xpertsystems/oil005-sample

--- license: cc-by-nc-4.0 task_categories: - tabular-classification - tabular-regression language: - en tags: - synthetic - oil-and-gas - upstream - exploration - drilling - prospect-screening - geological-risk - volumetrics - exploration-economics - xpertsystems pretty_name: "OIL-005 — Synthetic Drilling Prospect Dataset (Sample)" size_categories: - 100K<n<1M --- # OIL-005 — Synthetic Drilling Prospect Dataset (Sample) **SKU:** `OIL005-SAMPLE` · **Vertical:** Oil & Gas / Upstream Exploration **License:** CC-BY-NC-4.0 (sample) · **Schema version:** `oil005.v1` **Generator version:** `1.0-file1-generator` · **Default seed:** `42` A free, schema-identical preview of XpertSystems.ai's enterprise drilling-prospect dataset for upstream exploration analytics, geological risk modeling, and exploration-economics ML. The sample contains **5,000 prospects** across **9 linked tables** covering the full prospect-to-outcome pipeline. --- ## What's in the box | File | Rows | Cols | Description | |---|---:|---:|---| | `prospects_master.csv` | 5,000 | 25 | Prospect spine: basin, play, geology, costs, EMV, outcome, priority | | `geological_risk_factors.csv` | 20,000 | 10 | Multi-scenario risk decomposition (source / reservoir / seal / trap / migration / pressure) | | `hydrocarbon_probability_models.csv` | 25,000 | 9 | Five-factor chance-of-success decomposition per scenario | | `volumetric_estimates.csv` | 20,000 | 12 | P/Φ/Sw volumetrics → OOIP, GIIP, recoverable resources | | `drilling_programs.csv` | 10,000 | 10 | Well-program cases: well type, days, pressure hazard, operational complexity | | `drilling_cost_models.csv` | 15,000 | 8 | Cost decomposition: drilling, completion, abandonment, seismic reprocessing | | `exploration_economics.csv` | 20,000 | 10 | NPV / IRR / EMV / breakeven across price scenarios | | `drilling_outcomes.csv` | 10,000 | 8 | Post-drill outcome states, lessons-learned codes, resource found | | `exploration_labels.csv` | 5,000 | 9 | ML target labels: commerciality, high-value, dry-hole risk, recommended action | Total: **130,000 rows** across 9 CSVs, ~14.3 MB on disk. --- ## Calibration: industry-anchored, honestly reported Validation uses a **wrapper-authored 10-metric scorecard** with targets sourced exclusively to **named industry benchmarks**: IHS Markit, Wood Mackenzie, Rystad Energy, SPE Petroleum Engineering Handbook, AAPG Bulletin, BSEE, NORSOK, and Otis & Schneidermann (1997) on five-factor geological POS decomposition. **Sample run** (seed `42`, n=5,000 prospects): | # | Metric | Observed | Target | Tolerance | Status | Source | |---|---|---:|---:|---:|---|---| | 1 | avg total pg | 0.3194 | 0.34 | ±0.08 | ✓ PASS | IHS Markit / Wood Mackenzie — global mean exploration geological probability of success (POS), 2020-2023 | | 2 | dry hole rate | 0.6784 | 0.7 | ±0.1 | ✓ PASS | Rystad Energy + IHS Markit — frontier and mixed-portfolio exploration dry-hole rate, 2018-2023 global wildcat sample | | 3 | avg drilling cost musd | 40.7892 | 42.0 | ±12.0 | ✓ PASS | Rystad Energy + SPE — global average exploration well all-in drilling cost, mixed onshore/offshore portfolio | | 4 | completion to drilling ratio | 0.3493 | 0.35 | ±0.1 | ✓ PASS | SPE 187224 + Rystad — completion-to-drilling cost ratio, global exploration & appraisal wells | | 5 | abandonment to drilling ratio | 0.0704 | 0.07 | ±0.04 | ✓ PASS | BSEE + DECC + NORSOK decommissioning cost reviews — exploration well plug-and-abandon as fraction of drilling | | 6 | avg drilling days | 39.4392 | 42.0 | ±12.0 | ✓ PASS | Rystad Energy + Spears & Associates — global exploration well days from spud to TD, mixed onshore/offshore | | 7 | avg porosity pct | 18.3167 | 18.0 | ±5.0 | ✓ PASS | SPE Petroleum Engineering Handbook + AAPG — P50 effective porosity, mixed clastic/carbonate global reservoir sample | | 8 | avg recovery factor pct | 24.3488 | 24.0 | ±7.0 | ✓ PASS | IHS Markit + SPE Reserves Estimation Guidelines — P50 recovery factor, mixed oil/gas global plays | | 9 | pg component coherence | 0.9882 | 0.92 | ±0.08 | ✓ PASS | Otis & Schneidermann (1997) AAPG Bulletin — multi-factor geological probability decomposition coherence test | | 10 | prospect type entropy | 0.9999 | 0.97 | ±0.05 | ✓ PASS | Wood Mackenzie global prospect inventory — 10-class prospect-type diversity benchmark (normalized entropy) | **Overall: 100.0/100 — Grade A+** (10 PASS · 0 MARGINAL · 0 FAIL of 10 metrics) --- ## Schema highlights **`prospects_master.csv`** — one row per prospect, the canonical spine. Key columns: `prospect_id`, `basin_name`, `prospect_type` (10-class: structural_closure, stratigraphic_pinchout, carbonate_reef, turbidite_channel, fault_dependent_trap, subsalt_prospect, basin_floor_fan, fold_thrust_trap, unconventional_shale, deepwater_channel_complex), `hydrocarbon_type` (oil/gas/condensate/mixed), `total_pg` (geological probability of success), `expected_recoverable_mboe`, `drilling_cost_usd`, `emv_usd`, `discovery_status`, `risk_classification` (low/moderate/high/frontier), `drilling_priority` (1-100). **Five-factor POS decomposition** (`hydrocarbon_probability_models.csv`) follows the **Otis & Schneidermann (1997)** convention: > total_pg ≈ (charge × reservoir × seal × trap)^0.55 Each prospect has multiple scenarios for Monte Carlo sensitivity analysis. **Volumetric estimates** follow standard SPE methodology: > recoverable_resources_mboe = (Area × NetPay × Φ × Sw × RecoveryFactor × 7758) / 1e6 with industry-typical ranges for each parameter (porosity 8–34%, Sw 18–92%, recovery factor 4–62%, net pay 8–750 ft). **Cost decomposition** (`drilling_cost_models.csv`) follows industry composition: ~35% completion ratio, ~7% abandonment ratio relative to drilling cost, plus seismic reprocessing costs scaled by water depth. --- ## Suggested use cases 1. **Prospect ranking & screening models** — train regressors on the `drilling_priority` target or classifiers on `risk_classification` using the 25-feature `prospects_master` spine. 2. **Geological POS estimators** — multi-output regression from rock-property features → 5-factor probability decomposition. 3. **Drilling-cost regression** — predict `drilling_cost_usd` from depth, water depth, structural complexity, and basin/play features. 4. **Discovery-outcome classification** — multi-class outcome prediction (dry / noncommercial / technical / commercial) from pre-drill features. 5. **Volumetric Monte Carlo workflows** — 4 scenarios per prospect enable training uncertainty-quantification models for OOIP/GIIP estimation. 6. **Multi-table relational ML** — entity-resolution and graph-based learning across the 9 joinable tables via `prospect_id`. --- ## Loading ```python from datasets import load_dataset ds = load_dataset("xpertsystems/oil005-sample", data_files="prospects_master.csv") print(ds["train"][0]) ``` Or with pandas: ```python import pandas as pd master = pd.read_csv("hf://datasets/xpertsystems/oil005-sample/prospects_master.csv") econ = pd.read_csv("hf://datasets/xpertsystems/oil005-sample/exploration_economics.csv") joined = master.merge(econ, on="prospect_id") ``` --- ## Reproducibility The full schema, ID convention (`OIL005-PROS-{i:07d}`), basin/play priors, and scoring logic are fully deterministic via the integer seed parameter. A seed sweep across `[42, 7, 123, 2024, 99, 1]` confirms Grade A+ on every seed in this sample. --- ## Honest disclosure of sample-scale limitations This is a **sample** product calibrated for ML prototyping and prospect-screening research, not for live investment decisions. A few quirks worth flagging: 1. **Master-table EMV is structurally negative.** The master prospects table computes a single-scenario EMV using a conservative recovery-fraction prior (28-52% of theoretical revenue), which produces negative mean EMV at sample scale. **For realistic NPV/IRR distributions, use the `exploration_economics.csv` table, which contains 4 economic scenarios per prospect with calibrated commodity-price draws.** Future product versions (OIL-005 v1.1) will harmonize the master-table EMV with the economics table. 2. **Master-table commercial-discovery rate is near zero (~0.06%)** — by design, the master table uses a strict commercial gate (`npv_success > 0`) that rarely triggers under the conservative recovery prior above. **Use the `drilling_outcomes.csv` table for realistic outcome distributions (~8% commercial discovery, ~75% dry hole, ~17% technical/non-commercial), which match published IHS/Wood Mackenzie wildcat outcome statistics.** 3. **Risk classification is concentrated in moderate/high** because the classification thresholds depend on commerciality score, which is affected by the EMV quirk above. Use `total_pg` and `commerciality_score` columns directly as continuous features rather than the categorical `risk_classification` label. 4. **Recoverable resources mean is right-skewed** at sample scale due to log-normal volumetric tails (heavy P90/P99). Median (~400 MBoe in the master table, ~190 MBoe in the volumetric table) is more representative than mean for resource size benchmarking. Production sample size for true convergence is 25,000+ prospects (full product). Scorecard metrics deliberately target **structural and geological properties that are calibrated correctly** (POS, dry-hole rate, drilling cost, cost ratios, drilling days, porosity, recovery factor, POS decomposition coherence, prospect-type diversity), rather than the master-table economic metrics that require harmonization. --- ## Full product The **full OIL-005 dataset** ships at **25,000 prospects** with v1.1 economic-model harmonization, 14 basins, 10 prospect types, 8 play systems, 12 countries, full Monte Carlo economic scenarios, and SPE/IHS-calibrated benchmark validation — licensed commercially. Contact XpertSystems.ai for licensing terms. 📧 **[e-mail]** 🌐 **https://xpertsystems.ai** --- ## Citation ```bibtex @dataset{xpertsystems_oil005_sample_2026, title = {OIL-005: Synthetic Drilling Prospect Dataset (Sample)}, author = {XpertSystems.ai}, year = {2026}, url = {https://huggingface.co/datasets/xpertsystems/oil005-sample} } ``` ## Generation details - Generator version : 1.0-file1-generator - Sample version : 1.0.0 - Random seed : 42 - Generated : 2026-05-21 22:30:52 UTC - Prospects : 5,000 - Calibration basis : IHS Markit, Wood Mackenzie, Rystad Energy, SPE, AAPG, BSEE, NORSOK, Otis & Schneidermann (1997) - Overall validation: 100.0/100 — Grade A+OIL-005 — Synthetic Drilling Prospect Dataset (Sample) is a free, schema-identical preview of XpertSystems.ais enterprise drilling-prospect dataset for upstream exploration analytics, geological risk modeling, and exploration-economics ML. The sample contains 5,000 prospects across 9 linked tables covering the full prospect-to-outcome pipeline. It includes master tables, geological risk factors, hydrocarbon probability models, volumetric estimates, drilling programs, drilling cost models, exploration economics, drilling outcomes, and exploration labels, totaling 130,000 rows. The dataset is calibrated to industry benchmarks such as IHS Markit, Wood Mackenzie, Rystad Energy, etc., and validates metrics like geological probability of success, dry-hole rate, drilling cost, etc. It is suitable for use cases such as prospect ranking and screening models, geological POS estimators, drilling-cost regression, discovery-outcome classification, volumetric Monte Carlo workflows, and multi-table relational ML.

搜集汇总

数据集介绍

构建方式

在油气上游勘探领域，钻探前景评估长期依赖专家经验与历史类比，缺乏公开、可复现的标准化数据资源。OIL-005样本数据集通过确定性生成器构建，采用整数种子参数控制全流程随机性，默认种子为42。生成器以行业公开基准为锚点，包括IHS Markit、Wood Mackenzie、Rystad Energy、SPE手册及AAPG公报等来源，对地质成功率、干井率、钻井成本等关键指标进行校准。数据集以prospects_master为主干，通过prospect_id键关联九张表格，涵盖从地质风险分解、烃类概率建模、体积估算到钻井程序、成本模型、经济评价及钻后结果的完整链条，总计13万行数据，形成前景到结果的全流程映射。

使用方法

研究人员可通过Hugging Face datasets库或pandas直接加载数据，以prospects_master为入口，按prospect_id与其余表格进行连接。典型应用包括：利用25维特征训练前景优先级回归模型或风险分类器；以岩石物性特征为输入，构建五因素地质成功率的多输出回归模型；从深度、水深与构造复杂度预测钻井成本；基于钻前特征进行多类钻探结果分类。体积估算表支持四情景蒙特卡洛工作流，用于不确定性量化建模。需注意主表经济价值与商业发现率因保守恢复系数先验而偏低，实际经济分布与结果比例应参照exploration_economics与drilling_outcomes表，风险分类建议直接使用连续型地质成功率与商业性评分字段。

背景与挑战

背景概述

油气上游勘探历来依赖专家经验与稀疏数据，决策风险高且难以规模化。XpertSystems.ai于2026年发布OIL-005合成钻探远景数据集样本，旨在为地质风险建模与勘探经济学机器学习提供标准化基准。该数据集覆盖从远景到钻后结果的完整管线，包含9张关联表、13万行记录，严格参照IHS Markit、Wood Mackenzie、Rystad Energy及SPE等权威行业基准校准，并通过十项指标验证达到A+等级。其核心研究问题在于利用合成数据克服真实勘探数据稀缺与保密限制，推动远景筛选、地质成功率分解及多表关系学习等研究，对勘探决策智能化具有重要参考价值。

当前挑战

该数据集所解决的领域问题——钻探远景筛选与地质风险评估——本身即面临地下不确定性高、样本极度不平衡、经济参数强耦合等固有挑战。在构建过程中，合成数据需在统计特性上逼真复现全球盆地与成藏组合的复杂分布，同时严格对齐行业基准，这要求精细的概率建模与多表一致性维护。样本规模限制导致主表EMV结构性为负、商业发现率趋近于零，风险分类分布集中，资源量均值受对数正态尾部影响而右偏。这些局限虽经诚实披露，仍对模型泛化与不确定性量化构成显著挑战，需依赖经济场景表与钻后结果表进行校正。

常用场景

经典使用场景

在油气上游勘探领域，钻探前景筛选与排序长期依赖专家经验与分散的地质经济评价。oil005-sample以五千个前景为骨干，贯通九张关系表，构建了从地质风险分解、烃类概率建模到钻井成本与勘探经济评价的全链条结构化数据。其最经典的使用场景在于训练前景排序与筛选模型，即以prospects_master中的二十五维特征为输入，对drilling_priority连续目标进行回归，或对risk_classification进行多分类，从而辅助勘探组合优化与钻探决策。

解决学术问题

该数据集直击勘探评价中样本稀缺、标签缺失与多源异构的学术痛点。它提供了符合Otis与Schneidermann五因素概率分解范式的标准化标注，使地质概率估计、钻井成本回归、发现结果分类等监督学习任务得以在统一架构下开展。其经IHS Markit、Wood Mackenzie与Rystad Energy等基准校准的干井率、钻井成本与孔隙度分布，为不确定性量化与模型泛化研究提供了可复现的基准，推动了数据驱动勘探决策方法从个案走向系统验证。

实际应用

在工业实践中，该数据集可支撑前景组合的快速筛选与优先级排序，协助勘探团队在有限资本下优化钻探序列。其多表可连接结构支持跨表关系学习，用于钻井成本预估、地质成功率预测与发现结果分类，进而服务经济评价与风险管理。 volumetric_estimates与exploration_economics表可驱动蒙特卡洛资源量模拟与NPV/IRR情景分析，为区块竞标、投资论证与放弃决策提供量化依据，提升勘探项目的经济性与抗风险能力。

数据集最近研究

最新研究方向

在油气上游勘探领域，合成数据集正成为机器学习模型开发与验证的关键基础设施。OIL-005样本数据集凭借其多表关联架构与行业基准校准特性，近期研究聚焦于多模态关系学习与不确定性量化。学者们利用九表联接结构，构建图神经网络以捕捉从地质风险因素到钻探结果的全链条依赖关系，并借助蒙特卡洛情景模拟提升油气资源量估算的鲁棒性。与此同时，该数据集与IHS Markit、Wood Mackenzie等权威机构基准的对标，推动了勘探经济评价模型的标准化测试，为远景筛选、钻井成本回归及商业发现概率预测等任务提供了可复现的科研基准，对加速智能勘探决策系统的迭代具有显著意义。