摘要： 定量刻画网络流量的长相关特性是网络特性研究的重要基础。对当前常用的Hurst指数估计算法进行了详细归纳。在此基础上，以已知Hurst指数的分形高斯噪声(fGn)序列为主要研究对象，利用逆向方法，分别研究了周期信号以及高斯白噪声影响下的Hurst指数估计算法的估计性能。通过比较，发现没有任何一种Hurst指数估计算法能够广泛应用于复杂条件下网络流量序列的Hurst指数的准确估计，其主要原因是因为这些算法的主要思想都是在全域内运用了求和平均的方法，使得流量序列的高可变信息受损，导致估计误差增大。

关键词: 因特网, 网络流量, 长相关, Hurst指数

Abstract: Quantify long-range dependence is a fundamental building block of Internet traffic characteristics. The existing and widely used Hurst estimate algorithms are comprehensively summarized. Based on this, taking the fractal Gaussian noise (fGn) series with a prior known Hurst exponent into account, the performance of these Hurst algorithms impacting on the periodical signal and the Gaussian white noise respectively using reverse method is evaluated. The results show that there is no algorithm that can estimate the Hurst exponent accurately and reliably under the complexity environment. The main reason is that these algorithms apply average method in global domain to a certain degree, which may result that the high variability information of the network traffic is destroyed. Consequently, the estimate error of the different algorithm becomes larger.

Key words: Internet, Network traffic, Long-range dependence (LRD), Hurst exponent