Research On Analysis Method Of Network Traffic Anomaly Information

With the development of economy and science and technology and the informatization in recent years,the research on network traffic anomaly detection technology is of great significance.However,the current technology faces many challenges with the increasingly complex network environment.In order to adapt to the current network conditions to detect network traffic anomaly efficiently and accurately,this work proposes an offline network traffic anomaly detection method.This method mainly consists of two modules:(1)network traffic data feature preprocessing and data augmentation module;(2)multi-scale decomposition and multi-channel anomaly detection module.In the network traffic data feature preprocessing and data enhancement module,this work proposes a method of network traffic data feature selection and data augmentation which uses the Minimum Redundancy—Maximum Relevance(MRMR)technology,and PCA feature fusion and Weighted DTW Barycenter Averaging(W-DBA)network traffic data augmentation method.In the multi-scale decomposition and multi-channel anomaly detection module,this work proposes a novel network traffic anomaly detection algorithm combining multi-scale decomposition and multi-channel detection.The method includes a Complementary Ensemble Empirical Mode Decomposition(CEEMD)technology decomposes and characterizes traffic data into multi-scale decomposition components,and a multi-channel Generalized Likelihood Ratio Test(GLRT)detection module for detecting anomalies from multi-scale data.The multi-channel anomaly detection algorithm proposed in this paper overcomes the shortcomings of traditional methods that usually only work independently at each scale,and only analyze time-related information with insufficient pre-processing of network traffic data,by pre-processing and fully consider the internal frequency-time correlation in multiple scales during anomaly detection.Through the analysis of experimental results on three different data sets,it is proved that the proposed method has better detection performance than other traditional methods.