Research On Self-similarity Analysis Of Campus Network Traffic And Traffic Prediction Based On EMD Recurrent Neural Network

With the rapid development of network technology,the rapid increase of network traffic leads to the increase of Internet complexity and brings a series of threats to network security,such as network congestion.The research shows that the important characteristics of network traffic data,such as self-similarity,long-term correlation,multi-fragmentation and periodicity,have been widely used in many aspects,and self-similarity theory and traffic prediction occupy an important position in the field of network traffic research.Network traffic data depends to a large extent on its self-similarity and high nonlinearity,so it is difficult to fully capture these characteristics by the prediction model.Through the observation and analysis of the network traffic in the past and predicting the characteristics of the network traffic in the future,it provides a reliable basis for network monitoring,resource management,threat detection,abnormal activities and so on.At the same time,it can also enable managers to master the real-time characteristics of network traffic,and use this information to plan and configure the campus network reasonably to ensure the normal operation of the campus network.Therefore,in order to improve the ability to analyze and predict the network traffic,this paper makes an in-depth study of the campus network and analyzes its regular characteristics,using the actual network environment and traffic data to establish a more accurate campus network traffic prediction model.First of all,this paper studies the self-similarity characteristics of campus network traffic,combined with the overall campus network topology structure,through the analysis of the long-term,sudden and periodic of campus network traffic data,a method based on EMD empirical mode transformation and R/S analysis is applied to calculate the hurst value of campus network traffic self-similarity process,through the campus network traffic on different time scales.The results calculated bythis method are compared with those calculated by the ordinary R/S analysis method.Secondly,after verifying the self-similarity of campus network traffic and the adaptability of EMD in dealing with network traffic data,a hybrid prediction model of cyclic neural network traffic based on EMD and Adam optimization is proposed to predict the real traffic data of campus network for 6 months,and the performance of the model is evaluated and compared.From the experimental results,we can see that the campus network traffic has the characteristic of self-similarity,and the self-similarity Hurst value during the day is generally higher than the Hurst value at night.For the campus network,which is a periodic and regular local area network,the calculation of traffic self-similarity Hurst value directly can not better monitor the abnormal traffic.By preprocessing the collected time series information of initial network traffic with EMD transform,this transformation process can eliminate the trend items that affect the self-similarity of network traffic.Through this method,it has a certain effect on the calculation of self-similarity Hurst value,and the time series information of network traffic after EMD transformation will not change its original self-similarity characteristics.Because the network traffic data itself has many unique characteristics,such as self-similarity,long-term correlation and so on,when considering the autocorrelation characteristics,combined with the research on the self-similarity of the campus network,the advantages of denoising the trend items separated by LSTM and EMD are combined.The Adam optimizer is used to optimize the network model,and the final experimental study shows that the method combined with EMD empirical mode can improve the prediction accuracy of the model to a certain extent.