Benchmark

Note

These are preliminary results and will be updated up to the final submission deadline (September 20th 2024, see here). For example, new simulations for the spe11a r4 and spe11c r4 are still running to handle the mass issue and to include the 1000 years of initialization time, respectively.

All configuration files are located in the benchmark folder.

SPE11A

  • r1_Cart_1cm

    Uniform grid of 1 cm size

  • r2_Cart_1cm_capmax2500Pa

    Uniform grid of 1 cm size (maximum capillary pressure of 2500 Pa instead of 95000 Pa following the remarks in the benchmark description)

  • r3_cp_1cmish_capmax2500Pa

    Corner-point grid of ca. 1 cm size (maximum capillary pressure of 2500 Pa instead of 95000 Pa following the remarks in the benchmark description)

  • r4_Cart_1mm_capmax2500Pa

    Uniform grid of 1 mm size (maximum capillary pressure of 2500 Pa instead of 95000 Pa following the remarks in the benchmark description)

To run the cases in the terminal:

pyopmspe11 -i r1_Cart_1cm.txt -o r1_Cart_1cm -m all -g all -t 1 -r 280,1,120 -w 0.16666666666666666
pyopmspe11 -i r2_Cart_1cm_capmax2500Pa.txt -o r2_Cart_1cm_capmax2500Pa -m all -g all -t 1 -r 280,1,120 -w 0.16666666666666666
pyopmspe11 -i r3_cp_1cmish_capmax2500Pa.txt -o r3_cp_1cmish_capmax2500Pa -m all -g all -t 1 -r 280,1,120 -w 0.16666666666666666
pyopmspe11 -i r4_Cart_1mm_capmax2500Pa.txt -o r4_Cart_1mm_capmax2500Pa -m all -g all -t 1 -r 280,1,120 -w 0.16666666666666666

As mentioned in the CSP description, using the maximum value of 2500 Pa instead of 95000 Pa does not significantly impact the results (comparing r1 and r2), and this choice also reduces the simulation time. In addition, the corner-point grid results (r3) compare very well to the fine-scale simulations (r4).

Performance data

_images/benchmark_spe11a_performance.png

Sparse data

_images/benchmark_spe11a_sparse_data.png

Spatial maps

_images/massfracta.png

SPE11B

  • r1_Cart_10m

    Uniform grid of 10 m size (1 m dx size on left and right boundaries)

  • r2_cp_10mish

    Corner-point grid of ca. 10 m size (1 m dx size on left and right boundaries)

  • r3_cp_10mish_convective

    Corner-point grid of ca. 10 m size (1 m dx size on left and right boundaries) using a subgrid model for convective mixing for facies 2 and 5.

  • r4_Cart_1m

    Uniform grid of 1 m size

To run the cases in the terminal:

pyopmspe11 -i r1_Cart_10m.txt -o r1_Cart_10m -m all -g all -r 840,1,120 -t 5 -w 0.1
pyopmspe11 -i r2_cp_10mish.txt -o r2_cp_10mish -m all -g all -r 840,1,120 -t 5 -w 0.1
pyopmspe11 -i r3_cp_10mish_convective.txt -o r3_cp_10mish_convective -m all -g all -r 840,1,120 -t 5 -w 0.1
pyopmspe11 -i r4_Cart_1m.txt -o r4_Cart_1m -m all -g all -r 840,1,120 -t 5 -w 0.1

For the box A quantities, the convective model results (r3) compare very well to the fine-scale simulations (r4), which runs ca. 500 times faster. Details on the convective model will be available in Mykkeltvedt et al., under review. For the implementation in OPM Flow of this model, it is work in progress to handle better the zones in the geological model where dissolve CO2 accumulates.

Performance data

_images/benchmark_spe11b_performance.png

Sparse data

_images/benchmark_spe11b_sparse_data.png

Spatial maps

_images/massfractb.png

SPE11C

  • r1_Cart_50m-50m-10m

    Grid of [50, 50, 10] m size (1 m dx size on left and right boundaries and 1 m dy size on back and front boundaries)

  • r2_cp_50m-50m-8mish

    Corner-point grid of [50, 50, mean ca. 8] m size (1 m dx size on left and right boundaries and 1 m dy size on back and front boundaries)

  • r3_cp_50m-50m-8mish_convective

    Corner-point grid of [50, 50, mean ca. 8] m size (1 m dx size on left and right boundaries and 1 m dy size on back and front boundaries) using the convective model for facies 2 and 5.

  • r4_cp_8m-8mish-8mish

    Corner-point grid of [8, mean ca. 8, mean ca. 8] m size (1 m dx size on left and right boundaries and 1 m dy size on back and front boundaries)

To run the cases in the terminal:

pyopmspe11 -i r1_Cart_50m-50m-10m.txt -o r1_Cart_50m-50m-10m -m all -g all -r 168,100,120 -t 0,5,10,15,20,25,30,35,40,45,50,75,100,150,200,250,300,350,400,450,500,600,700,800,900,1000 -w 0.1
pyopmspe11 -i r2_cp_50m-50m-8mish.txt -o r2_cp_50m-50m-8mish -m all -g all -r 168,100,120 -t 0,5,10,15,20,25,30,35,40,45,50,75,100,150,200,250,300,350,400,450,500,600,700,800,900,1000 -w 0.1
pyopmspe11 -i r3_cp_50m-50m-8mish_convective.txt -o r3_cp_50m-50m-8mish_convective -m all -g all -r 168,100,120 -t 0,5,10,15,20,25,30,35,40,45,50,75,100,150,200,250,300,350,400,450,500,600,700,800,900,1000 -w 0.1
pyopmspe11 -i r4_cp_8m-8mish-8mish.txt -o r4_cp_8m-8mish-8mish -m all -g all -r 168,100,120 -t 0,5,10,15,20,25,30,35,40,45,50,75,100,150,200,250,300,350,400,450,500,600,700,800,900,1000 -w 0.1 -u opm

To run the case with more than 100 million cells (r4), it required improvements in the OPM Flow simulator (e.g., METIS support), as well as in the pyopmspe11 pre- and postprocessing functionality (and of course a big computer, the NORCE HPC cluster). See this gif for visualization in ParaView of the CO2 molar fraction (liquid phase) over time.

Performance data

_images/benchmark_spe11c_performance.png

Sparse data

_images/benchmark_spe11c_sparse_data.png

Spatial maps

_images/massfractc.png