Dresden 2026 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 6: Machine Learning in Dynamics and Statistical Physics I

DY 6.13: Vortrag

Montag, 9. März 2026, 12:45–13:00, HÜL/S186

POD-Subspace Reconstruction of Convective Reversal Dynamics from Limited Sensor Data — •Tim Kroll and Oliver Kamps — CDSC ,University of Münster

We introduce a data-driven modelling framework that leverages a hybrid LSTM-neural-network architecture to capture convection reversals from limited time-series measurements.. The method operates entirely in POD space, enabling efficient and accurate reconstruction of complex dynamical systems from limited observations by modelling non-orthogonal modes as a superposition of POD modes. The corresponding dynamics are modelled by an LSTM, incorporating knowledge about the history of the timeseries. We demonstrate its effectiveness on convection processes, showing that measurements from a single sensor - of either temperature T or velocity V - are sufficient to recover the full spatiotemporal dynamics, consisting of temperature, velocity or a combination of both, within the reduced representation. Furthermore this approach has potential to be applied in different scientific fields detached from convection or fluid dynamics.

Keywords: Machine Learning; Convection; Complex Systems; Neural Networks