ITcon Vol. 18, pg. 99-118, http://www.itcon.org/2013/6

Evaluating a data clustering approach for life-cycle facility control

submitted:June 11, 2012
revised:February 20, 2013
published:April 2013
editor(s):Amor R
authors:A. Christopher Bogen, Computer Scientist,
Engineering Research and Development Center, US Army Corps of Engineers
Email: Chris.Bogen@usace.army.mil

Mahbubur Rashid, Computer Scientist,
Engineering Research and Development Center, US Army Corps of Engineers
Email: Mahbubur.Rashid@usace.army.mil

E. William East, Civil Engineer
Engineering Research and Development Center, US Army Corps of Engineers
Email: Bill.W.East@usace.army.mil

James Ross, Computer Scientist
Engineering Research and Development Center, US Army Corps of Engineers
Email: James.E.Ross@usace.army.mil
summary:Data reported by sensors in building automation and control systems is critical for evaluating the as- operated performance of a facility. Typically these systems are designed to support specific control domains, but facility performance analysis requires the fusion of data across these domains. Since a facility may have several disparate, closed-loop systems, resolution of data interoperability issues is a prerequisite to cross-domain data fusion. In previous publications, the authors have proposed an experimental platform for building information fusion where the sensors are reconciled to building information model elements and ultimately to an expected resource utilization schedule. The motivation for this integration is to provide a framework for comparing the as- operated facility with its intended usage patterns. While the authors’ data integration framework provides representational tools for integrating BIM and raw sensor data, appropriate computational approaches for normalization, filtering, and pattern extraction methods must be developed to provide a mathematical basis for anomaly detection and “plan” versus “actual” comparisons of resource use. This article presents a computational workflow for categorizing daily resource usage according to a resolution typical of human-specified schedules. Simulated datasets and real datasets are used as the basis for experimental analysis of the authors’ approach, and results indicate that the algorithm can produce 90% matching accuracy with noise/variations up to 55%.
keywords:Building Information Modelling (BIM), machine learning, pattern detection, signal processing, building automation and control
full text: (PDF file, 1.681 MB)
citation:Bogen AC, Rashid M, East EW, Ross J (2013). Evaluating a data clustering approach for life-cycle facility control, ITcon Vol. 18, pg. 99-118, http://www.itcon.org/2013/6