Geostat NF RATE
geostat No Fault Rate Control 3D
Property Type
geostat
Fault Config
No Fault
Well Control
Rate Control
Grid Dimensions
64 x 64 x 8
Train Samples
800
Val Samples
100
Test Samples
100
Models Evaluated
4
Dataset protocol
Current data provenance for this dataset.
These details come from the Arena protocol metadata and should be updated there when the simulation or preprocessing pipeline changes.
Simulation target
Simulator: OPM Flow
Physics: Two-phase oil-water flow
Reported outputs: Pressure (bar), Water saturation (fraction)
Split and geometry
Split rule: Sequential split by unique grid/property pair; all cases sharing the same grid and property sample stay in the same split.
Variable-throw cases are mapped from simulator CPG grids to a conformal 32-layer ML grid; static website flow figures show raw simulator targets on the CPG grid.
Results
Loss
8 results
| Sat (frac) | Pres (bar) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Model | Loss | Epochs | rel-L²▲ | MRE | MAE | rel-L² | MRE | MAE | Params |
| 1 | UNet3D | AbsLp(p=2) | 200 | 0.1436 | 0.2628 | 0.038677 | 0.0302 | 0.0260 | 5.741697 | 20.8M |
| 2 | SegResNet | AbsLp(p=2) | 200 | 0.1505 | 0.2850 | 0.037936 | 0.0310 | 0.0264 | 5.822663 | 18.8M |
| 3 | UNet3D | MSE | 200 | 0.1555 | 0.3080 | 0.039246 | 0.0313 | 0.0264 | 5.805879 | 20.8M |
| 4 | SegResNet | MSE | 200 | 0.1625 | 0.3186 | 0.038700 | 0.0333 | 0.0278 | 6.039331 | 18.8M |
| 5 | SwinUNETR | AbsLp(p=2) | 200 | 0.1881 | 0.3405 | 0.046272 | 0.0370 | 0.0310 | 6.744236 | 16.3M |
| 6 | SwinUNETR | MSE | 200 | 0.2012 | 0.3854 | 0.042536 | 0.0414 | 0.0334 | 7.141094 | 16.3M |
| 7 | FNO3D | AbsLp(p=2) | 200 | 0.2176 | 0.4239 | 0.050780 | 0.0422 | 0.0361 | 7.765929 | 20.1M |
| 8 | FNO3D | MSE | 200 | 0.2413 | 0.4747 | 0.051482 | 0.0472 | 0.0398 | 8.519495 | 20.1M |
Metrics Over Time
Water Saturation (SWAT)
Pressure
Validation vs Test
Each point is one run (model × loss variant). Points on the y=x diagonal generalize from val to test consistently; points far above it overfit val. Six panels per field cover rel-L2, MAE, and the four MRE / MAPE variants from the published literature so the choice of denominator is exposed rather than hidden.
Training Configuration
Configuration for the best-performing model (UNet3D).
Training
Epochs200
Learning Rate0.0001
Weight Decay0.01
Max Grad Norm1
Batch Size32
Loss FunctionCombinedGradientLoss
SchedulerStepLR (step_size=40, gamma=0.95)
Input Features
POROPERMXSWAT_0PRESSURE_0DEPTH
Output Fields
SWATPRESSURE
Model Architecture
in channels5
out channels40
base channels32
kernel size3
depth3
num time steps1
squashed outputfalse
flatten outputtrue