基于时域划分的网络流量预测方法  

Network traffic prediction approach based on time-domain division

作　　者：夏明山[1,2] 王丽 XIA Mingshan;WANG Li(Dongguan Research Departmnet,Institute of High Energy Physics,Chinese Academy of Sciences,Beijing 100049,China;Neutron Science Department,Spallation Neutron Source Science Center,Dongguan Guangdong 523803,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院高能物理研究所东莞研究部,北京100049 [2]散裂中子源科学中心中子科学部,广东东莞523803 [3]中国科学院大学,北京100049

出　　处：《计算机应用》2024年第S01期183-187,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(11905239)。

摘　　要：网络流量的预测一般通过建立相应的分析模型分析时序序列的发展过程及趋势,缺乏对网络流量波动空间和周期特征的分析,无法对短时突变流量及短周期流量进行精准预测。为了提高网络流量的预测效果,提出一种基于时域特征分析的网络流量预测方法。该方法通过分析网络流量的周期特征,建立时域划分模型,使具有相同趋势及波动空间的网络流量重组,突出短时突变流量和周期趋势特征,增强数据规律,以提高网络流量预测精度。选取反向传播(BP)神经网络、长短期记忆(LSTM)神经网络及小波神经网络(WNN)模型,采用均方误差(MSE)作为衡量标准,分别验证时域划分模式和全时域模式网络流量预测效果。结果表明,时域划分模式时BP神经网络和WNN的MSE相比全时域模式更低,其中时域划分模式BP神经网络的MSE降低为全时域模式时的1/24,说明通过分析网络流量数据建立的时域划分模型能够提高网络流量预测性能,为大规模互联网环境下的网络流量预测分析提供一种分而治之的方法。The prediction of network traffic is generally based on the analysis of the development process and trend of time series by establishing corresponding analytical model,and lacks the analysis of the fluctuation spatial and periodic characteristics of network traffic,which can not achieve accurate prediction of short-term abrupt traffic and short-term periodic traffic.In order to improve the prediction effect of network traffic,a network traffic prediction method based on time domain feature analysis was proposed.By analyzing the periodic characteristics of network traffic,a time-domain division model was established to recognize the network traffic with the same trend and fluctuation space,the characteristics of short-term abrupt traffic and periodic trend were highlighted,and the data rules were enhanced to improve the prediction accuracy of network traffic.Back Propagation(BP)neural network,Long Short-Term Memory(LSTM)neural network and Wavelet Neural Network(WNN)models were selected and the Mean-Square Error(MSE)was used as evaluation to verify the network traffic prediction effects of both time-domain division mode and full time-domain mode respectively.Experimenatl esults show that the MSEs of BP neural network and WNN under the time-domain division mode are lower than those under the full time-domain mode,in which the MSE of BP neural network under the time-domain division mode is reduced to 1/24 of that under the full time-domain mode,which indicates that the time-domain division model established by analyzing the network traffic data can improve the network traffic prediction performance,and provides a divide and conquer method for network traffic prediction and analysis under large-scale Internet environment.

关 键 词：网络流量预测 特征分析 流量重组 时域划分 神经网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]