Reliable anomaly detection in water systems using the self-adjusting, label-free, data-driven algorithm (SALDA)

Reliable anomaly detection in water systems, particularly for leak detection in water distribution networks (WDNs), remains a critical challenge due to high data variability, operational uncertainties, and scarcity of labeled datasets. This study introduces SALDA, a self-adjusting, label-free, data-...

Full description

Saved in:

Bibliographic Details
Published in:	Journal of water process engineering Vol. 76; p. 108207
Main Authors:	Najafi, Hessam, Huang, Jinhui Jeanne
Format:	Journal Article
Language:	English
Published:	Elsevier Ltd 01.08.2025
Subjects:	Adaptive anomaly detection Dynamic time warping Label-free data-driven algorithm Water distribution network Z-number uncertainty modeling Z-number uncertainty modeling Adaptive anomaly detection Dynamic time warping Label-free data-driven algorithm Water distribution network
ISSN:	2214-7144, 2214-7144
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Reliable anomaly detection in water systems, particularly for leak detection in water distribution networks (WDNs), remains a critical challenge due to high data variability, operational uncertainties, and scarcity of labeled datasets. This study introduces SALDA, a self-adjusting, label-free, data-driven algorithm for anomaly detection that dynamically updates its baseline without requiring pre-labeled data. Composed of four interconnected modules, SALDA utilizes different thresholds and baselines, enabling the detection of both sudden and gradual changes in time series. Unlike conventional threshold-based approaches, SALDA leverages dynamic time warping (DTW) to optimally align real-time data with a continuously evolving baseline, thereby effectively capturing variations in the network. Additionally, Z-number-based uncertainty-aware thresholding further enhances detection reliability. Its computationally efficient, decentralized structure allows direct deployment on flow and pressure sensors, supporting real-time monitoring. The proposed approach was validated using both synthetic and real-world datasets from a district metered area (DMA)-based WDN in China, and 30 months of high-frequency (15-min interval) sensor data were analyzed. SALDA was benchmarked against conventional threshold-based techniques and clustering-based unsupervised methods, demonstrating up to 66 % higher detection accuracy while maintaining robustness across varying operational conditions. These findings underscore the potential of SALDA as a scalable, self-adjusting anomaly detection framework for enhancing the resilience of water systems. •Proposes SALDA with four modules for reliable anomaly detection in water systems.•Integrates Z-numbers to improve reliability and reduce threshold uncertainty.•Employs Dynamic Time Warping, ensuring more accurate alignment of time series.•Detects sudden and gradual leaks by adapting to changing operational conditions.•Validated on synthetic and real-world data from 174 flow and pressure sensors.
ISSN:	2214-7144 2214-7144
DOI:	10.1016/j.jwpe.2025.108207