An Leakage detection Method for Water Supply Network Based on Federal Learning Network

Abstract Detecting leaks in water supply networks is a significant global challenge that faces several obstacles. Traditional methods for leak detection often exhibit limitations including low accuracy and high costs due to the complexity of the process. Furthermore, real-world application of these methods is hindered by factors such as scarce leak events, inadequate data collection methods, expensive data procurement, and privacy concerns around water supply network data. To overcome these barriers, our study proposes a leakage detection strategy for water supply networks that leverages a weighted optimization federal learning framework. By merging conventional leak detection techniques with federal learning, we enrich the leak detection dataset, enhancing the model's anomaly recognition abilities and fortifying its resilience. Through the implementation of effective weight allocation strategies, our method attains promising outcomes in identifying leaks within water supply network datasets, showcasing the efficacy of federal learning in leakage detection. This approach streamlines accurate detection and classification across diverse leak scenarios, introducing a novel perspective in tackling water supply network leakage challenges.