面向报警根源分析的历史相似数据段查找

Searching historical similar data segments for alarm root-cause analysis

当工业现场发生报警时,操作员需要快速找到报警根源,从而采取正确的措施消除报警。为了更好地帮助操作员进行报警根源分析,提出一种多变量的历史相似数据段查找方法。首先,改进一种单变量相似数据段查找算法,消除其在查找相似数据段时存在的误差;然后,提出一种新的基于距离分布来确定时间窗口宽度的方法,用来定义被查找数据段的时长;最后,对多变量历史数据进行相似性查找,并对数据段的相似程度进行排序。通过TE(Tennessee Eastman)过程的仿真实验,验证了该方法相对于传统算法具有更好的有效性和快速性。

When an alarm occurs,operators need to find the root-causes quickly and initiate corrective actions to eliminate it.To help operators analyze the root-causes of an alarm,a method for searching multi-variate historical similar data segments was proposed in this paper.Firstly,an algorithm for searching univariate similar data segments was improved and the error in searching similar data segments was eliminated.Secondly,a new method based on the distance distribution for determining the time window width was proposed to define the length of the data segment to be searched.Finally,the similarity search was performed on the multi-variate historical data,and the similarity degrees of data segments were sorted.The simulation of TE(Tennessee Eastman)process has proved that the proposed method has better efficiency and faster than traditional algorithms.