The building adapter: Towards quickly applying building analytics at scale

Abstract

Building analytics can produce substantial energy savings in commercial buildings by automatically detecting wasteful or incorrect operations. However, a new building's sensing and control points need to be mapped to the inputs of an analytics engine before analysis is feasible and the process of mapping is a highly manual process - a key obstacle to scaling up building analytics. In this paper, we present new techniques to perform automatic mapping without any manual intervention. Our approach builds on and improves upon techniques from transfer learning: it learns a set of statistic classifiers of the metadata from a labeled building and adaptively integrates those classifiers to another unlabeled building, even if the two buildings have very different metadata conventions. We evaluate this approach using 7 days' data from over 2,500 sensors located in 3 commercial buildings. Results indicate that this approach can automatically label at least 36% of the points with more than 85% accuracy, while the best baseline achieves only 63% label accuracy on average. These techniques represent a first step towards technology that would enable any new building analytics engine to scale quickly to the 10's of millions of commercial buildings across the globe, without the need for manual mapping on a per-building basis.