Analysis And Prediction Of Internet Of Things Traffic Characteristics

Characteristic analysis of the network traffic and traffic modeling,as a frontier hot topic in the basic theoretical research of computer networks,are the basis of modern network communication system analysis,design,performance evaluation and dynamic adjustment.At present,the network is changing from the traditional Internet to the Internet of Thing(Io T).Io T traffic has different features and behaviors from the traditional Internet,including the number of connected devices and the amount of data are huge,so the traffic characteristics are bound to change.It is necessary to re-evaluate the traffic characteristics of the next-generation network characterized by the Internet of Things and big data.This thesis is dedicated to the analysis of traffic characteristics and traffic prediction of the Internet of Things.First,use relevant software to show the changes in Io T traffic,and through intuitive observation to get a preliminary understanding of the characteristics of Io T traffic.According to the combination of user roles into three types of traffic,things to things,things to human and human to human,and in-depth analysis of the characteristics of each type of traffic.Then analyze the characteristics of aggregate traffic,focusing on the influence of factors such as the variance,density and ratio of the aggregation source.Finally,based on the conclusions of the analysis of Io T traffic characteristics,a suitable model is selected to predict the traffic and verify the accuracy of the analysis results of the traffic characteristics.Firstly,a statistical analysis tool is used to visually analyze the characteristics of Io T traffic,and it is found that Io T traffic has both fractal characteristics and burstiness.Then use the R/S estimation method to estimate the Hurst parameters of Io T traffic under different time scales to judge the self-similarity and degree of Io T traffic.Studies have shown that the Io T traffic has multi-fractal characteristics.As the time scale increases,Io T traffic gradually degenerates from long correlation to short correlation.The heavy-tailed distribution is used to characterize the burst characteristics of Io T traffic.Four typical heavy-tailed distributions and exponential distributions are used to respectively fit the Io T traffic,and the regression evaluation index is used for error testing,and then the most accurate model is selected to describe the burst characteristics of Io T traffic.The results of the fitting comparison analysis found that the traffic of thing to people is more in line with the Lognormal distribution,and the Pareto distribution could more accurately describe the burst characteristics of the for other types of Io T traffic.Then,based on the method of combining simulated traffic and real traffic,the influence of factors such as the variance,density and proportion of the aggregated source on the fractal and heavy-tail characteristics of aggregated traffic is studied.The research results show that the aggregate flow characteristics are not only related to the H parameter of the aggregate source flow.As the variance,density and proportion of the aggregate source flow increase significantly,the aggregate flow characteristics will also change accordingly.The flow with large variance,density,and proportion in the aggregation source has a greater impact on the characteristics of aggregated traffic.Finally,based on the above analysis results of the multi-fractal characteristics of Io T traffic,the FARIMA model is selected for Io T traffic forecasting,and an improved EMDFARIMA traffic forecasting model is proposed.In this model,EMD is used to modal decomposition of the traffic time series,and then the K-Means algorithm is used to aggregate based on the correlation coefficient between the decomposed sequence and the original sequence.The FARIMA model is used to predict the flow of the aggregated components and the prediction results are linearly superimposed.The results of the experiment show that the prediction accuracy of the model is better than that of the traditional FARIMA model.