融合小波变换与贝叶斯LS-SVM的网络流量预测  

Combining wavelet transform and Bayesian least squares support vector machines to predict network traffic

作　　者：刘渊[1,2] 王鹏[1] 

机构地区：[1]江南大学信息工程学院 [2]江南大学数字媒体创意中心,江苏无锡214122

出　　处：《计算机应用研究》2009年第6期2229-2231,共3页Application Research of Computers

基　　金：国防基础研究基金资助项目(A1420061266);高等学校科技创新工程重大项目培育资金项目

摘　　要：为了提高网络流量预测的精度,研究了一种融合小波变换与贝叶斯LS-SVM的网络流量预测方法。首先将原始流量数据时间序列进行小波分解,并将分解得到的近似部分和各细节部分分别单支重构到原级别上;对各个重构后的序列分别用最小二乘支持向量机进行预测,将贝叶斯证据框架应用于最小二乘支持向量机模型参数的选择;将各个预测结果重构后得到对原始序列的预测结果。对比实验表明,该模型不仅具有较快的运行速度,而且具有较高的预测精度。In order to improve the precision of the network traffic prediction, this paper proposed a new method of network traf- fic prediction combining wavelet transform and least squares support vector machines （LS-SVM）. The original network traffic time series was decomposed into approximate series and several detail series. The result of single branch reconstruction of each decomposed series was more unitary than the original series in frequency, and it could be built traffic model with LS-SVM. The Bayesian evidence framework was applied to LS-SVM in order to determine the regularization parameters and kernel parameters effectively. The prediction of the original series could be obtained by the synthesis of each reconstructed series＇ prediction result. The simulation results indicate the high accuracy and soeed of this method.

关 键 词：网络流量预测 小波变换 支持向量机 最小二乘支持向量机 贝叶斯框架 

分 类 号：TP393.01[自动化与计算机技术—计算机应用技术]