面向网络故障和性能管理的告警聚类关联

对告警发生的时间特性进行了分析，给出了一种自动伸缩的关联窗口选择算法，保证了告警集合的完整性。采用非确定性网络实体关系模型构造概率编码本，引入虚拟伴生告警解决了孤立告警故障问题。最后提出了一种概率编码本矢量解码算法，利用预处理算法对故障空间进行简约，然后用矢量启发式算法寻找最可能的故障集合。算法仿真证明矢量解码算法与增量假定更新算法有相同的准确性，而矢量解码算法有更高的解码速度。

Alarm Clustering Correlation for Network Fault and Performance Management

After analyzing the temporal character of alarm, an auto-flexible correlation window selection algorithm is given to assure the integrity of alarm set. The non-deterministic network entity relationship models are used to construct probabilistic codebook, and the isolated alarm fault problem is solved by the idea of virtual concomitant alarm. Finally, a novel vector decode algorithm is proposed for probabilistic codebook. The fault space is predigested in terms of preprocessing algorithm, and the most possible fault set is obtained by heuristic vector algorithm. The simulation demonstrates that vector decode algorithm has the same accuracy as incremental hypothesis update algorithm and shows much higher decode speed.