Traffic behavior in a large-scale network is very perplexing and can be viewed as a complicated non-linear system.So far the research on traffic behavior doesn't have a well-rounded method.According to the character of non-linear network traffic

the traffic time series is decomposed into trend component

period component

mutation component and random component.With such decomposition

a complicated traffic can be simulated by compound of four simpler sub-series with different mathematical tools.In order to check our model

the long-term traffic behavior of the CERNET backbone network and NSFNET backbone network are analyzed using the decomposed model

and the results are compared with ARIMA model.According to the autocorrelation function value and prediction error function value

the decomposed model has the advantage of simplicity and high precision to describe the traffic marco-behavior.