Datasets:

xpertsystems
/

enr004-sample

license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
  - time-series-forecasting
tags:
  - synthetic-data
  - oil-and-gas
  - upstream
  - production
  - decline-curve
  - arps
  - arps-decline
  - type-curve
  - well-economics
  - reserves
  - eur
  - spe-prms
  - api-production-reporting
  - epa-subpart-w
  - methane-emissions
  - ghg-emissions
  - pvt
  - standing-correlation
  - vasquez-beggs
  - reservoir-engineering
  - petroleum-engineering
  - permian
  - eagle-ford
  - bakken
  - appalachian
  - marcellus
  - gulf-of-mexico
  - shale
  - unconventional
  - horizontal-drilling
  - hydraulic-fracturing
  - esp
  - rod-pump
  - gas-lift
  - wti
  - henry-hub
  - commodity-prices
  - ornstein-uhlenbeck
  - royalty
  - working-interest
  - net-revenue-interest
  - nri
  - gor
  - water-cut
  - bsw
  - bottomhole-pressure
pretty_name: ENR004 — Synthetic Upstream Oil & Gas Production Dataset (Sample)
size_categories:
  - 10K<n<100K
configs:
  - config_name: production
    data_files: enr004_production_data.parquet
  - config_name: wells
    data_files: enr004_wells_metadata.parquet
  - config_name: fields
    data_files: enr004_fields_metadata.parquet
  - config_name: facilities
    data_files: enr004_facilities.parquet

ENR004 — Synthetic Upstream Oil & Gas Production Dataset (Sample Preview)

XpertSystems.ai | Synthetic Data Factory | Energy & Climate Vertical

A four-table, physics-calibrated upstream oil & gas production dataset spanning 5 US basins (Permian, Eagle Ford, Bakken, Appalachian/Marcellus, GOM Offshore) with daily-resolution well-level production timeseries, Arps decline curve physics, Standing (1947) and Vasquez-Beggs (1980) PVT correlations, Ornstein-Uhlenbeck commodity prices, and EPA Subpart W methane intensity. Calibrated benchmark-first against SPE Petroleum Resources Management System (PRMS), API production reporting, EPA Subpart W, and EIA Drilling Productivity Reports.

This is the sample preview — 50 wells × 8 fields × 2 years × daily (~~15K production records). The full product covers 5,000 wells × 50 fields × 5 years (~~9M rows) with full geologic heterogeneity, complete basin/fluid/lift coverage, and 500 surface facilities.

Dataset summary

Table	Rows (sample)	What it contains
`production_data`	~15,699	Daily well-level production: oil/gas/water rates, GOR, WOR, BSW, cumulative volumes, reservoir/wellhead/bottomhole pressures, ESP/rod-pump/gas-lift parameters, WTI/Henry Hub prices, oil/gas revenues, LOE, netback, well status, methane/CO2/H2S/flare emissions, spill flags
`wells_metadata`	50	Per-well metadata: API number, basin, fluid type (Black Oil / Volatile Oil / Gas Condensate / Dry Gas / CBM), lift type (ESP / Rod Pump / Gas Lift / Plunger / Natural Flow), completion type (Horizontal / Deviated / Vertical), lateral length, perforation stages, proppant/fluid volumes, skin, PI, IP30/IP90, Arps decline parameters, EUR, working interest, royalty rate, NRI
`fields_metadata`	8	Reservoir block metadata: basin, formation, initial pressure, reservoir temperature, permeability, porosity, net pay, API gravity, sulfur content, bubble point, oil viscosity, FVF, OOIP/OGIP, drive type, aquifer strength
`facilities`	15	Surface facilities: GPF, separator trains, tank batteries, compressor stations, LACT units, saltwater disposal — with treating capacities, separator T/P, uptime, throughput utilization, meter factors

All four tables are provided in both CSV and Parquet. They join via field_id (production ↔ wells ↔ fields ↔ facilities).

Calibration sources

All ten validation metrics target named industry sources, not generator self-metrics:

SPE PRMS (Petroleum Resources Management System) — economic identities (NRI = WI × (1 - royalty)), reserves classification
Arps (1945) Decline Curve Analysis — hyperbolic/exponential/harmonic decline equations
Standing (1947) PVT correlation — solution GOR vs. pressure
Vasquez-Beggs (1980) FVF correlation — oil formation volume factor
API Production Reporting — well numbering (14-digit API), status codes, GOR/WOR bounds
EPA Subpart W — upstream methane emissions intensity
EIA Drilling Productivity Report — basin-level lateral lengths, completion statistics, fluid mix
EIA Spot WTI 2019-2024 — commodity price ranges for OU calibration
Ornstein-Uhlenbeck process — mean-reverting commodity price model

Validation scorecard (seed = 42)

10/10 PASS · Grade A+ (100%) across all six canonical seeds (42, 7, 123, 2024, 99, 1).

#	Metric	Observed	Target	Tol	Type	Source
1	`nri_structural_match_rate`	1.000	0.99	±0.01	FLOOR	SPE PRMS
2	`pressure_structural_fbhp_le_reservoir`	1.000	0.99	±0.01	FLOOR	Reservoir engineering
3	`cumulative_oil_monotonic_per_well_rate`	1.000	0.99	±0.01	FLOOR	Mass balance
4	`arps_decline_observed_per_well_rate`	1.000	0.95	±0.05	FLOOR	Arps (1945)
5	`api_gravity_mean_degrees`	38.81	38.0	±4.0	two-sided	EIA crude grades
6	`horizontal_lateral_length_mean_ft`	8,160	8,500	±2,000	two-sided	EIA DPR
7	`methane_oil_intensity_pct`	0.835	0.85	±0.30	two-sided	EPA Subpart W
8	`wti_price_mean_usd_per_bbl`	75.90	68.0	±20.0	two-sided	EIA WTI 2019-2024
9	`oil_revenue_non_negative_rate`	1.000	0.99	±0.01	FLOOR	Economics integrity
10	`gor_in_industry_bounds_rate`	1.000	0.99	±0.01	FLOOR	SPE GOR bounds

Schema highlights

`production_data` (~15,699 rows × 36 columns)

Identifiers (4): well_id, api_number, field_id, production_date.

Rates & ratios (8): oil_rate_bopd, gas_rate_mcfd, water_rate_bwpd, gross_rate_blpd, gor_scf_per_bbl, wor, bsw_pct, plus cumulative counterparts.

Cumulative production (3): cumulative_oil_bbl, cumulative_gas_mscf, cumulative_water_bbl — monotonically increasing per well.

Pressures (4): reservoir_pressure_psia (depletion-aware), flowing_wellhead_pressure_psia, flowing_bottomhole_pressure_psia, drawdown_psia.

Artificial lift (4): esp_frequency_hz, rod_pump_spm, rod_pump_fillage_pct, gas_lift_rate_mmscfd — populated per lift_type.

Economics (7): wti_price_usd_per_bbl, realized_oil_price_usd_per_bbl, henry_hub_usd_per_mmbtu, oil_revenue_usd, gas_revenue_usd, loe_total_usd_per_day, net_operating_cash_flow_usd, netback_usd_per_boe.

Well status (1): well_status ∈ {PRODUCING, SHUT_IN, P_AND_A}.

HSE / Environmental (5): methane_emissions_mcfd, co2_emissions_tons_per_day, flare_volume_mmscfd, h2s_concentration_ppm, spill_event_flag.

`wells_metadata` (50 rows × 30 columns)

well_id, api_number, field_id, basin, fluid_type, lift_type, completion_type, lateral_length_ft, n_perforation_stages, proppant_volume_mlb, fluid_volume_bbl, skin_factor, productivity_index_bopd_per_psi, spud_date, completion_date, first_production_date, ip_30_bopd, ip_90_bopd, decline_type (Hyperbolic / Exponential / Harmonic), initial_decline_rate_pct_yr, hyperbolic_b_factor, terminal_decline_rate_pct_yr, eur_oil_mbo, eur_gas_mmscf, working_interest_pct, royalty_rate_pct, net_revenue_interest_pct, workover_count, recompletion_flag.

`fields_metadata` (8 rows × 19 columns)

field_id, reservoir_id, basin, formation_name, initial_reservoir_pressure_psia, reservoir_temperature_degF, permeability_md, porosity_pct, net_pay_ft, area_acres, api_gravity, sulfur_content_pct, bubble_point_pressure_psia, oil_viscosity_cp, formation_volume_factor_bo, original_oil_in_place_mmbo, original_gas_in_place_bcf, reservoir_drive_type (Solution Gas / Water Drive / Gas Cap / Compaction / Combination), aquifer_strength_index.

`facilities` (15 rows × 12 columns)

facility_id, field_id, basin, facility_type (GPF / Separator Train / Tank Battery / Compressor Station / LACT Unit / Saltwater Disposal), oil_treating_capacity_bopd, gas_compression_hp, separator_inlet_pressure_psia, separator_temperature_degF, facility_uptime_pct, throughput_utilization_pct, lact_meter_factor, saltwater_disposal_bwpd.

Suggested use cases

Type curve / EUR forecasting — train Arps decline parameter estimators from early-time IP30/IP90 + completion features (lateral length, n_stages, proppant); predict EUR per well
Production rate forecasting — multi-step time-series prediction of oil_rate_bopd, gas_rate_mcfd, water_rate_bwpd conditioned on reservoir pressure, drawdown, and lift parameters
Well economics ML — net cash flow regression from WTI, oil rate, GOR, royalty, and LOE; build break-even price predictors
Reservoir pressure decline modeling — train ML to predict reservoir_pressure_psia from cumulative production and drive type
GOR / WOR progression modeling — classifier for fluid evolution phases (above bubble point, gas breakthrough, water breakthrough)
ESP / rod pump anomaly detection — unsupervised models on esp_frequency_hz and rod_pump_fillage_pct for downhole equipment failure prediction
Methane emissions intensity benchmarking — train regressors for methane_emissions_mcfd per BOE; useful for EPA Subpart W reporting ML augmentation
Basin classification — predict basin from completion params, GOR, API gravity, and decline parameters
Lift type selection — classifier for lift_type given reservoir conditions, fluid properties, and depth proxies
Commodity price stress testing — replay WTI/Henry Hub OU paths against well economics; quantile-based break-even analysis
Plug-and-abandon (P&A) prediction — survival model for time-to-P&A from IP30, decline rate, and price exposure
Multi-table relational ML — join production × wells × fields × facilities for full-stack value-chain optimization

Loading examples

from datasets import load_dataset

# Load the primary production table
prod = load_dataset("xpertsystems/enr004-sample", "production", split="train")
print(prod.shape)

import pandas as pd
from huggingface_hub import hf_hub_download

# Load all four tables and join
prod = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr004-sample", "enr004_production_data.parquet",
    repo_type="dataset",
))
wells = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr004-sample", "enr004_wells_metadata.parquet",
    repo_type="dataset",
))
fields = pd.read_parquet(hf_hub_download(
    "xpertsystems/enr004-sample", "enr004_fields_metadata.parquet",
    repo_type="dataset",
))

# Per-well decline ratio (last 30 days / first 30 days)
for wid, g in prod.groupby("well_id"):
    g = g.sort_values("production_date")
    producing = g[g["oil_rate_bopd"] > 1]
    if len(producing) >= 60:
        ratio = producing.tail(30)["oil_rate_bopd"].mean() / producing.head(30)["oil_rate_bopd"].mean()
        wells_meta = wells[wells["well_id"] == wid].iloc[0]
        print(f"{wid[:8]}  {wells_meta['basin']:<14} {wells_meta['fluid_type']:<14} "
              f"decline ratio = {ratio:.3f}")

# Arps decline curve fit per well
import numpy as np
import pandas as pd

def fit_exponential_di(g):
    """Estimate exponential decline rate per year from log-linear fit."""
    g = g.sort_values("production_date")
    producing = g[g["oil_rate_bopd"] > 1].reset_index(drop=True)
    if len(producing) < 30:
        return np.nan
    days = np.arange(len(producing))
    log_oil = np.log(producing["oil_rate_bopd"].values)
    m, _ = np.polyfit(days, log_oil, 1)
    return -m * 365  # fraction per year

fit_di = prod.groupby("well_id").apply(fit_exponential_di)
print(f"Fitted di mean: {fit_di.mean():.3f}/yr  (target: see wells.initial_decline_rate_pct_yr)")

Limitations and honest disclosures

This sample is calibrated for structural fidelity, not bit-exact reproduction of any specific basin's archive. Specifically:

The production table is NOT a uniform daily series for every well. Generator line 543-545 skips consecutive inactive days as a memory optimization. For long-shut-in or P_AND_A wells, only the transition rows are recorded. Use well_status and production_date explicitly when building daily-frequency time series.
unit_status (PRODUCING / SHUT_IN / P_AND_A) is computed from oil_rate_bopd vs. economic limit — there is no separate operational state model for unplanned shutdowns vs. mechanical issues vs. economic shut-ins. Use status as a coarse activity proxy, not an event log.
Decline curve log-linear R² at daily resolution is low (~0.02-0.05) even though the underlying Arps process is correctly implemented. The ±5% lognormal noise + 2% seasonal modulation + 3% downtime mask the smooth Arps trend at daily cadence. For decline-curve ML, aggregate to monthly first (the industry-standard cadence). The structural decline property (last 30 days < first 30 days) holds at 100% of qualifying wells.
The Ornstein-Uhlenbeck WTI process mean-reverts to $68 at θ=0.15/yr but the 2-year sample mean wanders $60-$87 across seeds. For volatility- sensitive backtests, replay with deliberate price shocks; for level exposure, use the realized prices per row directly.
base_gor is derived from bubble_point_pressure × 0.5 (generator line 466), then escalated by 1 + t × uniform(0.05, 0.25)/yr per well. This produces realistic GOR rise for depleting reservoirs above bubble point, but does NOT model the gas-cap-driven GOR collapse that occurs in solution-gas-drive reservoirs after they drop below bubble point.
h2s_concentration_ppm is derived from sulfur_content_pct × 5000 — a coarse proxy. Real H2S concentrations depend on bacterial sulfate reduction, formation chemistry, and souring history. Use as a hazard flag, not for materials selection or treating chemistry design.
spill_event_flag fires at 0.001/day — at sample scale that's ~15 events across the full table; per-basin spill rate analysis needs the full product scale (5K wells × 5 years).
recompletion_flag is per-well static — generator does not model the date/timing/effect of a recompletion on subsequent production.
Surface facilities are not linked 1:1 to specific wells — the facilities.field_id joins to the field, but wells-to-facility assignment is not modeled. Treat facilities as field-level summary.
Commodity price differentials are per-well constants drawn at well-creation time, not time-varying — the full product models basin-specific differential dynamics (e.g., Midland-Cushing, Appalachian-NYMEX).

The full ENR004 product addresses these by per-event SCADA shutdowns, monthly-aggregated decline analytics, basin-specific GOR/WOR phase modeling, well-facility linkage, and time-varying differentials — contact us for the licensed commercial release.

Companion datasets in the Energy & Climate vertical

ENR-001 — Synthetic Power Grid Operations Dataset (bus telemetry, line flows, dispatch, frequency, contingency)
ENR-002 — Synthetic Renewable Energy Generation Dataset (solar/wind/ hybrid SCADA, weather, forecast, PCC, BESS)
ENR-003 — Synthetic Electricity Demand & Load Forecasting Dataset (zone-level demand, multi-horizon forecasts, peak events, EV/DER, TOU, LMP)
ENR-004 — Synthetic Upstream Oil & Gas Production Dataset (you are here) — well-level production, decline curves, PVT, commodity prices, Subpart W methane

Use ENR-001 + ENR-002 + ENR-003 + ENR-004 together for a complete energy value-chain ML workflow: upstream production economics (ENR-004) → generation supply mix and renewables (ENR-002) → grid dispatch and delivery (ENR-001) → demand-side load and pricing (ENR-003).

For subsurface companion data (seismic, well logs, reservoir simulation, geological formations), see the OIL series (OIL-001 through OIL-004) in our Oil & Gas vertical.

For the broader catalog:

Materials & Energy — MAT-001
Insurance & Risk — 10 SKUs
Cybersecurity — 11 SKUs

Citation

@dataset{xpertsystems_enr004_sample_2026,
  author       = {XpertSystems.ai},
  title        = {ENR004 Synthetic Upstream Oil and Gas Production Dataset (Sample Preview)},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/xpertsystems/enr004-sample}
}

Contact

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

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