Research On The Classification Algorithm Of Electric Power Network Traffic Identification Based On Cluster Analysis

This thesis is supported by the science project of State Grid Corporation of China called Research and Application of Multi-Dimensional Sensing Technology over Smart Pipe for Information and Communications Network, whose aim of research is about the service-oriented traffic classification and sensing. The main topic of thesis is investigating the algorithm of network traffic classification based on EM algorithm, which can improve the performance of classification.According to the fact that the EM algorithm is sensitive to the initial value, the thesis put forward to a new improved EM algorithm based on the Z matrix and Tsallis entropy. After analyzing the advantage and disadvantage of the common component model and separate mixture model, we can combine both of them to create a restrict matrix Z. Based on above, the new algorithm taking advantage of the matrix Z and the q-DAEM which utilizes the Tsallis entropy is generated. Then the new algorithm is programmed to verify the actual effect of traffic classification. The eventual result in the Moore set shows that the new algorithm achieves better performance than the EM and DAEM algorithm with only a little amount of calculation.According to the fact that the Gaussian mixture model generated by EM is sensitive to the outlying values, this thesis presents a new spatial EM algorithm based on the median-based location and rank-based scatter estimator. Considering that the mean and covariance of the mixture elliptical distribution have a low breakpoint, which can be easily affected by the outlying values, we replace sample mean and covariance to the spatial median and rank covariance matrix. The eventual result in Moore set shows that the spatial EM algorithm can identify the network traffic of ATTACK whose rate of identification is relatively low in other algorithms.The whole thesis can be divided into six parts and its main content is as follows:In the first chapter, research background along with its implication is introduces, basic principle of traffic identification is elaborated and several common techniques are analyzed as well, in which the machine learning are specially listed. At the same time, the architecture of this thesis is described.In the second chapter, the EM algorithm is mainly presented and the drawbacks of the EM are pointed out, followed by some existed derived EM algorithm.In the third chapter, the improved EM algorithm based on the restrict matrix Z and Tsallis entropy is put forward, followed by verifications about its performance on the Moore set.In the fourth chapter, the spatial EM algorithm based on the median-based location and rank-based scatter estimator is put forward, followed by verification about its robustness and tolerance of outlying value based on the network flows of ATTACK on the Moore set.In the last chapter, research work of this thesis is concluded, the inadequate aspects are analyzed and future direction is also pointed out.