Department Systems Analysis, Integrated Assessment and Modelling

Neurale ODEs

Neurale ODEs Illustration

Deep-Learning-Methoden haben herkömmliche konzeptionelle Modelle bei der Modellierung von Niederschlägen und Abflüssen oft übertroffen, aber das Verständnis der internen Funktionsweise dieser Modelle bleibt eine Herausforderung. Während konzeptionelle Modelle klare Einblicke in hydrologische Prozesse bieten, fehlt bei Deep-Learning-Modellen diese Interpretierbarkeit. Um dieses Problem zu lösen, haben wir hydrologische neuronale Modelle mit gewöhnlichen Differentialgleichungen (ODEs) eingeführt, die die Vorhersagekraft von Deep Learning mit der Interpretierbarkeit von konzeptionellen Modellen kombinieren.
Bei neuronalen ODEs werden die internen Prozesse, die normalerweise durch Differentialgleichungen dargestellt werden, durch neuronale Netze ersetzt. Diese Verschmelzung ermöglicht es dem Deep Learning, mechanistische Modellierung zu integrieren, wodurch die interne Dynamik leichter zu interpretieren ist.

Team

Dr. Fabrizio Fenicia Tel. +41 58 765 5440 Send Mail
Marvin Höge

Marvin Höge

Publikationen

Improving hydrologic models for predictions and process understanding using neural ODEs

Deep learning methods have frequently outperformed conceptual hydrologic models in rainfall-runoff modelling. Attempts of investigating such deep learning models internally are being made, but the traceability of model states and processes and their interrelations to model input and output is not yet fully understood. Direct interpretability of mechanistic processes has always been considered an asset of conceptual models that helps to gain system understanding aside of predictability. We introduce hydrologic neural ordinary differential equation (ODE) models that perform as well as state-of-the-art deep learning methods in stream flow prediction while maintaining the ease of interpretability of conceptual hydrologic models. In neural ODEs, internal processes that are represented in differential equations, are substituted by neural networks. Therefore, neural ODE models enable the fusion of deep learning with mechanistic modelling. We demonstrate the basin-specific predictive performance for 569 catchments of the continental United States. For exemplary basins, we analyse the dynamics of states and processes learned by the model-internal neural networks. Finally, we discuss the potential of neural ODE models in hydrology.
Höge, M.; Scheidegger, A.; Baity-Jesi, M.; Albert, C.; Fenicia, F. (2022) Improving hydrologic models for predictions and process understanding using neural ODEs, Hydrology and Earth System Sciences, 26(19), 5085-5102, doi:10.5194/hess-26-5085-2022, Institutional Repository

Kontakt

Dr. Fabrizio Fenicia Tel. +41 58 765 5440 Send Mail