Abstract
The reliability of urban water distribution networks (WDNs) is critical for public health and safety. This study presents a novel approach to predicting WDN failures by leveraging Graph Neural Networks (GNNs) and incorporating coupled features of road and water networks, with an emphasis on traffic-related characteristics. Our framework employs Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), and GraphSAGE to capture the complex spatial dependencies and interactions between road infrastructure and water pipelines. We evaluate the performance of these models using a dataset from a Chinese city, focusing on metrics such as Area Under the Curve (AUC), accuracy, and recall. Our results indicate that GraphSAGE outperforms other models, demonstrating its effectiveness in leveraging neighborhood information for failure prediction. The analysis of feature importance highlights the significance of traffic-related attributes, such as the distance of pipelines from the center of intersections, road grades, and the angle of pipelines relative to roads, in addition to traditional factors like pipeline length, diameter, and age. By integrating these coupled features, our study offers a more accurate and comprehensive understanding of failure risks, providing valuable insights for proactive maintenance and management of urban WDNs.