Research on a Novel Unsupervised-Learning-Based Pipeline Leak Detection Method Based on Temporal Kolmogorov-Arnold...

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Traditional AI methods for pipeline leak detection face challenges in complete process detection and incur high costs due to GPU-intensive neural networks.
• Existing methods often lack interpretability and struggle with complex temporal data patterns.

Purpose of the Study:

• To develop a novel, cost-effective, and transparent automated leak detection system for urban water supply pipelines.
• To enhance the accuracy and interpretability of leak detection by integrating prior knowledge with reconstruction error theory.

Main Methods:

• A hybrid AI model combining the Kolmogorov-Arnold Network (KAN) with an autoencoder (AE) was developed to capture temporal dependencies and reconstruction capabilities.
• A novel unsupervised anomaly sequence labeling method was created, integrating prior knowledge with reconstruction error theory for improved leak detection.

Main Results:

• The proposed KAN-AE model and sequence labeling method achieved a segment-wise precision of 93.1% in field experiments on urban water supply pipelines.
• The new method demonstrated enhanced interpretability and accuracy compared to commonly used models and methods.

Conclusions:

• The study presents a robust and transparent solution for automated pipeline leak detection, suitable for large-scale deployment.
• This approach facilitates the cost-effective development of digital twin systems for urban pipeline leak emergency management.