Regensburg 2025 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 21: Poster Heterogeneous Catalysis
O 21.2: Poster
Montag, 17. März 2025, 18:00–20:00, P2
Bayesian Inference of Kinetic Models of Heterogeneous Catalysis by Normalizing Flows — •Andreas Panagiotopoulos1, Javed Mudassar2, Jens-Uwe Repke2, Georg Brösigke2, and Sebastian Matera1 — 1Fritz-Haber-Institut der MPG, Berlin — 2Technical University Berlin
Estimating kinetic parameters is typically done by classical fitting a model to experimental reactor data, which, however, suffers from a number of fundamental problems like ill-posedness, multiple possible solutions and the lack of reliable uncertainty estimates. By reformulating the problem in a probabilistic language, Bayesian inference cures these problems, but also requires to sample from a high-dimension probability distribution. Because of their high non-linearity and sensitivity, this becomes challenging for kinetic models and established sampling approaches become inefficient. We investigate Normalizing Flows in conjunction with Quasi Monte Carlo sampling to address this problem. In this approach, a bijective nonlinear parameter transformation is sequentially learned such that a uniform sampling from the transformed parameters leads to a good importance sampler of the Bayesian posterior. We investigate the performance of the proposed approach on an empirical model for methanol synthesis on Cu based catalysts using synthetic and experimental data.
Keywords: Bayesian inference; Normalizing Flows; parameter estimation; uncertainty quantification; kinetic model