基于EM算法对报警时间序列的分析预测

Analysis and prediction of alarm time series based on EM algorithm

报警之间有些是孤立的,有些相互之间存在着某种关联。尽管报警类型存在多样性,但其系统基本都是由幂律分配。报警间相关性的研究是报警预测和防范的重要手段。文章关注报警之间的事件时间间隔。通过使用一系列数学统计方法(如马尔科夫链、幂律指数函数等),对报警时间间隔进行分析,并对报警类型进行分类预测。区分出多个报警类型之间的关系,发现事件间时间分布与地震的分布相似。如果其不同于指数分布,则间接地证明了报警之间的非平凡时空相关性。结果表明,事件间时间统计对于描述报警的动力学至关重要。

Some alarms are isolated from each other,and some are related to each other.Despite the diversity of alarm types,the system is basically distributed by power-law.The study of correlation between alarm is an important means of alarm prediction and prevention.This article focuses on the event time interval between alarms.By using a series of mathematical statistical methods such as Markov chain,power-law function,the alarm time interval is analyzed,and the alarm type is classified and predicted.It is found that the time distribution between events is similar to the distribution of earthquakes.If it is different from the exponential distribution,it indirectly proves the non-trivial space-time correlation between alarms.Results show that the time statistics between events is crucial to describe the dynamics of alarms.