Calibração de rugosidade de tubulações de redes de distribuição de água, via método transiente inverso com aplicação de algoritmo genético

• Ano de defesa: 2009
• Autor(a) principal: Galiza, Roberto Von Paumgartten de
• Orientador(a): Araújo, John Kenedy de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Recursos hídricos; Distribuição de água; Algoritmos genéticos
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/17231

An important step in the modelling of hydraulic networks, calibration makes it possible to predict the behavior of physical parameters essential to decision-making. The present study used inverse transient analysis with a genetic algorithm to calibrate pipe roughness in a water distribution network with simulations based on variable parameters such as mutation, number of chromosomes, number of generations, number of nodes, transient time and type of demand variation (abrupt or smooth). Simulations were divided into cases, and each solution given by the computer model was evaluated by an objective function, based on the squared difference between estimated and actual results for the transient loads in the monitored node(s). An analysis of the solutions given by the model shows how the inverse method, the genetic algorithm and the choice of parameters influenced final results. A study of the cases revealed that for the network under study the efficiency of the method could not be improved by increasing isolated parameters of the genetic algorithm alone. In accordance with the inverse method employed, a set of roughness values was created by averaging the solutions found for the same parameters in ten successive runs of the genetic algorithm (seeds). The estimated hydraulic loads in the monitored node(s) were close to the actual loads, with calculated roughness values very similar to actual roughness values in some of the pipe sections. The efficiency of the method was evaluated by mean relative error (MRE) testing (comparing the mean relative errors of estimated and actual roughness values for each pipe). The best MRE result was 18.9%. The smallest relative error (1%) was observed in pipe 2, the biggest (59.1%) in pipe 10