企画セッション: リスクと不確実性の定量化

[2A13-21-01] Hierarchical Bayesian Inference for Parameter Uncertainty Quantification

Masaru Kitahara², *Takeshi Kitahara¹, Michael Beer² (1. Kanto Gakuin University, 2. Leibniz University Hannover)

Keywords:Hierarchical Bayesian, Stochastic Model Updating, Staircase Density function, Uncertainty Quantification

In the Bayesian approach, not a single set of model predictions but their probability distribution can be derived from the posterior distribution. However, the classical Bayesian approach generally relies on the assumption that the prediction error follows a Gaussian distribution. The hierarchical Bayesian approach, hence, has gained attention, where a specific distribution is assigned to model parameters and its hyper-parameters are updated. While the main issue of this approach is the choice of the distribution, we propose to employ the staircase density functions to flexibly approximate a wide of distributions. The proposed approach is demonstrated on numerical examples.