Storage of CO2 in geological formations, such as oil and gas reservoirs, is considered an import means to reduce emissions of CO2 into the atmosphere. Accurate modeling of the CO2 migration is an important tool to analyse the risk of leakage in potential injection sites. To account for uncertainties in the geological model we need to run the simulations several times, with changes in the parameters, for the risk analysis. Along with several simulations we also need longer time scales than one normally has when simulation flow in porous medium, since the CO2 should stay underground a lot longer than the time it takes to extract fossil fuels. Because of this, we do not only require an accurate model, but also a fast model for simulation.Using the vertical equilibrium assumption, and a vertically integrated model, has been shown to give good performance benefits with respect to the full 3D models used in the petroleum industry today, as well as being an accurate model.
In this thesis we will investigate how the GPU can be used as an accelerator unit for such vertically integrated models.We will compare the performance gained from using a GPU accelerated solver and a multi core solver, with respect to the performance from a serial solver. From this we will demonstrate that the GPU is a good accelerator unit for these model.
The solvers will be demonstrated to scale well both on simple grids with no faults, as well as on real world data with faults and geological traps for the CO2. Lastly we will compare the error obtained on the GPU by using single precision floating point numbers instead of the double precision used on the CPU, and show that this error is negligible.