Keywords

water-energy nexus; smart meter; water-related energy; demand management; input variable selection

Start Date

25-6-2018 10:40 AM

End Date

25-6-2018 12:00 PM

Abstract

Planning and management strategies for water and energy systems are key for meeting future demands under population growth, urbanization, changing economic and climate conditions, and emerging technologies. Given the water-energy nexus, interest has raised towards the design of coordinated water-energy interventions to manage urban water and energy end use demands - including that of the residential sector – and ultimately foster both water and energy conservation and use efficiency. On this respect, while sub-daily resolution data gathered via advanced metering infrastructures and intelligent sensors installed at the household/building scale enables recording water uses with a finer granularity than in the past, new models that adequately facilitate our understanding of water and energy demands and their inter-dependencies are needed. In this work, we propose an information extraction-based approach to estimate residential water-related electricity for heating purposes. Our approach relies only on the knowledge of fine resolution (e.g., 1 min sampling frequency) water and electricity data collected by two single-point, non-intrusive, water and electricity meters. We first process the data to detect water use events, compute time and consumption-based features for each event. We then use Iterative Input Selection, a variable selection algorithm for data-driven models, to determine the optimal subset of features needed to build a regression model that estimates the end-use electricity used for heating water, for each use event. We use extremely randomized trees as nonparametric regression models. Results from an application of the onto data collected from a single household in Canada show that our approach can estimate the water-related electricity used from the instant hot water unit at each consumption event with an accuracy of over 90%. In addition, we demonstrate that a joint analysis of water and electricity data collected via smart meters can help unpacking the electricity use related to specific water end- uses, such as clothes washers.

Stream and Session

Stream B: (Big) Data Solutions for Planning, Management, and Operation and Environmental Systems

Session B1: Modelling and Managing Urban Water and Energy Demands in the Era of Big Data

COinS
 
Jun 25th, 10:40 AM Jun 25th, 12:00 PM

An information extraction-based approach to estimate residential water-related energy from single-point smart meter data

Planning and management strategies for water and energy systems are key for meeting future demands under population growth, urbanization, changing economic and climate conditions, and emerging technologies. Given the water-energy nexus, interest has raised towards the design of coordinated water-energy interventions to manage urban water and energy end use demands - including that of the residential sector – and ultimately foster both water and energy conservation and use efficiency. On this respect, while sub-daily resolution data gathered via advanced metering infrastructures and intelligent sensors installed at the household/building scale enables recording water uses with a finer granularity than in the past, new models that adequately facilitate our understanding of water and energy demands and their inter-dependencies are needed. In this work, we propose an information extraction-based approach to estimate residential water-related electricity for heating purposes. Our approach relies only on the knowledge of fine resolution (e.g., 1 min sampling frequency) water and electricity data collected by two single-point, non-intrusive, water and electricity meters. We first process the data to detect water use events, compute time and consumption-based features for each event. We then use Iterative Input Selection, a variable selection algorithm for data-driven models, to determine the optimal subset of features needed to build a regression model that estimates the end-use electricity used for heating water, for each use event. We use extremely randomized trees as nonparametric regression models. Results from an application of the onto data collected from a single household in Canada show that our approach can estimate the water-related electricity used from the instant hot water unit at each consumption event with an accuracy of over 90%. In addition, we demonstrate that a joint analysis of water and electricity data collected via smart meters can help unpacking the electricity use related to specific water end- uses, such as clothes washers.