Smart meter-powered water end-use activity recognition: data, algorithms, and privacy-enhancing technologies to foster data sharing for research and practice

Leveraging smart water meter data to build knowledge on when and how water is used in different households at the fixture level yields valuable insights for demand management and efficiency programs for both customers and water providers. The literature on smart water meter development, end-use recognition methods, and smart meter-based water conservation has been growing over the last 30 years. Yet, many controversial aspects related to the uncertain technological, economic, and social benefits of high-resolution digital meters still limit the availability of household-level water usage data for research and practice. When it comes to water end-use recognition in particular, the following research questions are still open. First, on data and privacy: What limitations still hamper large-scale collection of smart meter data with complementary end-use information? How can we protect users’ privacy and avoid exposing sensitive information about the activities of metered households, especially if an adversary gains access to such data? Second, on end-use activity recognition algorithms: Can we bypass the need of intrusive calibration or time-consuming manual processing required for developing end-use classification algorithms? How many data points are sufficient to train these algorithms? What is the tradeoff between classification accuracy and training dataset size? In this work we build on recent research efforts developed on data across three continents to provide quantitative answers to the above questions and define guidelines for water providers willing to publish household water meter data for research and planning purposes while guaranteeing users’ privacy and minimal costs for data gathering campaigns.