几类带有非Lipschitz激励函数的神经网络鲁棒稳定性分析-计算数学专业论文.docxVIP

哈尔滨工业大学理学硕士学位论文 哈尔滨工业大学理学硕士学位论文 - - II - Abstract As we all know, neural networks are widely used in various fields, and the stability is the basis and precondition of applications. During the implementation of neural networks, the stability is destroyed because of two main factors, which are time delays and uncertainties. Therefore, in order to play the application value of the neural network, the discussion for robust stability of delayed neural network is extremely important. In this paper, we mainly study the robust stability for two classes of the delayed neural networks. Firstly,The robust exponential stability for neural network with general activation function is studied. We make a comparison with some previous results. We cancelled global Lipschitz condition of the activation function. The assumptions for activation function in our paper is more generally. Based on the topological degree theory, we prove existence, uniqueness of equilibrium point. By Lyapunov method, we provide global robust exponential stability of the neural network. And the advantages of this conclusion is illustrated by an example. Secondly, the global robust stability of neural networks with inverse-Lipschitz activation function is investigated. The neural networks with inverse-Lipschitz activation function have important application value and attract more and more attention from researchers. In this paper, a sufficient condition which ensures the global robust stability of the neural network is established. Some numerical examples are given to verity the effectiveness of the result. Keywords: delayed neural network, robust stability, topological degree theory, non-Lipschitz, Lyapunov function - - III - 目 录 摘 要 I ABSTRACTII 第 1 章 绪 论 1 1.1 神经网络的发展背景及意义 1 1.2 神经网络鲁棒稳定性的研究现状2 1.3 预备知识 4 1.3.1 神经网络模型4 1.3.2 神经网络的稳定性 5 1.3.3 拓扑度的几个基本定理 5 1.3.4 几个重要的不等式 6 1.4 本文的主要研究内容 6 第 2 章 一般激励函数神经网络的全局鲁棒指数稳定性 8 2.1 平衡点的存在唯一性 8 2.2 平衡点全局鲁棒指数稳定性研究12 2.3 数值算例 16 2.4 本章小结 20 第 3 章 反向