DONATE

IAHR Document Library


« Back to Library Homepage « Proceedings of the 39th IAHR World Congress (Granada, 2022)

A Machine Learning Based Method for Fast Estimation of Damages Due to Failure of off-Stream Reservoirs

Author(s): Nathalia Silva Cancino; Fernando Salazar; Marcos Sanz-Ramos; Ernest Blade

Linked Author(s): Marcos Sanz Ramos

Keywords: Machine Learning; IBER; Off-stream reservoirs; Dam breach; Floods

Abstract: With the modification of the regulations of the Hydraulic Public Domain of Spain in 2008, around 70.000 owners of irrigation off-stream reservoirs are obliged to elaborate a classification assessment on the potential damages due to failure. The conventional procedure involves a comprehensive 2D hydraulic modelling, for which high engineering knowledge, efforts and resources are needed, which are often unavailable for many owners. In this work, a simplified methodology based on Machine Learning techniques is proposed to easily estimate the damages in case of failure of an off-stream reservoir. It is solely based on the physical characteristics of the structure and the surrounding terrain and avoids the need for a detailed 2D hydraulic model. With the parametrization of the main physical characteristics of off-stream reservoirs and surrounding areas, 800 synthetic cases were generated with the Latin Hypercube Sampling method. An automatized tool was developed for the Iber software to generate a database of breach scenarios, compute the flood propagation and extract results on 200 specific points (gauges) for each model. The results on every location in terms of the maximum depth and velocity were classified based on the three-level of damage defined by the Spanish Technical Guide for dam risk classification: mild, undefined and severe. The resulting data included both the physical characteristics of every case and the classification of the level of affection on every gauge. A machine learning classifier was fitted using a training set including 75% of the synthetic data. The model was later used to predict the damage level on an independent test set including the remaining 25% of the data. The simplified method presented an accuracy of 0.80 on the synthetic test set. It was also validated by its application to a real case of breach simulation on a Spanish irrigation reservoir, for which a detailed 2D hydraulic calculation was available. The simplified method provided accurate estimates of the damage level, especially in those areas classified as mild and severe. The results show that the simplified method can be a useful tool to owners and administrations for having a preliminary estimate on the potential damage due to failure of off-stream reservoirs that can be the base for prioritizing investments in safety.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221036

Year: 2022

Copyright © 2024 International Association for Hydro-Environment Engineering and Research. All rights reserved. | Terms and Conditions