搜索

x
中国物理学会期刊

多双涡卷忆阻Hopfield神经网络动力学分析及其实现

Dynamical Analysis and Implementation of Multi-Double-Scroll Memristive Hopfield Neural Networks

PDF
导出引用
  • 多双涡卷忆阻 Hopfield 神经网络凭借其复杂的非线性混沌动力学特性,在图像加密、信息安全等领域展现出重要的应用价值。然而,构建多双涡卷的忆阻模型通常采用多项式的方法,其数学形式较为复杂,从而限制了其应用。为解决这一问题,本文提出了一种基于分段函数的忆阻模型,并将其作为可变突触权重引入至 Hopfield 神经网络中,替代固定权重的突触,进而构建了一种新型的双忆阻全连接 Hopfield 神经网络。理论分析表明,该系统能够生成数目可控的多双涡卷混沌吸引子,其中包含复杂的网格多双涡卷混沌吸引子,且涡卷数量可通过忆阻的内部参数灵活调整。此外,研究发现该系统可展现出具有初始偏移增强特性的共存吸引子,其数量也可由忆阻模型的内部参数进行调控。同时,基于 NIST SP 800-22 测试结果,验证了该网络生成的混沌序列具备良好的随机性与不可预测性,满足密码学应用要求,预示了其潜在的应用价值。最后,基于FPGA 完成了该系统的数字电路实现,并验证了其物理存在性与可行性。

     

    Recent research on multi-double-scroll memristive Hopfield neural networks has attracted considerable attention, and significant progress has been made. However, most existing models rely on a single memristor for regulation, yielding unidirectional multi-double-scroll chaotic attractors. Moreover, current systems are typically constructed using polynomial-based memristor models, in which the number of scrolls varies with the number of polynomial terms. Although such approaches allow programmable control of scroll count, their complex mathematical formulations hinder further application.
    To overcome this limitation, this study proposes a new memristor model based on piecewise functions, which is incorporated into a Hopfield neural network as both a self-synaptic and an inter-synaptic weight. Theoretical analysis shows that the proposed memristive Hopfield neural network can generate an arbitrary number of multi-double-scroll attractors, and the scroll count can be flexibly adjusted via the memristor’ s control parameters. In addition, the model can be extended by adding more memristive synapses to construct higher-dimensional networks, demonstrating flexibility and generality.
    The nonlinear dynamics of the system are investigated using bifurcation diagrams, Lyapunov exponent spectra, phase portraits, and basins of attraction. Results indicate that the system can produce grid-style multi-double-scroll chaotic attractors, where the total number of scrolls equals the product of the scroll counts generated along each of the two directions. Further analysis reveals that the system exhibits initialoffset-enhanced coexisting attractors: varying only the initial condition of the memristor yields multiple chaotic attractors with identical shapes but shifted positions. The number of these coexisting attractors can also be controlled via parameters, indicating the presence of super-multistability.
    The chaotic sequences generated by the network pass all 15 tests of the NIST SP 800-22 statistical suite, with p-values above 0.01, confirming their randomness and unpredictability for cryptographic applications such as image encryption and secure communication. Comparative results with recent studies show that the proposed system achieves p-values closer to 0.5 and relatively high pass rates across multiple tests, demonstrating satisfactory randomness.
    Finally, the proposed system is implemented on an FPGA platform using SOPC technology and the Euler discretization method. Oscilloscope measurements agree well with MATLAB numerical simulations, verifying its physical realizability and feasibility.

     

    目录

    /

    返回文章
    返回