Time-Series-Based Leakage Detection Using Multiple Pressure Sensors in Water Distribution Systems

Area of Science:

• Hydraulic Engineering
• Water Resource Management
• Data Science in Utilities

Background:

• Water distribution systems (WDS) face significant economic losses and water shortages due to undetected leaks.
• Time series monitoring data offers more reliable insights for leak detection compared to single-point data.
• Fluctuating pressure and flow rate data in WDS necessitate advanced analytical methods.

Purpose of the Study:

• To propose a novel threshold selection method for correlation coefficients using time series data for improved leak detection in WDS.
• To leverage leak scenario falsification to enhance the interpretation of time series monitoring data.
• To investigate the advantages of temporal data interpretation for identifying leaks in water networks.

Main Methods:

• Utilizing temporally varying correlations between data from multiple pressure sensors within WDS.
• Implementing a threshold selection method based on correlation coefficients and leak scenario falsification.
• Scanning multiple times using a defined scanning time window and testing the effect of window length on threshold selection.

Main Results:

• The proposed method demonstrates potential for improving leak detection performance in real-world WDS.
• A scanning time window length of 3-6 units was found to achieve better performance in case studies.
• Performance is influenced by various factors including modeling and measurement uncertainties.

Conclusions:

• Time series data analysis, particularly using temporal correlations, offers a promising approach for effective leak detection in WDS.
• The developed threshold selection method provides a more robust diagnostic tool for water utilities.
• Further research is needed to mitigate the impact of uncertainties on leak detection accuracy.