Flow time series decomposition to identify non-revenue water components in drinking water distribution systems: A data-driven approach

Abstract

Water utilities face challenges in managing non-revenue water, which encompasses unbilled authorised consumption, leaks, bursts, authorised consumption errors, and unauthorised consumption. Several approaches have been developed to address these issues. Most existing methods focus on estimating individual components of non-revenue water, rather than considering all aspects comprehensively. The installation of smart water meters has significantly reduced unmetered billed consumption, addressing issues related to the absence of water meters in some customer locations or difficulties in systematic meter reading. Water utilities can obtain a comprehensive view of non-revenue water over time by combining the billed metered consumption time series obtained with smart meters with the network flow time series. Partitioning the non-revenue water time series into several components, each representing a different pattern in the data, can help one better grasp the underlying patterns. In this paper, time series decomposition techniques reveal hidden non-revenue water components, allowing the water utilities to create a network strategy to reduce water losses. Several decomposition methods were applied, and the best reliable results were achieved with Singular Spectrum Analysis.