Detecção de vazamentos em redes hidráulicas utilizando Redes Neurais de Grafos

Abstract

This work approaches the problem of distributed monitoring of water distribution networks focusing on leakage detection, using computational simulation of water distribution networks and machine learning techniques. The implemented machine learning technique is based on Graph Neural Networks, which are structures with the capacity to take into account the spatial displacement of the measurement points in the network alongside the measurement data, thus providing insight regarding the leak location. A case study application was developed to evaluate the capacity of the algorithm. A hypothetical water distribution network was implemented in a software environment to enable data collection in diverse operation scenarios, especially leakages in different locations. The initial results demonstrate that the graph-based approach is a viable methodology for water leakage detection.