Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 7: Data analytics for dynamical systems I (Focus Session joint with DY and BP) (joint session SOE/BP/CPP/DY)
SOE 7.6: Topical Talk
Dienstag, 17. März 2020, 11:15–11:45, GÖR 226
Data driven modelling of spatio-temporal chaos in extended dynamical systems — •Ulrich Parlitz1,2, Sebastian Herzog1,3, Florentin Wörgötter3, Roland S. Zimmermann1,2, Jonas Isensee1,2, and George Datseris1 — 1Max Planck Institute for Dynamics and Self-Organization, Göttingen, Germany — 2Institut für Dynamik komplexer Systeme, Georg-August-Universität Göttingen, Germany — 3Drittes Physikalische Institut, Georg-August-Universität Göttingen, Germany
Many spatially extended nonlinear systems, an example being excitable media, exhibit complex spatio-temporal dynamics. We shall present machine learning methods to predict the temporal evolution of these systems or estimate their full state from limited observations. The applied techniques include Reservoir Computing [1] and a combination of a Convolutional Autoencoder with a Conditional Random Field [2,3], whose perfomance will be compared to Nearest Neighbours Prediction based on dimension reduced local states [4]. Examples for demonstrating and evaluating the methods employed include the Lorenz-96 model, the Kuramoto-Sivashinsky equation, the Barkley model, and the Bueno-Orovio-Cherry-Fenton model, describing cardiac (arrhythmia) dynamics.
[1] R. S. Zimmermann and U. Parlitz, Chaos 28, 043118 (2018)
[2] S. Herzog et al., Front. Appl. Math. Stat. 4, 60 (2018)
[3] S. Herzog et al., Chaos (to appear) (2019)
[4] J. Isensee, G. Datseris, U. Parlitz, J. of Nonlinear Sci. (2019)