Data-driven Approach using Unsupervised Learning for Detecting Anomalies in Facility Operations

As carbon emission reduction is being emphasized globally, the importance of efficient building operations to reduce energy use is emerging as an important factor. To optimize building operations (i.e., achieving better facility conditions with less energy use), this research proposes a methodology that leverages 1) the DBSCAN algorithm to not only cluster groups but identify anomalous behaviors and 2) the DBA algorithm to have a representative case of each cluster group. The methodology is demonstrated against real-world operational data (e.g., time-series temperature) of Building A operated by the Mastern Investment Group located in Seoul, South Korea. As a result, the anomalous cases are identified by the methodology, leading to interviews with the facility managers. After the cause of anomalies is defined and reasoned through discussion with the managers, a couple of strategies are suggested to the managers with the aim of managing the building in an efficient manner.