| With the rapid development of computer and Internet technology, in the information age today, people can not be separated from the network at all times, more and more diverse applications on the network, which is also filled with a lot of useless applications occupy a large amount of network bandwidth , For the rapid development of network application service identification technology. At present, the service identification technology based on deep packet inspection is relatively mature, but the complexity of communication protocols and more and more encryption services make the service identification technology based on deep data flow detection become a hotspot nowadays. Depth data stream detection technology at this stage is still in the research stage. In this paper, the theoretical analysis of various types of business characteristics, and the use of machine learning algorithms related to training and learning, and finally the feasibility of simulation modeling for validation.The main work of this paper is as follows:(1) According to the data flow characteristics of data available for analysis of data flow characteristics for the distribution of data flow characteristics, select the appropriate modeling data stream modeling characteristics. If the feature vector of the data stream can be clustered centrally, the K-mean algorithm is chosen for clustering, and the clustering process is optimized by using the rough fuzzy K-mean clustering algorithm according to the previous research results. Efficiency and accuracy. For the clustering invalid sample points, this paper carries on the secondary clustering to improve the accuracy of clustering, at the same time modeling and simulation to determine the accuracy of clustering has improved.(2) For the small sample business feature vector, which can not be clustered centrally, this paper proposes a support vector machine to cluster it. In the process of computing the most hyperplane, this paper uses genetic algorithm to calculate the most hyperplane, Through the simulation, it is proved that the genetic algorithm has the advantage over the pure mathematics.(3) For the other non-existent samples, the GMM algorithm and the neural network algorithm can be used to classify. In the process of clustering based on neural network, genetic algorithm can be used to optimize the neural network, To avoid falling into the local optimal, and achieved a more satisfactory results. |
