Descripción del proyecto
Mejora de la vigilancia y la respuesta a las epidemias con redes neuronales basadas en la física
Las redes neuronales basadas en la física (PINN, por sus siglas en inglés) integran a la perfección los datos y los modelos de la física matemática, de forma que permiten predecir una solución lejos de los puntos de datos experimentales incluso en escenarios complejos, inciertos y que se entienden parcialmente. Esta capacidad las hace especialmente adecuadas para predecir la evolución de una epidemia en varias escalas temporales y espaciales, lo cual facilita la elaboración de políticas eficaces. En el proyecto REACT, que cuenta con el apoyo de las Acciones Marie Skłodowska-Curie, se aprovecharán las PINN para integrar datos epidemiológicos del pasado con un modelo físico de propagación epidémica. Este método, combinado con las herramientas teóricas del sistema, permitirá la supervisión epidémica en bucle cerrado que aborda eficazmente las incertidumbres, al validar el algoritmo de estimación en tiempo real mediante la predicción de los datos futuros y adaptando el modelo epidémico en consecuencia.
Objetivo
The current COVID-19 crisis has highlighted the failure of existing epidemic monitoring techniques in timely predicting the epidemic situation and facilitating efficient policy recommendations. Because of being open-loop or linearization-based, these techniques cannot handle model and data uncertainties effectively. Designing a feedback mechanism to enable reliable, closed-loop epidemic monitoring is crucial but challenging because of the nonlinearity and heterogeneities of the epidemic spread process. The control mechanisms for epidemic mitigation are well-known, such as testing, lockdown, social distancing, etc. However, when, where, and to what extent should the health authority implement these policies depends on the accurate estimation and forecasting of the epidemic situation, which is very difficult with the classic observer design techniques. To alleviate the difficulties posed by these observers, an interdisciplinary approach of physics-informed neural network (PINN) in combination with system-theoretic tools is proposed in this project for closed-loop epidemic monitoring that can effectively cope with uncertainties. The task of PINN is to estimate the unknown nonlinearity (i.e. disease transmission rate) and epidemic parameters by using both the physics of epidemic spread (i.e. model) and the past epidemiological data. The closed-loop structure copes with the uncertainties and validates the estimation algorithm in real-time by predicting the future data and adjusting the epidemic model accordingly. The information received by the PINN-based observer will be utilized by the optimal controller to devise optimal policy recommendations under socio-economic constraints for epidemic mitigation. The closed-loop epidemic monitoring and control technique will be integrated to understand the geographic and demographic heterogeneities during epidemic outbreaks, which will significantly enhance the effectiveness of optimal policies.
Ámbito científico
Palabras clave
Programa(s)
- HORIZON.1.2 - Marie Skłodowska-Curie Actions (MSCA) Main Programme
Régimen de financiación
HORIZON-TMA-MSCA-PF-GF - HORIZON TMA MSCA Postdoctoral Fellowships - Global FellowshipsCoordinador
100 44 Stockholm
Suecia