【摘要】 网络应用程序的多样化对网络流量分类提出了新的挑战。如何在变化的环境中准确地识别已知类和新类流量，然后实现模型在线更新，最后将新类纳入已知类范畴成为了研究的要点。针对这一问题，本文提出了一种基于极限学习机(Extreme Learning Machine, ELM)的在线学习模型，使用基于ELM算法的距离度量选择辅助训练样本，根据距离度量阈值进行新类检测，采用串联识别新类的二分类器的方式包含新的流量类别，当串联的分类器数量达到设定值时重新训练模型。在真实网络流数据集上的测试结果显示，本文方法已知类F1和开集总体准确率NA均能达到0.9以上。与代表性文献方法相比，在分类性能和时间性能方面均有更好的表现。更多还原

【Abstract】 The diversity of network applications poses new challenges to network traffic classification. How to accurately identify the known class and the new class traffic in the changing environment, then realize the online model update, finally include the new class into the known class category has become the key point of research. To solve this problem, an online learning model based on Extreme Learning Machine(ELM) algorithm is proposed in this paper. The distance measurement based on ELM algorithm is used to select auxiliary training samples, and the new class detection is carried out according to the distance measurement threshold. The binary classifier of the new class is identified in series to include the new traffic class, and the model is retrained when the number of series classifiers reaches the set value. The test results on real network flow data sets show that the known F1 scores and open set overall accuracy NA of the proposed method can reach above 0.9. Compared with representative literature method, it has better performance in the classification and time consumption.更多还原