本文为英国杜伦大学（作者：Chen, Jian-Rong）的博士论文，共222页。

本文讨论了人工神经网络的一些基本理论问题及其在通信和控制系统中的应用。我们考虑了广泛应用于人工神经网络训练的反向传播算法的收敛性，提出了两步变步长技术来加速收敛。仿真结果表明，与传统的反向传播算法相比，该算法具有明显的改进。讨论了人工神经网络泛化性能与其结构和表示策略的关系。结果表明，网络结构代表了环境的先验知识，对泛化性能有很大的影响。本文还给出了一个关于多层感知器（MLP）型网络的隐单元数和容量的定理。在论文的应用部分，我们讨论了利用人工神经网络进行非线性系统辨识的可行性，分析了该方法的优缺点。本文还研究了人工神经网络在通信信道均衡中的应用以及宽带ATM（异步传输模式）通信网络中的呼叫接入控制问题。最后一章是结论和建议。

In this thesis some fundamental theoreticalproblems about artificial neural networks and their application incommunication and control systems are discussed. We consider the convergenceproperties of the Back-Propagation algorithm which is widely used for trainingof artificial neural networks, and two step size variation techniques areproposed to accelerate convergence. Simulation results demonstrate significantimprovement over conventional Back-Propagation algorithms. We also discuss therelationship between generalization performance of artificial neural networksand their structure and representation strategy. It is shown that the structureof the network which represent a priori knowledge of the environment has astrong influence on generalization performance. A Theorem about the number ofhidden units and the capacity of self-association ML P (Multi-Layer Perceptron)type network is also given in the thesis. In the application part of thethesis, we discuss the feasibility of using artificial neural networks fornonlinear system identification. Some advantages and disadvantages of thisapproach are analyzed. The thesis continues with a study of artificial neuralnetworks applied to communication channel equalization and the problem of callaccess control in broadband ATM (Asynchronous Transfer Mode) communicationnetworks. A final chapter provides overall conclusions and suggestions forfurther work.

1 引言

2 人工神经网络回顾

3 MLP与反向传播算法

4 MLP网络的推广与表示

5 用于非线性系统辨识的MLP网络

6 基于自组织神经网络的自适应均衡

7 基于学习网络的自适应ATM呼叫接入控制

8 结论与未来工作展望

更多精彩文章请关注公众号：