Research On Prediction Model Of Network Traffic

Along with the rise of mobile Internet, high-definition video, online games,cloud application services, network traffic has exploded, resulting in a variety ofnetwork problems. In order to provide a better quality of service, networkmaintenance and management is particularly important. Through the network trafficprediction, not only can make the real-time monitoring of network traffic data forpeople, providing important basis for operators to direct traffic, but also can providean effective basis for the network traffic control, bandwidth allocation, routing control,fault management. As the reason, before the occurring of the network overload,measures can be taken in advance, to ensure the normal service network. Therefore,the network traffic prediction play a decisive role in the modern communicationnetwork.This paper analyses the characteristic of several kinds of network trafficforecasting models, but these models in predicting performance have their defects.Network traffic data related to the long, self similarity, periodicity, sudden andmulti-scale features based on flow changes, the single forecast model can notaccurately describe the complexity of higher law. Combination forecasting modelcombined with the advantages of various single forecasting model to describe thecomplex flow characteristics, a large number of practical results show that, thecombination forecasting model to describe the complex flow characteristics is moreaccurate and comprehensive, It can effectively improve the prediction accuracy ofnetwork traffic prediction.Due to the character of nonlinear stationary, self similar and multi scale that thenetwork traffic time series exhibit, a new network flow combination forecastingmodel which takes advantages of empirical mode decomposition and least squaressupport vector regression machine is proposed. Firstly, through the empirical modedecomposition, the original flow with complex characteristics will be decomposedinto several time series which with a certain regularity, relatively simple and easy topredict. Then we use the least squares support vector regression models optimized bythe particle swarm, And last the final forecasting result of the original sequence wasobtained by support vector machine combination. The experimental results verifiedthe validity and feasibility of the model. The algorithm model can obviously improvethe prediction accuracy of the network traffic. The parameters of LS-SVM model selection plays a key role to the prediction effect, Further study of this paper, weintroduce the Global Edition artificial fish swarm algorithm to replace the particleswarm algorithm to optimize the parameters, After the experiment, the network trafficcombination prediction model based on the least squares support vector machineoptimized by the global edition artificial fish swarm has better prediction accuracy.