Time Series Data Decomposition-Based Anomaly Detection and Evaluation Framework for Operational Management of Smart Water Grid
- 1 September 2021
- journal article
- research article
- Published by American Society of Civil Engineers (ASCE) in Journal of Water Resources Planning and Management
- Vol. 147 (9), 04021059
- https://doi.org/10.1061/(asce)wr.1943-5452.0001433
Abstract
With the increasing adoption of advanced meter infrastructure (AMI), smarter sensors, and temporary and/or permanent data loggers, it is imperative to leverage data analytics methods with hydraulic modeling to improve the quality and efficiency of water service. One important task is to timely detect and evaluate anomaly events so that corresponding actions can be taken to prevent and mitigate the impact of possible water service disruption, which may be caused by the anomaly incidents including but not limited to pipe bursts and unauthorized water usages. In this paper, a comprehensive analysis framework is developed for anomaly event detection and evaluation by developing an integrated solution, which is implemented in multiple components including: (1) data-preprocess or cleansing to eliminate and correct error data records; (2) decomposition of time series data to ensure data stationarity; (3) outlier detection by statistical process control methods with stationary time series; (4) classification of system anomaly events by either correlation analysis of high-flow events with low-pressure events or high-flow outliers with low-pressure outliers; and (5) quantitative evaluation of the system anomaly events with field reported leak incidents. The solution framework has been applied to the water supply zone that is permanent monitored with the flow meter at the inlet and 12 pressure stations throughout the zone with more than 8,000 pipes. Analysis has been conducted with one-year monitoring data and 106 historical leak records, which are employed to validate 526 detected anomaly events. Among them, a 75% true positive rate has been achieved and 90% of 106 field events have been successfully detected with a lead time of more than 24 h. The results obtained indicate that the developed solution method is effective at facilitating the operational management of a smart water grid by maximizing the return of investment in continuously monitoring water distribution networks.Keywords
This publication has 22 references indexed in Scilit:
- Evolutionary Deep Learning with Extended Kalman Filter for Effective Prediction Modeling and Efficient Data AssimilationJournal of Computing in Civil Engineering, 2019
- Water Distribution System Burst Detection Using a Nonlinear Kalman FilterJournal of Water Resources Planning and Management, 2015
- Water distribution systems flow monitoring and anomalous event detection: A practical approachUrban Water Journal, 2015
- Real-time Burst Detection in Water Distribution Systems Using a Bayesian Demand Forecasting MethodologyProcedia Engineering, 2015
- Pattern matching and associative artificial neural networks for water distribution system time series data analysisJournal of Hydroinformatics, 2013
- Novelty detection for time series data analysis in water distribution systems using support vector machinesJournal of Hydroinformatics, 2010
- Development and Verification of an Online Artificial Intelligence System for Detection of Bursts and Other Abnormal FlowsJournal of Water Resources Planning and Management, 2010
- Sensor-fusion of hydraulic data for burst detection and location in a treated water distribution systemInformation Fusion, 2003
- A neural network approach to burst detectionWater Science & Technology, 2002
- A method for automatic validation of long time series of data in urban hydrologyWater Science & Technology, 2002