Otimização de manobras para localização de vazamentos em redes de distribuição de água: uma abordagem baseada em transitórios lentos

Abstract

In recent years, water scarcity has become a global challenge, affecting both human water supply and critical economic sectors such as hydroelectric power generation in Brazil. In this context, efficient water resource management is essential to reduce losses, lower operational costs, and ensure water availability for the population, considering that water distribution networks still exhibit high loss rates. Therefore, this study proposes a methodology for detecting and locating small-scale leaks (approximately 1 L/s) in water distribution networks, based on hydraulic transients generated by slow and controlled maneuvers, simulated using the Rigid Inertial Dynamic Model (MDIR). The proposed methodology comprises three main stages: (i) hydraulic modeling, in which a simulator was implemented in Python for both steady-state and transient flow regimes; (ii) maneuver assessment and sensor placement, conducted through sensitivity analyses of hydraulic variables (pressure and flow rate); and (iii) leak localization, performed through metaheuristic optimization (PSO), estimating both the leak position and intensity simultaneously through transient response calibration. In the three case studies (Two-Loop, BLA, and Modena networks), it was observed that pipes with higher velocities and flow rates exhibited greater sensitivity to maneuvers, allowing a reduction in the search space and consequently in the number of required simulations. The sensitivity analysis also guided sensor placement, identifying regions with higher sensitivity concentration and highlighting the need for clustering algorithms to ensure adequate spatial distribution and network coverage. The proposed sensor allocation approach outperformed those reported in the literature, providing greater sensitivity and spatial coverage, particularly in detecting small-scale leaks. Regarding localization, part of the simulated leaks was correctly identified, while others converged to neighboring nodes, with only minor variations in hydraulic parameters. Finally, when calibration errors of up to 10% were simultaneously introduced in demand and roughness, the results remained convergent to positions close to or within the sensor coverage area. Therefore, the proposed approach proved effective for detecting and locating leaks of approximately 1 L/s, achieving better performance when applied under minimum night flow conditions.