Abstract
Commercial and industrial buildings account for a considerable fraction of all the energy consumed in the U.S., and reducing this energy consumption has become a national grand challenge. Based on the large-scale deployment of sensors in modern commercial buildings, many organizations are applying data analytics to the thousands of sensing and control points to detect wasteful, incorrect and inefficient operations for energy savings. Scaling this approach is challenging, however, because the metadata about these sensing and control points is inconsistent between buildings, or even missing altogether. As a result, an analytics engine cannot be applied to a new building without first addressing the issue of mapping: creating a match between the sensor stream context and the inputs of a data analytics engine. This mapping process currently requires significant integration effort and anecdotally can take a week or longer for each building. Thus, metadata mapping is a major obstacle to scaling up building analytics.
The overarching goal of this research is to enable automatic inference of the sensor context such as its type, location, and relationships to others, so that building analytics can be quickly applied at scale. Bearing this goal in mind, we have developed a suite of techniques to infer the sensor context (i.e., the metadata), requiring minimal human intervention. At the core are fully automated techniques that infer two kinds of contextual information about each sensor: the type and its relationship with other sensors. The type inference technique leverages information from existing well mapped buildings to help the inference for a new building, while the relationship inference technique builds upon the intuition that connected equipment or co-located sensors are exposed to the same real world events, thus exhibiting correlated changes in their data. We have also designed methods to complement the automatic type inference technique in the absence of already mapped buildings or when faced with high-dimensional data. The techniques proposed in this dissertation represent a first step towards technology that would enable any new building analytics to quickly scale to the 10’s of millions of commercial buildings across the globe, with the minimal need of manual mapping on a per-building basis. With the advent of Internet of Things (IoT) and proliferation of sensory data, continuously inferring the context of the data at scale will become inevitable, and we believe the techniques presented in this dissertation are generally applicable to the broader picture of IoT.
