CAJ | 학술논문

针对现有告警信息相关性分析方法没有客观全面考虑各告警的重要程度，无法体现告警之间个体差异性等问题，该文提出一种基于小波神经网络的加权关联规则告警挖掘算法。综合告警级别、告警类型以及告警设备类型3个主要告警属性，将其作为小波神经网络的输入，通过对历史样本数据的学习确定连接权值，合理地评估各个告警属性重要程度，利用所得权值向量进一步挖掘告警加权关联规则。结果表明所提算法在权值确定时能够综合考虑告警信息的多个属性及历史经验，得到的权值更能合理地反映告警重要度，所得关联规则能够更加准确地反映告警之间的相关性。
침대현유고경신식상관성분석방법몰유객관전면고필각고경적중요정도，무법체현고경지간개체차이성등문제，해문제출일충기우소파신경망락적가권관련규칙고경알굴산법。종합고경급별、고경류형이급고경설비류형3개주요고경속성，장기작위소파신경망락적수입，통과대역사양본수거적학습학정련접권치，합리지평고각개고경속성중요정도，이용소득권치향량진일보알굴고경가권관련규칙。결과표명소제산법재권치학정시능구종합고필고경신식적다개속성급역사경험，득도적권치경능합리지반영고경중요도，소득관련규칙능구경가준학지반영고경지간적상관성。
The existing alarm information correlation analysis methods take no consideration of the alarm significance, so they are unable to reflect individual differences between the alarms. In order to solve the problem, a new alarm information relevance mining mechanism based on wavelet neural network is proposed in this paper. The three key attributes of alarm information, alarm level, alarm type and alarm equipment type are considered as the inputs of the wavelet neural network respectively. Further, the weight corresponds to the importance of individual attribute, which can be determined reasonably by training with history sample. Finally, the association rules can be mined accurately. Results show that the proposed algorithm can consider multiple influence factors and history sample comprehensively, and the obtained weight can scientifically reflect the importance of the alarms;moreover, the association rules can reflect the correlation between the alarms more accurately.