Partitioning Leakage Detection in Water Distribution Systems: A Specialized Deep Learning Framework Enhanced by Spatial–Temporal Graph Convolutional Networks

Abstract

Effective leakage detection is crucial for ensuring operational efficiency, reducing water loss, and maintaining infrastructure integrity in water distribution systems (WDSs). This study presents a specialized leakage detection approach enhanced by spatial–temporal graph convolutional networks (ST-GCN). This method combines large-scale network partition, optimized sensor placement, pilot-scale network partition, and the ST-GCN model, which captures both spatial and temporal dependencies. Then, two case studies are employed to evaluate the effectiveness of this method. The model achieved an average accuracy, precision, recall, and F1-score of 98.38, 98.89, 97.95, and 98.41% across multiple tests for Network A and of 98.51, 98.51, 98.56, and 98.53% for Network B, respectively, which demonstrate the model’s high performance. Furthermore, it compares the model’s simulation results with three existing methods. The enhanced ST-GCN model is superior to those of the other models in terms of accuracy, confirming its superior effectiveness in detecting leakages.