Research And Implementation Of Network Traffic Forecast Method Based On Empirical Mode Decomposition

With the gradual popularization and development of the Internet,the number of devices connected to the network is gradually increasing,new applications are continually emerging,the aggregate network traffic on the backbone link is rapidly growing,and it presents many unusual sophisticated features,which poses a considerable challenge for maintaining network stability and improving performance.To enhance network stability and improve operating efficiency and resource utilization,it is necessary to transition from the traditional problem-driven management model to the data-driven management model.By analyzing historical traffic data and using forecasting methods to know future needs and status in advance,more flexible and efficient configurations can be made.However,the current mainstream prediction methods have the problems of low prediction accuracy,high model complexity,and lack of adaptive capabilities adjustment.Therefore,this paper proposed a network traffic prediction model with high precision and self-adaptive adjustment ability because of the problems in network traffic prediction,combined with Empirical Mode Decomposition and machine learning algorithm ideas.This model can predict aggregated network traffic with complex characteristics on the backbone link and provide a theoretical basis for network bandwidth resource allocation decisions.The main research work of this paper includes the following three aspects:(1)A network traffic prediction model based on Empirical Mode Decomposition(EMD)and Residual Gated Recurrent Unit Neural Network(Res Gru NN)was proposed.The model uses EMD to decompose the complex time series of traffic into multiple narrow-band components,and build a component prediction model based on Res Gru NN for each component.Finally,the prediction results of each component prediction model are weighted and fused to obtain the result.(2)A component fusion weight setting method based on the orthogonal index system was proposed.First,the statistical characteristics of each component and the performance of the component prediction model are evaluated to get the initial evaluation index system.Then the principal component analysis is used to remove the correlation between each index to obtain an orthogonal index system,calculate the weight of each component based on the orthogonal index system,and perform weighted fusion on the prediction results of the component prediction model to improve the prediction accuracy of the prediction model under the influence of noise or randomness.(3)An incremental data decomposition and incremental learning method based on Gaussian Weight Sliding Window-Empirical Mode Decomposition(GWSW-EMD)was proposed.Through this method,the incremental data components can be obtained in time and reliably,and the incremental learning ability of each component prediction model is given,so that the network traffic prediction model can be adaptively adjusted according to the change of traffic characteristics,to ensure that the accuracy of prediction will not be significantly reduced by the change of traffic characteristics,improving the performance stability of the model.To sum up,this paper made an in-depth study on the network traffic prediction problem of backbone links with complex characteristics and proposed a network traffic prediction model based on EMD and Res Gru NN.On this basis,this paper proposed a component fusion weight setting method,and finally,put forward an processing method of network traffic incremental data to give the model incremental learning ability.This paper designed simulation experiments for each research result,and developed and deployed a network traffic prediction prototype system to verify the feasibility and effectiveness of the research results of this paper.