基于合并分层聚类的网络拓扑推断算法

摘要

针对HTE（Hierarchical Topology Estimation ）算法计算复杂度较高的问题及其在节点相关性估计方差较大条件下性能下降的问题，提出基于合并分层聚类的网络拓扑推断算法。该算法采用自底向上的合并分层聚类，每次聚类仅使用与最大相关节点对有关的数据，相对HTE算法降低了运算复杂度；建立了改进的有限混合模型，增加了参数推断的有效数据，提高了算法的参数估计精度。仿真结果表明相对HTE算法，本文方法可以更快地推断出网络拓扑，且在节点相关性估计方差较大条件下，有更高的拓扑推断精度。%Considering the high complexity of HTE (Hierarchical Topology Estimation )and its performance degradation under the condition of large correlation estimation variance ,a method based on agglomerative hierarchical lustering is proposed .The method employs bottom-up agglomerative hierarchical clustering ,which only uses the data related to the node pair with the largest correlation ,so it has lower computation complexity than HTE .A modified finite mixture model is established ,increasing the amount of effective data ,which improves the accuracy of parameter estimation .The simulation demonstrates that the proposed method infers the topology more rapidly ,with higher accuracy when the correlation estimation variance is large .