基于SVM的数字标签库特征子空间集成分类  

Integrated classification of digital label library feature subspaces based on SVM

作　　者：王猛 杨劲松 黄俊惠 WANG Meng;YANG Jinsong;HUANG Junhui(Yongyao Technology Branch of Ningbo Power Transmission and Transformation Construction Co.,Ltd.,Ningbo 315000,China)

机构地区：[1]宁波送变电建设有限公司永耀科技分公司,浙江宁波315000

出　　处：《电子设计工程》2024年第22期181-185,共5页Electronic Design Engineering

基　　金：浙江电力公司科技项目资助(GDKJRM20213062)。

摘　　要：数字标签库特征子空间中的数据集存在一定差异性,容易出现正负类样本决策误差,降低子空间集成分类精度。为此,该文提出了基于SVM的数字标签库特征子空间集成分类方法。构建基于SVM的集成学习模型,删除大量冗余子空间,采用最大均值差异方法衡量数字标签库特征子空间数据的分布差异,简化复杂的分类步骤。根据SVM的最优超平面求解原理,训练正负类样本决策集。计算特征子空间数据与给定边缘样本集距离,构建空间集成分类函数,实现标签库特征子空间集成分类。实验结果可知,该方法对Normal、Probe、Dos、U2R特征子空间分类结果与实验数据一致,具有精准的分类结果。The datasets in the characteristic subspace of the digital label library have certain differences,which are prone to positive and negative class sample decision errors and reduce the accuracy of subspace ensemble classification.To this end,a digital tag library feature subspace ensemble classification method based on SVM is proposed.Construct an integrated learning model based on SVM,delete a large number of redundant subspaces,and use the maximum mean difference method to measure the distribution differences of digital label library feature subspace data,simplifying the complex classification steps.According to the optimal hyperplane solution principle of SVM,the positive and negative sample decision sets are trained.The distance between the feature subspace data and the given edge sample set is calculated,and the spatial integration classification function is constructed to realize the integrated classification of the feature subspace of the tag library.The experimental results show that the method has accurate classification results for Normal,Probe,Dos and U2R feature subspaces,which are consistent with the experimental data.

关 键 词：SVM 数字标签库 特征子空间 集成分类 

分 类 号：TN302[电子电信—物理电子学]