基于多层感知器的流量分类方法研究  

Research on refined classification method based on multilayer perceptron

作　　者：郭丽[1] 刘磊[2] Guo Li;Liu Lei(Software College,Zhongyuan University of Technology,Zhengzhou 450007,China;School of Intelligent Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450015,China)

机构地区：[1]中原工学院软件学院,郑州450007 [2]郑州航空工业管理学院智能工程学院,郑州450015

出　　处：《电子测量与仪器学报》2019年第7期56-64,共9页Journal of Electronic Measurement and Instrumentation

基　　金：河南省科技攻关项目(172102210526,172102210593,182102210523);河南省教育厅科学研究重点项目(16A520099,17B520042);郑州航空工业管理学院青年基金(201103002)资助项目

摘　　要：为了解决现有流量分类方法识别准确率低、复杂度高、高速流量处理能力弱、对既定特征依赖程度高等问题,提出了一种新的流量分类方法。利用深度学习的思想识别复杂网络应用、感知应用内部服务。在多层感知器模型基础上,基于无监督学习算法训练多层感知器参数,基于有监督学习算法回溯调整,经过不断迭代优化分类器性能。为了获得更好的分类效果,引入了基于载荷熵值以及字节值的分类特征。实验结果表明,基于多层感知器的流量分类方法在复杂应用识别上具有较高的识别准确率,在应用服务感知上具有很高的精确率、更少的训练和测试时间。相比于其他机器学习算法,基于多层感知器的流量分类方法能够更加准确地识别复杂网络应用,更高效地感知应用内部服务。In order to solve the problem that the existing traffic classification method has low recognition accuracy,high complexity,weak high-speed traffic processing capability and high dependence on established features,a new traffic classification method is proposed using deep learning ideas to identify complex network applications and perceive application internal services.Based on the multi-layer perceptron model,the multi-layered perceptron parameters are trained with the unsupervised learning algorithm,and the classifier performance is optimized with the iterative learning algorithm.In order to obtain a better classification effect,classification features based on load entropy values and byte values are introduced.The experimental results show that the multi-layer perceptron-based traffic classification method has higher recognition accuracy in complex application recognition,and has higher accuracy and less training and testing time in application service perception.Compared with other machine learning algorithms,the multi-layer perceptron-based traffic classification method can better identify complex network applications and perceive application internal services.

关 键 词：多层感知器 流量分类 应用识别 服务感知 

分 类 号：TP393.06[自动化与计算机技术—计算机应用技术]