摘要
合理选择隐含层结点个数是BP神经网络构造中的关键问题,对网络的适应能力、学习速率都有重要的影响。在此提出一种确定隐结点个数的改进方法。该方法基于隐含层神经元输出之间的线性相关关系与线性无关关系,对神经网络隐结点个数进行削减,缩减网络规模。以零件工艺过程中的加工参数作为BP神经网络的输入,加工完成的零件尺寸作为BP神经网络的输出建立模型,把该方法应用于此神经网络模型中,其训练结果证明了该方法的有效性。
The number of hidden units in a BP neural network is significant in characterizing the performance of the network. It greatly influences generalization ability and learning speed. This paper proposes an improved method of determining the number of hidden nodes. The method, based on the linear correlation and linear independent relationship of the output of hidden layers, reduces the number of hidden units and the size of network. Taking the machining process parameters as the input of BP neural network and taking the completed part as the output, this paper applied the method to neural network model. The training results demonstrate the effectiveness of this method
出处
《系统仿真技术》
2014年第2期154-158,共5页
System Simulation Technology
基金
航空制造过程中工程变更传播机制与自修复响应方法研究资助项目(51205201)
关键词
BP神经网络
隐结点个数
线性相关
线性无关
BP neural network
the number of hidden nodes
linear correlation
linear independent