xpertsystems/oil003-sample

OIL-003 — Synthetic Reservoir Simulation Dataset (Sample Preview)

A free, schema-identical ~49K-row preview of the full OIL-003 commercial product from XpertSystems.ai.

Benchmark-calibrated synthetic reservoir simulation data spanning pressure decline, saturation evolution, well controls with Arps decline-curve production forecasts, and 5-scenario EOR sweep efficiency modeling — across 8 reservoir types and 5 drive mechanisms.

Designed for ML model development in production forecasting, depletion-stage classification, water-breakthrough prediction, EOR screening, and reservoir-health scoring.

---

What's in this sample

File	Rows	Columns	Description
reservoir_master.csv	10	15	Per-reservoir headers — basin, reservoir type, drive mechanism, fluid system, depth, pressure, temperature, area, thickness, initial saturations
grid_cells.csv	1,470	11	3D grid (7×7×3) per reservoir — porosity, perm tensor (kx, ky, kz), NTG, facies, transmissibility index
pressure_timesteps.csv	23,521	5	Cell-level pressure at each simulation timestep (16 timesteps × 1,470 cells)
saturation_timesteps.csv	23,521	6	Cell-level oil/water/gas saturation at each timestep (material-balance enforced)
well_controls.csv	161	9	Per-well BHP, oil/gas/water rates, injector rates, producer/injector roles
production_forecasts.csv	161	8	Cumulative oil/gas/water + EUR estimates per well per timestep
eor_scenarios.csv	51	7	5 EOR scenarios per reservoir — sweep efficiency, recovery factor uplift
reservoir_labels.csv	161	7	Depletion stage (early/mid/late life), water breakthrough flag, reservoir health score, recovery factor

Total: ~49,000 rows across 8 CSVs, ~3.2 MB.

---

Coverage

8 reservoir types — conventional sandstone, carbonate, tight oil, shale, offshore turbidite, naturally fractured, heavy oil, gas condensate

5 drive mechanisms — solution gas, water drive, gas cap, waterflood, combination drive

4 fluid systems — black oil, volatile oil, gas condensate, heavy oil

6 facies classes — clean sand, shaly sand, carbonate, tight sand, fractured zone, barrier shale

5 EOR scenarios per reservoir — none, waterflood, gas injection, polymer flood, CO2 injection

4 injection patterns — five-spot, line drive, peripheral, huff-and-puff

3 depletion stages — early life (P/Pi > 0.88), mid life (0.70–0.88), late life (< 0.70)

---

Calibration source story

The full OIL-003 generator is calibrated to reservoir engineering anchors drawn from:

• SPE Petroleum Engineering Handbook Vol. V (Reservoir Engineering) — porosity distributions, permeability ranges, recovery factor envelopes, net-to-gross conventions
• Craft, Hawkins & Terry — Applied Petroleum Reservoir Engineering (3rd ed.) — material balance, pressure decline physics, drive mechanism classifications
• Arps (1945) — hyperbolic decline curve analysis for well production forecasts
• SPE/WPC/AAPG/SPEE Petroleum Resources Management System (PRMS) — EUR estimation and reserves classification
• Lake — Enhanced Oil Recovery (1989) — EOR sweep efficiency baselines by recovery mechanism

The generator is benchmark-first by design — default parameters are centered on validation targets (porosity 0.18, horizontal permeability 185 mD, initial pressure 4800 psi, recovery factor 0.38, sweep efficiency 0.67), so primary calibration anchors are stable even at sample scale.

Sample-scale validation scorecard

Metric	Observed	Target	Tolerance	Status	Source
Avg reservoir porosity	0.192	0.18	±0.04	✅ PASS	SPE Hbk Vol. V
Avg horizontal permeability (mD)	208.2	185.0	±60.0	✅ PASS	SPE Hbk Vol. V
Avg initial pressure (psi)	4,918	4,800	±400	✅ PASS	Craft & Hawkins
Saturation balance error	2e-6	0.0	≤0.001	✅ PASS	Material balance (So+Sw+Sg=1)
Pressure decline consistency	1.000	≥1.000	±0.05	✅ PASS	Drive mechanism physics
Avg recovery factor	0.394	0.38	±0.08	✅ PASS	SPE/PRMS
Sweep efficiency mean	0.733	0.70	±0.10	✅ PASS	Lake, EOR (1989)
EOR scenario diversity	5	5	—	✅ PASS	OIL-003 schema
Reservoir type diversity	8	8	—	✅ PASS	OIL-003 schema
Production decline monotonicity	1.000	≥0.95	±0.05	✅ PASS	Arps (1945)

Grade: A+ (100/100) — verified across 6 random seeds (42, 7, 123, 2024, 99, 1).

---

Loading examples

Pandas — explore the reservoir master

import pandas as pd

res = pd.read_csv("reservoir_master.csv")
print(res[["reservoir_id", "basin", "reservoir_type", "drive_mechanism",
           "depth_ft", "pressure_initial_psi"]])

Pandas — pressure decline curve for one reservoir

import pandas as pd
import matplotlib.pyplot as plt

p = pd.read_csv("pressure_timesteps.csv")
p_avg = (p[p["reservoir_id"] == "RES-000001"]
         .groupby("simulation_day")["pressure_psi"].mean())
p_avg.plot(title="Average reservoir pressure vs simulation day")
plt.xlabel("Days"); plt.ylabel("Pressure (psi)")
plt.show()

Hugging Face Datasets

from datasets import load_dataset

ds = load_dataset("xpertsystems/oil003-sample", data_files={
    "reservoirs":  "reservoir_master.csv",
    "grid":        "grid_cells.csv",
    "pressure":    "pressure_timesteps.csv",
    "saturation":  "saturation_timesteps.csv",
    "wells":       "well_controls.csv",
    "forecasts":   "production_forecasts.csv",
    "eor":         "eor_scenarios.csv",
    "labels":      "reservoir_labels.csv",
})
print(ds)

Depletion-stage classifier baseline

import pandas as pd
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import train_test_split

labels = pd.read_csv("reservoir_labels.csv")
sat = pd.read_csv("saturation_timesteps.csv")
p = pd.read_csv("pressure_timesteps.csv")

sat_agg = sat.groupby(["reservoir_id", "simulation_day"])[
    ["oil_saturation", "water_saturation", "gas_saturation"]].mean().reset_index()
p_agg = p.groupby(["reservoir_id", "simulation_day"])["pressure_psi"].mean().reset_index()

feats = labels.merge(sat_agg, left_on=["reservoir_id", "timestep"],
                     right_on=["reservoir_id", "simulation_day"]) \
              .merge(p_agg, on=["reservoir_id", "simulation_day"])

X = feats[["oil_saturation", "water_saturation", "gas_saturation", "pressure_psi"]]
y = feats["depletion_stage"]
Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.25,
                                       stratify=y, random_state=42)
clf = GradientBoostingClassifier(random_state=42).fit(Xtr, ytr)
print(f"Depletion-stage classifier accuracy: {clf.score(Xte, yte):.3f}")

---

Suggested use cases

• Production forecasting — Arps-decline regression on well_controls + production_forecasts (oil/gas/water rates)
• Depletion-stage classification — 3-class (early/mid/late life) supervised learning on pressure + saturation features
• Water-breakthrough prediction — binary flag prediction from cell-level saturation evolution
• EOR scenario screening — which EOR method maximizes sweep efficiency given reservoir properties
• Recovery-factor regression — predict final RF from reservoir master + grid heterogeneity
• Reservoir-health scoring — regression target combining pressure ratio, oil saturation, and water cut
• 3D grid heterogeneity modeling — geostatistics, conditional simulation, upscaling experiments
• Decline-curve fitting — practice exponential/hyperbolic/harmonic fits against ground-truth Arps parameters
• Time-series modeling — depth-ordered LSTM/Transformer on per-reservoir simulation trajectories

---

Sample vs. full product

Aspect	This sample	Full OIL-003 product
Reservoirs	10	12,000+ (configurable)
Grid resolution	7×7×3 = 147 cells/res	120×120×40 = 576,000 cells/res
Time step	120 days	30 days
Simulation horizon	1,800 days (~5 yr)	3,650 days (10 yr)
Total cells	~1,470	Up to 5M (sampled from billions)
Total timesteps	~23K	Up to hundreds of millions
Schema	identical	identical
Calibration	identical	identical
License	CC-BY-NC-4.0	Commercial license

The full product includes higher grid resolution (120×120×40 vs 7×7×3), finer time steps (30 days vs 120), longer simulation horizons (10 yr vs 5 yr), and far more reservoirs (12K+ vs 10) covering all 8 basins and reservoir types at production scale. Contact us for the full product.

---

Limitations & honest disclosures

• Sample is preview-only. 10 reservoirs is enough to demonstrate schema, calibration anchors, and physical relationships, but is not statistically sufficient for production-grade model training. Use the full product for serious work.
• Grid resolution is much coarser than industry standard. This sample uses 7×7×3 cells per reservoir; commercial reservoir simulators (Eclipse, CMG, INTERSECT) use 100×100×40+ grids. The full product matches industry resolution.
• Time step is coarser than the generator's prod mode. This sample uses 120-day steps; the prod mode default is 30-day steps. As a result, water-breakthrough detection lands at ~1,250 days vs the generator's 920-day target — the underlying breakthrough physics are unchanged, but the temporal resolution at which we detect Sw crossing 0.42 is wider. We validated this with a structural saturation-balance metric (error < 1e-5) instead of an absolute breakthrough-day metric.
• Decline curves are Arps-only. The generator uses hyperbolic Arps decline (Arps, 1945) for production forecasts. Real wells often follow modified hyperbolic + exponential tail (Robertson 1988, Duong 2010), particularly for unconventionals. Use synthetic data for ML pretraining; tune final models against real production data.
• EOR sweep efficiency is parametric, not from compositional simulation. Sweep values come from EOR-type baselines (Lake 1989); they do not capture compositional/PVT effects of CO2 miscibility, polymer rheology, or surfactant-polymer interactions. Use for scenario screening and ranking, not absolute prediction.
• Faulting and fracture geometry are flag-only. faulted_flag and fractured_flag indicate presence; detailed fault/fracture network geometries are not in this dataset.
• No PVT tables. Fluid properties are categorical (fluid_system); explicit Bo/Bg/μo/Rs curves are not in this sample. The full product includes PVT tables.

---

Citation

If you use this dataset, please cite:

@dataset{xpertsystems_oil003_sample_2026,
  author       = {XpertSystems.ai},
  title        = {OIL-003 Synthetic Reservoir Simulation Dataset (Sample Preview)},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/xpertsystems/oil003-sample}
}

---

Contact

• Web: https://xpertsystems.ai
• Email: pradeep@xpertsystems.ai
• Full product catalog: Cybersecurity, Insurance & Risk, Materials & Energy, Oil & Gas, and more

Sample License: CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0) Full product License: Commercial — please contact for pricing.