Regensburg 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 4: Focus Session: Nonlinear Dynamics and Stochastic Processes – Advances in Theory and Applications I

DY 4.6: Vortrag

Montag, 17. März 2025, 11:00–11:15, H43

Which methods are best suited for predicting chaotic time series? — •Ulrich Parlitz — Max Planck Institute for Dynamics and Self-Organization, Göttingen, Germany

Since the pioneering work in the 1980s on approximating the time evolution of dynamical systems using delay embedding, many methods for predicting univariate and multivariate chaotic time series have been proposed and published. Recently, the prediction of chaotic time evolution has also been used to demonstrate the performance of novel machine learning algorithms, but in many cases only low-dimensional chaos is considered, as generated by the classical Lorenz-63 system. In this talk, different prediction methods will be contrasted and compared in terms of their prediction power and complexity when applied to low- and high-dimensional chaotic time series.

Keywords: chaotic dynamics; nonlinear time series analysis; machine learning; reservoir computing; delay reconstruction