Description du projet
Une nouvelle manière de calculer le comportement des tempêtes
La prédiction d’événements météorologiques graves comme de fortes précipitations ou des tempêtes est importante, car elle laisse aux communautés le temps de mieux se préparer. Elle fournit également des informations essentielles à l’industrie du transport aérien en permettant d’anticiper les retards et les mesures à prendre. Le projet PERSEVERE, financé par l’UE, examine l’utilisation des réseaux neuronaux, en particulier les réseaux de neurones informés par la physique (PINN pour «physics-informed neural network») qui combinent des modèles physiques et l’apprentissage axé sur les données, afin de prédire le statut des fluides de l’atmosphère dans des régions qui affectent le trafic aérien. PERSEVERE s’intéressera également au développement de réseaux antagonistes génératifs spéciaux pour améliorer et compléter le résultat des PINN en stimulant la précision de la prédiction et en intégrant des variables supplémentaires comme la température et l’humidité de l’air.
Objectif
The novel development of Physics-Informed Neural Networks (PINNs), which incorporate the constraints given by physics laws into the training process, as excellent means of computing fluidic fields and their characteristics such as velocity and pressure has open the gates to numerous applications. One of them is the data enhancement of experiments, since PINNs can reconstruct by means of applying the Navier-Stokes equations as loss function the full fluid domain in areas where experiments are limited by technology. On the other hand, the development of Generative Adversarial Networks (GANs) as robust networks with excellent precision but excessive computational costs leaves the door open to further investigate new applications where PINNs and GANs can be combined to amplify their strengths and reduce their weak points. One of those applications regards the forecast of severe weather conditions, where PINNs are useful to compute the fluidic behavior of storms approaching a certain location, whereas GANs can incorporate many additional parameters, such as wind speed, humidity, temperature and electric content, which may be essential to determine if in the following 48h a certain location is going to suffer from severe weather conditions.
The estimation and forecast of storms is essential to the air transport industry, since the losses incurred due to delays and deviations of air traffic caused by the presence of storms have been reported to be over $38.5 billion in USA. The development of a computational architecture which is able to determine if severe weather events are going to take place within the next 48h is therefore of crucial importance. There exists no model nor application in which the combined strengths of PINNs and GANs have been put into practice, one to rapidly estimate the fluidic behaviour of a moving storm, the second to calculate the properties of the field with high precision.
Champ scientifique
Programme(s)
- HORIZON.1.2 - Marie Skłodowska-Curie Actions (MSCA) Main Programme
Régime de financement
HORIZON-TMA-MSCA-PF-EF - HORIZON TMA MSCA Postdoctoral Fellowships - European FellowshipsCoordinateur
28903 Getafe (Madrid)
Espagne