几类延时反馈神经网络的稳定性研究-神经网络数学理论及应用专业论文.docxVIP

摘 摘 要 本文主要研究了几类延时反馈神经H络的垒局渐近稳定性、全局指数稳定性和全局鲁 棒稳定性．同时也讨论了延时神经阿络的周期解的存在唯一性和指数收敛性． 在第一章第一节中，利用Brouwel度理论和Lyapunov方法研究了定常延时神经网络 的全局指数稳定性，同时，利用压缩映像原理对论了周期解的存在唯一性和指数收敛性； 在第二节中，利用同胚性质、Lasalle不变原理等方法研究r一类具有分布延时的神经网络 的全局渐近稳定性和全局指数稳定性；最后，在第三节中讨论了具有时变延时的神经网络 的全局渐近稳定性． 在第二章中．利用压缩映像原理．同且￡性质、LaSalle不变原理和通过构造新的Lyapunov 泛函等研究了一类延时神经网络的全局鲁棒稳定性．并得到了一些新的在工程上易于验证 的判据． 本文所得到的结果只要求激励函数满足全局Lipschitz条件，有时也假定激励函数是单 调非减的t但是不要求激励函数是有界的、可导的，也不要求连接权矩阵是对称的．另外． 本文研究的延时神经网络模型包含了一些著名的神经网络模型，如延时细胞神经网络、延时 Hopfield神经网络和舣向联想记忆网络，因此我们的结果是以前一些结果的推广和完善． 关键字：神经网络，延时，全局渐近稳定．全局指数稳定．全局鲁棒稳定，周期解． AbstractIll Abstract Ill this pal)e1．we investigat．E、th·global；ksymlltotical stalAlit)the global exponential sta— bility and the global robust stability tlf several kinds of recurrent flf-ural networks with delays In the same time，tile existence，uniqueness and the exponential(：onvergence of tile periodic solution of the neural networks with delays are also discussed In Section 1 of Cbapter 1，by utilizing tile 13rouwer degree theotY and Lyapunov approach， tile global exponential stability,is studied for a class of neural net works with constant delays． And tile existence．uniqueness and tile exponential convergence¨f the periodic solution are discussed by using contraction mapping principle；In Section 2，by oleans of homeomorphism theory and LaSalle invariant principle etc．，tile global asymptotical stability and the global exponential stability are investigated for a class of neural networks with distributed delays； Finally，the global asymptotical stability are discussed for a class of neural networks with time—varying delays in Section 3 In Chapter 2，by using contraction mapping principle，]lomeomorphism property,LaSalle invmiant principle aJld eoustlueting new Lyapmmv fuuctionals，the global robust stability of a class of neural networks is discussed．and a set of new Cl iteria．which are easy t()be checked in engineering，are derived． The results obtained in this paper only require the activation furlctions to be Lipschitz