Development of a hybrid Migha-ANN model for calibration of absolute roughness in water distribution networks

ABSTRACT

This work aimed to develop a new hybrid method for the calibration of absolute roughness in water distribution networks, starting from the combination of the Alternative Hydraulic Gradient Iterative Method (Migha) with a Multilayer Perceptron Artificial Neural Network (ANN). Different configurations were tested for the model, in different scenarios that differed in the number of observed pressures used. The results found suggest that the combination of methods was favorable to roughness calibration, with mean absolute errors lower than 0.016 mm in the model that calculated only one value per material for the parameter. The good performance was also evident when analyzing the pressures and flows returned by the calibrated network simulation, which showed average relative errors of 0.12 and 0.36%, respectively, in the scenario with 14% of nodes with known pressures, errors considerably lower than those obtained by the initial configuration of the network and the exclusive use of Migha.

Keywords:
water distribution networks; calibration; roughness; alternative hydraulic gradient iterative method; artificial neural network