Aplicação da rede neural mapas auto-organizáveis e algoritmo genético NSGA - II à setorização de rede de distribuição de água

Abstract

Water is an essential resource for life, yet a significant portion is wasted in urban distribution networks. In Brazil, water losses reach approximately 40%, with 60% of these attributed to physical losses, such as leaks. Controlling these losses is crucial and represents a complex challenge that demands innovative solutions and advanced tools. Nevertheless, many strategies are still developed heuristically. In this context, this study proposes a method for the sectorization of water distribution networks, applied to the benchmark Balerma network. The method is structured in two phases: clustering, using self-organizing maps (SOM), and optimization, leveraging the NSGA – II genetic algorithm. In the clustering phase, the SOM identified the sectors or District Metering Areas (DMAs), achieving quantization and topographic errors of 0.0819 and 0.0700, respectively, which indicate a well-trained neural network. In the optimization phase, the placement of isolation valves and flow measurement devices led to a 4.55% reduction in the original average pressure and a 1.21% decrease in the estimated network losses. Additional simulations, considering a minimum service pressure of 10 meters, achieved a 21.60% reduction in the network's average pressure and an estimated 46.47% reduction in water losses (6.02 percentage points less than the original average losses). The proposed approach proved to be promising, combining robust computational techniques with low computational effort and offering a feasible solution for loss control in water distribution systems. Despite being applied to a naturally challenging network model, due to its topography and service characteristics, the method demonstrated significant potential for improving network performance.