基于KPCA优化ESN的网络流量预测方法  

Network traffic prediction method based on KPCA optimized ESN

作　　者：田中大[1] 李树江[1] 王艳红[1] 高宪文[2] 

机构地区：[1]沈阳工业大学信息科学与工程学院,辽宁沈阳110870 [2]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《电机与控制学报》2015年第12期114-120,共7页Electric Machines and Control

基　　金：国家自然科学基金(61034005);辽宁省博士科研启动基金(20141070)

摘　　要：为了提高网络流量的预测精确度,提出一种核主成分分析(KPCA)优化回声状态网络(ESN)的网络流量预测方法。首先利用相空间重构对网络流量序列进行处理,提高序列的可预测性,然后对网络流量序列进行核主成分分析,提取序列中的有效信息,通过实验方法确定回声状态网络的储备池参数,最后利用回声状态网络对网络流量进行预测。与标准回声状态网络、差分自回归滑动平均模型(ARIMA)、以及最小二乘支持向量机(LSSVM)预测模型进行了仿真对比,结果表明提出的方法具有更高的预测精确度以及更小的预测误差,同时一定程度上减少了预测时间。In order to improve the prediction accuracy of network traffic, a network traffic prediction method based on kernel principal component analysis proposed. Firstly, network traffic series was processed struction, then the effective information was extracted （KPCA） optimized echo state network （ESN） was to improve the predictability by phase space recon- through kernel principal component analysis. The reservoir parameters of echo state network were determined through the experiment method. Finally, net- work traffic was predicted through the echo state network. The proposed method is compared with stand- ard echo state netowrk, auto regressive integrated moving average （ARIMA）, and least squares support vector machine （LSSVM） predictive model. The simulation results show that the proposed method has higher prediction accuracy with smaller predictive error, at the same time the prediction time is reduced.

关 键 词：网络流量 预测 回声状态网络 核主成分分析 相空间重构 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]