Contamination source identification in water distribution networks using convolutional neural network.

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Contamination source identification (CSI) is critical for water distribution systems (WDSs) security.
• Limited water quality sensors and contaminant detection challenges hinder effective CSI.
• Consumer complaints offer an alternative data source but face accuracy issues due to complaint delay uncertainty.

Purpose of the Study:

• To develop a robust methodology for CSI in WDSs using consumer complaint data.
• To address the challenge of uncertain complaint delay times impacting CSI accuracy.
• To leverage deep learning for improved pattern recognition in consumer complaint data for CSI.

Main Methods:

• Constructed complaint matrices to visualize spatiotemporal complaint characteristics during intrusion events.
• Proposed a novel methodology employing Convolutional Neural Networks (CNNs) for pattern recognition.
• Utilized CNNs to identify inherent patterns linking complaints to specific contaminant intrusion nodes.

Main Results:

• The CNN-based methodology demonstrated effectiveness in identifying contamination sources in WDSs of various scales.
• The approach showed high accuracy and robustness even with significant uncertainties in complaint delay times.
• CNN outperformed a back-propagation artificial neural network in CSI accuracy and robustness.

Conclusions:

• Consumer complaint data, when analyzed with advanced methods like CNNs, can be effectively used for CSI in WDSs.
• The proposed CNN framework offers a significant improvement for accurate and robust contamination source identification.
• This methodology enhances water quality security by providing a reliable approach to pinpoint contamination origins.