基于分层概率图的动态网络在线异常检测方法
张岩庆;杨国正;田杰
【期刊名称】《计算机应用研究》 【年(卷),期】2016(033)010
【摘要】针对当前动态网络异常检测方法只能检测点异常、边异常和子图异常等局部异常事件的缺点，提出了基于分层概率图的在线异常检测方法。首先以窗口为单位将动态网络表征为树状图，然后运用概率分布的方法量化动态网络中连接关系的不确定性，最后结合统计假设检验的方法确定网络窗口是否发生异常。将该方法运用于AS级Internet的动态网络。实验结果表明，分层概率图模型不仅能够准确表征动态网络拓扑结构的演变，而且可以有效检测动态网络中的异常事件。%The existing anomaly detection methods on dynamic networks can only focus on a local region of the networks,such as the anomalies of nodes,edges or subgraphs,therefore,this paper introduced an online anomaly detection method based on hierarchical probability graph (HPG).This method firstly formalized each window of dynamic networks as a dendrogram,then quantified the uncertainty of relationships among nodes in the dynamic networks as probability distributions,and finally deter-mined whether or not a network window was anomalous by statistical hypothesis test.Applying this method to dynamic networks of AS level Internet,the experimental results show that HPG can not only represent the evolution of dynamic network topology accurately,but also detect anomalous event effectively.
基于分层概率图的动态网络在线异常检测方法
