Loading [MathJax]/jax/output/SVG/jax.js
Research article

Synchronization analysis of delayed quaternion-valued memristor-based neural networks by a direct analytical approach

  • Received: 11 January 2024 Revised: 12 March 2024 Accepted: 22 April 2024 Published: 24 May 2024
  • This issue discusses the asymptotic synchronization and the exponential synchronization for memristor-based quaternion-valued neural networks under the time-varying delays. Some criteria for synchronization of the memristor-based quaternion-valued neural networks are given by exploiting the set-valued theory, the differential inclusion theory, some analytic techniques, as well as constructing novel controllers, It is worth noting that the synchronization problem about the memristor-based quaternion-valued neural networks were studied by the direct analysis method in this paper. Finally, the main theoretical results were verified by numerical simulations.

    Citation: Jun Guo, Yanchao Shi, Shengye Wang. Synchronization analysis of delayed quaternion-valued memristor-based neural networks by a direct analytical approach[J]. Electronic Research Archive, 2024, 32(5): 3377-3395. doi: 10.3934/era.2024156

    Related Papers:

    [1] Jun Guo, Yanchao Shi, Weihua Luo, Yanzhao Cheng, Shengye Wang . Exponential projective synchronization analysis for quaternion-valued memristor-based neural networks with time delays. Electronic Research Archive, 2023, 31(9): 5609-5631. doi: 10.3934/era.2023285
    [2] Chao Yang, Juntao Wu, Zhengyang Qiao . An improved fixed-time stabilization problem of delayed coupled memristor-based neural networks with pinning control and indefinite derivative approach. Electronic Research Archive, 2023, 31(5): 2428-2446. doi: 10.3934/era.2023123
    [3] Yong Zhao, Shanshan Ren . Synchronization for a class of complex-valued memristor-based competitive neural networks(CMCNNs) with different time scales. Electronic Research Archive, 2021, 29(5): 3323-3340. doi: 10.3934/era.2021041
    [4] Xiangwen Yin . A review of dynamics analysis of neural networks and applications in creation psychology. Electronic Research Archive, 2023, 31(5): 2595-2625. doi: 10.3934/era.2023132
    [5] Xuerong Shi, Zuolei Wang, Lizhou Zhuang . Spatiotemporal pattern in a neural network with non-smooth memristor. Electronic Research Archive, 2022, 30(2): 715-731. doi: 10.3934/era.2022038
    [6] Wanshun Zhao, Kelin Li, Yanchao Shi . Exponential synchronization of neural networks with mixed delays under impulsive control. Electronic Research Archive, 2024, 32(9): 5287-5305. doi: 10.3934/era.2024244
    [7] Shuang Liu, Tianwei Xu, Qingyun Wang . Effect analysis of pinning and impulsive selection for finite-time synchronization of delayed complex-valued neural networks. Electronic Research Archive, 2025, 33(3): 1792-1811. doi: 10.3934/era.2025081
    [8] Tianyi Li, Xiaofeng Xu, Ming Liu . Fixed-time synchronization of mixed-delay fuzzy cellular neural networks with $ L\acute{e}vy $ noise. Electronic Research Archive, 2025, 33(4): 2032-2060. doi: 10.3934/era.2025090
    [9] Yawei Liu, Guangyin Cui, Chen Gao . Event-triggered synchronization control for neural networks against DoS attacks. Electronic Research Archive, 2025, 33(1): 121-141. doi: 10.3934/era.2025007
    [10] Bin Zhen, Ya-Lan Li, Li-Jun Pei, Li-Jun Ouyang . The approximate lag and anticipating synchronization between two unidirectionally coupled Hindmarsh-Rose neurons with uncertain parameters. Electronic Research Archive, 2024, 32(10): 5557-5576. doi: 10.3934/era.2024257
  • This issue discusses the asymptotic synchronization and the exponential synchronization for memristor-based quaternion-valued neural networks under the time-varying delays. Some criteria for synchronization of the memristor-based quaternion-valued neural networks are given by exploiting the set-valued theory, the differential inclusion theory, some analytic techniques, as well as constructing novel controllers, It is worth noting that the synchronization problem about the memristor-based quaternion-valued neural networks were studied by the direct analysis method in this paper. Finally, the main theoretical results were verified by numerical simulations.



    Memristor was proposed by Chua in [1], as the fourth basic electronic component besides resistors, inductors, and capacitors. The memristor value changes with the circuit current, and after the circuit is powered off, its resistance value is the value at the moment of power off. In other words, memristor is a class of nonlinear resistors with memory functions. Due to the fact that nanotechnology was far from mature at that time, the physical realization of memristors was extremely difficult. Therefore, the research of memristors failed to achieve a major breakthrough. Until 2008, the Hewlett-Packard [2] laboratory developed memristors. Since then, memristor and its applications have attracted the attention of numerous scholars [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18].

    Previous research results indicated that the memristors has the function of simulating brain synapses. Due to the above properties of memristors, researchers use memristors as the connection weights of traditional neural networks to obtain memristor-based neural networks [3, 4, 5]. Obviously, the model of memristor-based neural networks is a special type of nonlinear system. Recently, the study about the above system's dynamical behavior attracted a lot of research attention and achieved many interesting results. Currently, the research on memristor-based neural networks mainly concentrates in either the real number domain or the complex number domain [7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]. However, the related research in the field of quaternion is relatively rare.

    Quaternion, which is a class of divisible algebras, was proposed by Hamilton [19] in 1843. The exchange law of quaternion multiplication is not, which makes the study of quaternion is slower than the real number or the complex number, and is more difficult. However, due to the development of modern mathematics, applications of quaternion in digital image processing, face recognition, quantum mechanics and other fields have been discovered [20, 21, 22, 23, 24]. A series of results have been achieved, especially in three-dimensional Data modeling and processing in space and four-dimensional space [21, 22]. For example, for rotation and affine transformation in three-dimensional space, the quaternion representation is not only more compact and effective, but also can effectively avoid the defects of matrix and Euler representations. Therefore, quaternion is increasingly attracting the attention of scholars.

    The quaternion was introduced into the traditional neural network, and quaternion-valued neural networks (QVNNs) have been built. QVNNs show improved performances in color night vision [25], 3D wind forecasting[26, 27], image compression[28], and so on. As the extension of real-valued neural networks(RVNNs) and complex-valued neural networks(CVNNs), QVNNs have some notable advantages such as a low dimensionality and a high efficiency in handling multi-dimensional data. Using the three imaginary parts of the quaternion to denote the three primary colors in the color space, QVNNs not only do not need to deal with the three primary colors separately, but also show the correlation between the three primary colors. Furthermore, QVNNs show an improved performance than RVNNs or CVNNs in handling certain optimization and estimation problems [29].

    By introducing the quaternion algebra into memristor-based neural networks(MNNs), the quaternion-value memristor-based neural networks(QVMNNs) is obtained. The connection weights, state variables, and activation function values of QVMNNs are all derived from the quaternion domain. Recently, some interesting conclusions have been shown with the in-depth research of QVMNNs [30, 31, 32, 33, 34, 35, 36]. In [31], exponential synchronization for QVMNNs with delayed was studied by quantized intermittent control tactics. In [33], the synchronization for fuzzy QVMNNs was discuss by the Lexicographical order method. Finite-time anti-synchronization about the inconsistent markovian QVMNNs under reaction-diffusion terms was investigated in [34]. The exponential synchronization conditions of the delayed inertial QVMNNs were given in the form of linear matrix inequality(LMIs) in [35]. In [36], Wei and Cao studied fixed-time synchronization for the quaternion-valued memristor-based neural networks by dividing the system into real and imaginary parts.

    Inspired by the above research, it is worth discussing the asymptotic synchronization and the exponential synchronization for the QVMNNs under time delays. Different from the above research methods, this issue directly tackles the quaternion memristive neural network, which naturally poses a problem when determining which quaternion is big or small. To solve this problem, this paper will adopt the vector ordering approach, which supplies the theoretical basis to determine the "magnitude" of two different quaternions. On this basis, we propose a direct method to discuss the asymptotic synchronization and the exponential synchronization for memristor-based quaternion-valued neural networks under the time-varying delays, which simplifies the proof process.

    The main works about this issue are arranged as follows. In Section 2, the model is built and the basics that will be used later are introduced. Several conditions are obtained for the asymptotic synchronization and the exponential synchronization for the QVMNNs under time-varying delays through the controllers in Section 3. Two numerical examples are used to verify the accuracy for the conclusions in Section 4. The last section draws a conclusion.

    Notations: In this article, R is the real field and Q is the quaternion field. co[a,b] expresses the closure for the convex hull Q manufactured by the quaternion a,b. C(1)([ς,0],Rn) shows the class of continuous functions from [ς,0] to Rn.

    The quaternion is the hypercomplex number consisting of one real part and three imaginary parts. For uQ, this is written as

    u=uR+uIi+uJj+uKk

    where uR,uI,uJ,uKR. Moreover, i,j,k are the imaginary parts, which follow the Hamilton principle:

    i2=1,j2=1,k2=1,ij=ji=,ki=ik=jk,jk=kj=i.

    The conjugate of u is denoted by¯u=uRuIiuJjuKk. The modulus of u is written as

    |u|=¯uu=(uR)2+(ul)2+(ul)2+(uK)2.

    u1=nm=1|um| is the norm of x. For two quaternions u1=uR1+uI1i+uJ1j+uK1k and u2=uR2+uI2i+uJ2j+uK2k, the addition between them is defined as

    u1+u2=uR1+uR2+(ul1+ul2)i+(uJ1+uJ2)j+(uK1+uK2)k.

    According to the Hamilton rule, the product of two quaternions is

    u1u2=(uR1uR2ul1ul2uJ1uJ2uK1uK2)+(uR1ul2+ul1uR2+uJ1uK2uK1uJ2)i+(uR1uJ2+uJ1uR2+uK1uI2uI1uK2)j+(uR1uK2+uK1uR2+uI1uJ2uJ1uI2)k.

    Consider the QVMNNs under time-varying delays as follows:

    ˙ωp(t)=cpωp(t)+nq=1apq(ωp(t))fq(ωq(t))+nq=1bpq(ωp(t))fq(ωq(tτ(t)))+I,    t0, (2.1)

    where ωp(t)Q stands for the state vector, p=1,2,,n,. C is the self-feedback matrix, C=diag{c1,c2,,cn}, and bpq(ωp(t)), apq(ωp(t)) are the connection weight matrices. f(ω(t)):QnQn denotes the activation function. IQn is the external input. Moreover, τ(t) denotes a transmission delay 0τ(t)τ. The initial conditions of system (2.1) are selected as ω(s)=ϕ(s),τs0, where ϕ(s)C(1)([τ,0],Qn).

    Assumption 2.1. There are some positive numbers lq,FqR that satisfy

    |fq(v)fq(u)|mq|vu|,    |fq(u)|<Fq,

    for all u,vQ.

    Assumption 2.2. [37] It can be defined that:

    apq(up(t))={ˆapq=aR1pq+aI1pqi+aJ1pqj+aK1pqk,    |up(t)|Tpˇapq=aR2pq+aI2pqi+aJ2pqj+aK2pqk,    |up(t)|<Tp,bpq(up(t))={ˆbpq=bR1pq+bI1pqi+bJ1pqj+bK1pqk,    |up(t)|Tpˇbpq=bR2pq+bI2pqi+bJ2pqj+bK2pqk,    |up(t)|<Tp,

    where the positive number Tp is the switching jumps.

    Remark 1. According to Assumption 2.2, bpq(up(t)) and apq(up(t)) are piecewise functions; therefore, the Quaternion-Valued Memristor-Based neural networks of system (2.1) is a discontinuous system.

    Lemma 2.1. [38] Let u,vQ, ϵ>0 be a constant. Then, the following inequality is true:

    vu+ˉuˉvϵˉuu+1ϵvˉv.

    From the theory of the set valued map and differential inclusion[39], system (2.1) could be rewritten as follows:

    ˙ωp(t)cpωp(t)+nq=1co[apq,aTTpq]fq(ωq(t))+nq=1co[bTpq,bTTpq]fq(ωq(tτ(t)))+Jp (2.2)

    where ˜apq=max{|ˆapq|,|ˇapq|}, aTTpq=max{ˆapq,ˇapq}, aTpq=min{ˆapq,ˇapq}, ˜bpq=max{|ˆbpq|,|ˇbpq|}, bTTpq=max {ˆbpq,˜bpq}, and bTpq=min{ˆbpq,˜bpq}.

    Then, there are apq co[aTpq,aTTpq], bpqco[bTpq, bTTpq], which satisfies the following:

    ˙ωp(t)=cpωp(t)+nq=1apqfq(ωq(t))+nq=1bpqfq(ωq(tτ(t)))+J. (2.3)

    Considering (2.1) as a drive system, then the following system can be used as a response system:

    ˙vp(t)=cpvp(t)+nq=1apq(vp(t))fq(vq(t))+nq=1bpq(vp(t))fq(vq(tτ(t)))+up(t)+J, t0, (2.4)

    where up(t) is the controller.

    From the differential inclusion and theory of the set valued map, system (2.4) can also be rewritten as:

    ˙vp(t)cpvp(t)+nq=1co[aTpq,aTTpq]fq(vq(t))+nq=1co[bTpq,bTTpq]fq(vq(tτ(t)))+Jp+up(t). (2.5)

    Equivalently, there are apqco[aTpq,aTTpq], bpqco[bTpq,bTTpq], which satisfies the following:

    ˙vp(t)=cpvp(t)+nq=1apqfq(vq(t))+nq=1bpqfq(vq(tτ(t)))+Jp+up(t). (2.6)

    Let e(t)=(e1(t),,en(t))T=v(t)ω(t), the following error system can be obtained:

    ˙ep(t)=cpep(t)+nq=1[apqfq(vq(t))apqfq(ωq(t))]+nq=1[bpqfq(vq(tτ(t)))bpqfq(ωq(tτ(t)))]+up(t). (2.7)

    Definition 2.1. If there are two constants π>0 and γ1 that satisfy the following inequality,

    e(t)γeπtsupτs0e(tτ(t)),  t0.

    then the systems (2.1) and (2.4) are globally exponentially synchronized.

    Lemma 2.2. [40] Assume α1 and α2 are two constants such that λ1>λ2>0. y(t) is a nonnegative continuous function, which is defined on [t0τ,+). The following inequality is true:

    D+(y(t))λ1y(t)+λ2ˉy(t),for  all  tt0,

    where ˉy(t)=suptτsty(s), and D+(y(t))=¯limh0+y(t+h)y(t)h is the upper-right Dini derivative. For tt0, one can get y(t)ˉy(t0)eγ(tt0)., which is the only positive solution to γ=λ1λ2eγτ.

    In the following section, some sufficient criterion are acquired about the global exponential stability for the memristor-based neural network under time-varying delays. Then, the following main results are established.

    Theorem 3.1. Under the Assumptions 2.1 and 2.2, the systems (2.1) and (2.4) are exponentially synchronized by the controller (3.1):

    up(t)=kp(t)ep(t)ηp. (3.1)

    If there are two constants ε1>0, ε2>0, then the following is satisfied:

    2cp+2kp1ε1m2p1ε1˜apqˉ˜apq1ε2˜bpqˉ˜bpq>0,ε2maxpm2p<1τ,

    where kp(t)Rn,

    ηp=nq=1(aTTpqaTpq)Fq+nq=1(bTTpqbTpq)Fq.

    Proof. Construct the following auxiliary function:

    V(t)=np=1ˉep(t)ep(t)+ttτ(t)ˉep(s)ep(s)ds. (3.2)

    Before continuing, the following estimation of ˙ep(t) and ˙ˉep(t) can be given:

    ˙ep(t)=cpep(t)+nq=1[apq(t)fq(vq(t))apq(t)fq(ωq(t))]+nq=1[bpq(t)fq(vq(tτ(t)))bpq(t)fq(ωq(tτ(t)))]+up(t)=cpep(t)+nq=1(apq(t)apq(t))fq(ωq(t))+nq=1apq(t)[fq(vq(t))fq(ωq(t))]+nq=1bpq(t)[fq(vq(tτ(t)))fq(ωq(tτ(t)))]+nq=1(bpq(t)bpq(t))fq(ωq(tτ(t)))]kp(t)ep(t)ηp
    ˙ˉep(t)=cpˉep(t)+nq=1[ˉfq(vq(t))ˉapq(t)ˉfq(ωq(t))ˉapq(t)]+nq=1[ˉfq(vq(tτ(t)))ˉbpq(t)ˉfq(ωq(tτ(t)))ˉbpq(t)]+ˉup(t)=cpˉep(t)+nq=1ˉfq(vq(t))[ˉapq(t)ˉapq(t)]+nq=1[ˉfq(vq(t))ˉfq(ωq(t))]ˉapq(t)+nq=1[ˉfq(vq(tτ(t)))ˉbpq(t)ˉfq(ωq(tτ(t)))ˉbpq(t)]kp(t)ˉep(t)ˉηp.

    Next, the derivative of V(t) along (1.7) is computed as follows:

    ˙V(t)=np=1˙ˉep(t)ep(t)+np=1ˉep(t)˙ep(t)+np=1ˉep(t)ep(t)(1˙τ(t))np=1ˉep(tτ(t))ep(tτ(t))=np=1{cpˉep(t)+nq=1[ˉfq(vq(t))ˉapq(t)ˉfq(ωq(t))ˉapq(t)]+nq=1[ˉfq(vq(tτ(t)))ˉbpq(t)ˉfq(ωq(tτ(t)))ˉbpq(t)]+ˉup(t)}ep(t)+np=1ˉep(t){cpep(t)+nq=1[apq(t)fq(vq(t))apq(t)fq(ωq(t))]+nq=1[bpq(t)fq(vq(tτ(t)))bpq(t)fq(ωq(tτ(t)))]+u(t)}+np=1ˉep(t)ep(t)(1˙τ(t))np=1ˉep(tτ(t))ep(tτ(t))np=1{cpˉep(t)+nq=1(ˉaTTpqˉaTpq)Fq+nq=1[ˉfq(vq(t))ˉfq(ωq(t))]ˉapq(t)+nq=1(ˉbTTpqˉbTpq)Fq+ˉup(t)+nq=1[ˉfq(vq(tτ(t)))ˉfq(ωq(tτ(t)))]ˉbpq(t)}ep(t)+np=1ˉep(t){cpep(t)+nq=1(aTTpqaTpq)Fq+nq=1apq(t)[fq(vq(t))fq(ωq(t))]+nq=1(bTTpqbTpq)Fq+up(t)+nq=1bpq(t)[fq(vq(tτ(t)))fq(ωq(tτ(t)))]}+np=1ˉep(t)ep(t)(1τ)np=1ˉep(tτ(t))ep(tτ(t)). (3.3)

    Lemma 2.1 shows that there are two positive constants ε1,ε2, which satisfy the following:

    [ˉfq(vq(t))ˉfq(ωq(t))]ˉapq(t)e(t)+ˉe(t)apq(t)[fq(vq(t))fq(ωq(t))]ε1[ˉfq(vq(t))ˉfq(ωq(t))][fq(vq(t))fq(ωq(t))]+1ε1ˉe(t)˜apqˉ˜apqe(t)ε1m2qˉeq(t)eq(t)+1ε1ˉe(t)˜apqˉ˜apqe(t) (3.4)
    [ˉfq(vq(tτ(t)))ˉfq(ωq(tτ(t)))]ˉbpq(t)e(t)+ˉe(t)bpq(t)[fq(vq(tτ(t)))fq(ωq(tτ(t)))]ε2[ˉfq(vq(tτ(t)))ˉfq(ωq(tτ(t)))][fq(vq(tτ(t)))fq(ωq(tτ(t)))]+1ε2ˉe(t)˜bpqˉ˜bpqe(t)ε2m2qˉeq(tτ(t))eq(tτ(t))+1ε2ˉe(t)˜bpqˉ˜bpqe(t). (3.5)

    Therefore, together with systems (3.4) and (3.5), the following can be obtained:

    ˙V(t)np=1{2cp+2kp1ε1m2p1ε1˜apqˉ˜apq1ε2˜bpqˉ˜bpq}ˉep(t)ep(t)+[ε2maxpm2p(1τ)]ˉeq(tτ(t))eq(tτ(t))<0. (3.6)

    Then, the error variable e(t) will exponentially converge to zero; in other words, systems (2.1) and (2.4) will achieve exponentially synchronized synchronization.

    Theorem 3.2. For two given Assumptions 2.1 and 2.2, the systems (2.1) and (2.4) can achieve asymptotically synchronization with the controller (3.7), if there are some constants ε1, ε2 that satisfy the following inequality:

    up(t)=kpep(t)ηp, (3.7)

    where

    2kp2cp+ε1m2p+1ε1nq=1˜apqˉ˜apq+1ε2nq=1˜bpqˉ˜bpq,ηp=nq=1(aTTpqaTpq)Fq+nq=1(bTTpqbTpq)Fq.

    Moreover, the control gains satisfy the following:

    ρ1=minp(2cp+2kpε1m2p1ε1nq=1˜apqˉ˜apq1ε2nq=1˜bpqˉ˜bpq),ρ2=ε2maxpm2p,    ρ1>ρ2>0.

    Proof. Construct the suggested function V(t), which is defined by

    V(t)=np=1ˉep(t)ep(t).

    The derivative of V(t) along (2.7) is computed as follows:

    ˙V(t)=np=1˙ˉep(t)ep(t)+np=1ˉep(t)˙ep(t)=np=1{cpˉep(t)+nq=1[ˉfq(vq(t))ˉapq(t)ˉfq(ωq(t))ˉapq(t)]+nq=1[ˉfq(vq(tτ(t)))ˉbpq(t)ˉfq(ωq(tτ(t)))ˉbpq(t)]+ˉu(t)}ep(t)+np=1ˉep(t){cpep(t)+nq=1[apq(t)fq(vq(t))apq(t)fq(ωq(t))]+nq=1[bpq(t)fq(vq(tτ(t)))bpq(t)fq(ωq(tτ(t)))]+u(t)}np=1(2cp+2kpε1m2p1ε1nq=1˜apqˉ˜apq1ε2nq=1˜bpqˉ˜bpq)ˉep(t)ep(t)+ε2np=1m2pˉeq(tτ(t))eq(tτ(t))ρ1V(t)+ρ2V(tτ(t)). (3.8)

    where ρ1, ρ2 are two constants.

    According to Lemma (2.2), it can be inferred that

    V(t)=maxτθ0V(θ)expγt (3.9)

    where γ is the solution of Eq (3.10)

    xyexpγθγ=0. (3.10)

    Consequently, error system (2.7) is globally asymptotically stable. In other words, the stabilization control of the systems (2.1) and (2.4) can be achieved by controller (3.7).

    Remark 2. The QVMNNs was divided into a real part and three imaginary parts in the existing work[36]. Moreover, the direct method is used to discuss the QVMNNs in this paper, which is more realistic. The results are presented in the shape of easily verifiable algebraic inequalities.

    Then, the following two numerical simulations are dedicated to verify the validity of the given theoretical results.

    Example 1. Consider the two-neuron quaternion-valued memristor-based neural networks (2.1) with c1=c2=1; the connection weights are as follows:

    a11(ω1(t))={2.12.7i+1.9j2.5k,|ω1(t)|<1,2.21.6i+2.2j1.5k,|ω1(t)|1,a12(ω1(t))={0.20.4i0.2j0.3k,|ω2(t)|1,0.60.9i0.5j0.8k,|ω2(t)|<1,a21(ω2(t))={1.50.3i+1.6j0.4k,|ω2(t)|1,1.0+0.7i+1.1j+0.6k,|ω2(t)|<1,a22(ω2(t))={1.30.1i1.2j0.3k,|ω2(t)|1,0.70.3i0.8j0.2k,|ω2(t)|<1,,b11(ω1(t))={1.6+2.5i1.5j+2.3k,|ω1(t)|1,0.1+3.0i1.4j+2.9k,|ω1(t)|<1,b12(ω1(t))={1.40.9i0.1j0.6k,|ω1(t)|1,0.51.5i0.3j1.7k,|ω1(t)|<1,b21(ω2(t))={1.01.2i1.3j1.3k,|ω2(t)|1,0.90.3i0.6j0.1k,|ω2(t)|<1,b22(ω2(t))={1.20.6i+1.1j0.4k,|ω2(t)|1,0.60.7i+0.4j0.8k,|ω2(t)|<1,.

    In the response system (2.4), the activation functions are selected as f(ω)=0.1tanh(ω). Obviously, when F1=F2=0.1, the activation functions satisfy Assumption 2.1. Figures 1 and 2 show trajectories for the error variables eπ1, eπ2, π=R,I,J,K about systems (2.1) and (2.4) without a controller. The error system cannot converge to zero. Thus, the systems (2.1) and (2.4) cannot be synchronized in this situation.

    Figure 1.  The error state variables eπ1, π=R,I,J,K, without controller.
    Figure 2.  The error state variables eπ2,π=R,I,J,K, without controller.

    According to the above parameters, the following is directly calculated from Theorem 3.1:

    2q=1(a+1qa1q)Fq+2q=1(b+1qb1q)Fq=0.21+0.2i+0.14j+0.33k,2q=1(a+2qa2q)Fq+2q=1(b+2qb2q)Fq=0.180.02j+0.23j+0.05k.

    Therefore, with controller (3.1), it can pick out η1=0.21+0.2i+0.14j+0.33k, η2=0.180.02j+0.23j+0.05k, k1=k2=1, ε1=ε2=2 τ(t)=0.650.25cos(t). Then, the conditions for the Theorem 3.1 are held. Figures 3 and 4 depict the error variables eπ1, eπ2,π=R,I,J,K between systems (2.1) and (2.4) under the controller (3.1). Hence, the error system tends to 0. That is to say that systems (2.1) and (2.4) achieve asymptotic synchronization.

    Figure 3.  The error state variableseπ1,π=R,I,J,K, with the controller.
    Figure 4.  The error state variables eπ2,π=R,I,J,K, under the controller.

    Example 2. Consider the two-neuron QVMNNs (2.1) with the following:

    a11(ω1(t))={2.02.7i+2.0j2.5k,|ω1(t)|<1,2.31.6i+2.3j1.5k,|ω1(t)|1,a12(ω1(t))={0.10.4i0.1j0.3k,|ω2(t)|1,0.50.9i0.5j0.7k,|ω2(t)|<1,a21(ω2(t))={1.60.3i+1.5j0.4k,|ω2(t)|1,1.1+0.7i+1.0j+0.6k,|ω2(t)|<1,a22(ω2(t))={1.20.1i1.3j0.2k,|ω2(t)|1,0.80.3i0.7j0.3k,|ω2(t)|<1,b11(ω1(t))={1.5+2.6i1.5j+2.3k,|ω1(t)|1,0.1+3.1i1.4j+3.0k,|ω1(t)|<1,b12(ω1(t))={1.40.9i0.1j0.6k,|ω1(t)|1,0.51.5i0.5j1.6k,|ω1(t)|<1,b21(ω2(t))={1.21.1i1.3j1.3k,|ω2(t)|1,0.80.2i0.6j0.1k,|ω2(t)|<1,b22(ω2(t))={1.30.5i+1.2j0.4k,|ω2(t)|1,0.50.8i+0.4j0.7k,|ω2(t)|<1,.

    Moreover, c1=c2=2, the time-delays are chosen as τ(t)=0.6+0.3sin(t), and the active function is fq(ωq(t))=0.1tanh(ωq(t)), which satisfy the requirements in Assumption 2.1 with Fq=0.1, mq=0.1, q=1,2. Figures 5 and 6 depict the error variables eπ1, eπ2,π=R,I,J,K between systems (2.1) and (2.4) without a controller. Set ε1=ε2=5; then, from the conditions of Theorem 3.2 and the above given parameters, the following results can be obtained:

    2k12c1+ε1m21+1ε12q=1˜a1qˉ˜a1q+1ε22q=1˜b1qˉ˜b1q=5.776,2k22c2+ε1m22+1ε12q=1˜a2qˉ˜a2q+1ε22q=1˜b2qˉ˜b2q=1.862.
    Figure 5.  The error state variables eπ1,π=R,I,J,K, without controller.
    Figure 6.  The error state variables eπ2,π=R,I,J,K, without controller.

    Therefore, one could select k1=3,k2=1 and the controller can be designed as follows:

    ρ1=minp(2cp+2kpε1m2p1ε1nq=1˜apqˉ˜apq1ε2nq=1˜bpqˉ˜bpq)=0.138,ρ2=ε2maxpm2p=0.05,

    which means ρ1>ρ2>0.

    From the above calculation results, it can be seen that all conditions of Theorem 3.2 are met. Figures 7 and 8 depict the error variables eπ1, eπ2,π=R,I,J,K between systems (2.1) and (2.4) under the controller (3.7). It can be seen from the above discussion that the drive system (2.1) and the response system (2.4) are synchronized, which means that the control technique achieves the desired effect.

    Figure 7.  The error state variables eπ1,π=R,I,J,K, with the controller.
    Figure 8.  The error state variables eπ2,π=R,I,J,K, with the controller.

    In this paper, by using analytical techniques, and constructing two novel controllers, several control strategies were obtained to investigate the asymptotic synchronization and the exponential synchronization of quaternion-valued memristor-based neural networks. At the same time, the direct analysis method was given to discuss the synchronization problem, which simplified the proof process. In the end, numerical simulations were supplied to display the main theoretical results.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    Funding was provided by the Key Laboratory of Numerical Simulation of Sichuan Provincial Universities KLNS-2023SZFZ002; the Scientific Research Foundation of Chengdu University of Information Technology KYTZ202184, KYQN202324 and KYTD202243.

    The authors declare there is no conflicts of interest.



    [1] L. Chua, Memristor-the missing circuit element, IEEE Trans. Circuit Theory, 18 (1971), 507–519. https://doi.org/10.1109/TCT.1971.1083337 doi: 10.1109/TCT.1971.1083337
    [2] D. B. Strukov, G. S. Snider, D. R. Stewart, R. S. Williams, The missing memristor found, Nature, 453 (2008), 80–83. https://doi.org/10.1038/nature06932 doi: 10.1038/nature06932
    [3] Y. Li, Y. Zhong, L. Xu, J. Zhang, X. Xu, H. Sun, et al., Ultrafast synaptic events in a chalcogenide memristor, Sci. Rep., 3 (2013), 1619. https://doi.org/10.1038/srep01619 doi: 10.1038/srep01619
    [4] Y. Pershin, M. Di Ventra, Experimental demonstration of associative memory with memristive neural networks, Neural Netw., 23 (2010), 881–886. https://doi.org/10.1016/j.neunet.2010.05.001 doi: 10.1016/j.neunet.2010.05.001
    [5] F. Merrikh-Bayat, S. Shouraki, Memristor-based circuits for performing basic arithmetic operations, Procedia Comput. Sci., 3 (2011), 128–132. https://doi.org/10.1016/j.procs.2010.12.022 doi: 10.1016/j.procs.2010.12.022
    [6] Z. Q. Wang, H. Y. Xu, X. H. Li, H. Yu, Y. C. Liu, X. J. Zhu, Synaptic learning and memory functions achieved using oxygen ion migration/diffusion in an amorphous InGaZnO memristor, Adv. Funct. Mater., 22 (2012), 2759–2765. https://doi.org/10.1002/adfm.201103148 doi: 10.1002/adfm.201103148
    [7] G. Zhang, J. Hu, Y. Shen, New results on synchronization control of delayed memristive neural networks, Nonlinear Dyn., 81 (2015), 1167–1178. https://doi.org/10.1007/s11071-015-2058-5 doi: 10.1007/s11071-015-2058-5
    [8] J. Hu, J. Wang, Global uniform asymptotic stability of memristor-based recurrent neural networks with time delays, in The 2010 International Joint Conference on Neural Networks (IJCNN), (2010), 1–8. https://doi.org/10.1109/IJCNN.2010.5596359
    [9] J. Cheng, L. Xie, D. Zhang, H. Yan, Novel event-triggered protocol to sliding mode control for singular semi-Markov jump systems, Automatica, 151 (2023), 110906. https://doi.org/10.1016/j.automatica.2023.110906 doi: 10.1016/j.automatica.2023.110906
    [10] A. Wu, Z. Zeng, X. Zhu, J. Zhang, Exponential synchronization of memristor-based recurrent neural networks with time delays, Neurocomputing, 74 (2011), 3043–3050. https://doi.org/10.1016/j.neucom.2011.04.016 doi: 10.1016/j.neucom.2011.04.016
    [11] J. Cheng, Y. Wu, Z. Wu, H. Yan, Nonstationary filtering for fuzzy Markov switching affine systems with quantization effects and deception attacks, IEEE T. Syst. Man Cybern.: Syst., 52 (2022), 6545–6554. https://doi.org/10.1109/TSMC.2022.3147228 doi: 10.1109/TSMC.2022.3147228
    [12] A. Wu, S. Wen, Z. Zeng, Synchronization control of a class of memristor-based recurrent neural networks, Inf. Sci., 183 (2012), 106–116. https://doi.org/10.1016/j.ins.2011.07.044 doi: 10.1016/j.ins.2011.07.044
    [13] G. Zhang, Y. Shen, L. Wang, Global anti-synchronization of a class of chaotic memristive neural networks with time-varying delays, Neural Netw., 46 (2013), 1–8. https://doi.org/10.1016/j.neunet.2013.04.001 doi: 10.1016/j.neunet.2013.04.001
    [14] X. Li, R. Rakkiyappan, G. Velmurugan, Dissipativity analysis of memristor-based complex-valued neural networks with time-varying delays, Inf. Sci., 294 (2015), 645–665. https://doi.org/10.1016/j.ins.2014.07.042 doi: 10.1016/j.ins.2014.07.042
    [15] N. Li, W. Zheng, Bipartite synchronization for inertia memristor-based neural networks on coopetition networks, Neural Netw., 124 (2020), 39–49. https://doi.org/10.1016/j.neunet.2019.11.010 doi: 10.1016/j.neunet.2019.11.010
    [16] Y. Shi, J. Cao, G. Chen, Exponential stability of complex-valued memristor-based neural networks with time-varying delays, Appl. Math. Comput., 313 (2017), 222–234. https://doi.org/10.1016/j.amc.2017.05.078 doi: 10.1016/j.amc.2017.05.078
    [17] H. Wang, S. Duan, T. Huang, L. Wang, C. Li, Exponential stability of complex-valued memristive recurrent neural networks, IEEE Trans. Neural Netw. Learn. Syst., 28 (2017), 766–771. https://doi.org/10.1109/TNNLS.2015.2513001 doi: 10.1109/TNNLS.2015.2513001
    [18] Y. Cheng, Y. Shi, Synchronization of memristor-based complex-valued neural networks with time-varying delays, Comput. Appl. Math., 41 (2022), 388. https://doi.org/10.1007/s40314-022-02097-6 doi: 10.1007/s40314-022-02097-6
    [19] W. Hamilton, Elements of Quaternions, Longmans, Green, & Company, London, 1866.
    [20] S. Pei, C. Cheng, A novel block truncation coding of color images by using quaternion-moment-preserving principle, in 1996 IEEE International Symposium on Circuits and Systems (ISCAS), (1996), 684–687. https://doi.org/10.1109/ISCAS.1996.541817
    [21] M. Xiang, B. S. Dees, D. P. Mandic, Multiple-model adaptive estimation for 3-D and 4-D signals: a widely linear quaternion approach, IEEE T. Neur. Netw. Lear., 30 (2019), 72–84. https://doi.org/10.1109/TNNLS.2018.2829526 doi: 10.1109/TNNLS.2018.2829526
    [22] J. Wang, Y. Li, J. Li, X. Luo, Y. Shi, S. Jha, Color image-spliced localization based on quaternion principal component analysis and quaternion skewness, J. Inf. Secur. Appl., 47 (2019), 353–362. https://doi.org/10.1016/j.jisa.2019.06.004 doi: 10.1016/j.jisa.2019.06.004
    [23] T. Barfoot, J. Forbes, P. Furgale, Pose estimation using linearized rotations and quaternion algebra, Acta Astronaut., 68 (2011), 101–112. https://doi.org/10.1016/j.actaastro.2010.06.049 doi: 10.1016/j.actaastro.2010.06.049
    [24] C. Zou, K. Kou, Y. Wang, Quaternion collaborative and sparse representation with application to color face recognition, IEEE T. Image Process., 25 (2016), 3287–3302. https://doi.org/10.1109/TIP.2016.2567077 doi: 10.1109/TIP.2016.2567077
    [25] T. Isokawa, T. Kusakabe, N. Matsui, F. Peper, Quaternion neural network and its application, in Knowledge-Based Intelligent Information and Engineering Systems, Springer, Berlin, 2003. https://doi.org/10.1007/978-3-540-45226-3_44
    [26] B. C. Ujang, C. C. Took, D. P. Mandic, Quaternion-valued nonlinear adaptive filtering, IEEE Trans Neural Netw., 22 (2011), 1193–1206. https://doi.org/10.1109/TNN.2011.2157358 doi: 10.1109/TNN.2011.2157358
    [27] L. Luo, H. Feng, L. Ding, Color image compression based on quaternion neural network principal component analysis, in 2010 International Conference on Multimedia Technology, (2010), 1–4. https://doi.org/10.1109/ICMULT.2010.5631456
    [28] H. Kusamichi, T. Isokawa, N. Matsui, Y. Ogawa, K. Maeda, A new scheme for color night vision by quaternion neural network, in Proceedings of the 2nd International Conference on Autonomous Robots and Agents, (2004), 1315.
    [29] S. Qin, J. Feng, J. Song, X. Wen, C. Xu, A one-layer recurrent neural network for constrained complex-variable convex optimization, IEEE T. Neur. Netw. Lear., 29 (2018), 534–544. https://doi.org/10.1109/TNNLS.2016.2635676 doi: 10.1109/TNNLS.2016.2635676
    [30] Y. Shi, X. Chen, P. Zhu, Dissipativity for a class of quaternion-valued memristor-based neutral-type neural networks with time-varying delays, Math. Method. Appl. Sci., 46 (2023), 18166–18184. https://doi.org/10.1002/mma.9551 doi: 10.1002/mma.9551
    [31] T. Zhang, J. Jian, Quantized intermittent control tactics for exponential synchronization of quaternion-valued memristive delayed neural networks, ISA Trans., 126 (2022), 288–299. https://doi.org/10.1016/j.isatra.2021.07.029 doi: 10.1016/j.isatra.2021.07.029
    [32] Z. Tu, D. Wang, X. Yang, J. Cao, Lagrange stability of memristive quaternion-valued neural networks with neutral items, Neurocomputing, 399 (2020), 380–389. https://doi.org/10.1016/j.neucom.2020.03.003 doi: 10.1016/j.neucom.2020.03.003
    [33] R. Li, J. Cao, Dissipativity and synchronization control of quaternion-valued fuzzy memristive neural networks: Lexicographical order method, Fuzzy Set. Syst., 443 (2022), 70–89. https://doi.org/10.1016/j.fss.2021.10.015 doi: 10.1016/j.fss.2021.10.015
    [34] X. Song, J. Man, S. Song, C. Ahn, Finite/Fixed-time anti-synchronization of inconsistent markovian quaternion-valued memristive neural networks with reaction-diffusion terms, IEEE T. Circuits-I, 68 (2021), 363–375. https://doi.org/10.1109/TCSI.2020.3025681 doi: 10.1109/TCSI.2020.3025681
    [35] D. Lin, X. Chen, G. Yu, Z. Li, Y. Xia, Global exponential synchronization via nonlinear feedback control for delayed inertial memristor-based quaternion-valued neural networks with impulses, Appl. Math. Comput., 401 (2021), 126093. https://doi.org/10.1016/j.amc.2021.126093 doi: 10.1016/j.amc.2021.126093
    [36] R. Wei, J. Cao, Fixed-time synchronization of quaternion-valued memristive neural networks with time delays, Neural Netw., 113 (2019), 1–10. https://doi.org/10.1016/j.neunet.2019.01.014 doi: 10.1016/j.neunet.2019.01.014
    [37] R. Li, X. Gao, J. Cao, K. Zhang, Exponential stabilization control of delayed quaternion-valued memristive neural networks: vector ordering approach, Circ. Syst. Signal Pr., 39 (2020), 1353–1371. https://doi.org/10.1007/s00034-019-01225-8 doi: 10.1007/s00034-019-01225-8
    [38] Z. Tu, J. Cao, A. Alsaedi, T. Hayat, Global dissipativity analysis for delayed quaternion-valued neural networks, Neural Netw., 89 (2017), 97–104. https://doi.org/10.1016/j.neunet.2017.01.006 doi: 10.1016/j.neunet.2017.01.006
    [39] A. F. Filippov, Differential Equations with Discontinuous Right-hand Sides, Springer Science & Business Media, Berlin, 1988.
    [40] J. Cao, J. Wang, Absolute exponential stability of recurrent neural networks with Lipschitz-continuous activation functions and time delays, Neural Netw., 17 (2004), 379–390. https://doi.org/10.1016/j.neunet.2003.08.007 doi: 10.1016/j.neunet.2003.08.007
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(897) PDF downloads(38) Cited by(0)

Figures and Tables

Figures(8)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog