Large Scale Particle Swarm Optimization For Internet Video Traffic Feature Selection

With the rapid development of the Internet,the transmission range of network video is becoming wider and wider.The various information contained in network video also potentially affects people's cognition and views.Therefore,it is particularly important to effectively identify Internet video traffic and control the transmission of unhealthy video.However,in the process of identifying Internet video traffic,regardless of the size of the data packets collected from video streams or other data at streaming levels,there are hundreds of thousands of features of the data.And these features contain a large number of irrelevant information and redundant information,which affect the effectiveness of the identification.But,the method of feature selection can deal with irrelevant information and redundant information effectively.Optimization algorithm is a very effective algorithm for feature selection.It can evaluate each dimension of feature of data,then remove irrelevant information and redundant information,and select the most effective information for identification and classification.But the general optimization algorithm can not deal with large-scale data and the optimization process is long.The main purpose of this paper is to study a large-scale particle swarm optimization algorithm with fast convergence speed for feature selection of network video stream data,so as to validate the effectiveness of the selected features by identifying the different types of video traffic data.The main research content is to collect different types of Internet video stream data,and to study a large-scale optimization algorithm for fast and effective selecting the features of video stream data,so as to realize the identification of different types of video.The main research points and innovations of this paper are embodied in the following three aspects:(1)The collection and initial feature extraction of different types of Internet video trafficBecause there is no open Internet traffic data set of different types of video in the field of video stream identification,the video traffic data collected by users themselves is used in this paper.In this paper,the original Internet traffic data packets of different types of video are collected,and the original data are pre-processed by filtering the irrelevant data packets.Finally,the statistical features of byte coding and packet arrival time are extracted from theraw stream data,and the feature data sets are obtained.(2)Study of large-scale optimization algorithm based on particle swarm optimizationFeature selection is a complex combinatorial optimization problem,and the size of feature dataset used in this paper is large.Therefore,a large-scale global optimization algorithm with fast convergence speed and good global search ability is proposed.In this paper,the principle of maximizing the fitness difference is proposed,and based on this principle,the particle ranking paired learning strategy is proposed.In order to further improve the performance of the algorithm,a method of combination of biased center learning strategy and sorting pairing learning strategy is proposed.Finally,the performance of the proposed algorithm is tested on two widely used benchmark test function sets,CEC2010 and CEC2013 function sets,and the performance of the sorting pairing learning strategy and biased center learning strategy are analyzed.(3)Study of feature selection method for video stream data based on Optimization AlgorithmBased on the large-scale global optimization algorithm proposed in this paper,a numerical classifier(Imbalance Data Gravitation based Classification Model,IDGC model)is selected to select the features and validate the effectiveness of the features of video stream data.In the process of feature selection,each feature of video stream data sample is given a weight;then,feature selection is carried out by using optimization algorithm to optimize these feature weights and using training data set to train model;finally,the validity of the effectiveness of feature selection is verified by video identification of test data set.