A parallelized algorithm to speed up 1D free-surface flow simulations in irrigation canals

A parallelized algorithm to speed up 1D free-surface flow simulations in irrigation canals

Abstract A parallel algorithm for 1D free-surface flow simulations in irrigation canals is shown. The model is based on the Hartree method applied to Saint-Venant equations. Due to the close-to-steady flow nature in irrigation canals, external and internal boundary conditions are linearized to prese...

Full description

Saved in:
Bibliographic Details
Published in:Journal of hydroinformatics 2020-11, Vol.22 (6), p.1620-1639
Main Authors: Bessone, Lucas, Soler-Guitart, Joan, Gamazo, Pablo
Format: Article
Language:eng
Subjects:
Accuracy
Algorithms
Automation
Benchmarks
Boundary conditions
Canals
Cell size
Equations
Flow simulation
Flow velocity
Free surfaces
Hydraulics
Irrigation
Irrigation canals
Linearization
Methods
Parallel processing
Performance evaluation
Simulation
Simulators
Steady flow
Surface flow
Work platforms
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract A parallel algorithm for 1D free-surface flow simulations in irrigation canals is shown. The model is based on the Hartree method applied to Saint-Venant equations. Due to the close-to-steady flow nature in irrigation canals, external and internal boundary conditions are linearized to preserve the parallel character. Gate trajectories, off-take withdrawals, and external boundary conditions are modeled as piece-wise functions of time, so there are discontinuities. To achieve a fully parallelized algorithm, an explicit version of the Hartree method is chosen, and external and internal boundary conditions are linearized around operation point. This approach is used to build a computer simulator, written in C-CUDA language. Two tests by ASCE Committee on Canal Automation Algorithms have been used to evaluate accuracy and performance of the algorithm. The Maricopa Stanfield benchmark has been used to prove its accuracy, and the Corning Canal benchmark to evaluate performance in terms of processing time. Surprisingly, solving a 12 hr-long prediction horizon with a cell size of about Δx= 10 m is less than 1 s on a Nvidia K40 card. Results were compared with a serial and a multi-CPU version of the same algorithm. The implementation that showed the best performance on different platforms is the one that uses GPU.
ISSN:1464-7141
1465-1734