BA Lars Jose
Bayesian parameter identification and uncertainty quantification
This thesis deals with the offline identification and uncertainty quantification of unknown model parameters from measured input and output trajectories. With regard to multi-disciplinary industrial development projects, only numerical methods are considered, which are suitable for the parameter identification of black-box simulation models. Since prior knowledge about the distribution of the sought parameter values is often available in the industry, Bayesian approaches are the main focus of this thesis. Parametric and non-parametric methods for the estimation of the likelihood function are introduced, as this is a crucial step in
Bayesian parameter identification and uncertainty quantification. The Bayesian approaches are compared with an established maximum-likelihood parameter estimator and uncertainty quantification based on the Fisher information matrix, in theory and numerical studies.