Surrogate-based active learning for parameter inference in geosciences via Bayesian sparse² multi-adaptivity enhanced by information theory
Final Report Abstract
The outcome of this project addresses challenges that arise for uncertainty quantification and surrogateaided model calibration in the context of non-linear, convection-dominated problems. A key example is the modeling of carbon dioxide (CO2 ) storage in geological formations. The methodological key part of the project was to develop new, adaptive manners of surrogate construction. Ingredients to the methodological development were the arbitrary polynomial expansion, adaptive refinement into local stochastic elements, combinations with Gaussian processes, fully Bayesian formulations that automatically induce sparsity, and active-learning strategies based on information-theoretic criteria. With this, we completed all planned method developments and went beyond originally planned project goals. All developed techniques were tested on a CO2 sequestration benchmark. In particular, the regularizations, adaptive refinements, active learning strategies and sparsity concepts lead to an improvement of surrogate accuracy and surrogate robustness at smaller computational costs for training. As final proof of success, we chose a tailored surrogate modeling approach from within our new developments, hybridized it with a deterministic, optimization-based technique for model calibration, and then performed a Bayesian parameter inference for a large-scale model that describes CO2 sequestration in the real-world Ketzin pilot site. All codes and data are provided in openly accessible repositories according to FAIR principles.
Publications
-
Bayesian3 Active Learning for the Gaussian Process Emulator Using Information Theory. Entropy, 22(8), 890.
Oladyshkin, Sergey; Mohammadi, Farid; Kroeker, Ilja & Nowak, Wolfgang
-
Uncertainty quantification using Bayesian arbitrary polynomial chaos for computationally demanding environmental modelling: conventional, sparse and adaptive strategy, Computational Methods in Water Resources XXIII, Stanford, USA, December 14-17, 2020
S. Oladyshkin, F. Beckers, I. Kröker, F. Mohammadi, A. Heredia, M. Noack, B. Flemisch, S. Wieprecht, W. Nowak
-
Arbitrary multi-resolution multi-mavelet-based polynomial chaos expansion for data-driven uncertainty quantification, InterPore German Chapter Meeting, February 2, 2021
I. Kröker, S. Oladyshkin
-
Arbitrary multi-resolution polynomial chaos expansion for uncertainty quantification in geoscientific applications, UNCECOMP 2021, 4th International Conference on Uncertainty Quantification in Computational Sciences and Engineering, 28-30 June 2021, Streamed from Athens, Greece
I. Kröker, S. Oladyshkin
-
Uncertainty quantification and global sensitivity analysis for coupled porous-medium and free-flow problems, InterPore German Chapter Meeting, February 2, 2021
I. Kröker, I. Rybak, S. Oladyshkin
-
Arbitrary multi-resolution multi-wavelet-based polynomial chaos expansion for data-driven uncertainty quantification. Reliability Engineering & System Safety, 222, 108376.
Kröker, Ilja & Oladyshkin, Sergey
-
Bayesian parameter inference on arbitrary multi-resolution polynomial chaos expansion based surrogate models, SIAM Conference on Uncertainty Quantification (UQ22), April 12 - 15, 2022, Atlanta, Georgia, U.S. (hybrid)
I. Kröker and S. Oladyshkin
-
Optimal Exposure Time in Gamma-Ray Attenuation Experiments for Monitoring Time-Dependent Densities. Transport in Porous Media, 143(2), 463-496.
Gonzalez-Nicolas, Ana; Bilgic, Deborah; Kröker, Ilja; Mayar, Assem; Trevisan, Luca; Steeb, Holger; Wieprecht, Silke & Nowak, Wolfgang
