Abstract
This paper deals with the pinning synchronization problem for a class of neural networks with time-varying delay and nonsymmetrical coupling. The weighted average of all the node states is selected as the controlled synchronization state. A pinning feedback controller is proposed, and some sufficient conditions are derived to guarantee the global asymptotical synchronization by use of the Lyapunov–Krasovskii function method. Consequently, the adaptive pinning synchronization is also investigated. Finally, a numerical example is given to show the effectiveness of the main results.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
Over the past few decades, neural networks have received considerable attention for their important applications in many fields of image processing, pattern recognition and associative memories, etc [1–3]. Since the switching speed of information processing and the inherent neuron communication is finite, time delays are always unavoidably existent in neural networks, which may lead to instability or significantly deteriorated performances. Then it is necessary to take into account the time delay in neural networks [4–6].
Also, much concern has been paid to the synchronization of coupled neural networks for its potential applications, such as secure communication, and so on [7–13]. He and Cao [9] discussed the global synchronization of coupled neural networks with one single time-varying delay coupling, and established several sufficient criteria. Song et al. [10] proposed a pinning synchronization control scheme for a class of linearly coupled neural networks. Tang and Wong [11] considered the distributed synchronization problem of coupled neural networks by randomly occurring control method. Employing discontinuous impulsive control scheme, [12] put forward some criteria for the synchronization of coupled networks by impulsive control. Liu and Chen [13] investigated the cluster synchronization problem of the directed networks by intermittent pinning controller. In fact, not all the complex networks can synchronize by themselves. Those, who can not achieve synchronization by themselves, need introducing proper controllers to force them to synchronize. Among some existing control methods, the pinning control is a very important one.
Pinning control strategy applies the local feedback injections to a small fraction of nodes to achieve the global performances of the entire networks, which is effective and practical, especially for the large-size networks. Nowadays, the pinning synchronization of complex networks has been widely investigated, and meanwhile several selection rules of pinned nodes have been proposed in the existing literatures. Wang and Chen [14] concluded that the most-highly connected nodes can be pinned to get the better performance for the undirected networks. Chen et al. [15] showed that even using one single pinned node could control the whole networks as long as the coupling strength was large enough. Song and Cao [16] proposed the pinned nodes selection rule according to the out-degree and in-degree of the nodes, and the synchronization problem was studied for undirected and directed networks. In addition, the pinning control problems of complex networks were investigated, and some sufficient pinning conditions were established in [17–21].
As far as the synchronization state is concerned, most previous studies on synchronization problem of complex networks choose a special solution of the homogeneous systems as the controlled synchronization state [10, 16, 17, 19, 20, 22, 23], usually described by \(\dot{s}(t)=f(t,s(t))\). As pointed out in [24, 25], the synchronization of the networks not only depends on the topology, but also is heavily related to the dynamic of each node. Then, the contributions and influences of all nodes in synchronization seeking processes of the dynamical networks should be taken into account. In [26], authors considered the average of all the node states \(s(t)=\sum _{i=1}^{N}\frac{1}{N}x_i(t)\) to be the synchronization state. For the purpose of practical control strategy, authors in [24, 27, 28] further investigated the synchronization of complex networks to a desired synchronization state, which was in term of the weighted average of all node states in the networks, that was \(s(t)=\sum _{i=1}^{N}\xi _ix_i(t)\) with \(\sum _{i=1}^{N}\xi _i=1~(\xi _i>0)\). In some degree, there research results are important and valuable to the realistic dynamical behaviors of complex works. To the best of authors’ knowledge, the synchronization of delayed coupled neural networks in the above sense has not been fully studied by pinning control yet, and it is just the motivation of this paper.
Based on the aforementioned discussions, this paper aims to analyze the synchronization problem for coupled neural networks with the time-varying delay. The synchronization in a different sense is focused on via pinning control and adaptive pinning control respectively. By employing the Lyapunov–Krasovskii (L–K) function theory and an appropriate pinning staggery, the pinning controlled synchronization conditions are derived. Compared with the existing results, the main novelty of this paper lies in two aspects. First, the synchronization state, in terms of the weighted average of all node states, is introduced in coupled neural networks with different time-varying delays. Second, the more simple linear matrix inequalities conditions are established, and then the effectiveness of the proposed method can be checked conveniently by numerical results.
The remainder of this paper is organized as follows. In Sect. 2, the considered coupled neural networks model is formulated and the pinning synchronization problem is introduced. The pinning synchronization conditions are obtained for concerned networks in Sect. 3. Section 4 considers the adaptive pinning synchronization. A numerical example is provided to illustrate the validity of the proposed method in Sect. 5. Some conclusions are made in Sect. 6, together with some potential future study.
Throughout this paper, the superscript \(^T\) stands for the transpose of a matrix. \(\mathbb {R}^{n}\) and \(\mathbb {R}^{n\times n}\) denote the n-dimension Euclidean space and set of all \(n\times n\) real matrices, respectively. A real symmetric matrix \(P>0~\) (\(\ge \)0) denotes P being a positive definite (positive semi-definite) matrix. For a matrix \(P\in \mathbb {R}^{n\times n}\), \(P^s=\frac{P+P^T}{2}\) is its symmetric part, \(\lambda _{max}(P)\) and \(\lambda _{min}(P)\) represent its maximum and minimum eigenvalues, respectively. \(\mathrm{diag}\{\ldots \}\) denotes a block-diagonal matrix, and \(\Vert \cdot \Vert \) represents the Euclidean norm of a vector and its induced norm of a matrix. \(I_n\) stands for an n-dimensional identity matrix. Matrices, if not explicitly stated, are assumed to have compatible dimensions. The symmetric terms in a symmetric matrix are denoted by \(*\).
2 Problem Formulation
Consider an array of linearly coupled neural networks consisting of N identical nodes in which the dynamics of the ith \((1\leqslant i \leqslant N)\) node is described by
where \(x_i(\cdot )=[x_{i1}(\cdot ),\ldots , x_{in}(\cdot )]^T\in \mathbb {R}^n\) is the state vector, \(C=\mathrm{diag}\{c_1,\ldots ,c_n\}\) is a positive diagonal matrix. \(A=(a_{ij})_{n\times n}\) and \(B=(b_{ij})_{n\times n}\) represent the connection weight matrix and the delayed connection weight matrix, respectively. \(f(x_i(\cdot ))=[f_{1}(x_{i1}(\cdot )),\ldots ,f_{n}(x_{in}(\cdot ))]^T\in \mathbb {R}^n\) denotes the neuron activation function. \(J(t)=[J_1(t),\ldots ,J_n(t)]^T\in \mathbb {R}^n\) is the external input vector. \(\tau _1(t)\) and \(\tau _2(t)\) are the time-varying delays, which can be constant or bounded and differentiable functions. The positive constants \(c_1\) and \(c_2\) are the strengths for the constant coupling and delayed coupling, respectively. \( G_1=(G_{1,ij})_{N\times N}\) is the constant coupling configuration matrix defined to be diffusive, i.e., \(G_{1,ij}\geqslant 0~(i\ne j)\) and \(G_{1,ii} =-\sum _{j=1,j\ne i}^{N}G_{1,ij}\). The delayed coupling configuration matrix \( G_2=(G_{2,ij})_{N\times N}\) is also diffusive with \(G_{2,ij}\geqslant 0~(i\ne j)\) and \(G_{2,ii} =-\sum _{j=1,j\ne i}^{N}G_{2,ij}\). The coupling matrices \(G_1\) and \(G_2\) are nor required to be symmetric or irreducible.
Assumption 2.1
\(\tau _1(t)\) and \(\tau _2(t)\) are bounded and differentiable with \(\bar{\tau }_1=\sup _{t\geqslant 0}\tau _1(t)\) and \(\bar{\tau }_2=\sup _{t\geqslant 0}\tau _2(t)\). Meanwhile, \(\dot{\tau }_1(t)\leqslant \mu _1<1\) and \(\dot{\tau }_2(t)\leqslant \mu _2<1\) for all \(t\geqslant 0\) with \(\mu _1>0\) and \(\mu _2>0\).
Assumption 2.2
There exists a constant matrix L such that
Remark 2.1
Assumptions 2.1 and 2.2 impose restrictions on the time-varying delays and the activation function, and are widely used in the literature [16, 20].
The initial conditions for coupled networks (1) are given by
where \(\bar{\tau }=\max \{\bar{\tau }_1,\bar{\tau }_2\}\), \(C([-\bar{\tau }, 0],\mathbb {R})\) denotes the set of continuous functions from \([-\bar{\tau }, 0]\) to \(\mathbb {R}\). Since the synchronization problem is considered, we now give the definition.
Definition 2.1
The coupled neural networks (1) is said to be globally asymptotically synchronizable to the goal trajectory s(t) if the following relations
hold for the initial conditions (3).
We will address the pinning synchronization of (1) by feedback controllers, and the controlled networks model of the ith \((1\leqslant i \leqslant N)\) node follows
This paper aims to design and implement an appropriate pinning controller for (4) such that all the states of the controlled networks can globally asymptotically synchronize to a desired synchronization s(t), which is presented in terms of a weighted average of all the states in the networks. More specifically, that is
where \(x_i(t)\) is the state of the delayed dynamical networks (4), and \(\xi _k\) are chosen to satisfy \(\sum _{k=1}^N\xi _k=1\) with \(\xi _k>0\), \(i=1,\ldots , N\). Such synchronization strategy takes into account the contributions and confluences of all the nodes of the dynamical networks, compared with a special solution of the homogeneous systems being the synchronization state in most existing literatures. Moreover, the synchronization state s(t) may also be an equilibrium point, a periodic orbit, or a chaotic attractor, among others [24, 27].
The following pinning feedback controller is adopted,
where the feedback control gain \(d_i>0\) if the ith node is pinned, otherwise \(d_i=0\). The next section will provide several sufficient conditions for global asymptotical synchronization, where some special cases are also considered in corollaries.
3 Sufficient Conditions for Global Asymptotical Synchronization
For convenience, we denote that \(\Lambda =\mathrm{diag}\{\xi _1,\ldots ,\xi _N\}\), \(\Omega _1=(\Lambda ^{\frac{1}{2}}G_1\Lambda ^{-\frac{1}{2}})\otimes \Gamma _1\) and \(\Omega _2=(\Lambda ^{\frac{1}{2}}G_2\Lambda ^{-\frac{1}{2}})\otimes \Gamma _2\). By defining the synchronization error \(e_i(t)=x_i(t)-s(t)\), the error dynamical system is governed as follows
where
Before giving the main results, two lemmas are provided first.
Lemma 3.1
([29, Schur Complement]). The linear matrix inequality \( \left[ \begin{array}{cc}S_{11} &{} S_{12} \\ *&{} S_{22}\end{array}\right] <0\) is equivalent to \(S_{22}<0\) and \(S_{11}-S_{12}S_{22}^{-1}S_{12}^T<0,\) where \(S_{11}=S_{11}^T\), \(S_{22}=S_{22}^T\).
Lemma 3.2
([10]). For a symmetric matrix \(M\in \mathbb {R}^{N\times N}\) and a diagonal matrix \(D=\mathrm{diag}\{\bar{d}_1,\ldots ,\bar{d}_l,\underbrace{0,\ldots ,0}_{N-l}\}\) with \(\bar{d}_i>0, i=1,\ldots ,l~(1\leqslant l\leqslant N)\). Set \(M-D=\left[ \begin{array}{cc}E-\widetilde{D} &{}F\\ F^T &{} M_l\end{array}\right] \), where \(M_l\) is the minor matrix of M by removing its first l row-column pairs, E and F are matrices with appropriate dimensions, \(\widetilde{D}=\mathrm{diag}\{\bar{d}_1,\ldots ,\bar{d}_l\}.\) If \(\bar{d}_i>\lambda _{\max }(E-FM_l^{-1}B^T),i=1,\ldots ,l\), then \(M-D<0\) is equivalent to \(M_l<0\).
Theorem 3.1
Under Assumptions 2.1–2.2, the pinning controlled networks (4) with controller (6) can globally asymptotically synchronize to the desired synchronization state s(t) given by (5), if there exist positive scalars \(\epsilon \) and \(\gamma _2\) such that the following inequality holds
where \(\Phi =-C+\frac{AA^T}{2}+\frac{BB^T}{2}+\frac{L^TL}{2}+(\gamma _1+\gamma _2)I_n\) with \(\gamma _1=\frac{\lambda _1+\epsilon }{1-\mu _1}\) and \(\lambda _1=\frac{1}{2}\lambda _{\max }(L^TL)\).
Proof
Firstly, a Lyapunov–Krasovskii function is constructed for (7) as follows
where \(\gamma _1\) and \(\gamma _2\) are positive scalars to be specialized below. According to Assumptions 2.2 and using Kronecker product technique, the derivative of V(t) along the trajectory of system (4) can be calculated as
Noticing that \(\sum _{i=1}^{N}\xi _ie_i(t)=0\), we have \(\sum _{i=1}^{N}\xi _ie_i^T(t)\Pi (t)=0\). Based on Assumption 2.2 and the fact \(2x^Ty\leqslant x^Tx+y^Ty\), \(\forall x,y\in \mathbb {R}^n\), one can make the following estimation
Similarly, it follows that
By denoting \(\nu (t)=[\sqrt{\xi _1}e_1^T(t),\ldots ,\sqrt{\xi _N}e_N^T(t)]^T\), the transformations can be made as
and
Submitting the above inequalities (10)–(13) into (9) yields that
Choose \(\gamma _1=\frac{\lambda _1+\epsilon }{1-\mu _1}\) such that \(\lambda _1-\gamma _1(1-\mu _1)=-\epsilon <0.\) Then, the above inequality (14) degenerates into
where \(\eta ^T(t)=\left[ \begin{array}{cc}\nu ^T(t)&\nu ^T(t-\tau _2(t))\end{array}\right] ^T\). Furthermore, from the condition (8) in Theorem 3.1 and (15), it follows that \(\dot{V}(t)<0\), which indicates that the error system (7) is globally asymptotically stable, i.e., \(\lim _{t\rightarrow \infty }e_i(t)=0\). Therefore, the pinning controlled networks (4) globally asymptotically synchronize to the desired synchronization state s(t) in (5). Now, the proof follows. \(\square \)
Remark 3.1
The condition (8) in Theorem 3.1 is given in form of linear matrix inequality (LMI) with respect to the invariants \(\epsilon \) and \(\gamma _2\), which can be checked by Matlab Toolbox. In addition, the minimization of the estimations of control gains \(d_i\) can be conducted. For example, it is assumed that the same control gain d is chosen, and then \(D=d\Sigma \) in Theorem 3.1, where \(\Sigma =\mathrm{diag}\{\sigma _1,\ldots ,\sigma _N\}\) represents the pinning matrix, and \(\sigma _i=1\) when the ith node is pinned, otherwise \(\sigma _i=0\). The minimization of the estimation of control gain d can be formulated as
In fact, the single control gain also can be optimized by similar method.
When there does not exist the uncoupled discrete delay or the coupling delay in networks model, the controlled system (4) will degenerate into the following forms respectively:
Based on Theorem 3.1, we can derive the corresponding results for the pinning synchronization of the above systems (16) and (17). Now, the following corollaries are proposed, and the detailed proofs are omitted.
Corollary 3.1
Under Assumptions 2.1–2.2, the pinning controlled networks (16) with controller (6) can globally asymptotically synchronize to the desired synchronization state s(t) in (5), if there exists a positive scalar \(\gamma _2\) such that the following inequality holds
with \(\bar{\Phi }=-C+\frac{AA^T}{2}+\frac{L^TL}{2}+\gamma _2I_n\).
Corollary 3.2
Under Assumption 2.1–2.2, the pinning controlled networks (17) with controller (6) can globally asymptotically synchronize to the desired synchronization state s(t) in (5), if there exists a positive scalar \(\epsilon \) such that the following inequality holds
where \(\bar{\bar{\Phi }}=-C+\frac{AA^T}{2}+\frac{BB^T}{2}+\frac{L^TL}{2}+\gamma _1I_n\) with \(\gamma _1=\frac{\lambda _1+\epsilon }{1-\mu _1}\) and \(\lambda _1\) is defined in Theorem 3.1.
Remark 3.2
By setting \(B=0,~\tau _1(t)=0,~\gamma _1=0\) in Theorem 3.1, we can get Corollary 3.1 on the pinning synchronization problem for system (16). Moreover, letting \(G_2=0,~\Gamma _2=0,~\gamma _2=0\), Corollary 3.2 can be concluded for system (17). Furthermore, choosing \(\mu _1=\mu _2=0\) in Theorem 3.1, one can also derive the corresponding results for the constant delays case. With \(\xi _i=\frac{1}{N}~(1\leqslant i\leqslant N)\), the achieved results are the ones for the case that the average of all node states is the desired synchronization state.
Corollary 3.3
Under Assumptions 2.1–2.2, the pinning controlled networks (4) with controller (6) can globally asymptotically synchronize to the desired synchronization state \(s(t)=\frac{1}{N}\sum \limits _{k=1}^{N}x_k(t)\), if there exist positive scalars \(\epsilon \) and \(\gamma _2\) such that the following inequality holds
where \(\Phi \) is the same as the one in Theorem 3.1.
Remark 3.3
Compared with the recent literatures [24] and [27], which focus on the weighted average synchronization problems, the results in this paper have a better application to some extent. First, only the coupling delay is considered in [24], and it is a constant delay. As for the networks in [27], the coupling delay is not involved. Second, from a computational point of view, the criteria in [24, 27] are related to some eigenvalues, while our proposed conditions are in form of linear matrix inequalities, which can be checked conveniently by Matlab LMI Toolbox.
4 Adaptive Pinning Synchronization
Based on the above pinning synchronization analysis, we will further study the adaptive pinning synchronization of the networks (4). Without loss of generality, let the nodes \(i_1,\ldots ,i_l\) be selected as the pinned nodes and rearrange the order of the nodes such that the first l nodes are pinned or controlled. Therefore, it indicates that \(d_i=0\) for \(i=l+1,\ldots , N\) in the pinning controller (6). If \(d_i\) for \(i=1,\ldots ,l\) are designed to be time-variable, the details are as follows
where \(q_i\) are positive constants. The controller \(u_i(t)\) in (18) is said to be an adaptive pinning controller. In the following content, we will derive some conditions under which the coupled neural networks (4) can achieve the global synchronization to the desired state s(t) in (5) with the adaptive pinning controller (18). To avoid unnecessary duplication with Theorem 3.1, we will omit some details and give a sketch of the proof.
Theorem 4.1
Under Assumptions 2.1–2.2, the coupled neural networks (4) can globally asymptotically synchronizes to the desired orbit s(t) in (5) with the adaptive pinning controller (18), if there exist scalars \(\epsilon >0\) and \(\gamma _2>0\) such that the following inequality holds
where \(\Phi \) is the one in Theorem 3.1.
Proof
Construct a Lyapunov function for coupled networks (4) as follows
where \(q_i>0\) are defined in (18), and \(d_i^{*}>0\) are constants to be determined below. Using Kronecker product technique and Assumption 2.1, the time derivative of \(\bar{V}(t)\) along the trajectory of system (4) can be computed and estimated as
where \(D^{*}=\mathrm{diag}\{d_1^{*},\ldots ,d_l^{*},\underbrace{0,\ldots ,0}_{N-l}\}\), \(\nu (t)\) and \(\lambda _1\) are the ones in Theorem 3.1. Letting \(\gamma _1=\frac{\lambda _1+\epsilon }{1-\mu }\) and from (20), it is known that
where \(\eta (t)\) has been defined in Theorem 3.1, and \(\bar{D}=\mathrm{diag}\{d_1^{*},\ldots ,d_l^{*},\underbrace{0,\ldots ,0}_{2N-l}\}\).
On the other hand, denote the matrix \(M=\left[ \begin{array}{cc} I_N\otimes \Phi +c_1\Omega _1^s &{} \frac{1}{2}c_2\Omega _2 \\ *&{} -\gamma _2(1-\mu _2)I_{nN}\end{array}\right] \), which is symmetric, and let
where \(\widehat{D}=\widehat{\bar{D}}\otimes I_n\) with \(\widehat{\bar{D}}=\mathrm{diag}\{d_1^{*},\ldots ,d_l^{*}\}\), and \(\widehat{A}, \widehat{B}\) are matrices with appropriate dimensions. According to Lemma 2.1, if the positive constants \(d_i^{*}\) satisfy \(d_i^{*}>\lambda _{\max }(\widehat{A}-\widehat{B}(M)^{-1}_{nl}\widehat{B}^T) , (i=1,\ldots ,l)\), then \((M)_{nl}<0\) can give that \(M-\bar{D}\otimes I_n<0\). Together with (19) and (21), we have \(\dot{V}(t)\leqslant 0\) and \(\dot{V}(t)=0\) if and only if \(e(t)=e(t-\tau (t))=0_{nN}\). Therefore, the pinning controlled coupled neural networks (4) can globally asymptotically synchronize to the trajectory s(t) in (5) by the adaptive pinning controller (18) if the condition (19) holds. This completes the proof. \(\square \)
Remark 4.1
In the adaptive pinning synchronization design, we can rearrange network nodes in ascending order such that the first l nodes are pinned candidates. Based on the out-degree and in-degree of all the nodes, the pinned nodes can be selected in this paper. The details can be seen in [10] and [16].
5 Numerical Example
We consider the pinning synchronization of coupled neural networks (4) with ten nodes (i.e., \(N=10\)), and the relevant parameters are given by [10, 30]:
The activation function is descried as \(f(x_i(t))=[\tanh (x_{i1}(t)),\tanh (x_{i2}(t))]^T\). It can be seen from Assumption 2.1 that \(L=\mathrm{diag}\{1.0,1.0\}\). The coupling matrix \(G_1\) is given by (22), which is determined by a directed topology (See Fig. 2 in [10]), and the delayed coupling matrix is set to be \(G_2=0.5G_1\). The coupling strengths are chosen as \(c_1=120,~c_2=5.\) The weighted average values \(\xi _i\) are chosen randomly in (0, 1). Letting \(e_i(t)=x_i(t)-s(t)\) with \(s(t)=\sum _{i=1}^{10}\xi _ix_i(t)\), and the synchronization error is measured by \(E(t)=\sqrt{\frac{1}{10}\sum _{i=1}^{10}e_i^T(t)e_i(t)}\).
First, we set the time delays are constant scalars \(\tau _1(t)=\tau _2(t)=1.0\) with \(\mu _1=\mu _2=0\), and the inner connecting matrices are defined as \(\Gamma _1=\Gamma _2=I_{n}\). For the case, the pinning synchronization to the isolate node is studied in [10]. Now, we deal with the pinning synchronization to the weighted average of all node states. Employing Theorem 4.1 and pinning the nodes 3, 4, 9, we can check the synchronization of the concerned system to s(t) by LMI Matlab toolbox with controller (18), and obtain that the feasible solutions are \(\epsilon =1.5894\) and \(\gamma _2=24.4658\). The state trajectories, the synchronization error E(t) and the adaptive feedback gains are illustrated in Fig. 1. It shows that the coupled neural networks with parameters in (22) can synchronize to the state s(t) under the adaptive pinning controller (18).
For the case that the time delays are different time-varying functions \(\tau _1(t)=0.7-0.3\cos (t)\) and \(\tau _2(t)=0.5-0.2\sin (t)\) and resorting to Matlab LMI toolbox, solving the inequality (19) for setting \(\mu _1=0.3\) and \(\mu _2=0.2\) leads to some feasible solutions \(\epsilon =1.9892\) and \(\gamma _2=21.8983\). The corresponding simulation results are given in Fig. 2.
Remark 5.1
Since the weighted average synchronization problem is concerned, the results in [10, 30] may not be employed here. In addition, for the case that \(\tau _1(t)\) and \(\tau _2(t)\) are different time-varying functions, the methods in [24, 27] may not work.
6 Concluding Remarks
This paper has investigated the problem of pinning synchronization for a class of coupled neural networks with time-varying delays. The Lyapunov–Krasovskii function method is developed to derive some sufficient criteria, which can guarantee that the pinning controlled networks globally asymptotically synchronize to the desired state that is in terms of the weight average of all node states in networks. Further study can focus on some advanced analysis technique to obtain the less conservative results, the synchronization of switched networks, and other control methods for complex networks.
References
Narendra K, Parthasarathy K (1990) Identification and control of dynamical systems using neural networks. IEEE Trans Neural Netw 1(1):4–27
Bishop C (1995) Neural networks for pattern recognition. Oxford University Press, New York
Watta PB, Wang K, Hassoun MH (1997) Recurrent neural nets as dynamical boolean systems with application to associative memory. IEEE Trans Neural Netw 8(5):1268–1280
Arunkumar A, Sakthivel R, Mathiyalagan K (2015) Robust reliable \(H_{\infty }\) control for stochastic neural networks with randomly occurring delays. Neurocomputing 149:1524–1534
Sakthivel R, Vadivel P, Mathiyalagan K, Arunkumar A, Sivachitra M (2015) Design of state estimation for bidirectional associative memory neural networks with leakage delays. Inf Sci 296:263–274
Wu YY, Cao JD, Alofic A, AL-Mazrooeic A, Elaiwc A (2015) Finite-time boundedness and stabilization of uncertain switched neural networks with time-varying delay. Neural Netw 69:135–143
Cao J, Li P, Wang W (2006) Global synchronization in arrays of delayed neural networks with constant and delayed coupling. Phys Lett A 353:318–325
Lu J, Ho DWC, Liu M (2007) Globally exponential synchronization in an array of asymmetric coupled neural networks. Phys Lett A 369:444–451
He W, Cao JD (2009) Global synchronization in arrays of coupled networks with one single time-varying delay coupling. Phys Lett A 373(31):2682–2694
Song Q, Cao JD, Liu F (2012) Pinning synchronization of linearly coupled delayed neural networks. Math Comput Simul 86:39–51
Tang Y, Wong WK (2013) Distributed synchronization of coupled neural networks via randomly occurring control. IEEE Trans Neural Netw Learn Syst 24(3):43–447
Li P, Cao J, Wang Z (2007) Robust impulsive synchronization of coupled delayed neural networks with uncertainties. Phys A 373:261–272
Liu X, Chen T (2011) Cluster synchronization in directed networks via intermittent pinning control. IEEE Trans Neural Netw 22(7):1009–1020
Wang X, Chen G (2002) Pinning control of scale-free dynamical networks. Phys A 310:521–531
Chen T, Liu X, Lu W (2007) Pinning complex networks by a single controller. IEEE Trans Circuits Syst I(54):1317–1326
Song Q, Cao J (2010) On pinning synchronization of directed and undirected complex dynamical networks. IEEE Trans Circuits Syst I(57):672–680
Yu W, Chen G, Lü J (2009) On pinning synchronization of complex dynamical networks. Automatica 45(2):429–435
Zhou J, Lu J (2008) Pinning adaptive synchronization of a general complex dynamical network. Automatica 44(4):996–1003
Zheng C, Cao J (2014) Robust synchronization of coupled neural networks with mixed delays and uncertain parameters by intermittent pinning control. Neurocomputing 141:153–159
Li L, Cao J (2011) Cluster synchronization in an array of coupled stochastic delayed neural networks via pinning control. Neurocomputing 74:846–856
Sorrentino F, Bernardo M, Garafalo F, Chen G (2007) Controllability of complex networks via pinning. Phys Rev E 75 (4), Article ID: 046103, 6 p
Wang Q, Yu C, Gao H (2014) Synchronization of identical linear dynamic systems subject to input saturation. Syst Control Lett 64:107–113
Wang Q, Yu C, Gao H (2014) Semiglobal synchronization of multiple generic linear agents with input saturation. Int J Robust Nonlinear Control 24(18):3239–3254
Zhou J, Wu Q, Xiang L, Cai S, Liu Z (2011) Impulsive synchronization seeking in general complex delayed dynamical networks. Nonlinear Anal 5:513–524
Cai S, Zhou J, Xiang L, Liu Z (2008) Robust impulsive synchronization of complex delayed dynamical networks. Phys Lett A 372:4990–4995
Zhou J, Chen T (2006) Synchronization in general complex delayed dynamical networks. IEEE Trans Circuits Syst I(53):733–744
Wu Q, Zhang H (2015) Drive network to a desired orbit by pinning control. Kybernetika 51(1):150–172
Cai ZM, Zhou PP, Liu ZR (2015) Synchronization analysis of hybrid-coupled delayed dynamical networks with impulsive effects: a unified synchronization criterion. J Frankl Inst 352:2065–2089
Boyd SP, El Ghaoui L, Feron E, Balakrishnan V (1994) Linear matrix inequalities in system and control theory, vol 15. Society for industrial and applied mathematics (SIAM), Philadelphia
Lu W, Chen T (2004) Synchronization of coupled connected neural networks with delays. IEEE Trans Circuits Syst I 51(12):2491–2503
Acknowledgments
This research was supported in part by the Projects of International Cooperation and Exchanges NSFC (61520106009), in part by the National Natural Science Foundation of China (61403027), in part by the Fundamental Research Funds for the Central Universities (FRF-TP-15-071A1), and in part by the Science Foundation of China Postdoctoral under Grant 2014M561543.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Li, Q., Guo, J., Wu, Y. et al. Weighted Average Pinning Synchronization for a Class of Coupled Neural Networks with Time-Varying Delays. Neural Process Lett 45, 95–108 (2017). https://doi.org/10.1007/s11063-016-9514-1
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11063-016-9514-1