Time-frequency Analysis Based On Network Traffic Anomaly Detection

With computer network scale and application fields enlarging day by day, network plays an important role in people's daily work and life. As the growth of the Internet popularize intensity, network traffic anomalies ,which are caused by network attacks, worm virus, malicious download, equipment failure ,etc, increasingly impact the network performance, some unusual offensive flow disrupts the normal operation of the network order, poses a serious threat to network security .In such situations ,how to detect the network traffic anomalies accurately and in time, ensure the normal operation of the network ,provide users with a good network environment ,become a topic of concern .If we regard the traffic flow as a time-varying signal, traffic flow anomalies not only perform in time domain, but also in frequency domain .And it can display some inherent characteristics in frequency domain. Most of the existing traffic anomaly detection methods discuss only the characteristics in time domain, but the characteristics in frequency domain are less involved. To remedy the deficiency of the analysis methods in time domain and detect the anomalies more accurately, we take research on the network traffic anomaly detection based on time-frequency analysis, and demonstrate the inherent characteristics of the traffic anomalies in time-frequency domain in greater depth.This paper summarizes the existing network traffic anomaly detection methods and models related, takes further study in applying time-frequency analysis methods to the anomaly detection, and put forward some new thinking. With the integration of time-frequency analysis, pattern-recognition and statistical analysis algorithm, we present the respective anomaly detection models.First, realized a Wegener-Willie distribute based network traffic anomaly detection algorithm. We make use of Wegener-Willie distribute to analyze the inherent time-frequency distribution characteristics of the traffic flow signal. Then According to the experience of analysis on historical flow, we construct a normal flow training sample aggregation and a abnormal flow training sample aggregation. At last we...