摘要
在电力通信告警相关性分析中,往往需要将原始告警数据库转化为事务型数据库,针对传统均匀滑动时间窗口提取告警事务效率低的问题,文中提出一种基于近邻传播聚类的滑动时间窗口法,通过将近邻传播聚类算法应用于告警数据预处理过程中,将原始告警数据转化成能够用于滑动时间窗口法处理的若干时间段。实验证明,基于近邻传播聚类的滑动时间窗口法具有更高的事务提取效率,解决了固定时间段划分方法的不合理性,有利于告警相关性分析,在告警事务提取中具有更广泛的应用前景。
In power communication alarm correlationanalysis, the original alarm database needs to be transformed into a transactional database. Traditional sliding time window has low efficiency on extracting alarm transactions, so sliding time window method based on propagation clustering is proposed. With this method, affinity propagation clustering is applied to alarm pretreatment, the alarm sequence would be considered to a series of data objects which is turned into some segments with higher similarity. The performance testing of the algorithm indicates that this method has higher extract efficiency, which solves the irrationality of fixed time window. The method is helpful for alarm correlation analysis and has a wider application prospect for extracting alarm transactions.
出处
《电子设计工程》
2016年第16期142-145,共4页
Electronic Design Engineering
基金
北京市自然科学基金项目(4142049)
关键词
告警事务
关联分析
近邻传播聚类
时间窗口
滑动步长
alarm transactions
correlation analysis
affinity propagation clustering
time window
sliding step