Comparative Evaluation of Machine Learning Techniques for Resilience Analysis & Leak Detection in Water Distribution System

Abstract

Leakage in water distribution systems remains a critical challenge, leading to significant economic losses, environmental impacts, and operational inefficiencies. Traditional methods of leak detection have been limited in their ability to handle large-scale systems with varying operational conditions. This study aims to improve the detection and analysis of leaks in water distribution networks use by machine learning techniques, specifically K Nearest Neighbors (KNN) and Support Vector Machines (SVM). These algorithms are evaluated for their ability to identify outliers in hydraulic data, including pressure and flow measurements, which are indicative of leaks. In addition to leak detection, the research also focuses on analyzing the effects of single and multiple leakages on network performance. The Pressure Deficiency Index (PDI) is used to assess the impact of leakage scenarios on the pressure distribution within the system. This approach allows for a comprehensive understanding of how leaks influence network behavior under various conditions. Furthermore, the study compares the resilience and reliability of the network by evaluating its behavior under both single and multiple leakage conditions using the Modified Resilience Index (MRI). Experiments are conducted using the LeakDB benchmark dataset, which includes data from two distinct water distribution networks with different demand scenarios. The results demonstrate that the leak detection accuracy using KNN and SVM ranges from 40% to 100%, depending on the selected algorithm, data characteristics, and leakage scenarios. This paper highlights the potential of combining machine learning techniques with resilience and reliability metrics to optimize leak detection and improve the overall performance and sustainability of water distribution systems.