网络非平稳流量多尺度时间序列预测数学建模  

Mathematical Modeling of Multi-Scale Time Series Prediction for Network Non-Stationary Traffic

作　　者：吴蕾[1] 曾慧平[1] 王海威[1] WU Lei;ZENG Hui-ping;WANG Hai-wei(Science and Technology College of NCHU,Gongqingcheng Jiangxi 332020,China)

机构地区：[1]南昌航空大学科技学院,江西共青城332020

出　　处：《计算机仿真》2021年第8期356-359,434,共5页Computer Simulation

基　　金：江西省教育厅科学技术研究项目(青年项目)(GJJ171520)。

摘　　要：网络流量具有时变性与非线性等特征,线性时间序列预测性能不佳,提出网络非平稳流量多尺度时间序列预测数学建模。利用离散低通滤波器确定流量分解系数,经过初始化处理,对滤波器做插零完成小波分解;使用支持向量机方法设置回归函数,确保函数最小化,并将低维空间中非线性回归问题转换为高维空间线性回归问题,在初始低维空间做核函数计算获取高维空间内积,引入双曲核函数建立支持向量机每一步的预测模型;重构小波分解后的时间序列,利用预测模型求解回归函数,即可实现对整体流量多尺度时间序列的预测。实验结果表明,上述方法提高预测精度,减少预测延时。The time-varying and nonlinear network traffic result in poor prediction performance of linear time series.Therefore,this paper proposes a multi-scale time series prediction model of network non-stationary traffic.A discrete low pass filter was used to determine the decomposition coefficient of the flow.After initialization,the filter was zeroed to complete the wavelet decomposition.Through the support vector machine,the regression function was set to ensure the minimization of the function.The nonlinear regression in low dimensional space was transformed into linear regression in high dimensional space.In the initial low dimensional space,the kernel function was computed to obtain the inner product of the high dimensional space.According to the hyperbolic kernel function,each step prediction model of SVM was established.After wavelet decomposition,the time series were reconstructed.The prediction model was applied to solve the regression function,achieving the multi-scale time series prediction of the overall flow.The experimental results show that this method can improve prediction accuracy and reduce prediction delay.

关 键 词：网络非平稳流量 多尺度时间序列 预测模型 小波分解 支持向量机 

分 类 号：TP318[自动化与计算机技术—计算机软件与理论]