Hyperparameter Optimization of a Convolutional Neural Network Model for Pipe Burst Location in Water Distribution...

Area of Science:

• Hydraulic engineering
• Machine learning applications
• Water resource management

Background:

• Water distribution networks (WDN) are critical infrastructure prone to pipe bursts.
• Accurate and timely detection of pipe bursts is essential for minimizing water loss and service disruption.
• Traditional methods for pipe burst detection can be inefficient and inaccurate.

Purpose of the Study:

• To present a hyperparameter optimization process for a convolutional neural network (CNN).
• To apply the optimized CNN for identifying pipe burst locations in water distribution networks (WDN).
• To evaluate the model's performance under varying noise levels and sensor proximities.

Main Methods:

• Hyperparameter optimization for CNN, including early stopping, dataset size, normalization, batch size, optimizer, and learning rate regularization.
• Utilized a case study from a real-world WDN.
• Evaluated the model with distinct measurement noise levels and pipe burst locations.

Main Results:

• Identified optimal CNN parameters: convolutional 1D layer (32 filters, kernel size 3, stride 1), 5000 epochs, 250 datasets (normalized 0-1), batch size 500, Adam optimizer with learning rate regularization.
• The parameterized model's pipe burst search area dispersion is influenced by sensor proximity to the burst and measurement noise levels.

Conclusions:

• The optimized CNN provides a robust method for pipe burst localization in WDN.
• Model performance is sensitive to data quality (noise) and sensor network configuration.
• Further research can refine the model for enhanced accuracy in complex WDN scenarios.