Abstract: Forecasting of network traffic plays a significant role in many domains such as congestion control, network management and diagnose, and router design. In accordance with modern network, traditional Auto Regressive Moving Average（ARMA） model fails to describe the characteristic of network traffic very well. Therefore, Generalized Auto Regressive Conditional Heteroskedasticity（GARCH） model is studied for network traffic. The simulation shows that GARCH model is well fitted on the real data of network traffic. Meanwhile, the accuracy of the forecast based on the model is much better than that of ARMA model.

Key words: network traffic forecast, non-linear, Generalized Auto Regressive Conditional Heteroskedasticity（GARCH） model

摘要： 网络流量预测在拥塞控制、网络管理与诊断、路由器设计等领域都具有重要意义。根据当今网络流量的特点，传统的ARMA模型在描述网络流量数据特性时有一定的局限性，从而影响网络流量预测的精度。针对这个问题，研究了使用广义自回归条件异方差模型（GARCH）对网络流量数据进行建模的方法，通过仿真实验表明，该模型可以较好地描述网络流量数据的异方差性，同时其预测精度较之传统的ARMA模型的预测精度也得到了大幅提升。

关键词: 网络流量预测, 线性, 广义自回归条件异方差模型