Research On Network Traffic Prediction Technology Based On Mode Decomposition

The prediction problem of network traffic time series is a important issue in the field of computer network research.Since the network traffic time series has the characteristics of self-similarity,multi-fractal,chaotic,and catastrophic,the modeling of network traffic prediction in recent years starts change to chaos analysis and multi-scale reconstruction and combination prediction model.The thesis first takes the mutation of network traffic as the research issue.Aiming at the mutation problem of network traffic sequence,a local threshold network traffic de-noising model is proposed.Firstly,the current research on the de-noising of network traffic is analyzed.Based on the shortages of wavelet threshold de-noising method and empirical mode decomposition abandoning high-frequency component de-noising methods,a network traffic de-noising model based on CEEMDAN-DE is proposed.Firstly,the model adaptively decomposes the original network traffic sequence,then uses the dispersion entropy to analyze the complexity of the sub-sequence,and choose the noise signal set based on the threshold value of dispersion entropy balance point.Then,the de-noising set is divided by discard,de-noising and reserve to effectively suppress high-frequency noise while keeping the high-frequency information and weak-frequency information in the original signal from being lost.Finally,the processed IMFs are integrated to get denoised traffic sequence.In this thesis,the CEEMDAN-DE network traffic de-noising model is simulated on the simulation signal,by comparing with the wavelet rigrsure hard threshold de-noising,the wavelet sqtwolog soft threshold de-noising and the CEEMDAN high frequency abandon,the effectiveness of the method is demonstrated.The thesis focuses on the self-similarity and chaos of network traffic.For the prediction of network traffic,a multi-scale chaotic network traffic combination prediction model based on IVMD-AVE is proposed.In this thesis,firstly,the parameter optimization method of the K value of IMF,penalty factor ? and Lagrange multiplier step r in the process of signal analysis based on decomposition frequency threshold balance point and decomposition RMSE minimization is proposed.Then the IVMD method is used to process the decomposition of the network traffic sequence,and the IMFs of the IVMD are reconstructed into high frequency,intermediate frequency,and low frequency sequences based on the average value of entropy.Finally,the Elman neural and PSO-LSSVM are used to integrate the predict the different frequency sequences and integrat the prediction results.The IVMD-AVE network traffic prediction model is combined with the CEEMDAN-DE network traffic local threshold de-noising model,and simulation experiments are performed on the actual network traffic data set.Compared with single or combined prediction model such as Elman neural network and VMD-Elman-SVM,verify that this model has good adaptability and can improve the effect of network traffic prediction.