摘要
为了准确挖掘离群点,降低离群点对通信数据造成的影响,对IPv6远程监控网络无状态通信数据多尺度离群点挖掘算法进行了研究。通过IPv6远程监控网络获得无状态通信数据,依据提取的无状态通信数据的季节性、趋势性和自相似性特征,运用傅里叶变换将无状态通信数据分为两类。再用K均值法对两类分别进行聚类,确定无状态通信数据的邻域,将其作为基础。采用卷积神经网络对无状态通信数据进行离群点挖掘,初始化卷积神经网络;根据卷积神经网络输出值,判别该网络是否符合停止条件,反复重复卷积神经网络的运算步骤,挖掘全部离群点,实现无状态通信数据多尺度离群点挖掘。实验结果表明,无状态通信数据类别的个数越少,挖掘效率越高;所提方法能准确挖掘IPv6远程监控网络无状态通信数据多尺度离群点的个数,准确分析离群原因。
In order to accurately mine outliers and reduce the impact of outliers on communication data,a multi-scale outlier mining algorithm for stateless communication data in IPv6 remote monitoring network was investigated.The stateless communication data were obtained through an IPv6 remote monitoring network,and based on the seasonali-ty,trend,and self-similarity characteristics of the extracted stateless communication data,the Fourier transform was used to divide the stateless communication data into two classes.Then,the K-mean method was used to cluster the two classes to determine the neighborhood of the stateless communication data,which was used as the basis for out-lier mining using a convolutional neural network on the stateless communication data.The convolutional neural net-work was initialized,and according to the output value of the convolutional neural network,it was determined whether the network met the stopping condition.The operation steps of the convolutional neural network were re-peated,all the outlier points were mined,and the multi-scale outlier mining of stateless communication data was achieved.The experimental results showed that the fewer the number of stateless communication data categories,the higher the mining efficiency;the proposed method can accurately mine the number of multiscale outliers of stateless communication data in IPv6 remote monitoring network and accurately analyze the reasons for the outliers.
作者
刘琨
张晓涵
曹汝坤
李帅
LIU Kun;ZHANG Xiaohan;CAO Rukun;LI Shuai(China Green Development Investment Group Co.,Ltd.,Beijing 100020,China)
出处
《电信科学》
2023年第8期118-126,共9页
Telecommunications Science
