Estudo da calibração de redes de distribuição de água submetidas ao regime transiente utilizando algoritmos genéticos e diferentes funções objetivo

Abstract

In this dissertation it was carried out on how the calibration of the roughness of pipes of water distribution networks with different objective functions takes place. It is widely known the difficulties that are experienced when there is no knowledge of the physical parameters of a network, especially when it is subject to transient regimes. When applying Genetic Algorithms in two networks submitted to a transient with different objective functions, it was aimed to obtain parameters, such as permanent and transient hydraulic flows and loads, to apply and obtain the values of three objective functions previously selected and modified to better represent the objectives of the study, when used to calibrate the roughness from nodes and selected sections of the studied networks, the best three solutions of each one being analyzed. Two hypothetical networks with known parameters were used here, applying a hydraulic transient by demand variation. The three objective functions chosen represent the quadratic difference between the actual and simulated loads (FO1), the quadratic difference between the actual and simulated flows (FO2) and the two previous differences summed using individual weights for each (FO3). Four nodes were selected in each of the nets, each with a striking characteristic, being the largest and the smallest distance from the node to the supply reservoir of the network, a node with a large number of tubes connected to it and a node that was at the end of the net. The results show that FO1 obtained a more important performance in 3 of the 4 groups analyzed, when considering the Relative Mean Error (RMS) in the roughness calibration, besides the values of hydraulic load and flow in permanent and transient regime, as well it can be seen that the hydraulic load is better shown to represent these points studied. When looking at the three best solutions of each objective function, it is noted that there is no direct relationship between the lower function value and the production of a smaller error, especially in the roughness calibration, besides that the roughness error is considerably higher than the hydraulic load, for example. It is noted that the application of the selected objective functions still cause a considerable RMS in the roughness calibration, even if this error does not propagate through the load, then it is to be noted that changes need to be made in these functions to better represent the reality, although show a good beginning for the study and that the advantages that the genetic algorithm brings to the calibration processes are always positive and should be explored.