基于贝叶斯网络的报警系统管理方法

Bayesian network based method to improve alarm system performance for process industries

本文提出了一种基于贝叶斯网络的报警系统管理方法,通过不断采集报警系统产生的数据,深入学习分析报警数据的特性,从而建立监控变量之间定量关系的贝叶斯网络模型,具体包括:i)从报警数据学习监控变量定性的相关关系,建立监控变量相关关系的有向无环图网络,ii)学习监控变量的相关参数及相互影响参数,iii)计算相关监控变量之间的连接强度。基于该贝叶斯网络模型,通过推理分析,报警系统能够快速定位引起报警的根本原因并采取措施,减少报警泛洪,进而不断调优报警网络性能,为获得更好的报警管理效果提供依据。

Alarm flooding is a significant problem in chemical process industries. In this paper, a Bayesian network based method is proposed to improve the alarm system performance. The proposed method comprises of four features： i） Bayesian network based model is learned directly from the DCS alarm record data, ii） variables correlations are represented as a directed acyclic graph; iii） all possible states of monitored variables are accounted carefully; and iv） quantitative strength among variables are defined. These features will benefit to root causes analysis, reduce redundant and false alarms, tune monitored variables thresholds, and finally improve the alarm system performance. Furthermore, in emergence situation, the DCS alarm record data can be rapidly analyzed and find the root cause, so as to identify the fault just in time and take desired actions to eliminate the unwanted deviations propagation and prevent the accident. The proposed method is explained with a simple tank case study step by step. An industrial DCS alarm system is studied to demonstrate the effectiveness of the proposed method.