As people move through the building, they generate paths seen as activations of different sensors and actuators. Individually, a path can be described a sequence of events; each event corresponds to a significant change in the state or value of a sensor or actuator. As paths repeat and overlap, a subsequence of events can be described as rules of the form: A then B then C. For example, (A) Cross the door, (B) sit at the desk, (C) turn on the computer. Finally, by examining the rules, flat and hierarchical groupings can be derived. A flat group can be of the form: all desks and computers in the room with a door (A), and a hierarchical group of the form: room devices (door, ceiling lights, window blinds, people counter), desk devices( presence detector, desk lamp, computer).
Natural clustering of office building sensor, actuator and context variables. The clusters were created by association using the building management system configuration as the data source. The variables from two offices are shown. To the left, a meeting room, to the right, a four desk office. The desks are positioned as petals around the office room common variables. The meeting room does not have a clear variable separation, despite the configurations defining two areas of interest.
Hierarchical group association of office building sensor, actuator and context variables for a 4 person office. The room is at the top of the hierarchy and has variables which associate at this level. For example, the door state sensor, total number of people in the room, room ceiling lights. The cells represent individual desk areas, and subcells represent areas within the desk; typically, in front of the screen or to the side.
General three step representation of the rule mining framework. The time series of office building sensor, actuator and context variables is converted into events. The events are then mined to extract rules of the form A then B, the final step interprets the rules an creates flat or hierarchical groups.
Hierarchical representation of office building sensor, actuator and context variables for a 4 person office. The tree structure is constructed by analysing which is the order of activation as users enter the space. For example, to activate computer work (pink circles), presence in the room is activated (top green node); then, presence at the desk(second level green nodes); finally, distance right and distance left (grey nodes).
"Mining hierarchical relations in building management variables", Pervasive and Mobile Computing, ScienceDirect, February 2016.
"Mining relations and physical grouping of building-embedded sensors and actuators", Proceedings of the International Conference on Pervasive Computing and Communications (PerCom '15), IEEE, March 2015.
"Data Mining-based Localisation of Spatial Low-resolution Sensors in Commercial Buildings", Proceedings of the 3rd ACM International Conference on Systems for Energy-Efficient Built Environments (BuildSys '16), ACM, November 2016.