zur Startseite

Research - Dept. of Hydromechanics and Modelling of Hydrosystems

Probabilistic Risk Assessment for CO2 Storage Scenarios via Massive Stochastic Model Reduction
Project manager:Dr. Sergey Oladyshkin, Prof. Dr.-Ing. Wolfgang Nowak, M.Sc.
Deputy:apl. Prof. Dr.-Ing. Holger Class, Prof. Dr.-Ing. Rainer Helmig
Research assistants:Dr. Sergey Oladyshkin
Duration:1.1.2009 - 31.12.2011
Funding:SimTech Cluster of Excellence

This project is part of the research area:
Modeling of multiphase-multicomponent processes for the sequestration of CO2 in the subsurface

Publications: Link


Large-scale industrial CO2 injection into deep geologic formations bears an inherent risk of leakage back into atmosphere. The potential of CO2 injection as large-scale interim solution will vastly depend on our ability to quantify its uncertainties and risks. Up to date, field experience is limited to medium-scale test sites, and no probabilistic risk assessment has been applied. Current numerical simulation models are inadequate for stochastic simulation techniques, because they are too expensive for stochastic approaches based on repeated simulation. Even single deterministic simulations require parallel high performance computing. Because the involved multiphase flow processes of CO2 in porous media have a significantly nonlinear character, the problem is too non-linear for quasi-linear and other simplified stochastic tools. In the proposed approach, stochastic simulation and probabilistic risk assessment of CO2 storage scenarios is based on massive stochastic model reduction via polynomial chaos expansion in variable model parameters. The model response surface is projected onto a higher-order orthogonal basis of polynomials, allowing for non-linear propagation of model uncertainties onto the predicted risk. The variable parameters include uncertain model parameters, such as porosity, permeability, temperature and geometry, and a list of design and control parameters. The chosen degree of the polynomial balances between computational effort and accuracy. After an initial computational effort for model reduction, the reduced model is vastly faster than the original. Probabilistic risk assessment can then be performed at ease. The same reduced model will aid in follow-up tasks, such as the optimization of site exploration and engineering design. The long-time goal is application to site management, including real-time control.