基于BP神经网络聚类算法的P2P流量识别

摘要

Based on the supervised machine learning algorithm's BP neural network algorithm and the unsupervised machine learning algorithm' s k-means clustering algorithm,a semi-supervised BP neural network clustering algorithm was proposed to identify P2P traffic.The algorithm has advantages of supervised and unsupervised machine learning algorithms and can identify the traffic accurately,i.e.taking a small amount of offiine traffic samples for the marking and classification,and then taking the classification results as cluster centers to do a large number of online traffic clustering identification.This can improve the efficiency and ensure the accuracy of the results.The BP neural network was employed to have the collected small amount of traffic data in each stream classified according to standard deviation of the packet size,the transformation frequency,the average number of packets and the total byte number so that the characteristics mean of classification results can be obtained to guide the clustering of a large number of online data.The higher network testing accuracy proves the feasibility of this algorithm model.%在研究有监督机器学习算法中的BP神经网络算法和无监督的机器学习算法中的k-means聚类算法的基础上,提出一种半监督的BP神经网络聚类算法对P2P流量进行识别.该算法具有有监督和无监督的机器学习算法的优点,能快速地进行精确的流量识别,即取少量离线的流量样本进行标记与分类,然后利用分类结果为聚类中心对大量在线流量进行聚类识别.这样既提高了效率,又能保证结果的准确性.利用BP神经网络对所采集的少量流量数据中每个流按包大小标准差、变换频率、平均值、包数目和总字节数5个特征进行分类,得出分类结果的特征均值,对大量的在线数据进行指导聚类.多次实际网络测试结果的准确率很高,证明该算法模型是可行的.