基于组合模型的网络流量预测

Network traffic forecasting based on combined model

为了更好地对网络流量进行分析和管理,提出一种基于小波变换、自回归滑动平均模型(ARMA)和极限学习机(ELM)的组合预测模型W-ARMA-ELM.原始数据通过小波分解产生近似序列和细节序列,通过对分解序列的自相关性和偏自相关分析,平稳序列使用ARMA预测,而非平稳序列使用ELM预测.使用兰州大学教育网、网通流量数据和英国学术主干网流量数据三组不同的网络流量数据来检验组合模型W-ARMAELM的预测性能.实验结果表明提出的组合方法要比单一的ARMA和ELM预测效果要好.同时指出使用自相关和偏自相关分析相结合的方法对分解后的子序列进行平稳性判定有助于选择合适的组合模型从而提高预测精度.

In order to analyze and manage for network traffic,a combined forecasting model named WARMA-ELM which is based on wavelet transform,autoregressive moving average(ARMA)and extreme learning machine(ELM)was proposed.The original data were decomposed into to an approximate series and some details series through wavelet transform.After using autocorrelation and partial autocorrelation analysis,the stationary series were predicted by ARMA,while non-stationary series were forecasted by ELM.Three group network traffic data including Lanzhou University Education Network traffic data and Netcom,British academic backbone traffic data were used to test the prediction performance of W-ARMA-ELM.Experimental results show that the forecasted results of our proposed method are better than the single ARMA and ELM model.Meanwhile,it can be seen that the use of autocorrelation and partial autocorrelation analysis for decomposed series to determine decomposed series′stationary will help choose the right combination models to improve prediction accuracy.