Leak Detection in Water Distribution Networks via Pressure Analysis Using a Machine Learning Ensemble

Abstract

Water distribution networks (WDNs) are vital infrastructure which serve as a means for public utilities to deliver potable water to consumers. Naturally, pipelines degrade over time, causing leakages and pipe bursts. Damaged pipelines allow water to leak through, incurring significant economic losses. Mitigating these losses are important, especially in areas with water scarcity, to allow consumers to have adequate water supply. Globally, as the population increases, there is a need to make water distribution efficient, due to the rising demand. Thus, leak detection is vital for reducing the system loss of the network and improving efficiency. Monitoring WDNs effectively for leakage is often a challenging task due to the size of the area it covers, and due to the need to detect leaks as early as possible. Traditionally, this is done via pipeline inspection or physical modeling. However, such techniques are either time-consuming, resource intensive, or both. An alternative is machine learning (ML), which maps the relationship between pipeline data to detect leakages. This allows for a faster, yet reasonably accurate model for detection and localization. Machine learning techniques could be utilized together as an ensemble, which allows these techniques to work in conjunction with each other. Wavelet decomposition will be performed on the data to allow for a smaller dataset, as well as utilizing possible hidden features for the machine learning model.