Services on Demand
Journal
Article
Indicators
- Cited by SciELO
- Access statistics
Related links
- Cited by Google
- Similars in SciELO
- Similars in Google
Share
Revista científica
Print version ISSN 0124-2253On-line version ISSN 2344-8350
Abstract
LADINO-MORENO, Edgar-Orlando; GARCIA-UBAQUE, César-Augusto and GARCIA-VACA, María-Camila. Leak Estimation in Pressure Pipes for Drinking Water Systems through Artificial Neuronal Networks and Epanet. Rev. Cient. [online]. 2022, n.43, pp.2-19. Epub Feb 17, 2022. ISSN 0124-2253. https://doi.org/10.14483/23448350.18275.
This work deals with the estimation of a leak for a main pipe system without branches. An algorithm and a neural network with 4 input variables are proposed, a hidden layer with 25 neurons and 3 output variables. The data was obtained through a nested loop in Visual Basic (Excel®) establishing 35,837 leak scenarios for a 30 m pipe that conducts water with a kinematic viscosity of 0.000001 (m2/s), a diameter equal to 0.15222 m, roughness of 0.0000015 m, pressure drop of 3.5 m and two accessories (k 1 , k 2 ) that add up to 1.5. Two flowmeters and two virtual pressure gauges were installed in the hydraulic system at the beginning and end of the pipeline. Also, Epanet® and Hydroflo® (Tahoe Design Software) are used to structure the hydraulic model and validate the initial data. Matlab R2021a was used to analyze the Backpropagation and Bayesian Regularization learning algorithms adopting the log sigmoid transfer function. The mean square error and the coefficient of determination R 2 were implemented as a control function. The neural model obtained presented a mean square error of 1.44E-06 and a relative error equal to 0.0055% for the training data. The cross-validation of the neural network was carried out from 5,973 independent input data.
Keywords : artificial neural network; drinking water; leakage; Levenberg-Marquardt; water resources management..