摘要：

当前流量预测模型难以准确刻画互联网流量的多重特性,并且存在构建时间长、预测精度低的问题。为此,设计基于提升小波分解的网络流量混合预测模型(WLGC)。该模型利用提升小波将流量时间序列快速分解为分别具有低频和高频特性的近似时间序列和细节时间序列,近似时间序列利用最小二乘支持向量机(LSSVM)预测并通过广义回归神经网络(GRNN)进行误差校准,细节时间序列在半软阈值降噪后利用自适应混沌预测方法对其预测,最后使用提升小波重构得到时间序列的预测值。仿真实验结果表明,该模型可有效提高预测精度。

关键词: 流量预测, 提升小波, 最小二乘支持向量机, 广义回归神经网络, 阈值降噪, 混沌预测

Abstract:

Current traffic prediction models can not accurately depict the multi-properties of network traffic.Apart from this,model construction is time-consuming and prediction accuracy is low.To address the problem,a lifting-wavelet-based hybrid prediction model for network traffic called WLGC is proposed.In WLGC model,a lifting wavelet is adopted to quickly decompose traffic time series into low-frequency approximate time series and high-frequency detailed time series.Least Squares Support Vector Machine(LSSVM) is leveraged to predict the approximate time series and General Regression Neural Network(GRNN) is leveraged to calibrate the prediction error.The adaptive chaotic prediction method is used to predict the detailed time series after the semi-soft threshold denoising.Finally,the inverse lifting wavelet transform is performed to get the predicted values of the original time series.Simulation results verify the validity of the proposed method and the prediction accuracy is increased compared with current prediction methods.

Key words: traffic prediction, lifting wavelet, Least Squares Support Vector Machine(LSSVM), General Regression Neural Network(GRNN), threshold denoising, chaotic prediction

中图分类号: 

• TP18