Cluster analysis-based anomaly detection in building automation systems

Abstract Faults in heating, ventilation, and air-conditioning control networks substantially affect energy and comfort performance in commercial buildings. As these control networks are comprised of many sensors and actuators, it is challenging to identify, often subtle, anomalies caused by these faults. In this paper, we develop a cluster analysis method for anomaly detection. The proposed method consolidates the building automation system data into a small number of distinct patterns of operation. These distinct patterns help energy managers discover and interpret anomalies through visualization of these patterns. The method was demonstrated with a year’s worth of building automation system data from 247 thermal zones and an air handling unit. Anomalies associated with zone temperature and airflow control were identified in about one-third of these zones. At the air handling unit-level, we identified anomalies related with three different faults: the use of economizer mode with perimeter heating, and leaky outdoor and return air dampers. The use of economizer mode with perimeter heating affected 39% to 52% of the total operation period and caused the outdoor air damper to remain fully open and the heat recovery unit to remain off during most of the heating season.