Avaliação da codificação real e binária na identificação das rugosidades de tubulações aplicando algoritmo genético com o método transiente inverso

Abstract

Scorned for years, either because in the past water distribution networks were simpler or because the study was too complex, the study of the hydraulic transient became of great importance due to the risk it caused to its consumers with the effect of the water hammer. With the population increasing and always needing water for their consumption, the water distribution networks are constantly growing, and with this the increase of pumps and valves is necessary to reach the minimum and maximum pressure values, in addition to the quantity and quality of water. To analyze a transient hydraulic system, it is necessary to obtain a model that represents the mathematical description of its characteristics. The analysis of the transient flow incorporates an improvement in the model, an economically more viable operation and possibly a lower operating cost in the distribution networks. However, in order to simulate the network behavior with good precision, the physical parameters that involve the system must be known, resulting in pressure values and flow rates very close to the real one. The process of estimating the physical parameters of the hydraulic network is called calibration. Among the network parameters, the most difficult to specify is the absolute roughness due to industrial processes and the degree of surface finish, age of the pipe, etc. Predicting its modification over time, due to the alteration of the wall surface, often caused by the water quality itself, poses the designer before the difficult problem of determining the friction factors of the pipes, requiring experience and common sense. Using the genetic algorithm, with elitism, mutation and crossing operators in their real and binary representation, in the inverse transient method, in which the parameters are considered as variables, a comparison was obtained between the objective functions of the genetic algorithm and the average errors of both representations. Two networks were analyzed and in both cases, values close to the desired roughness were obtained, but with the real representation obtaining better values.