利用有限混合模型的高效网络拓扑推断算法  

Efficient topology inference algorithm using the finite mixture model

作　　者：张润生[1] 刘健[2] 李艳斌[1] 

机构地区：[1]中国电子科技集团公司第五十四研究所,河北石家庄050081 [2]中国空间技术研究院通信卫星事业部,北京100094

出　　处：《西安电子科技大学学报》2015年第4期133-139,共7页Journal of Xidian University

基　　金：国家科技支撑计划资助项目(2011BAH24B04);中国博士后科学基金面上资助项目(20110490989)

摘　　要：针对现有高效网络拓扑推断算法的性能对设定门限值非常敏感的问题,提出了利用有限混合模型的高效层析成像网络拓扑推断算法.该算法首先从叶子节点集合中任选一个节点,测量该节点与其余节点的相关性集合,然后对测量数据建立有限混合模型,对叶子节点进行粗略分类,推断出相应的内部节点;接下来对粗略分类结果中的每一个节点类重复上述过程,如此迭代直至推断出所有内部节点为止.试验结果表明,该算法可达到现有算法取最优门限时的拓扑推断精度,且该算法与现有的高效层析成像拓扑推断算法相比,只需测量更少的节点相关性数据.The performance of the existing efficient topology inference algorithm is highly sensitive to the threshold . To address the problem , a finite mixture model based topology inference algorithm is proposed . Firstly , a leaf node is selected from the original leaf‐node set , and then the similarities between the node and the other leaf nodes are measured , after which the original leaf‐node set is roughly divided into several subsets using the finite mixture model based on the measured similarities . The internal nodes corresponding to each subset could be inferred afterwards . Subsequently , the above procedures are applied for each subset obtained from rough division , and the process is iterated until all of the internal nodes are found . Analysis and simulation show that the proposed algorithm needs less correlation data than the existing algorithm , and performs almost as well as the existing algorithm with the optimum threshold .

关 键 词：拓扑推断 网络层析成像 有限混合模型 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]