Mining Sequential Alarm Pattern Based on the Incremental Causality PrefixSpan Algorithm

Alarm flood is the main reason of alarm system failure in the modern industry; therefore, it is an important issue to suppress the sequential alarm caused by anomalous propagation in an alarm flood. First, the alarm event is defined with time stamp, and the causal alarm pattern is defined as a closed frequent alarm pattern with a time constraint. An incremental causality PrefixSpan algorithm is proposed to address two objectives simultaneously: 1) effectively mine the causal alarm patterns from the known alarm data and 2) dynamically update them without rescanning the entire database when a new alarm flood occurs. The proposed algorithm is, therefore, particularly suitable for real-time operation. The experiments for the benchmark synthetic dataset and the Tennessee Eastman process are finished to verify the proposed method.