Research On The Algorithm Of Power Communication Network Traffic Classification Based On SVM

This thesis is supported by the project called Research and Application of Key Technologies of Power Information and Communication Network Flow Prediction and Pipeline Intelligence, whose research direction is the service-oriented traffic identification and main research in the thesis is the study of network traffic classification technology based on support vector machine (SVM). The investigation and introduction of SVM algorithm is to improve classification performance and the sensing ability of Power Information and Communications Network.According to the characteristics of kernel function in the case of non-linear classification, kernel polarization algorithms are studied and a feature weighting algorithm based on kernel polarization named KPFW is proposed. With the analysis of multi-classification SVM algorithm, combined with the algorithm of KPFW, an algorithm of feature weighting which is applicable for multi-classification problems is put forward. With the consideration of error accumulation in traditional directed acyclic graph SVM (DAG-SVM), the policy called fuzzy decision policy is raised, and the policy is applied in DAG-SVM algorithm, which is called improved DAG-SVM algorithm. These two methods mentioned above are integrated to verify the actual effects of network traffic classification on the target data sets, which is collected by Moore and other researchers. The results of experiment indicate that the proposed approaches classify traffic types at a relatively high accuracy and a relatively good stability.The whole thesis can be divided into six parts and the main contents are presented blow respectively.In the first chapter, research background accompanied by research implications are introduced, basic principle of traffic classification is elaborated and the comparison between several common traffic classification technologies are presented. At the same time, the architecture of the thesis is described.In the second chapter, SVM theory is mainly presented and the drawbacks of SVM in the application in actual data sets are analyzed. Through the analysis above, the multi-classification SVM algorithms are introduced. In addition, the generation method of various multi-classification SVM and the drawbacks as well as the scope of application of each algorithm are summarized.In the third chapter, the application of kernel methods in SVM makes non-linear problems solved conveniently. After the analysis of characteristics of kernel polarization, a feature weighting algorithm based on kernel polarization named KPFW is proposed. Under the condition of multi-classification, which is different from binary classification problems, KPFW is applied in multi-classification circumstances. To verify the effectiveness, the test is conducted on UCI data sets.In the fourth chapter, traditional DAG-SVM algorithm is firstly introduced, and the improved DAG-SVM based on fuzzy decision policy is proposed, followed by verification about its performance on data sets in Matlab itself.In the fifth chapter, the two approaches (KPFWL and improved DAG-SVM) are combined. To verify actual performance of traffic classification system, the experiments are conducted on Moore data sets.In the last chapter, research work of the thesis is concluded and the direction of future research is pointed out.