【作者】 刘期烈；

【导师】 张翠芳；

【作者基本信息】 西南交通大学 ， 通信与信息系统， 2004， 硕士

【摘要】 混沌被誉为非线性系统中的一朵奇葩，混沌控制在实际应用中显示出广阔的前景。人们对混沌控制展开了深入研究，提出了多种控制方法。但是，由于混沌系统动力学行为的复杂性和独特性，迄今，混沌系统的控制理论还很不成熟，特别是对系统数学模型难以准确建立、先知条件较少的情况下，传统的控制方法难以奏效。自20世纪80年代以来，神经网络研究取得很大进展，许多网络模型已被广泛地应用于模式识别、图像处理、语言处理等领域，神经网络能够学习任意连续非线性函数，决定了它们在处理高度非线性和严重不确定系统的控制方面存在巨大的潜力。 本文研究了电子和电路系统存在的混沌现象的神经网络控制问题，其中包括两个方面的内容：混沌系统的神经网络辨识技术和基于神经网络的混沌控制。 首先，在分析了在系统辨识过程中应用比较多的基本BP算法及其改进算法的优缺点的基础上，将附加动量项和自适应变学习率两种算法的优点相结合，给出了一种新的改进的BP算法，并将其用于BP神经网络的Lorenz混沌系统辨识研究，仿真结果表明所提出的算法的辨识效果优于单独附加动量项或自适应变学习率的BP算法。 其次，给出了神经网络应用于混沌系统控制的几种结构，并采用了模型参考自适应控制结构，利用B-样条神经网络对Henon混沌系统进行了控制仿真研究，其结果表明，该方法比BP神经网络控制混沌更有效。 最后，为了对未知多变量的混沌系统进行有效的控制，采用了动态神经网络控制未知多变量Lorenz混沌系统，利用计算机仿真证明了将动态神经网络引入混沌控制的可行性。更多还原

【Abstract】 The control of chaotic systems has capacious application foreground, a great number of method for the control of chaotic systems has been brought out. However, control theory of chaotic systems is still not perfect so far because of the complexity and particularity of the dynamic behavior in chaotic systems. Especially when the mathematic model of the system cannot be actually constructed and little preconditions be attained, classic control methods not act that well. Since 1980’s, the research of neural networks has made great progresses, which has already proved that the neural networks ean approach any continuous nonlinear function. Characteristics of neural networks make it has considerable potential when be used to control a system which is highly nonlinear or uncertain.In this paper, research concentrates on the neural network control of chaotic systems in electronics and circuits systems, including two fields: chaotic systems identification and chaotic control.Firstly, based on the analysis of advantages and disadvantages of basic back-propagation neural network and ameliorating arithmetic in the process of systems’ identification, a new ameliorating arithmetic which is the combination of additional momentum and adaptive momentum method has been presented, and then identifying Lorenz chaotic system is carefully studied by using of the new method. Results of the numerical simulation demonstrate that the newly presented scheme is of great effectiveness.Secondly, several kind of neural network controlling structures which apply to chaotic systems are given out, adopted reference model and self-adaptive controlling structure are consulted. This paper putmuch energy into the simulation of B-spline neural network in Lorenz chaotic system. Results of computer simulation clearly show that this method is much more effective as compared whit back-propagation neural networks.At last, in order to control unknown multivariable chaotic systems effectively, dynamical neural network is employed to control the unknown multivariable Lorenz chaotic system. Computer simulation elementary demonstrated the application feasibility of dynamical neural network in chaotic control.更多还原