An integrated logit model for contamination event detection in water distribution systems

Area of Science:

• Environmental Engineering
• Water Resource Management
• Data Science

Background:

• Contamination event detection in water distribution systems is a critical challenge.
• Existing methods often process water quality indicators independently, leading to suboptimal alarm integration.
• Current systems frequently rely on simple heuristics for alarm fusion, limiting performance.

Purpose of the Study:

• To develop an advanced methodology for robust contamination event detection in water distribution systems.
• To improve the integration of multiple water quality indicator alarms.
• To enhance the accuracy and reliability of water quality monitoring systems.

Main Methods:

• Utilized a statistically oriented discrete choice model for integrating individual water quality indicator alarms.
• Employed the maximum likelihood method for estimating the discrete choice model.
• Jointly calibrated the event detection system components, including the discrete choice model, using genetic algorithms on a training dataset.
• Focused on optimizing the probability fusion process for different indicators.

Main Results:

• The developed approach demonstrated improved performance in detecting contamination events.
• Significantly reduced the number of false positive alarms compared to previous studies.
• Showcased a higher probability of detecting events through effective alarm integration.
• Validated the methodology using real-world water quality data.

Conclusions:

• The integration of single alarms from various water quality indicators is crucial for effective event detection.
• A discrete choice model provides a statistically sound framework for improving the performance of alarm fusion.
• The proposed methodology offers a more accurate and reliable solution for safeguarding water distribution systems against contamination.