Online pattern matching and prediction of incoming alarm floods

Alarm floods are serious hazards for industrial process monitoring. In this paper, we propose an online algorithm to provide early prediction of an incoming alarm flood by matching an online alarm sequence with a pattern database and conducting similarity calculation. It overcomes three main challenges in online time-stamped pattern matching: the partial information (future events are unknown in the online sequence), the high computational efficiency requirement, and the segmentation of the online alarm sequence. New elements introduced in this algorithm include chattering and sequence window filters, modified time distance and similarity measurements, a modified gap penalty, and an incremental dynamic programming strategy. Potentially, the proposed algorithm could serve as the state identification stage in applications of predictive alarming, online alarm attribute modification, and online alarm suppression. A dataset from a real chemical plant is used to test both efficiency and accuracy of the proposed algorithm.