Prediction of Water Leakage in Pipeline Networks Using Graph Convolutional Network Method

This study aims to predict leaks in water-carrying pipelines by monitoring pressure drops. Timely detection of leaks is crucial for prompt intervention and repair efforts. In this research, we represent the network structure of pipelines using graph representations. Consequently, we propose a machine learning model called Graph Convolutional Neural Network (GCN) that leverages graph-type data structures for leak prediction. Conventional machine learning models often overlook the dependencies between nodes and edges in graph structures, which are critical in complex systems like pipelines. GCN offers an advantage in capturing the intricate relationships among connections in pipelines. To assess the predictive performance of our proposed GCN model, we compare it against the Support Vector Machine (SVM) model, a widely used traditional machine learning approach. In this study, we conducted experimental studies to collect the required pressure and flow data to train the GCN and SVM models. The obtained results were visualized and analyzed to evaluate their respective performances. The GCN model achieved a performance rate of 94%, while the SVM model achieved 87%. These results demonstrated the potential of the GCN model in accurately detecting water leaks in pipeline systems. The findings hold significant implications for water resource management and environmental protection. The knowledge acquired from this study can serve as a foundation for predicting leaks in pipelines that transport gas and oil.