基于机器学习的流量识别和路由控制系统的设计与实现

发布时间：2018-06-19 23:34

本文选题：流量识别 + 路由控制　；参考：《北京邮电大学》2014年硕士论文

【摘要】：随着Internet的高速发展,互联网业务从以单纯的数据业务为主逐步发展为语音、视频、数据等业务,涵盖了搜索、即时通信、网购、金融、游戏等领域。互联网极大地方便和丰富了人们的生活、学习、工作,同时也为网络运营商带来了极大的挑战。当前互联网在业务的QoS支持上,需要解决两个关键技术是：(1)如何识别不同的业务类型；(2)如何针对不同的业务类型,进行不同的控制。这两个关键技术可归纳为流量识别和流量的路由控制。本文围绕着流量识别和流量的路由控制这两个关键技术,进行深入的研究。本文在研究过程中主要做了如下工作：(1)研究和比较当前常用的流量识别技术,包括基于端口的流量识别技术、基于深度包检测的流量识别技术、基于行为模式的流量识别技术和基于机器学习的流量识别技术,其中对性能最优的基于有监督机器学习的流量识别技术进行了更进一步的研究和分析；(2)基于SVM算法和Adaboost算法提出了一个实时流量识别算法；(3)基于TCP/IP协议栈,设计数据平面和控制平面相分离的路由架构；(4)结合本文提出的实时流量识别算法和路由架构,实现具有流量识别功能的路由控制系统,并完成系统的测试。本文在提出基于SVM和Adaboost的实时流量识别算法时,进行了实验对比。实验结果表明,这种方法在实时流量识别方面的准确率比原始的SVM算法和基于决策树的Adaboost算法都高,具有可行性。结合这个实时流量识别算法和所设计的路由架构,本文实现了路由控制系统,并搭建了测试环境完成了场景测试,证明了系统的可行性。
[Abstract]:With the rapid development of Internet, Internet business has gradually developed from simple data services to voice, video, data and other services, covering search, instant messaging, online shopping, finance, games and other fields. The Internet greatly facilitates and enriches people's life, study and work, but also brings great challenges to network operators. In the current QoS support of Internet services, two key technologies need to be solved: 1) how to identify different types of services / 2) how to control different types of services. These two key technologies can be summarized as flow identification and traffic routing control. This paper focuses on the two key technologies of traffic identification and traffic routing control. In this paper, we mainly do the following work: 1) Research and compare the current commonly used traffic identification technology, including port based traffic identification technology, depth packet detection based traffic identification technology, Behavioral pattern based traffic identification technology and machine learning based traffic identification technology are further studied and analyzed, in which the best performance based on supervised machine learning traffic identification technology is further studied and analyzed. (2) based on SVM algorithm and Adaboost algorithm, a real-time traffic identification algorithm is proposed. Based on TCP / IP protocol stack, a routing architecture with data plane and control plane is designed. The routing control system with the function of flow identification is implemented, and the test of the system is completed. In this paper, a real-time traffic recognition algorithm based on SVM and Adaboost is proposed. Experimental results show that the accuracy of this method in real-time traffic recognition is higher than the original SVM algorithm and Adaboost algorithm based on decision tree. Combined with the real-time traffic identification algorithm and the designed routing architecture, this paper implements the routing control system, and builds a test environment to complete the scene test, which proves the feasibility of the system.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06;TP181

【参考文献】