Abstract: For distinguishing different kinds of data streams in the network, Autoregressive Conditional Duration（ACD） model is used to analyze the micro-characters of network flows in the time-domain, and the feasibility of the model is testified. One important advantage of ACD model is, without binning the unevenly sampled time series into evenly-spaced, it could obtain the features of network flows in the time-domain directly. Based on the statistical analysis of experimental data sets, the packets’ arriving time series in network flows can be fitted in with ACD model, furthermore ACD（2, 1） shows a good performance in modeling network flows of different applications.

Key words: autoregressive conditional duration, network flow, unevenly sampling, time series

摘要： 对网络中不同类型的数据流，应用自回归条件持续期模型（ACD），分析其中存在的时域微观特性，并研究ACD模型对网络数据流时序建模的适用性。使用ACD模型为具有随机到达过程的网络数据流时间序列建模，其优点是能够在不损失原始非等间隔时间序列特性的条件下，直接分析得到数据流的时域微观性质。在对实验数据集统计特性进行研究的基础上，得出数据包到达过程适用ACD模型的基本依据，采用ACD（2，1）模型对不同类型的网络数据流时间序列进行建模，结果表明其具有较好的拟合程度。

关键词: 自回归条件持续期, 网络数据流, 非等间隔采样, 时间序列