具有3个时滞的递归神经网络系统的稳定性分析

Stability analysis of recurrent neural network systems with three delays

摘要

研究由3个神经元组成的具有3个时滞的递归神经网络模型的稳定性。首先对系统在平衡点附近进行线性化处理, 求得其特征方程为含有两个指数项的超越方程。其次利用指数型多项式零点分布定理和特征根分析方法, 讨论系统稳定性切换的临界条件。最后建立保证系统稳定时参数需满足的充分性条件, 并给出3个时滞参数的临界取值。

Abstract

The paper investigates the stability of a recursive neural network model composed of three neurons with three delays. Initially, the system is linearized around its equilibrium point, and the characteristic equation, which contains two exponential terms, is derived. Subsequently, the critical conditions for stability switch are discussed by zero-point distribution properties of exponential polynomials and methods of eigenvalue analysis. Finally, the sufficient conditions are given to ensure the system stable, and the critical values of time-delay are deduced.

关键词

递归神经网络 / 多时滞 / 特征根分析方法 / 稳定性

Key words

recursive neural network / multiple time delays / eigenvalue analysis / stability

参考文献

[1] 吉月辉, 周海亮, 车适行, 等. 基于递归神经网络的再入飞行器最优姿态控制[J]. 控制理论与应用, 2021, 38(3): 329-338.

[2] JI Yuehui, ZHOU Hailiang, CHE Shixing, et al. Optimal attitude control of reentry aircraft based on recurrent neural networks[J]. Control Theory and Applications, 2021, 38(3): 329-338.

[3] 李冠军, 徐进. 时滞递归神经网络全局稳定性的新判据[J]. 华中师范大学学报(自然科学版), 2009, 43(3): 384-387.

[4] LI Guanjun, XU Jin. A new criterion for global stability of time—delay recurrent neural networks[J] Journal of Huazhong Normal University (Natural Science Edition), 2009, 43(3): 384-387.

[5] FEI Yu, LI Rongli, MENG Xiaofang, et al. Bifurcations of a fractional—order four—neuron recurrent neural network with multiple delays[J]. Computational Intelligence and Neuroscience, 2022, (1): 1779582.

[6] SAHU N, VERMA A K. Analysis of equilibria of a recurrent neural network involving transcendental function[J]. International Journal of Applied Information Systems, 2012, 4(5): 6-12.

[7] BAO Gang, PENG Yue, ZHOU Xue, et al. Region stability and stabilization of recurrent neural network with parameter disturbances[J]. Neural Processing Letters, 2020, 52(3): 1-14.

[8] LIU Peng, ZHENG Weixing, ZENG Zhigang. on complete stability of recurrent neural networks with time—varying delays and general piecewise linear activation functions[J]. IEEE Transactions on Cybernetics, 2019, 50(5): 1-15.

[9] RUIZ V, ARANDA—ESCOLÁSTICO E, SALT J L, et al. Stability and synchronization of switched multi—rate recurrent neural networks[J]. IEEE Access, 2021, 9: 45614-45621.

[10] GAO Bo, ZHANG Weinian. Equilibria and their bifurcations in a recurrent neural network involving iterates of a transcendental function[J]. IEEE Transactions on Neural Networks, 2008, 19(5): 782-794.

[11] LI Yingguo. Stability and bifurcation analysis of a three—dimensional recurrent neural network with time delay[J]. Journal of Applied Mathematics, 2012: 1-13.

[12] ZHAO Dongxia, WANG Junmin. Exponential stability and spectral analysis of a delayed ring neural network with a small world connection[J]. Nonlinear Dynamics, 2012, 68: 77-93.

[13] 鲍芳霞, 赵东霞, 庞玉婷. 具有3个时滞的环形神经网络系统的稳定性分析[J]. 重庆理工大学学报(自然科学), 2022, 36(8): 318-326.

[14] BAO Fangxia, ZHAO Dongxia, PANG Yuting. Stability analysis of a circular neural network system with three time delays[J]. Journal of Chongqing University of Technology (Natural Sciences), 2022, 36(8): 318-326.

[15] RUAN Shigui, WEI Junjie. On the zeros of transcendental functions with applications to stability of delay differential equations with two delays[J]. Dynamics of Continuous, Discrete and Impulsive Systems, 2003, 10(6): 863-874.

基金资助

山西省基础研究计划资助项目(202403021221124)