Data-driven Reduced Order Modeling for Reactive Transport in Nuclear Waste Repository Assessments

Abstract:

Dynamic mode decomposition (DMD) is a powerful data-driven technique for construction of reduced-order models (ROMs) of complex dynamical systems. We explore different ROM methodologies for the complex and coupled thermal, hydrological, mechanical and chemical (THMC) processes near engineered barrier systems (EBS). First, machine learning techniques are explored to reduce the hydrological and geochemical parameters space and identify the most significant observables in the system. The DMD algorithm, informed by the observables, is then used to construct a ROM to capture the temporal evolution of the distribution coefficient (Kd) of EBS as a function of the most relevant hydrological and geochemical parameters. Finally, the derived ROM for the time-varying distribution coefficient is provided for the overall performance assessment (PA) at a larger scale. This integration allows us to evaluate how the uncertainties in EBS parameters (at the scale of meters) could influence the overall performance of a repository (at the scale of kilometers), as well as to identify key parameters and time evolutions that dictate the overall performance.