Abstract: The network traffic has the features of relevance and nonstationarity. In order to make the traffic flow prediction model have the features of adaptability and correlation, this paper puts forward a new method by time point to create the model, and the time series combined with flow series. And it involves multivariate linear regression to evaluate arguments, uses the EWMA to process arguments and establishes the multivariate linear AR model. Experimental results show that compared with the AR model and ARMA model, the multivariate linear auto regression model is more accurate.

Key words: network flow, time series, prediction model, linear regression, AR model

摘要： 为使流量预测模型具有自适应性和相关性，以时间点为基础进行建模，结合时间序列与流量序列，引入多元线性自回归(AR)思想进行参数估算，对多次估算所得参数值建立指数加权移动平均数模型进行二次估算，在此基础上，建立多元线性自回归模型。实验结果证明，与AR模型、ARMA模型相比，基于多元线性AR模型的预测结果更准确。

关键词: 网络流量, 时间序列, 预测模型, 线性回归, 自回归模型

CLC Number: 

• TP393