Failure Modeling of Small and Medium-Sized Water Distribution Networks Based on a Hybrid Ranking Model

To ensure water supply safety and reduce environmental impact, it is essential to conduct failure modeling on water distribution networks (WDNs) to prevent pipe failures. However, the limited and imbalanced pipe break data in small and medium-sized WDNs make it challenging to develop a universal model. This study proposes a hybrid ranking model to address these challenges by integrating a novel comprehensive indicator (CI) with advanced data-pairing techniques. The CI quantifies localized environmental risks through spatial aggregation of historical break patterns, compensating for unmeasured factors such as soil corrosion. To mitigate data imbalance, an automatic pairing network generates balanced sample pairs by combining failed and not-failed pipes, expanding the training set while preserving data authenticity. The hybrid ranking model employs a dual-branch convolutional neural network (CNN) to capture nonlinear relationships between pipe attributes, hydraulic parameters, and CI-driven environmental risks. The model was validated on a real-world WDN in China, and achieved exceptional performance, with accuracy, precision, recall, F1 score, and area under curve (AUC) all exceeding 98%. Sensitivity analysis identified pipe material, leakage rate, and CI as the most critical factors influencing failure probability. Unlike conventional methods, this approach does not require extensive environmental monitoring data, making it particularly suitable for resource-constrained utilities.