Prediction of water levels in open channels: Innovative numerical modeling using neural networks

This project aims to implement a deep neuronal networks architecture - Pinn (Physics Reported Neural Network) - as a relatively fast and low cost of numerical resolution of the 1D balance equations with the aim of predicting the dynamics involved in the water -free water surface (drafts) in mountain rivers. The modeling of the water -free sheet in mountain rivers, presents a challenge due to the particularities involved in relation to irregularity in geometry, as well as the variety of flow types present in the journey. The above emphasizes the importance of the development of different numerical techniques that allow predicting, in a relatively rapid and low costs, the phenomenon under study. This will allow analyzing and planning properly the consequences that are due to the variation of drafts in rivers incidible decisively in two sustainable development objectives (SDGs): clean water and sanitation, and action by climate. In recent years, Machine Learning (ML) techniques have shown great potential as substitute models to approximate the behavior of both artificial and natural systems. Pinn uses the concise information found in the differential equations in such a way that they act as a regularizing, limiting the response space of the neural networks and making them converge to the correct answer. This implies a lower data requirement with a physical interpretation in the results provided by Pinn. It is necessary to provide the numerical rigor necessary to the Pinn architecture to be implemented in this project, as part of the solvency analysis of the method in relation to: convergence rate, sensitivity and predictability.

XIX UNIVERSITY COMPETITION FOR RESEARCH PROJECTS