Reservoir Probability Prediction Eval

reservoir-probability-prediction-eval

Generalization with Reverse-Calibration of Well and Seismic Data Using Machine Learning Methods for Complex Reservoirs Predicting During Early-Stage Geological Exploration Oil Field — Ivlev et al. (2023) (arXiv:2304.03048, 2023)

What this evaluates

Evaluates a machine learning model's ability to predict the probability of hydrocarbon reservoir presence in a 3D geological space using seismic and well log data. It probes the model's capacity for binary lithological classification and probabilistic calibration under early-stage exploration conditions with limited well data.

Datasets

• Achimov sedimentary complex field dataset — total ?; splits: train (-1), test (-1)

Metrics

• classification quality (primary) — range: other
  • Not explicitly defined in the text; refers generally to the performance of the binary classification forecast and the calibration of predicted probabilities against known lithological labels.

Input / output format

Input: 3D seismic survey attributes and well log data (lithological interpretations from 9 wells: 4 vertical/subvertical, 5 horizontal).

Output: A three-dimensional cube of calibrated probabilities indicating the likelihood of each voxel belonging to the reservoir class, plus a derivative map of reservoir thicknesses.

Scoring recipe

# Pseudo-code based on described protocol
predictions = model.predict(seismic_data, well_data)
quality_scores = []
for dataset in different_data_sets:
    preds = model.predict(dataset.seismic, dataset.well_logs)
    # Compare against known lithology labels
    score = evaluate_classification_quality(preds, dataset.labels)
    quality_scores.append(score)
return quality_scores

Common pitfalls

• The paper does not specify the exact classification metric used (e.g., accuracy, F1, AUC), making replication of the headline 'forecast quality' difficult.
• Dataset splits and exact sample sizes are not provided, only the total number of wells (9) and their orientation types.
• Reverse-calibration is described conceptually as using geophysical uncertainty as a hyperparameter, but the exact mathematical implementation is not detailed in the provided text.

Evidence (verbatim from paper)

The study includes the following sequence of actions: creation of data sets for training, selection of features, reverse-calibration of data, creation of a population of classification models, evaluation of classification quality, evaluation of the contribution of features in the prediction, assembling the population of models by stacking method. Assessment of changes in the quality of the forecast depending on the use of different data sets was carried out.

Citation

@misc{ivlev2023reversecalibration,
  title={Generalization with Reverse-Calibration of Well and Seismic Data Using Machine Learning Methods for Complex Reservoirs Predicting During Early-Stage Geological Exploration Oil Field},
  author={Ivlev et al. (2023)},
  year={2023},
  note={arXiv:2304.03048}
}

• arXiv: 2304.03048