Water Pipeline Leak Detection and Localisation in Water Distribution Networks

Water Distribution Networks (WDNs) suffer from significant water losses due to pipeline leaks, leading to financial losses and exacerbating global water scarcity concerns. This paper aims to improve the accuracy of leak detection and the identification of leakage locations in WDNs. Leakage in WDNs is a global challenge that impacts water utilities economically and environmentally. Traditional methods for leak detection are time-consuming, resource-intensive, and prone to inaccurate results and false alarms due to randomly placed sensors. Detecting concealed background leaks in WDNs is particularly challenging. To address these limitations, this paper proposes an enhanced ensemble supervised Machine Learning (ML) algorithm that integrates Graph Theory (GT), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). By combining multiple ML algorithms, the proposed model considers various factors affecting leak detection and localization, resulting in more accurate and reliable assessments of leak presence and location. The combination of EPANET and MATLAB provides a powerful tool for evaluating the performance of the proposed SVM-ANN-GT algorithm and comparing it with traditional methods. The results of the proposed SVM-ANN-GT algorithm can be used as a benchmark for accuracy and effectiveness. Three algorithms, SVM-ANN-GT, SVM, and ANN, were evaluated using EPANET. The SVM-ANN-GT algorithm achieved the highest average leak detection accuracy of 96%, in the order of performance, then followed by the SVM at 85% and ANN at 80%. The superior performance of SVM-ANN-GT can be attributed to its strategic sensor placement based on graph theory principles, which the other algorithms lacked, which then lead to lower level of accuracy.