Resumo:
The scarcity of natural resources, especially water and energy, put the population at risk of water supply. Therefore, they become necessary procedures to guarantee the optimal operation of a water distribution network. The use of optimization techniques such as genetic algorithms guarantees the optimal operation of the network, however, just finding the optimal points is not enough, since demand oscillations affect fluctuations in those presented, also presenting a learning capacity that guarantees the adaptability of the network. distribution of water according to demand, meeting the concept of smart cities. Thus, to give the water distribution network this learning capability, Artificial Neural Networks (ANN) of the “perceptron” type of a hidden layer were used, one of which is simple to act only at a demand point, predicting the values of the operational parameters (RNF, valves and pumps), as a proof of concept and then a more advanced ANN was developed, with seven hidden classes, in order to predict the operational specifications for the last 72 hours. The Water Distribution Networks used were two theoretical networks with 13 nodes, 2 Fixed Level Reservoirs (RNF), 2 pumps and 3 valves, varying only the mesh position. For both distribution networks, the predictions reached a good result in most points based on Norma NBR 12218/2017. As a conclusion, Artificial Neural Networks are proven to have a good capacity to predict operation when applied to water distribution networks due to their inherent complexity, having the potential for better results with future adjustments.