基于SVM的P2P流量早期识别研究

发布时间：2018-04-03 09:01

本文选题：P2P流量识别　切入点：早期流量　出处：《河北大学》2017年硕士论文

【摘要】：P2P应用在提高服务质量的同时,也为网络流量的管理带来了新的挑战。出于流量管理的需要,对P2P流量识别的研究越来越受到人们的关注。网络应用的早期流量,蕴含了丰富的识别信息。对网络应用早期流量识别的研究对流量识别准确度和实时性的提高意义重大。SVM(支持向量机)是一种优秀的机器学习算法,具有良好的学习和泛化能力,非常适合流量识别领域的应用。在本文中基于SVM对P2P流量早期识别进行了相关研究。本文的研究内容包括:1.对P2P应用开展研究。通过对相关理论的灵活应用,提出了一个基于SVM对P2P流量早期识别的框架。并通过开展实验,对常见的几种P2P应用流量进行识别,验证了该框架可行性和有效性。2.对框架中网络数据预处理过程进行设计,通过采集,数据净化,网络流还原,特征统计等步骤,实现了数据由捕获数据包到学习样本的转换。并对学习机泛化识别模型的过程进行了优化。3.对P2P的早期流量特征进行研究,就P2P早期流量的多连接性,动态性及双向性特征对特征选择的指导意义进行了论证。对基于数据包,网络流和数据包、网络流混合特征三种特征选择策略进行实验对比。验证了基于数据包、网络流混合特征的P2P早期流量识别,在准确度、实时性及稳定性的综合性能优于基于数据包和基于网络流的特征选择策略。更适合实际问题的解决。
[Abstract]:P2P applications not only improve the quality of service, but also bring new challenges to network traffic management.The early traffic of network application contains abundant identification information.The study of early traffic identification in network application is of great significance to improve the accuracy and real-time of traffic identification. SVM (support Vector Machine) is an excellent machine learning algorithm with good learning and generalization ability.Very suitable for traffic identification applications.In this paper, the early identification of P2P traffic based on SVM is studied.The research contents of this paper include: 1.Research on P2P applications.A framework for early identification of P2P traffic based on SVM is proposed through the flexible application of relevant theories.The feasibility and effectiveness of this framework are verified by experiments to identify the traffic of several P2P applications.The preprocessing process of network data in the framework is designed. Through the steps of data acquisition, data purification, network flow restoration and feature statistics, the data conversion from capture data packet to learning sample is realized.The process of generalized recognition model of learning machine is optimized. 3. 3.The characteristics of early P2P traffic are studied, and the guiding significance of multi-connectivity, dynamic and bidirectional characteristics of early P2P traffic to feature selection is demonstrated.Three feature selection strategies based on data packet, network flow, data packet and network flow are compared experimentally.It is verified that the performance of P2P early traffic identification based on data packet and network flow hybrid features is superior to that of feature selection strategy based on data packet and network flow in accuracy, real-time and stability.More suitable for the solution of practical problems.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.06;TP181

【相似文献】