网络管理告警数据库中时序规则挖掘的一种新方法

提出一种基于“最小发生的双时间窗口约束”时序规则挖掘新方法。该方法依据“双时间窗口”约束和“最小发生”判据,可判别在一个时间窗内的哪些告警事件导致了另一个时间窗内告警集合事件的产生,快速寻找出不同网络设备告警与其它网络设备告警之间的关联知识。通过对采集某省级IP网的告警数据库进行规则挖掘实验,表明该方法可以准确、快速地挖掘出海量网络告警数据库中大量有意义的时序规则,这些规则可作为选验知识来指导网络智能化故障定位、诊断及预测。

A NEW METHOD FOR MINING TIME-SERIES RULES IN NETWORK MANAGEMENT ALARM DATABASES

A new method "double time window algorithm using minimum occurrence" for mining time series rules in network management alarm databases is provided. Based on the restriction of "double time window" and "minimum occurrence", this method could decide which alarms in a time window make up other alarms in another time window, to get the complex time series correlation knowledge among the different network devices and alarms. The experiments to the large amount of alarm databases of one province IP network show that many interesting time series rules could be acquired efficiently. Furthermore, this kind of knowledge could be used to guide the intelligent network fault localization, diagnosis and forecasting.