基于模糊聚类的网络故障预报

Network Fault Predicting Based on Fuzzy Clustering

文章针对网管告警数据库中时间序列存在的连续性、不确定性和模糊性问题,提出了一种基于模糊聚类的时间序列规则挖掘新方法。该方法引入模糊聚类理论,可预测出一些告警(联合)事件的发生将导致哪些告警(联合)事件的随后产生。通过对某校园网的告警数据库进行规则挖掘实验,表明该方法可以准确、有效地挖掘出隐含在海量网管告警数据库中大量有意义的时序规则,规则中的概率参数(置信度和支持度)可作为网络管理的先验知识用来指导网络故障的诊断和预报。

For the problems of continuity,uncertainty and fuzziness in the time-series of the network management alarm database,this paper puts forward a new mining time-series method based on fuzzy clustering.The method applies the fuzzy clustering theory to predict which alarms cause other alarms.The experiment based on a campus network alarm database shows that many significant time-series rules could be acquired accurately and efficiently from a large amount of network management alarm database,and the probability parameter,confidence and support in this paper,in those rules could be used to guide the fault diagnosis and forecasting of the intelligent network.