Network Traffic Model Based On Wavelet Decomposition And Arima

With the arrival of information age, Internet is developing rapidly, while various new network application are emerging one after another, which led to the emergence of various network problems, and to brought the network monitoring huge challenges. Network detection is the important means to guarantee the normal operation of network, and it has important significance that apply the right and effective network traffic model in the network monitoring. The current network flow has performed various complex characteristics like the comparability, the multifractal sex etc., which led that traditional flow model has failed to meet the demand.This paper analyzes the characteristics of the traditional network traffic model, we know that those mature model shows very good prediction effect aiming at smooth network traffic. However, the current network flow showed no steady characteristics the emergence of universal for various complex characteristics have been discovered in the current network flow. In order to solve difficulty of applying traditional models on the current network flow with more complex characteristics, this paper introduces wavelet transform technique to use the multi-resolution characteristics to decompose the network traffic series with more complex factor into many sub-series on different scales, which will be convenient to be separated. The idea of using wavelet decomposition decompose the non-stationary time series into proper different frequency band on multiple stationary time series (sub-series), and then in these stationary sub-series make relative convenient and efficient handling instead of the whole complex signal, and finally unified handling restored to the original scales, improves prediction accuracy of the real network traffic.While introducing wavelet technology bring forecasting precision, because of the single time series decomposed into many groups of time sequence in different scales, which lead time complexity to be deteriorated seriously, and affected its practical application. Based on detailed analysis and study of this wavelet method, this paper puts forward some measures for improving the wavelet model, which would not impact prediction effect premise seriously, and shorten the time complexity of the algorithm as far as possible. First, we analysis the characteristics of many groups of son sequences,which are obtained from network traffic wavelet decomposition,and merge flow son sequence with the similar spectrum and similar characteristics, in order to reduce the number of sons sequence and reduce forecasting times. This method is reasonable for the premise of similar sequence characteristics are both smooth, stationary sequences,which still is smooth on algebra operations, and then consequently still can get reasonable final results. Experiment proof that the forecasting accuracy was not affected. Then, part the sub-series from merging into details items in high frequency spectrum, the outline in low frequency spectrum, and the middle periodic items, among which three different sequences we will adopt different processing measures to further reduce algorithm time complexity. This paper adopts this way discribed before, and make three components adopt a small amount of historical data, and low sampling frequency to not affect the prediction accuracy improvement from wavelet transform. The experimental results show that these measures improve the time complexity, under the situation of forecasting results have not been seriously affected, compared just introducing wavelet decomposition technique. Therefore, the network traffic prediction method based on wavelet transform and traditional ARIMA model of is feasible.