基于深度数据流检测的业务识别

发布时间：2018-06-07 18:33

本文选题：深度数据流检测 + 机器学习　；参考：《北京邮电大学》2017年硕士论文

【摘要】：随着计算机和互联网技术发展日新月异,在信息时代的今天,人们每时每刻都离不开网络,网络上的应用越来越多样化,其中也充斥着很多无用的应用占据了大量的网络带宽,针对网络应用业务识别技术飞速发展。目前,基于深度数据包检测的业务识别技术相对来说较为成熟,但是通信协议的复杂化和越来越多的加密业务,使得基于深度数据流检测的业务识别技术成为现在发展的一个热点。基于深度数据流检测技术现阶段主要还处于研究阶段。本文针对多种类型的业务特征进行理论分析,并利用机器学习算法进行相关的训练学习,最后仿真建模进行验证可行性。本文的主要工作如下:(1)根据数据流量分析流量可能存在的可利用数据特征,针对数据流特征的分布规律,选取适合建模的数据流特征进行建模。若数据流特征向量可进行中心聚类,则选取K-mean算法进行聚类,并根据前人的研究结果,利用粗糙模糊K-mean聚类算法对聚类过程进行优化,以提高聚类的效率和准确率。对于聚类无效的样本点,本文进行二次聚类以提高聚类的准确率,同时进行建模仿真,以确定聚类的准确率有所提高。(2)针对无法中心聚类的小样本业务特征向量,本文拟采用支持向量机对其进行聚类研究,在计算最有超平面过程中,本文拟采用遗传算法计算求得最有超平面,通过仿真,证实遗传算法相对于纯粹的数学计算存在着特有的优势。(3)对于其他不存在明显分布的业务特征样本,本文从理论上分析,可采用GMM算法和神经网络算法进行分类,在基于神经网络的聚类过程中,可使用遗传算法来优化神经网络,避免陷入局部最优,取得了较理想了效果。
[Abstract]:With the rapid development of computer and Internet technology, in the information age today, people can not do without the network every moment, the network applications are more and more diversified, and many useless applications occupy a large amount of network bandwidth. Aiming at the rapid development of network application service identification technology. At present, the technology of service identification based on deep packet detection is relatively mature, but the complexity of communication protocols and more encryption services, The technology of service identification based on deep data stream detection has become a hot spot. The detection technology based on deep data flow is still in the research stage. In this paper, various types of business characteristics are analyzed theoretically, and the machine learning algorithm is used for training learning. Finally, simulation modeling is used to verify the feasibility. The main work of this paper is as follows: (1) according to the data flow analysis of the available data characteristics of traffic, according to the distribution of data flow characteristics, the suitable data flow characteristics are selected to model. If the feature vectors of data flow can be clustered in the center, K-mean algorithm is selected to cluster. According to the previous research results, rough fuzzy K-mean clustering algorithm is used to optimize the clustering process in order to improve the efficiency and accuracy of clustering. For the invalid sample points, this paper carries out quadratic clustering to improve the clustering accuracy, and modeling and simulation to determine that the clustering accuracy has been improved. 2) for the small sample business feature vector which can not be clustered centrally, In this paper, support vector machine (SVM) is used to cluster it. In the process of calculating the most hyperplane, the genetic algorithm is used to calculate the most hyperplane. It is proved that genetic algorithm has a special advantage over pure mathematical computation. For other business feature samples that do not have obvious distribution, this paper theoretically analyses that GMM algorithm and neural network algorithm can be used to classify. In the clustering process based on neural network, genetic algorithm can be used to optimize the neural network to avoid falling into local optimum, and the results are satisfactory.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP393.0

【相似文献】