Speaker
Description
Flow and reactive transport in porous media are very important to improve our understanding of physical and chemical processes related to various geoscience and environmental applications such as enhanced geothermal systems, in-situ critical mineral and element recovery, unconventional resources recovery, and environmental fate and transport. One of the overarching challenges in improving prediction of flow and transport processes in porous media is how confidently we can estimate heterogenous permeability (and porosity) fields and associated parameters. Recent advances in machine learning (ML) involving advanced architectures and learning methods show promising results to enhance our ability to estimate heterogeneous subsurface properties and improve inverse modeling approaches. However, most of these ML methods have been evaluated with relatively simple synthetic cases. In this work state-of-the-art 3D tracer concentration datasets collected from non-reactive tracer transport experiments in a 3D sandbox setting using magnetic resonance imaging are utilized. Various ML workflows including Inverse physics-informed neural operator and ensemble smoother-multiple data assimilation approach with deep generative prior models are trained and evaluated to estimate 3D permeability fields and dispersivity distribution using spatio-temporal tracer concentrations in 3D sandbox experiments. These estimated fields with uncertainty quantification will be compared with traditional inverse modeling results. This work will provide outstanding benchmark datasets that can be used for validation of machine/deep learning approaches. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.
| Country | USA |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
