基于改进指数平滑模型的拥塞网络时延预测研究

Delay Prediction Research for Congestion Network Based on Improved Exponential Smoothing Model

由于网络拥塞环境的时延序列非平稳,传统时延预测方法大多使用历史值加权建立预测模型,此类模型存在滞后偏差、参数固定、预测精度低等缺陷。针对上述问题,提出了基于改进指数平滑模型的拥塞网络时延预测方法,即SV-ES(Shifting Velocity-Exponential Smoothing method)。该方法在传统指数平滑模型基础上,根据序列趋势变化特征生成指数式加权的偏移速度因子,采用二次指数平滑法修正结果,生成预测模型。基于预测模型的残差序列,使用误差函数提取序列回归信息,动态调整衰减因子,实现模型参数自适应更新。实验结果表明,在拥塞网络环境中,SV-ES时延预测结果的精度比ES提升了9.01%,比ARIMA提升了1.28%。

This paper proposes a congestion network delay prediction method based on the improved exponential smoothing model,namely SV-ES(Shifting Velocity-Exponential Smoothing method).Based on the traditional exponential smoothing model,this method defines the trend characteristics of node migration velocity segment fitting,and uses the quadratic exponential smoothing method to modify the migration velocity factor to generate a prediction model.Based on the residual sequence of the prediction model,the error function is used to extract the sequence regression information,and the attenuation factor is dynamically adjusted to realize the adaptive update of the model parameters.The experimental results show that SV-ES can effectively predict the congestion network delay,and the prediction accuracy is improved compared with the prediction results based on ES model and ARIMA model.