CApriori: A Parallel Association Rule Algorithm for Time Series Data

China ADS front-end demo linac (CAFe) uses a distributed control system based on experimental physics and industrial control system (EPICS), and the EPICS software system Alarms is used to monitor the alarm states of process variable(PV) in real time. Alarms stores a large amount of alarm data of time series representing alarm events, and the cause of failure can be determined by analyzing the correlation between alarm events. The traditional association rule algorithm is limited by the minimum support and can only get the association rules among frequent alarm events. Therefore, this paper proposes a parallel association rules algorithm, called CApriori, based on Spark, the big data computing engine, for processing the large amount of time series alarm data to find the association rules between low-support alarm events. In the second stage of the CApriori, distance correlation is introduced to remove candidate sets that of high frequency but low correlation. The proposed algorithm is applied to the data generated by the CAFe alarm system, and the results show that CApriori can find the association rules between the alarm events with high correlation and low support, which provides a basis for the intelligent fault diagnosis of the accelerator.