Research On Campus Network Traffic Prediction Based On Wavelet Neural Network

Internet services are running and developing at a very fast speed,and constantly changing and innovating in the direction of diversification.The network structure is becoming more and more complex.The network traffic prediction model can provide necessary reference for bandwidth allocation,routing control and error adjustment in network management.It is especially important to improve network performance for providing better quality of service(QoS).Therefore,the analysis and prediction technologies of network traffic have been a popular research topic in related fields.Firstly,on the basis of studying the complex and non-linear characteristics of network traffic,this paper filters and collects the real network traffic data from different equipment lines in the network center of domestic universities;then,considering the correlation of time series,the missing and abnormal parts of the original network traffic data are identified and corrected comprehensively,and the standard normalization method is used to complete the pretreatment of network traffic samples at the same time.Secondly,after thoroughly analyzing the current situation,advantages and disadvantages of various network traffic prediction algorithms,this paper chooses the wavelet neural network(WNN)as the basis of research,then an intelligent wavelet neural network(IWNN)prediction model is proposed by adding momentum term and dynamic self-learning factor in the process of parameter reverse correction..Experiments based on actual campus network traffic data fully show that the predicted results of IWNN are superior to BP neural network and WNN in terms of the fitting degree and the convergence ability.Thirdly,in order to improve the weak stability and slightly higher error of the single IWNN model,a campus network traffic prediction model based on IWNN integrated by AdaBoost(AdaBoost-IWNN)is constructed by using the method of neural network ensemble in this paper,and the overall generalization performance of the model is improved by using independent and diversified individual learners.The experimental results show that the AdaBoost-IWNN model has higher prediction accuracy and error stability than the IWNN model.Finally,aiming at the unstable epochs in the training process of the IWNN-based learners,an adaptive differential evolution(ADE)algorithm with self-tuning control parameters is proposed,which is introduced into the IWNN model as a pre-optimization strategy for weights and wavelet coefficients,by reasonably utilizing the global optimization ability of the ADE algorithm,it makes up the shortcomings of gradient descent algorithm in the IWNN model,which is sensitive to construction parameters and easy to fall into local minimum solution.Through a series of simulation prediction experiments,it is proved that the AdaBoost-ADE-IWNN prediction model has higher generalization ability and better comprehensive performance,it is a precise and reliable network traffic prediction algorithm model.