Abstract
This paper studies the exponential synchronization problem for a class of stochastic perturbed chaotic neural networks with both Markovian jump parameters and mixed time delays. The mixed delays consist of discrete and distributed time-varying delays. At first, based on a Halanay-type inequality for stochastic differential equations, by virtue of drive-response concept and time-delay feedback control techniques, a delay-dependent sufficient condition is proposed to guarantee the exponential synchronization of two identical Markovian jumping chaotic-delayed neural networks with stochastic perturbation. Then, by utilizing the Jensen integral inequality and a novel Lemma, another delay-dependent criterion is established to achieve the globally stochastic robust synchronization. With some parameters being fixed in advance, these conditions can be solved numerically by employing the Matlab software. Finally, a numerical example with their simulations is provided to illustrate the effectiveness of the presented synchronization scheme.
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
In 1990, Pecora and Carroll [10] addressed the synchronization of chaotic systems using a drive-response concept. The idea is to use the output of the drive system to control the response system so that they oscillate in a synchronized manner. Research on the synchronization of chaotic activity has broadened considerably in the last few decades. Besides the original master–slave mechanism for chaos synchronization, a wide variety of approaches have been presented for the synchronization of chaotic systems which include linear feedback control, nonlinear feedback control, impulsive control method, and adaptive design control, among many others. Synchronization in chaotic systems has been utilized in many applications. It was used to understand self-organization behavior in the brain as well as in ecological systems and has been applied to secure communications, among others [13–15, 17, 23, 27, 28].
Meanwhile, many neural networks may experience abrupt changes in their structure and parameters caused by some phenomena such as component failures or repairs, changing subsystem interconnections, and abrupt environmental disturbances. In this situation, there exist finite modes in the neural networks, and the modes may be switched (or jumped) from one to another at different times. These kinds of systems are known as Markovian jump neural networks. When noise disturbances are considered in Markovian jump neural networks, this class of neural networks is usually called Markovian jump stochastic neural networks or stochastic neural networks with Markovian switching. It is known that a Markovian jump stochastic neural network is more complicated and comprises a general stochastic neural network as its special case. Owing to the practical importance, many papers have recently devoted to study the stability analysis issue for Markovian jump stochastic neural networks [4, 17, 18, 21, 24, 27]. However, up to now, the synchronization problem for stochastic chaotic neural networks with Markovian switching has received little research attention, despite its practical importance. This situation motivates our present investigation.
This paper studies the exponential synchronization problem for a class of stochastic perturbed chaotic neural networks with both Markovian jump parameters and mixed delays. The mixed delays consist of discrete and distributed time-varying delays. Firstly, by using drive-response concept, a Halanay-type inequality for stochastic differential equations and time-delay feedback control techniques, a delay-dependent sufficient condition is obtained to guarantee the exponential synchronization of two identical Markovian jumping chaotic-delayed neural networks with stochastic perturbation. Next, by means of the Jensen integral inequality and a novel Lemma, another delay-dependent result is established to achieve the globally stochastic robust synchronization. With some parameters being fixed in advance, these conditions are expressed in terms of linear matrix inequalities, which can be solved numerically by employing the Matlab software. Finally, a numerical example is provided to illustrate the effectiveness of the presented synchronization scheme.
Notations Throughout this paper, W T,W −1 denote the transpose and the inverse of a square matrix W, respectively. W > 0(<0) denotes a positive (negative) definite symmetric matrix, I denotes the identity matrix with compatible dimension, the symbol “*” denotes a block that is readily inferred by symmetry. The shorthand \(\hbox{col}\{M_1,M_2,\ldots,M_k\}\) denotes a column matrix with the matrices \(M_1,M_2,\ldots,M_k\). sym(A) is defined as \(A+A^T,\hbox{diag}\{\cdot\}\) stands for a diagonal or block-diagonal matrix. For \({\tau>0, \mathcal{C}\left([-\tau,0];\mathbb{R}^n\right)}\) denotes the family of continuous functions ϕ from [ −τ, 0] to \({\mathbb{R}^n}\) with the norm \(||\phi||=\sup_{-\tau\leq s\leq 0}|\phi(s)|\). Moreover, let \({(\Upomega,\mathbb{F},\mathbb{P})}\) be a complete probability space with a filtration \({\{\mathbb{F}_t\}_{t\geq 0}}\) satisfying the usual conditions and \({\mathbb{E}\{\cdot\}}\) representing the mathematical expectation. Denote by \({\mathcal{C}^p_{\mathbb{F}_0}\left([-\tau,0];\mathbb{R}^n\right)}\) the family of all bounded, \({\mathbb{F}_0}\)-measurable, \({\mathcal{C}\left([-\tau,0];\mathbb{R}^n\right)}\)-valued random variables ξ = {ξ(s): − τ ≤ s ≤ 0} such that \({\sup_{-\tau\leq s\leq 0}\mathbb{E}|\xi(s)|^p<\infty}\). \(||\cdot||\) stands for the Euclidean norm; Matrices, if not explicitly stated, are assumed to have compatible dimensions.
2 Problem description and preliminaries
In this paper, we consider the following stochastic neural networks with both discrete and distributed time-varying delays
where \({x(t)=(x_1(t),x_2(t),\ldots,x_n(t))^T\in\mathbb{R}^n}\) is the state vector associated with n neurons, \({\tilde{\beta}(x(t))=\left(\tilde{\beta}_1(x_1(t)), \tilde{\beta}_2(x_2(t)),\ldots,\tilde{\beta}_n(x_n(t)) \right)^T\in\mathbb{R}^n}\) is the behaved function. \(A(t,\eta(t))=A(\eta(t))+\Updelta A(t,\eta(t)),B(t,\eta(t))=B(\eta(t))+\Updelta B(t,\eta(t)),G(t,\eta(t))=G(\eta(t))+\Updelta G(t,\eta(t))\) are the interconnection matrices representing the weight coefficients of the neurons. A(η(t)), B(η(t)), G(η(t)) are known real constant matrices. \(\Updelta A(t,\eta(t)),\Updelta B(t,\eta(t)),\Updelta G(t,\eta(t))\) are the time-varying structured uncertainties. \({\tilde{f}(x(t))=\left(\tilde{f}_1(x_1(t)),\tilde{f}_2(x_2(t)), \ldots,\tilde{f}_n(x_n(t))\right)^T\in\mathbb{R}^n}\) denotes the neural activation function. The bounded functions τ(t, η(t)), ϱ(t, η(t)) represent unknown time-varying delays with \(0\leq\tau(t,\eta(t))\leq \bar{\tau}(\eta(t))\leq\bar{\tau},\dot{\tau}(t,\eta(t))\leq \tau_d(\eta(t)),0\leq\varrho(t,\eta(t))\leq \bar{\varrho}(\eta(t))\leq\bar{\varrho}\), where \(\bar{\tau}(\eta(t)),\bar{\tau},\bar{\varrho}(\eta(t)),\bar{\varrho}\) are positive scalars, \(\hat{\tau}=\max\{\bar{\tau},\bar{\varrho}\}\). \({\mathfrak{I}}=(I_1,I_2,\ldots,I_n)^T\) is an external input, \(\varphi_1(t)\) is a real-valued initial vector function that is continuous on the interval \(\left[-\hat{\tau},0\right]\). {η(t), t ≥ 0} is a homogeneous, finite-state Markovian process with right continuous trajectories and taking values in finite set \(\mathcal{N}=\{1,2,\ldots,N\}\) based on given probability space \({(\Upomega,\mathbb{F},\mathbb{P})}\) and the initial model η 0. \(\tilde{\rho}(t,x(t),x(t-\tau(t,\eta(t))))\) is called the noise intensity vector. ω(t) is a one-dimensional Brownian motion defined on space \({(\Upomega,\mathbb{F},\mathbb{P})}\) with \({\mathbb{E}\{\hbox{d}\omega(t)\}=0, \mathbb{E}\{[\hbox{d}\omega(t)]^2\}=\hbox{d}t.}\) Let \(\Uppi=[\pi_{ij}]_{N\times N}\) denote the transition rate matrix with transition probability:
where \(\delta>0,\lim_{\delta\rightarrow 0^+}\frac{o(\delta)}{\delta}=0\) and π ij is the transition rate from mode i to mode j satisfying π ij ≥ 0 for i ≠ j with
For convenience, each possible value of η(t) is denoted by \(i(i\in\mathcal{N})\) in the sequel. Then, we have
Throughout this paper, we make the following assumptions:
Assumption 1
The noise intensity vector is assumed to be of the form:
where
with C i , D i being known real constant matrices.
Assumption 2
The admissible parameter uncertainties are assumed to be of the following form:
where \(E_i,H_{ji}(j=1,\ldots,5)\) are known real constant matrices with appropriate dimensions, and \({\Upphi_i(t)\in\mathbb{R}^{n\times n}}\) is the time-varying uncertain matrix satisfying \(\Upphi_i(t)^T\Upphi_i(t)\leq I\) for any t > 0.
Assumption 3
Each \(\tilde{\beta}_j(\zeta)\) is differentiable and satisfies the following condition
where λ j , δ j are known positive real constants.
For notational simplicity, we denote \(\Updelta=\hbox{diag}\{\delta_1,\ldots,\delta_n\},\Uplambda=\hbox{diag}\{\lambda_1,\ldots,\lambda_n\}\).
Assumption 4
Each neural activation function \(\tilde{f}_j(\cdot)(j=1,2,\ldots,n)\) is bounded, differentiable and satisfies the following condition
where γ j , σ j are known real constants.
We denote \(\Upgamma=\hbox{diag}\{\gamma_1,\gamma_2,\ldots,\gamma_n\}, \Upsigma=\hbox{diag}\{\sigma_1,\sigma_2,\ldots,\sigma_n\},\Uptheta=\hbox{diag}\{\theta_1,\theta_2,\ldots,\theta_n\}\), where \(\theta_j=\max\{\vert\gamma_j\vert,\vert \sigma_j\vert\}\).
The system (1) is considered as a drive system. Based on the drive-response concept for synchronization of coupled chaotic systems, which was initially proposed by Pecora and Carroll in [10], the corresponding response system of (1) is given in the following form:
where \({y(t)=(y_1(t),y_2(t),\ldots,y_n(t))^T\in\mathbb{R}^n}\) is the state vector associated with n neurons, \({u_i(t)=(u_{i1}(t),\ldots,u_{in}(t))^T\in\mathbb{R}^n}\) is the state feedback controller given to achieve the exponential synchronization between drive-response system, \(\varphi_2(t)\) is a real-valued continuous vector function on the interval \([-\hat{\tau},0]\).
In order to investigate the problem of exponential synchronization for the chaotic-delayed neural networks with stochastic perturbation, e j (t) = y j (t) − x j (t) is defined as the synchronization error, where x j (t) and y j (t) are the ith state variables of drive system (1) and response system (2), respectively. Therefore, the error dynamical system between (1) and (2) is given as follows:
where \(e(t)=(e_1(t),e_2(t),\ldots,e_n(t))^T, \beta(e(t))=\left(\tilde{\beta}_1(y_1(t))-\tilde{\beta}_1(x_1(t)), \tilde{\beta}_2(y_2(t))-\tilde{\beta}_2(x_2(t)),\ldots, \tilde{\beta}_n(y_n(t))-\tilde{\beta}_n(x_n(t))\right)^T, f(e(t))=\left(\tilde{f}_1(y_1(t))-\tilde{f}_1(x_1(t)), \tilde{f}_2(y_2(t))-\tilde{f}_2(x_2(t)),\ldots, \tilde{f}_n(y_n(t))-\tilde{f}_n(x_n(t))\right)^T, f(e(t-\tau_i(t)))=\left(\tilde{f}_1(y_1(t-\tau_i(t))) -\tilde{f}_1(x_1(t-\tau_i(t))),\,\tilde{f}_2(y_2(t-\tau_i(t))) -\tilde{f}_2(x_2(t-\tau_i(t))),\ldots,\tilde{f}_n(y_n(t-\tau_i(t))) -\tilde{f}_n(x_n(t-\tau_i(t)))\right)^T,\rho_i(t)\doteq\rho(t,e(t), e(t-\tau_i(t)))=\tilde{\rho}(t,y(t),y(t-\tau_i(t))) -\tilde{\rho}(t,x(t),x(t-\tau_i(t))) =C_i(t)e(t)+D_i(t)e(t-\tau_i(t))\).
In this paper, the control input vector with state feedback is designed as follows:
From Assumptions 2,3, we obtain that
Therefore, it follows from [3] that system (3) admits a trivial solution e(t) = 0.
To prove our main theorem, we need the following preliminaries.
Definition 1
Let \({\psi:\mathbb{R}\rightarrow\mathbb{R}}\) be a continuous function, the upper right Dini derivative of ψ(t) is defined as
Definition 2
The drive system (1) and the response system (2) are said to be exponentially synchronized if, for a suitably designed feedback controller, there exist constants υ ≥ 1 and ϑ > 0 such that
for any t ≥ 0, and the constant ϑ is defined as the exponential synchronization rate.
The development of the work in this paper requires the following lemmas.
Lemma 1
(see [8, 12, 16, 25]). Let A, B and C be real constant matrices with appropriate dimensions, matrix \(\Upphi(t)\) satisfies \(\Upphi(t)\Upphi(t)^T\leq I\). Then for any matrix P > 0 and scalar \(\varepsilon>0\), we have the following inequalities:
-
(1)
A T B + B T A ≤ A T P −1 A + B T PB;
-
(2)
\((A+B\Upphi(t)C)^TP^{-1}(A+B\Upphi(t)C)\leq A^T(P-\varepsilon BB^T)^{-1}A+\varepsilon C^TC\,if\,P-\varepsilon BB^T>0. \)
Lemma 2
(see [5], Jensen integral inequality). For any positive symmetric constant matrix \({M\in\mathbb{R}^{n\times n}}\), scalars r 1 < r 2 and vector function \({\varpi:[r_1,r_2]\rightarrow\mathbb{R}^n}\) such that the integrations concerned are well defined, the following matrix inequality holds:
From Lemma 2 of [26], we can easily establish the following Lemma 3, which plays an important role in obtaining our delay-dependent stability result.
Lemma 3
Set λ j , μ j be scalars satisfying \(\lambda_j\leq 1, \lambda_j+\mu_j\leq 4(j=1,2),\Upsigma_l(l=1,\ldots,4)\) be any nonnegative symmetric matrices, θ(t), ν(t) be real functions such that \({\theta:\mathbb{R}^+\rightarrow(\underline {\theta} ,\bar{\theta}), \nu:\mathbb{R}^+\rightarrow(\underline {\nu},\bar{\nu})}\), then we have
If we set λ j = μ j = 1(j = 1,2) in (6), noticing that
then we have the following lemma which is used in [11, 20].
Lemma 4
(See [11, 20]). Set \(\Upsigma_1,\Upsigma_2,\Upsigma_3,\Upsigma_4\) be any nonnegative symmetric matrices, θ(t), ν(t) be real functions such that \({\theta:\mathbb{R}^+\rightarrow(\underline {\theta},\bar{\theta}), \nu:\mathbb{R}^+\rightarrow(\underline{\nu},\bar{\nu})}\), then we have
Remark 1
Usually, we choose λ j = 1, μ j = 3(j = 1, 2) in Lemma 3. Apparently the result derived from Lemma 3 is less conservative than Lemma 4.
3 Main result
As well known, Itô’s formula plays important role in the stability analysis of stochastic Markovian systems and we cite some related results here [1]. Consider a general stochastic Markovian delay system
on t ≥ t 0 with initial value \({z(t_0)=z_0\in\mathbb{R}^n}\), where ϑ > 0 is time delay, \({f:\mathbb{R}^+\times\mathbb{R}^n\times\mathbb{R}^n \times\mathcal{N}\rightarrow\mathbb{R}^n}\) and \({g:\mathbb{R}^+\times\mathbb{R}^n\times\mathbb{R}^n \times\mathcal{N}\rightarrow\mathbb{R}^{n+m}}\). Let \({\mathcal{C}^{2,1}\left(\mathbb{R}^+\times\mathbb{R}^n \times\mathbb{R}^n\times\mathcal{N},\mathbb{R}^+\right)}\) denote the family of all nonnegative functions V(t, z, v, η(t)) on \({\mathbb{R}^+\times\mathbb{R}^n\times\mathbb{R}^n\times\mathcal{N}}\) which are continuously twice differentiable in z, v and once differentiable in t. Let £ be the weak infinitesimal generator of the random process \(\{z(t),\eta(t)\}_{t\geq t_0}\) along the system (7) (see [9, 19]), i.e.,
then, by the Dynkin’s formula [24, 28], one can get
In order to get our main results, we propose the following Halanay-type inequality for stochastic differential equations:
Lemma 5
[2, 7]. Let constants a > b ≥ 0, d > 0. Assume that there exists a positive continuous function V(t, x) satisfying the following inequality
then
where \(\upsilon\in(0,a-b]\) is the unique positive solution of the equation: υ = a − b e υd.
First, we consider model (1) with G i (t) = 0, i.e., the error dynamical system between (1) and (2) is given as follows:
Before presenting our first result, for simplicity, we introduce a new vector as
Let \(\varpi_j (j=1,2,\ldots,5)\) be row vectors with block matrix entries, i.e., the j-th block is an identity matrix and the others are zero blocks, such that \(e(t)=\varpi_1\xi_i(t), \beta(e(t))=\varpi_2\xi_i(t)\), and so on.
Now, we begin to state our result for error system (9).
Theorem 1
(See Appendix 1 for a proof). Assume that Assumptions 1–4 hold and \(\bar{\tau}_i,\bar{\tau},\tau_{di}\) are given scalars. The drive system (1) and the response system (2) with G i (t) = 0 can be exponentially synchronized for any \(0\leq\tau_i(t)\leq\bar{\tau}_i\leq\bar{\tau}, \dot{\tau}_i(t)\leq\tau_{di}\), if there exist symmetric definite positive matrices P i , F i , U, K i , diagonal positive matrices Q i , R i , S i , W, J i , Z i , M i , L i , and positive scalars \(\varepsilon_i,\epsilon_i,\iota,\nu\), such that the following matrix inequalities hold
where
with
Remark 2
If we set U = W = 0 in Theorem 1, then we can obtain a criterion to verify the exponential synchronization of system (1) and system (2) with \(\dot{\tau}(t,\eta(t))\) are not known or the time-varying delays τ(t, η(t)) are not differentiable.
Remark 3
Based on Schur complements, inequality (13) is equivalent to the following linear matrix inequality which can be solved numerically by employing the Matlab software:
where
with
Remark 4
For the inequalities (10–12), note that several nonlinear terms, such as, ν P i and \(\iota K_i\), coexist in the LMIs (10–12); thus, the conventional LMI solvers (feasp, mincx, and GEVP) are not applicable directly. In view of this fact, we present the following procedure to solve the inequalities (10–13).
-
1.
Initialize the system parameters, including \(A_i,B_i,C_i,D_i,E_i,H_{ji}(j=1,2,4,5),\pi_{il}, \Upsigma,\Upgamma,\Updelta,\Uplambda,\Uptheta,\bar{\tau}_i,\bar{\tau},\tau_{di}\) \((i,l=1,2,\ldots,N)\), and two given coefficients of accuracy κ > 0 and \(\tilde{\kappa}>0\).
-
2.
To obtain the theoretic maximum value of ν that satisfies the LMIs (10)-(13), we can solve the following generalized eigenvalue minimization problem (GEVP) (see [6, 22]):
minimize ν 1 > 0
s.t. inequalities (13) and \(\bar{P}_{i}\leq \nu_1F_{i}\) hold.
Denote the obtained minimum value by \(\underline {\nu}_1\), then we obtain the maximum value \(\bar{\nu}=1/\underline {\nu}_1\) of ν. Set \(\nu=\bar{\nu}\) (i.e., choose \(\bar{\nu}\) as the first value of ν in the beginning of the new iterative operation).
-
3.
Check whether ν < 0; if yes, go to step 9). Otherwise, go to step 4).
-
4.
Fix ν and use the following GEVP technique to obtain the maximum value of \(\iota\):
minimize \(\iota_1>0\)
s.t. inequalities (10), (13) and the following inequalities hold
$$ \bar{\tau}_i(K_i+\Uptheta L_i\Uptheta)\leq\iota_1( \nu P_i-K_i-\Uptheta L_i\Uptheta). $$
Denote the obtained minimum value by \(\underline{\iota}_1, \) then we obtain the maximum value \(\bar{\iota}=1/\underline {\iota}_1\) of \(\iota\). By setting \(\iota=\bar{\iota}\) (i.e., choose \(\bar{\iota}\) as the first value of \(\iota\) in the beginning of the new iterative operation).
-
5.
Check whether \(\iota<0\); if yes, go to step 7). Otherwise, go to step 6).
-
6.
Check whether the LMIs (10–13) are feasible; if yes, go to step 8). Otherwise, set \(\iota=\iota-\kappa\), and then go to step 5).
-
7.
Set \(\nu=\nu-\tilde{\kappa},\) go to step 3).
-
8.
Terminate the program and output the maximum values \(\nu,\iota\).
-
9.
Terminate the program and print “The LMIs (10–13) are infeasible”.
Before presenting our result of error system (3), we introduce another new vector as
Let \(\varsigma_j\ (j=1,2,\ldots,12)\) be row vectors with block matrix entries, i.e., the j-th block is an identity matrix and the others are zero blocks, such that \(e(t)=\varsigma_1\zeta_i(t), \chi_i (t)=\varsigma_{12}\zeta_i(t)\), and so on.
Now, based on Lemma 3, we can propose the following delay-dependent stability criterion for error system (3):
Theorem 2
(See Appendix 2 for a proof). Assume that Assumptions 1–4 hold and \(\bar{\tau}_i,\bar{\tau},\tau_{di}\) are given scalars. The drive system (1) and the response system (2) can be exponentially synchronized for any \(0\leq\tau_i(t)\leq\bar{\tau}_i\leq\bar{\tau},\ \dot{\tau}_i(t)\leq\tau_{di}\), if there exist symmetric definite positive matrices P i , U, K i , T j , diagonal positive matrices Q i , R i , S i , W, J i , Z i , M i , L i , real matrices X 1i , X 2i , X 3i and positive scalars \(\varepsilon_i,\epsilon_i,\iota,\nu\), such that the following matrix inequalities hold
where
with
Remark 5
Similar to Remark 2, if we set U = W = 0 in Theorem 1, then we can obtain a criterion to verify the exponential synchronization of system (1) and system (2) with \(\dot{\tau}(t,\eta(t))\) are not known or the time-varying delays τ(t, η(t)) are not differentiable.
Remark 6
Similar to Remark 3, by Schur complements, inequalities (15) can be equivalently transformed into linear matrix inequalities which could be solved numerically by employing the Matlab software.
4 Illustrative example
In this section, we give a example to demonstrate the effectiveness of our theoretic results.
Example 1
Consider system (1) with n = 2 and the following parameters:
The behaved functions are \(\tilde{\beta}_1(x)=\tilde{\beta}_2(x)=1.5x+0.5\sin x,\) the activation functions are \(\tilde{f}_1(x)=\tilde{f}_2(x)=\tanh(x)\), and the time-varying delays are \(\tau_1(t)=\tau_2(t)=0.8+0.2\sin t\). Then Assumptions 1–4 are satisfied with \(\Updelta=2I,\Uplambda=\Upsigma=\Uptheta=I,\Upgamma=0\) and \(\bar{\tau}_1=\bar{\tau}_2=\bar{\tau}=1, \tau_{d1}=\tau_{d2}=0.2\).
In this paper, the transition rate matrix is given as follows
and the control input vector with state feedback is designed as (4) with
Set \(\nu=0.2,\iota=0.1\), solving the LMIs (10–13) in Theorem 1 by resorting to the Matlab LMI Control Toolbox, we have one feasible solution as follows
Figure 1 shows the neural network model has a chaotic attractor with initial values \(x_1(t)=0.2,x_2(t)=0.5,\ t\in [-1,0]\). The initial values of the response system are taken as \(y_1(t)=-1.3,y_2(t)=2.1, t\in [-1, 0].\) Figures 2 and 3 depict the phase trajectories of the drive system and response system, respectively. Figure 4 shows the error states. By numerical simulation, we can see that the dynamical behaviors of response system (2) synchronize with master system (1) as shown in Figs. 2, 3.
5 Conclusion
In this paper, the exponential synchronization problem has been investigated for a class of stochastic perturbed chaotic neural networks with discrete and distributed time-varying delays as well as Markovian jump parameters. Based on a Halanay-type inequality for stochastic differential equations, the Jensen integral inequality and a novel Lemma, two delay-dependent sufficient condition are proposed to guarantee the exponential synchronization of two identical Markovian jumping chaotic-delayed neural networks with stochastic perturbation. With some parameters being fixed in advance, these conditions are expressed in terms of linear matrix inequalities, which can be solved numerically by employing the Matlab software. Finally, a numerical example with simulations is provided to illustrate the effectiveness and usefulness of the presented synchronization scheme.
References
Arnold L (1972) Stochastic differential equations: theory and applications. Wiley, New York
Baker CTH, Buckwar E (2005) Exponential stability in p-th mean of solutions, and of convergent Euler-type solutions, of stochastic delay differential equations. J Comput Appl Math 184(2):404–427
Friedman A (1976) Stochastic differential equations and applications. Academic Press, New York
Fu J, Zhang H, Ma T (2009) Delay-probability-distribution-dependent robust stability analysis for stochastic neural networks with time-varying delay. Prog Nat Sci 19:1333–1340
Gu K (2000) An integral inequality in the stability problem of time-delay systems. In: Proceedings of 39th IEEE conference on decision and control, Sydney, Australia, pp 2805–2810
Guan X, Chen C (2004) Delay-dependent guaranteed cost control for T-S fuzzy systems with time delays. IEEE Trans Fuzzy Syst 12(2):236–249
Li X, Cao J (2007) Delay-independent exponential stability of stochastic Cohen-Grossberg neural networks with time-varying delays and reaction-diffusion terms. Nonlinear Dyn 50:363–371
Liu Z, Zhang H, Zhang Q (2010) Novel stability analysis for recurrent neural networks with multiple delays via line integral-type L-K functional. IEEE Trans Neural Netw 21(11):1710–1718
Mao X (2002) Exponential stability of stochastic delay interval systems with Markovian switching. IEEE Trans Autom Contr 47(10):1604–1612
Pecora L, Carroll T (1990) Synchronization in chaotic systems. Phys Rev Lett 64:821–824
Qian W, Li T, Cong S, Fei S (2010) Improved stability analysis on delayed neural networks with linear fractional uncertainties. Appl Math Comput 217:3596–3606
Sanchez EN, Perez JP (1999) Input-to-state stability (ISS) analysis for dynamic NN. IEEE Trans Circuits Syst I 46:1395–1398
Song Q (2010) Synchronization analysis in an array of asymmetric neural networks with time-varying delays and nonlinear coupling. Appl Math Comput 216:1605–1613
Song Q, Cao J (2012) Passivity of uncertain neural networks with both leakage delay and time-varying delay. Nonlinear Dyn 67:1695–1707
Song Q, Cao J (2011) Synchronization of nonidentical chaotic neural networks with leakage delay and mixed time-varying delays. Adv Differ Equ 16
De Souza CE, Li X (1999) Delay-dependent robust control of uncertain linear state-delayed systems. Automatica 35:1313–1321
Tang Y, Fang J, Miao Q (2009) On the exponential synchronization of stochastic jumping chaotic neural networks with mixed delays and sector-bounded non-linearities. Neurocomputing 72:1694–1701
Wang Z, Zhang H, Jiang B (2011) LMI-based approach for global asymptotic stability analysis of recurrent neural networks with various delays and structures. IEEE Trans Neural Netw 22(7):1032–1045
Yuan C, Lygeros J (2005) Stabilization of a class of stochastic differential equations with Markovian switching. Syst Control Lett 54:819–833
Yue D, Tian E, Zhang Y, Peng C (2009) Delay-distribution-dependent stability and stabilization of T-S fuzzy systems with probabilistic interval delay. IEEE Trans Syst Man Cybern Part B 39:503–516
Zhang H, Liu Z, Huang GB, Wang Z (2010) Novel weighting-delay-based stability criteria for recurrent neural networks with time-varying delay. IEEE Trans Neural Netw 21(1):91–106
Zhang H, Ma T, Huang H-B, Wang Z (2010) Robust global exponential synchronization of uncertain chaotic delayed neural networks via dual-stage impulsive control. IEEE Trans Syst Man Cybern B 40(3):831–844
Zhang H, Quan Y (2001) Modeling identification and control of a class of nonlinear system. IEEE Trans Fuzzy Syst 9(2):349–354
Zhang H, Wang Y (2008) Stability analysis of Markovian jumping stochastic Cohen-Grossberg neural networks with mixed time delays. IEEE Trans Neural Netw 19(2):366–370
Zhang H, Wang Z, Liu D (2008) Global asymptotic stability of recurrent neural networks with multiple time-varying delays. IEEE Trans Neural Netw 19(5):855–873
Zheng CD, Shan QH, Zhang H, Wang Z (2013) On stabilization of stochastic Cohen-Grossberg neural networks with mode-dependent mixed time-delays and markovian switching. IEEE Trans Neural Netw Learn Syst 24(5):800–811
Zheng CD, Zhou F, Wang Z (2012) Stochastic exponential synchronization of jumping chaotic neural networks with mixed delays. Commun Nonlinear Sci Numer Simulat 17(3):1273–1291
Zhou W, Tong D, Gao Y, Ji C, Su H (2012) Mode and delay-dependent adaptive exponential synchronization in pth moment for stochastic delayed neural networks with Markovian switching. IEEE Trans Neural Netw Learn Syst 23(3):662–668
Acknowledgments
This work was supported by the National Natural Science Foundation of China 61034005, 61074073, 61273022, Program for New Century Excellent Talents in University of China (NCET-10-0306), and the Fundamental Research Funds for the Central Universities under Grants N110504001 and N100104102.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1
1.1 Proof of Theorem 1
Consider the following Lyapunov-Krasovskii functional:
where \(Q_i=\hbox{diag}\{q_{1i},q_{2i},\ldots,q_{ni}\}, R_i=\hbox{diag}\{r_{1i},r_{2i},\ldots,r_{ni}\}, S_i=\hbox{diag}\{s_{1i},s_{2i},\ldots,s_{ni}\}\).
It can be easily verified that V 1(t, e t , i) is a nonnegative function over \([-\bar{\tau},+\infty)\). Evaluating the time derivative of V 1(t, e t , i) along the trajectory of system (9), we have that
where
From Assumptions 3 and 4, we get that
In addition, we derive that
For any \(j=1,2,\ldots,n,\) it follows from (5) that
Thus, we have that
For any \(j=1,2,\ldots,n,\) from (5) we obtain that
Therefore, the following matrix inequalities hold for any positive diagonal matrices J i , Z i with compatible dimensions
According to Assumption 1 and Lemma 1, for any positive scalar \(\varepsilon_i\) we have that
From (5), the following inequalities hold for any positive diagonal matrix M i with compatible dimension
where
with
Now, by (14), it is easy to see that there exists a scalar α > 1 such that
where
Applying Schur complements to (27) results in
Using Assumption 1 and Lemma 1, for any positive scalar \(\epsilon_i\) we have that
This together with (28) provides that
By this inequality and (26), it is easy to see that
Taking the mathematical expectations on both sides of (17), from above inequality we have that
By integrating above inequality from t − τ(t) to t, we obtain that
It follows that
In view of (10) and (11), we have that
Noticing that
Therefore, from (16) we have that
This together with (30–31) yields that
Moreover, \({\mathbb{E}\{V_1(t,e_t,i)\}\geq e(t)^TP_ie(t)}\), therefore, it follows from (5) and (12) that
Thus, we obtain that
Substituting (32–34) into (29) derives that
Noting that α > 1, applying Lemma 5 to above inequality results in
where κ is the unique positive solution of the following equation:
Therefore, we arrive at the conclusion that
The proof is completed.
Appendix 2
1.1 Proof of Theorem 2
Define the following Lyapunov-Krasovskii functional:
where V 1(t, e t , i) ie defined in (16) and
It can be easily verified that V(t, e t , i) is a nonnegative function over \([-\hat{\tau},+\infty)\). Evaluating the time derivative of V(t, e t , i) along the trajectory of system (3), we have that
where
For any t with \(0<\tau_i(t)<\bar\tau_i\) and \(0<\varrho_i(t)<\bar\varrho_i\), from Lemma 2 we have the following inequalities
Set λ j = 1, μ j = 3, based on Lemma 2 we get from (38–39) that
It is easy to verify that Eq. (40) holds for any t with \(0\leq\tau_i(t)\leq\bar\tau_i\) and \(0\leq\varrho_i(t)\leq\bar\varrho_i.\)
On the other hand, by the Leibniz-Newton formula, we get that
Therefore, the following equalities hold for any real matrices X ji (j = 1, 2, 3) with compatible dimensions
From Lemma 1, the following matrix inequalities hold for any positive scalar \(\epsilon_i\)
By (22–25), (36–37) and (40–45), and taking the mathematical expectations on both sides of (35), we obtain that
From (15), there exists a positive scalar α 0 such that
Similar to the proof of Theorem 1 in [27], it implies that the error system (3) is globally exponentially stable. This completes the proof.
Rights and permissions
About this article
Cite this article
Zheng, CD., Zhang, H. & Wang, Z. Exponential synchronization of stochastic chaotic neural networks with mixed time delays and Markovian switching. Neural Comput & Applic 25, 429–442 (2014). https://doi.org/10.1007/s00521-013-1507-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-013-1507-7