A comprehensive investigation of knowledge discovered from historical operational data of a typical building energy system

Association rule mining has been widely utilized for extracting hidden operation patterns from huge volumes of building operational data. It usually generates a large number of association rules. However, most of them cannot reveal useful patterns. It is still unclear about how to develop effective methods to extract useful association rules. The major barrier is that characteristics of association rules mined from building operational data are still unknown. To reveal the characteristics, this study analyzes 101,787 association rules mined from the historical operational data of a typical chiller plant. The association rules are classified into useful ones and useless ones based on domain knowledge. Three insights are revealed successfully. Firstly, the useful association rules are related to component/sensor faults, control strategies, abnormal operation patterns, and normal operation patterns. And the useless association rules result from lacking physical meanings, insufficient information, and transient operation patterns. Secondly, the useful association rules only account for a very small proportion (4.64%) of all the association rules. Thirdly, three widely-used statistic indexes, i.e., support, confidence and lift, are ineffective in distinguishing between the useful and useless association rules actually in this field. Based on the three insights, five future research directions are proposed.

Automated summary

Association rule mining is widely used to extract hidden operation patterns from building operational data, but most of the generated rules are not useful. An analysis of over 100,000 association rules from a chiller plant reveals that useful rules are related to faults, control strategies, abnormal and normal operation patterns, but they only make up a small percentage. Common statistics indexes are ineffective in distinguishing between useful and useless association rules.