV4(t)vanishes att=tk. Thus, we haveV(tk-)≥V(tk+).
Define the infinitesimal operatorLofV(t)as follows:
(18)LV(t)=limΔ→0+Δ-1[E{V(t+Δ)∣e(t)}-V(t)].
Taking the derivative of (16) along the solution of system (7) for∀t∈[tk,tk+1), it yields
(19)LV1(t)≤2eT(t)×P(g-(e(t))+c10(G⊗D)e(t)+c20(G⊗A)e(t-τ(t))+Ke(tk))+eT(t)Q1e(t)-eT(t-p)Q1e(t-p)+eT(t)Q2e(t)-eT(t-h)Q2e(t-h)+eT(t)Q3e(t)-(1-μ)eT(t-τ(t))Q3e(t-τ(t)),LV2(t)=e˙T(t)Ze˙(t)-∫t-pte˙T(s)Z1e˙(s)ds-h∫t-hte˙T(s)Z2e˙(s)ds,LV3(t)=(p-(t-tk))e˙T(t)Ue˙(t)-∫tkte˙T(s)Ue˙(s)ds+(p-(t-tk))eT(tk)Re(tk)-(t-tk)eT(tk)Re(tk),LV4(t)=p2e˙T(t)We˙(t)-π24[x(t)-x(tk)]TW[x(t)-x(tk)].
If (14) is satisfied, then by utilizing Lemma 6, we have
(20)-p∫t-pte˙T(s)Z1e˙(s)ds≤-[e(t)-e(tk)e(tk)-e(t-p)]T[Z1S1*Z1]×[e(t)-e(tk)e(tk)-e(t-p)],-h∫t-hte˙T(s)Z2e˙(s)ds≤-[e(t)-e(t-τ(t))e(t-τ(t))-e(t-h)]T[Z2S2*Z2]×[e(t)-e(t-τ(t))e(t-τ(t))-e(t-h)].
On the other hand, the following inequality is true for any matrix N with appropriate dimensions:
(21)-∫tkte˙T(s)Ue˙(s)ds≤(t-tk)ξT(t)NU-1NTξ(t)+2ξT(t)N(e(t)-e(tk)),
where
(22)ξ(t)=[(t)eT(tk)eT(t-p)eT(t-τ(t))eT(t-h)g-T(e(t))]T.
LetΩ=Ω1+(c1(t)-c10)Ω2+(c2(t)-c20)Ω3. Becausec1(t)andc2(t)are mutually independent random variables, it can be obtained from (7) that
(23)E{e˙T(t)Ze˙(t)}=E{ξT(t)ΩTZΩξ(t)}=ξT(t)Ω-1ξ(t),E{e˙T(t)Ue˙(t)}=E{ξT(t)ΩTUΩξ(t)}=ξT(t)Ω-2ξ(t),
where Ω-1=Ω1TZΩ1+δ12Ω2TZΩ2+δ22Ω3TZΩ3,Ω-2=Ω1TZ3Ω1+δ12Ω2TZ3Ω2+δ22Ω3TZ3Ω3.
In addition, based on Assumption 3, for anyε>0, we have
(24)y(t)=ε[e(t)g-(e(t))]T[U-V-*I][e(t)g-(e(t))]≤0.
Combining (18)–(24) and taking mathematical exceptions on both sides of (16) give that
(25)E{LV(t)}≤ξT(t)Φξ(t),
where Φ=Σ+Ω-1+(p-(t-tk))(Ω-2+R-)+(t-tk)(NU-1NT-R-).
Noting thatΦis a convex combination oft-tkandp-(t-tk), soΦ<0if and only if
(26)Σ+Ω-1+pNU-1NT-pR-<0,Σ+Ω-1+p(Ω-2+R-)<0.
From the Schur complement, (12) and (13) can ensureΦ<0. This means thatE{LV(t)}≤-ρ∥e(t)∥2for a sufficiently smallρ>0. We can conclude that system (7) is asymptotically stable in the mean square sense. This completes the proof.
Remark 9.
Inspired by [20, 22], the characteristic of sampling instants has been considered in the construction of the Lyapunov functional. The discontinuous termsV3(t)andV4(t)can make full use of the sawtooth structure characteristic of sampling instants and play the key role in the reduction of conservatism. In the process of taking the derivative ofV(t), reciprocally convex approach and convex combination technique were employed, which were beneficial to lead less conservativeness. Moreover, the derived synchronization criterion is formulated in terms of LMIs that can be easily verified by using available software.
Next, we will consider how to design the desired sampled-data controllers. Based on Theorem 8, a set of sampled-data controllers are presented as follows.
Theorem 10.
Under Assumptions 3-4, the complex dynamical networks (1) with random coupling strength is globally asymptotically synchronized in mean square by the sampled-data controllers (5) if there exist matricesP=diag{P1,P2,…,PN}>0,Q1>0,Q2>0,Q3>0,Z1>0,Z2>0,U>0,R>0,W>0,S1,S2,N,X=diag{X1,X2,…,XN}, and a scalarε>0such that (14) and the following LMIs are satisfied:
(27)[Σ^-pR-ΥpN*-X^10**-pU]<0,[Σ^+pR-Υ*-X^2]<0,
where
(28)X^1=diag{-2P+Z,-2P+Z,-2P+Z},X^2=diag{-2P+Z+pU,-2P+Z+pU,-2P+Z+pU},Υ=[Υ1Tδ1Υ2Tδ2Υ3T],Υ1=[c10P(G⊗D)X0c20P(G⊗A)0P],Υ2=[P(G⊗D)00000],Υ3=[000P(G⊗A)00],Σ^=[Σ11Σ^12Σ13Σ14Σ15Σ16*Σ22Σ23-N4-N5-N6**Σ33000***Σ44Σ450****Σ550*****-εI],Σ^12=X+14π2W+Z1-S1-N1T+N2,
and the other terms follow the same definitions as those in Theorem 8. Moreover, the desired controllers gain matrices are given by
(29)Ki=Pi-1Xi,i=1,2,…,N.
=diag{I,I,I,I,I,I,I,P(Z+pU)-1,P(Z+pU)-1,P(Z+pU)-1}. Note that -PZ-1P≤-2P+Z and -P(Z+pU)-1P≤-2P+Z+pU are true for Z>0 and U>0. Then, performing a congruence transformation of J1 to (12) and performing a congruence transformation of J2 to (13), respectively, and considering the relation X=PK, we can obtain that if (27) holds, then (12) and (13) hold. This completes the proof.
Ifc1(t)=c10andc2(t)=c20, the random coupling strengths reduce to constant, and the error system (7) can be rewritten as following simple form:
(30)e˙(t)=g-(e(t))+c10(G⊗D)e(t)+c20(G⊗A)e(t-τ(t))+Ke(tk),i=1,2,…,N.
Based on Theorems 8 and 10, by eliminatingδ1andδ2, we can easily get the following results.
Corollary 11.
Under Assumption 3, for given controller gain matricesKi, the error system (30) with sampled-data controllersKican achieve synchronization, if there exist matrices P>0,Q1>0,Q2>0,Q3>0,Z1>0,Z2>0,U>0,R>0,W>0,S1,S2,N, and a scalarε>0 such that (10) and the following LMIs are satisfied:
(31)[Σ-pR-Ω1TpN*-Z0**-pU]<0,[Σ+pR-Ω^1T*-Z-pU]<0,
where the other terms follow the same definitions as those in Theorem 8.
Corollary 12.
Under Assumption 3, the complex dynamical network (1) with random coupling strengths is globally asymptotically synchronized in mean square by the sampled-data controllers (5) if there exist matricesP=diag{P1,P2,…,PN}>0,Q1>0,Q2>0,Q3>0,Z1>0,Z2>0,U>0,R>0,W>0,S1,S2,N,X=diag{X1,X2,…,XN}, and a scalarε>0 such that (10) and the following LMIs are satisfied:
(32)[Σ^-pR-Υ1TpN*-2P+Z0**-pU]<0,[Σ^+pR-Υ1*-2P+Z+pU]<0,
where the other terms follow the same definitions as those in Theorem 10.
Remark 13.
Since the characteristic of sampled-data control system is fully considered, the conservatism of Corollary 12 is much less than those not taking delay characteristic into account [27, 28], which will be verified by numerical example in next section.
Remark 14.
It is worth pointing out that the main result here can be extended to some more general complex dynamical networks with probabilistic time-varying coupling delay [18] or distributed coupling delay. Owing to the space limit, it is omitted here.
4. Numerical Examples
In this section, two numerical examples are given to show the validity of the proposed results.
Example 1.
Consider complex network model (1) with three nodes. The out-coupling matrix is assumed to beG=(Gij)N×Nwith
(33)G=[-1010-1111-2].
The time-varying coupling delay is chosen as τ(t)=0.2+0.05sin(10t). A straight-forward calculation givesh=0.25andμ=0.5. The nonlinear function f is taken as
(34)f(xi(t))=[-0.5xi1+tanh(0.2xi2)+0.2xi20.95xi2-tanh(0.75xi2)].
It can be found that f satisfies (2) with
(35)U=[-0.50.200.95],V=[-0.30.200.2].
The inner-coupling matrices are given asD=0and
(36)A=[1001].
Let the coupling strengthc2(t)be a constant; that is,c2(t)=c. For differentc, Table 1 lists the maximum sampling interval p obtained by Corollary 12 and [27–29]. From this table, we can see that our result has less conservatism than the existing ones.
Furthermore, choosingc=0.5and applying MATLAB LMI toolbox to solve the LMIs in Corollary 12, the gain matrices of the desired controllers can be obtained as follows:
(37)K1=[-0.6578-0.0978-0.0172-1.2316],K2=[-0.6578-0.0978-0.0172-1.2316],K3=[-0.4543-0.1223-0.0185-1.1988].
In the numerical simulation, assume that the initial values arex1(0)=[35]T,x2(0)=[-2-1]T,x3(0)=[21]T, ands(0)=[-53]T. The state trajectories of the synchronization error and the control inputsui(t)are given in Figures 1 and 2, respectively. Clearly, the synchronization errors are globally asymptotically stable in mean square under the proposed sampled-data scheme.
Maximum sampling interval p for different coupling strength c.
c
0.5
0.75
[27]
0.5409
0.1653
[28]
0.5573
0.2277
[29]
0.9016
0.8957
Corollary 12
0.9795
0.9121
Synchronization error states.
Sampled-data control inputs.
Example 2.
The isolated node of the dynamical networks and the coupling delay are the same as Example 1. The inner-coupling matrices are given as
(38)D=A=[0.1000.1],
and the outer-coupling matrix
(39)G=[-3102000-2011010-3101020-2001111-4011001-3].
We assume thatc1(t)andc2(t)are two mutually independent random variables satisfying normal distribution withc10=5,c20=1,δ1=0.5, andδ2=0.15. According to the property of normal distribution, almost all the values ofci(t)satisfyc1(t)∈(ci0-3δi,ci0+3δi); that isc1(t)∈(3.5,6.5)andc2(t)∈(1.55,2.45). Figures 3 and 4 depict the random coupling strengthsc1(t)andc2(t), respectively.
Letp=0.05; based on Theorem 10, we can get the corresponding sampled-data controller gain matrices
(40)K1=[-0.1314-0.1212-0.0709-1.1841],K2=[-0.4319-0.1243-0.0753-1.3982],K3=[0.1045-0.1144-0.0747-0.9176],K4=[-0.5565-0.1190-0.0665-1.5593],K5=[0.5769-0.1184-0.0688-0.4136],K6=[0.1801-0.1154-0.0846-0.8069].
In the numerical simulation, assume that the initial values arex1(0)=[32]T,x2(0)=[-1-3]T, x3(0)=[24]T,x4(0)=[5-1]T,x5(0)=[-43]T,x6(0)=[41]T, ands(0)=[3-4]T. The state trajectories of the synchronization error and the control inputsui(t)are given in Figures 5 and 6, respectively.
Random coupling strength c1(t).
Random coupling strength c2(t).
Synchronization error states.
Sampled-data control inputs.
5. Conclusions
In this paper, the sampled-data synchronization problem has been considered for a kind of complex dynamical networks with time-varying coupling delay and random coupling strengths. The sampling period and random coupling strengths considered here are assumed to be time varying but bounded and to obey normal distribution, respectively. By capturing the characteristic of sampled-data system, a novel discontinuous Lyapunov functional is defined. By using reciprocally convex approach and convex combination technique, a mean square synchronization criterion is proposed based on LMIs. The corresponding desired sampled-data controllers are designed. Numerical examples show the effectiveness of the proposed result.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
The work is Supported by the National Natural Science Foundation of China (Grant nos. 61203049 and 61303020), the Doctoral Startup Foundation of Taiyuan University of Science and Technology (Grant no. 20112010), and Shanxi Education Department Foundation (Grant no. 20121068).
StrogatzS. H.Exploring complex networks200141068252682762-s2.0-003582615510.1038/35065725AlbertR.BarabásiA.-L.Statistical mechanics of complex networks2002741479710.1103/RevModPhys.74.47MR1895096ZBL1205.82086WangX. F.ChenG. R.Complex networks: small-world, scale-free and beyond2003316202-s2.0-094227688010.1109/MCAS.2003.1228503WangX. F.ChenG. R.Synchronization in scale-free dynamical networks: robustness and fragility2002491546210.1109/81.974874MR1874226WangY.WangZ. D.LiangJ. L.A delay fractioning approach to global synchronization of delayed complex networks with stochastic disturbances200837239606660732-s2.0-5034909677410.1016/j.physleta.2008.08.008ZhouJ.LuJ.-A.LüJ.Adaptive synchronization of an uncertain complex dynamical network200651465265610.1109/TAC.2006.872760MR2228029LuJ. Q.HoW. D. C.CaoJ. D.A unified synchronization criterion for impulsive dynamical networks20104671215122110.1016/j.automatica.2010.04.005MR2877227ZBL1194.93090YuW. W.ChenG. R.LüJ. H.On pinning synchronization of complex dynamical networks200945242943510.1016/j.automatica.2008.07.016MR2527339ZBL1158.93308SongQ.CaoJ. D.On pinning synchronization of directed and undirected complex dynamical networks201057367268010.1109/TCSI.2009.2024971MR2729966LuJ.HoW. D. C.Globally exponential synchronization and synchronizability for general dynamical networks20104023503612-s2.0-7794977290810.1109/TSMCB.2009.2023509YaoJ.GuanZ.-H.HillD. J.Passivity-based control and synchronization of general complex dynamical networks20094592107211310.1016/j.automatica.2009.05.006MR2889275ZBL1175.93208ParkM. J.KwonO. M.ParkJ. H.LeeS. M.ChaE. J.Synchronization criteria for coupled stochastic neural networks with time-varying delays and leakage delay201234951699172010.1016/j.jfranklin.2012.02.002MR2917550ZBL1254.93012XuY. H.ZhouW. N.FangJ. A.SunW.PanL.Topology identification and adaptive synchronization of uncertain complex networks with non-derivative and derivative coupling201034781566157610.1016/j.jfranklin.2010.07.007MR2720944ZBL1202.93023WuC. W.Synchronization and convergence of linear dynamics in random directed networks20065171207121010.1109/TAC.2006.878783MR2239561LeeT. H.ParkJ. H.JiD. H.KwonO. M.LeeS. M.Guaranteed cost synchronization of a complex dynamical network via dynamic feedback control2012218116469648110.1016/j.amc.2011.11.112MR2879127ZBL1238.93070WangJ.-L.WuH.-N.Local and global exponential output synchronization of complex delayed dynamical networks201267149750410.1007/s11071-011-9998-1MR2869217ZBL1242.93008YangX. S.CaoJ. D.LuJ. Q.Synchronization of Markovian coupled neural networks with nonidentical mode-delays and random coupling strengths2012231607110.1109/TNNLS.2011.2177671YangX. S.CaoJ. D.LuJ. Q.Synchronization of coupled neural networks with random coupling strengths and mixed probabilistic time-varying delays201323182060208110.1002/rnc.2868MR3147017ZBL1278.93021FridmanE.SeuretA.RichardJ.-P.Robust sampled-data stabilization of linear systems: an input delay approach20044081441144610.1016/j.automatica.2004.03.003MR2153059ZBL1072.93018FridmanE.A refined input delay approach to sampled-data control201046242142710.1016/j.automatica.2009.11.017MR2877089ZBL1205.93099LamB. H. K.Output-feedback sampled-data polynomial controller for nonlinear systems201147112457246110.1016/j.automatica.2011.08.009MR2886876ZBL1228.93077LiuK.FridmanE.Wirtinger's inequality and Lyapunov-based sampled-data stabilization201248110210810.1016/j.automatica.2011.09.029MR2879416ZBL1244.93094LuJ. G.HillD. J.Global asymptotical synchronization of chaotic Lur'e systems using sampled data: a linear matrix inequality approach20085565865902-s2.0-4664910406510.1109/TCSII.2007.916788WuZ. G.ShiP.SuH. Y.ChuJ.Exponential synchronization of neural networks with discrete and distributed delays under time-varying sampling20122391368137610.1109/TNNLS.2012.2202687WuZ. G.ShiP.SuH. Y.ChuJ.Stochastic synchronization of Markovian jump neural networks with time-varying delay using sampled-data20134361796180610.1109/TSMCB.2012.2230441ZhangC. K.HeY.WuM.Exponential synchronization of neural networks with time-varying mixed delays and sampled-data2010741–32652732-s2.0-7864945260510.1016/j.neucom.2010.03.020LiN.ZhangY. L.HuJ. W.NieZ. Y.Synchronization for general complex dynamical networks with sampled-data20117458058112-s2.0-7865075283010.1016/j.neucom.2010.11.007WuZ.-G.ParkJ. H.SuH.SongB.ChuJ.Exponential synchronization for complex dynamical networks with sampled-data201234992735274910.1016/j.jfranklin.2012.09.002MR2992095ZBL1264.93013WuZ. G.ShiP.SuH. Y.ChuJ.Sampled-data exponential synchronization of complex dynamical networks with time-varying coupling2013241177118710.1109/TNNLS.2013.2253122ParkP.KoJ. W.JeongC.Reciprocally convex approach to stability of systems with time-varying delays201147123523810.1016/j.automatica.2010.10.014MR2878269ZBL1209.93076