LAUSR

Abstract Association rule mining is one of the most effective methods to reveal hidden operation patterns of building energy systems from massive amounts of building operation data. However, the mined association rules are always numerous, and most of them are worthless. It is very time-consuming for experts to extract valuable ones from the mined association rules. There is still a lack of effective post mining solutions for filtering out the association rules without physical meanings and the redundant association rules automatically. To fill in this gap, a post mining method is proposed in this study. A graph-based distance correlation indicator is proposed to remove the association rules without physical meanings. In addition, a rule generalization-based fusion approach is proposed to remove the redundant association rules. The method is evaluated using the one-year operation data of a chiller plant of a public building. A total of 149,588 raw double-variable association rules are obtained. Results show that the proposed post mining method can remove 99.72% of the raw association rules. Only 425 association rules are left finally to be further checked. Control strategies, anomalous operation patterns and device faults are revealed from the 425 association rules successfully.