Dataset · Zenodo · May 18, 2026

Reference dataset and MRST simulation script for "Entropy-Satisfying Physics-Informed Neural Networks for the Two-Dimensional Buckley–Leverett Equation"

Timur Imankulov, Yerzhan Kenzhebek, Samson Dawit Bekele, Danila Stolyarov, Darkhan Akhmed-Zaki, Panfilov Irina

Abstract

This deposit contains the reference datasets and the MRST simulation scripts used to generate them for the paper Entropy-Informed Physics-Informed Neural Networks for the Two-Dimensional Buckley–Leverett Equation. Simulation setup. Quarter five-spot waterflood on a 1000 m × 1000 m homogeneous, isotropic reservoir, solved with the implicit-pressure explicit-saturation (IMPES) operator-splitting scheme: incompTPFA for the pressure step, explicitTransport for the saturation step. The injector is rate-controlled (500 m³/day, water composition) at one corner of the domain; the producer is bottom-hole-pressure-controlled (100 bar) at the opposite corner. Quadratic Corey relative-permeability curves with no connate water or residual oil. Grid: 100 × 100 cells. Total simulated time: 500 days = 1.25 pore volumes injected, stored in 500 timesteps of 1 day each (plus the t = 0 state). Two cases are provided, differing only in oil viscosity: 1 cP (mobility ratio M = 1) and 3 cP (M = 3). Full parameter list in Table 1 of the paper. Contents. bl2d.m (M = 1) and m3_bl2d.m (M = 3) - the MATLAB / MRST scripts that run the simulations and write the CSV files. dataset.csv and m3_dataset.csv - per-cell water saturation and pressure for each of 501 stored timesteps (5 010 000 data rows per case; columns: x, y, t, P, Sw). well_cells.csv and m3_well_cells.csv - cell indices and coordinates of the injector and producer. well_report.csv and m3_well_report.csv - per-timestep bottom-hole pressures, volumetric rates and cumulative volumes for both wells. README.md - full documentation: file schemas, units, run instructions and physical parameters. Requirements to reproduce. MATLAB R2019a or later and MRST (MATLAB Reservoir Simulation Toolbox, https://www.sintef.no/projectweb/mrst/), incomp module only. The M = 1 files were produced with MRST 2024a and the M = 3 files with MRST 2026a. Licences. Code (bl2d.m, m3_bl2d.m): MIT. Data (all .csv files): CC-BY 4.0. Reuse, including for commercial purposes, is permitted under these licences, provided the accompanying paper is cited. Citation. Imankulov, T., Kenzhebek, Y., Bekele, S. D., Stolyarov, D., Akhmed-Zaki, D., & Panfilova, I. Entropy-Informed Physics-Informed Neural Networks for the Two-Dimensional Buckley–Leverett Equation (manuscript submitted for publication).

Model Reduction and Neural NetworksReservoir Engineering and Simulation MethodsEnhanced Oil Recovery TechniquesBuckley-Leverett equationphysics-informed neural networkswaterflooding

Reference dataset and MRST simulation script for "Entropy-Satisfying Physics-Informed Neural Networks for the Two-Dimensional Buckley–Leverett Equation"

Abstract

More by Samson Dawit Bekele