样本处理方法对BP网络泛化能力影响的比较研究  

A Comparative Study of the Effect of Sample Processing Method on the Generalization Ability of BP

作　　者：邬友[1] 夏玫 

机构地区：[1]内蒙古师范大学科技处,内蒙古呼和浩特010022 [2]呼和浩特市统计局,内蒙古呼和浩特010010

出　　处：《内蒙古师范大学学报（自然科学汉文版）》2015年第4期520-522,共3页Journal of Inner Mongolia Normal University(Natural Science Edition)

基　　金：内蒙古自然科学基金资助项目(2014MS0119)

摘　　要：泛化能力是评价BP神经网络的重要指标之一,没有良好的泛化能力,网络就不能够应用和推广.网络样本质量是决定BP神经网络能否训练成功的关键要素.对目前常用的几种样本处理方法进行总结和归纳,比较了每种处理方法的优劣.Generalization ability is one important index for BP neural network.The network cannot implemented or promoted without a good generalization ability.Among numerous factors affecting the BP neural network generalization ability,network sample quality is a key to determining the success of the BP neural network training.In this paper,a number of commonly used sample processing methods are summa-rized and classified.Comparisons are made among various approaches in terms of their advantages and disadvantages.Future research has been proposed in the sample processing methods.

关 键 词：BP 神经网络 样本空间 归一化 主成分分析 聚类 神经网络集成 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]