Pattern Mining of Alarm Flood Sequences Using an Improved PrefixSpan Algorithm with Tolerance to Short-Term Order Ambiguity
Qunxiong Zhu, Chengyan Jin, Yan-Lin He, Yuan Xu
Abstract
The alarm system monitors industrial plants in real-time to ensure safe operation. The scale of modern plants is expanding rapidly, processes are becoming increasingly complicated, and the cost of alarm configuration in modern control systems is decreasing. However, alarm systems suffer from low performance. A large number of alarms are often indicated to operators within a short period, known as alarm floods. Analysis and mining of similar patterns among different alarm floods is an efficient approach. Alarm flood analysis facilitates cause analysis of historical flood data, thus locating poorly configured alarms and predicting incoming alarm floods. The strongly correlated alarms almost appear simultaneously, and the order is uncertain. A pattern containing strongly correlated alarms cannot be directly extracted according to the order of occurrence. In this paper, an improved PrefixSpan algorithm is proposed to identify similar patterns with a tolerance to short-term order ambiguity, which treats alarm floods as time-stamped sequences. The effectiveness of the proposed algorithm is verified with Tennessee Eastman process simulations.