Despite the fact that residential buildings represent about 40% of electricity consumption in the United States (US), until recently most electric utilities only collected meter data once per month and rarely made it available to consumers or researchers. Now, however, electric utilities have begun deploying advanced metering infrastructure (AMI), capable of automatically reporting consumption at hourly or sub-hourly intervals. This new trend is being driven by growing interest in understanding patterns of residential energy usage, as energy supplies diversify and the drive to reduce U.S. emissions accelerates. Our multi-disciplinary group of researchers took a further step to understand patterns of energy usage in the urban residential sector by installing a micro-metering system capable of taking 10-second interval power data in nearly 400 apartments across a dozen buildings in New York City. Because of the frequency of the observations, the large sample size of residential units, and the 12-month study duration, this dataset provides unprecedented insight into how customers in urban multifamily buildings are using electricity.
A major part of the motivation behind this project, which was supported, in part, by a U.S. Department of Energy “Building Energy Efficiency Frontiers and Innovations Technologies” (BENEFIT) grant, was the inclusion of electricity usage feedback to residential consumers in messages that varied in content and presentation throughout the study period. Our group’s team of natural language processing (NLP) researchers used algorithms to automatically generate and send customized feedback messages to those residents in the 400 monitored apartments who expressed interest in knowing their electricity consumption. By analyzing a resident’s real-time response to these messages, we were able to identify the kinds of messaging that was most effective at motivating individual customers to reduce their electricity consumption, including at times when the local grid was most stressed.
One distinguishing quality of this dataset is its inclusion of both real and reactive power observations. If real power is pushing a wheel barrow, then reactive power is lifting the handles so that you can get it moving — not useful on its own, but necessary for doing any real work. In practice, the relationship between real and reactive power on an individual circuit tells us what kinds of devices are consuming energy. Appliances with large motors like window air-conditioners have a different trace than microwaves or TVs. Our group’s disaggregation researchers used this feature of the data to identify different kinds of electricity loads so that we could better understand the end uses that were being served.
Something that fascinated us about the collected data was the wide variation in electricity use patterns among purportedly similar customers living in the same set of buildings. When we look at electricity usage on the scale of a neighborhood or city, we see very regular daily and seasonal variations that can be explained in terms of the weather, meal times, and when residents are home from work or school. When we zoom in on a specific collection of customers, those trends no longer stand out. Instead we see thousands of individual appliances being turned on or off in response to a particular consumer’s wants and needs. Understanding the ways in which real consumers’ habits deviate from well-explained average behavior is useful in applications ranging from the design of novel utility tariffs to assessing the potential for distributed energy resources (DERs).
There are many ways that more granular data of household electricity use may benefit the research and practice community. Our team has only scratched the surface of this dataset’s potential. By sharing it, we hope that others will find applications that we haven’t yet tried or even imagined.
The data descriptor can be found here.