BP网络结构、参数及训练方法的设计与选择

Design and Selection of Construction, Parameters and Training Method of BP Network

根据已有的网络设计及其改进方案,对一个人工神经网络进行具体的设计,通过其详细的设计步骤与过程,对网络隐含层神经元数、初BP始权值、学习速率等参数在网络设计过程中的关系与影响,以及不同的改进算法在网络训练中所起的作用给予进一步揭示,使人们从中得到更BP多的启迪,以便使更多的人能够设计出效率更高、精度更好的神经网络。

The purpose of this paper is to design an artificial neural network according to the method and improved ways of training BP network. Through the detail design steps and procedures, the paper reveals further the relationships and effects among the number of hidden layer, initial weights and learning ratio of BP network. And the same time, the part of different fast training algorithms played in the network training is given in order to make more people design more efficient and better neural networks.