Data Mining to Uncover Heterogeneous Water Use Behaviors From Smart Meter Data

Knowledge on the determinants and patterns of water demand for different consumers supports the design of customized demand management strategies. Smart meters coupled with big data analytics tools create a unique opportunity to support such strategies. Yet, at present, the information content of smart meter data is not fully mined and usually needs to be complemented with water fixture inventory and survey data to achieve detailed customer segmentation based on end use water usage. In this paper, we developed a data‐driven approach that extracts information on heterogeneous water end use routines, main end use components, and temporal characteristics, only via data mining existing smart meter readings at the scale of individual households. We tested our approach on data from 327 households in Australia, each monitored with smart meters logging water use readings every 5 s. As part of the approach, we first disaggregated the household‐level water use time series into different end uses via Autoflow. We then adapted a customer segmentation based on eigenbehavior analysis to discriminate among heterogeneous water end use routines and identify clusters of consumers presenting similar routines. Results revealed three main water end use profile clusters, each characterized by a primary end use: shower, clothes washing, and irrigation. Time‐of‐use and intensity‐of‐use differences exist within each class, as well as different characteristics of regularity and periodicity over time. Our customer segmentation analysis approach provides utilities with a concise snapshot of recurrent water use routines from smart meter data and can be used to support customized demand management strategies.

Key Points:

We investigate which water use behaviors can be identified from nonintrusive, single‐point, smart meter data We identify and cluster primary water use behaviors of single‐family households from disaggregated end use data We reveal the main water end uses driving different behaviors, usage patterns, and regularity, to support customized demand management