Abstract: The forecast of network traffic is important to the security and availability of network. However, the traditional prediction methods based on uniform time weight lack generalizing ability, which results in the low prediction accuracy. The time weight of each history traffic data is calculated based on the time interval to the prediction point. The time weighted support vector regression model w-SVR is utilized to predict the network traffic. The prediction accuracy is improved attributing to the generalizing ability of w-SVR and the unique weight of each training data. The experimental results show that the prediction errors rate of w-SVR is decreased by 37.4% and 65.6% compared with ANN and AR model while its standard deviation is decreased by 46.2% and 53.3%.

Key words: network traffic, prediction, Support Vector Machine（SVM）, network security

摘要： 网络流量预测对于网络的安全和可用性至关重要，但是，传统的网络流量预测方法使用平均时间加权的方法进行预测，缺泛化能力导致预测精度低。基于每一个网络流量历史数据到预测点的时间间隔计算其时间权重，使用带时间权重的加权支持向量回归模型w-SVR预测网络流量。该模型因为其泛化能力和为每个训练数据设置单独的权重而提高了网络流量预测的准确性。模拟实验显示w-SVR模型相对于ANN和AR模型，预测错误率分别降低了37.4%和65.6%，而标准误差降低了46.2%和53.3%。

关键词: 网络流量, 预测, 支持向量机, 网络安全