Keywords
residential water, data disaggregation, clustering, water efficiency and conservation
Start Date
15-9-2020 2:00 PM
End Date
15-9-2020 2:20 PM
Abstract
Residential water consumption is understudied globally, especially regarding end uses of water both indoors and outdoors. Many areas of the United States, for example, collect meter-level data on a monthly timescale, and still others lack residential water meters entirely. To better understand end uses of water in the residential environment, additional data collection is needed, along with methods to make sense of data with better temporal resolution. As a demonstration of one possible approach, we installed a custom smart water metering system collecting data on total flow rate, temperature, and pressure at 1-second intervals, based on the data acquisition setup, from a single-family residential home in central Illinois, United States, starting in February 2018. Using one year of 1-second resolution data, we created a method to disaggregate meter-level data into end uses based on event detection as non-zero flow over a duration. Using our algorithm, we then disaggregated concurrent events using derivatives of the total aggregate flow rate and unsupervised machine learning k-means clustering approaches to compare with known end-use signatures for in-home fixtures and appliances. Analysis of the disaggregated data informs an estimate of water end uses, highlighting the dominance of the shower, clothes washer, and faucets in particular. Flow rates for disaggregated end uses reflect differences in fixture/appliance operations compared to manufacturers’ ratings and differences in behavior with human-controlled events (e.g., showers). Results can inform customized water conservation and efficiency recommendations at the household level. The disaggregation algorithm and unsupervised machine learning classification approach are readily transferrable to future residential smart water meter installations with similar temporal resolution data.
Meter to End-Use: Disaggregating Residential Water Data
Residential water consumption is understudied globally, especially regarding end uses of water both indoors and outdoors. Many areas of the United States, for example, collect meter-level data on a monthly timescale, and still others lack residential water meters entirely. To better understand end uses of water in the residential environment, additional data collection is needed, along with methods to make sense of data with better temporal resolution. As a demonstration of one possible approach, we installed a custom smart water metering system collecting data on total flow rate, temperature, and pressure at 1-second intervals, based on the data acquisition setup, from a single-family residential home in central Illinois, United States, starting in February 2018. Using one year of 1-second resolution data, we created a method to disaggregate meter-level data into end uses based on event detection as non-zero flow over a duration. Using our algorithm, we then disaggregated concurrent events using derivatives of the total aggregate flow rate and unsupervised machine learning k-means clustering approaches to compare with known end-use signatures for in-home fixtures and appliances. Analysis of the disaggregated data informs an estimate of water end uses, highlighting the dominance of the shower, clothes washer, and faucets in particular. Flow rates for disaggregated end uses reflect differences in fixture/appliance operations compared to manufacturers’ ratings and differences in behavior with human-controlled events (e.g., showers). Results can inform customized water conservation and efficiency recommendations at the household level. The disaggregation algorithm and unsupervised machine learning classification approach are readily transferrable to future residential smart water meter installations with similar temporal resolution data.
Stream and Session
false