Optimizing Leak Detection and Location in Water Distribution Networks Using SVM-RF Algorithm

Water Distribution Networks (WDNs) suffer significant water losses from pipeline leaks, impacting both the economy and global water scarcity concerns. This study seeks to enhance the precision of leak detection and identify leakage locations in WDNs, addressing challenges posed by background leakage. Existing techniques for leak detection in WDNs have limitations, particularly in detecting background leakage, leading to resource-intensive and time-consuming processes with potential inaccuracies. To minimize the loss of water, it is of utmost importance to identify and decrease excessive pressure in specific areas of the WDN. This paper proposes the SVM-RF algorithm, an ensemble supervised Machine Learning (ML) approach merging Support Vector Machines (SVM) and Random Forest (RF). The SVM-RF model, leveraging automatic feature extraction and dual classifiers, offers more accurate leak assessments based on flow rate, pressure, and temperature data. A combined probability scheme enhances efficiency by integrating diverse classifier models with distinct prediction outputs. The performance of different algorithms in leak detection accuracy was assessed using the EPANET software. Among the algorithms evaluated, the SVM-RF algorithm demonstrated superior performance, achieving an impressive average leak detection accuracy of 97%. In addition, the RF algorithm achieved an accuracy of 83%, while the SVM algorithm achieved 77% accuracy.