GAS TRAP PREDICTION FROM 3D SEISMIC AND WELL TEST DATA USING MACHINE LEARNING - paper approach and results

Aug 5, 2026

Upload by Hazel

Gas trap prediction is developed by combining 3D seismic attributes with gas well testing evidence. The workflow uses binary classification with training on well-derived labels: gas well tests are interpreted to estimate the radius of radial gas filtration (IARF), productive horizons are correlated near the wellbore, and volumes are built for positive and negative classes. Seismic wavefield attributes are extracted, sampled per spatial point, relevant features are selected, and an ensemble metamodel with logistic regression estimates gas-reservoir membership probabilities. Quality is validated on blind test wells with an F1 score of 0.893846.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Predicting gas traps from seismic and well tests

Materials and methods

Interpreting well tests and estimating IARF

Correlating productive horizons in near-wellbore space

Building training/validation/test datasets

Feature extraction and selection

Training ensemble models and evaluating probabilities

Conclusion

Training dataset construction and class definitions

Calibrated probability cube and blind-well validation

GAS TRAP PREDICTION FROM 3D SEISMIC AND WELL TEST DATA USING MACHINE LEARNING Dmitry Ivlev [dm.ivlev@gmail.com](dm.ivlev@gmail.com) The aim of this work is to create and apply a methodological approach for predicting gas traps from 3D seismic data and gas well testing. Materials and methods. Data from 12 drilled wells, 1368 seismic field attributes, covering an area of 456 km2 in the TVD depth interval from 1200 to 2400 m were used in this work. The proposed approach is based on binary classification algorithms with training on well data. The study includes the following sequence of operations: interpretation of gas well test results to determine the radius of radial gas filtration (IARF); correlation of the top and bottom of productive horizons based on seismic data in the near-wellbore space; creating volumes of space for positive and negative class; dividing the sample into training and validation; creation of a separate test sample for the metamodel; creation of a feature space-extraction of seismic wavefield attributes; creation of data sampling - assignment of a vector of seismic attributes to each point of space within the volumes of classes; selection of features; basic model training; generalized assessment of trait contribution; creation of an ensemble of classification models using a metamodel-logistic regression; prediction of the probability of space belonging to gas reservoirs; evaluation of forecast quality on a test sample. Conclusion. The paper formalizes an approach to creating a training dataset by selecting volumes with established gas saturation and filtration properties within a seismic wavefield. The volumes divide the studied space into positive and negative classes. Positive class is a volume of gas-saturated sands, identified by the results of detailed correlation of the gas sandstone top and bottom, within which there is a region with a boundary along the radius of continuous radial gas filtration. The radius was estimated by interpreting gas well test data after deconvolution of test regimes. The negative class includes volumes of explored space within a one-hundred-meter radius along the wellbore with proven futility based on test results and log interpretation. Each point in the space within the classes is assigned seismic field attributes. The training data set created in this way is used in the technological stack of sequential application of data processing methods and ensemble machine learning algorithms. As a result, the cube of calibrated probabilities of belonging of the studied space to gas reservoirs is obtained. The high efficiency of this approach is shown on a delayed test sample from three wells (blind wells) . The final value of the gas sandstone prediction quality metric f1 score was 0.893846. Keywords: machine learning, well data, seismic attributes, facies prediction, rock properties prediction, augmentation methods, ensemble learning, feature selection, evaluation of feature contribution to prediction, geophysics, well testing, deconvolution of well testing. Introduction Well testing and 3D seismic surveys are standard surveys in the oil and gas industry. Information from these types of surveys provides insight into the characteristics of the rock in the interwell space. The surveys utilize different physical principles to obtain information. Measurements of pressure in the perforation interval and fluid withdrawal volumes during well testing can reconstruct fluid filtration conditions and properties over a radius of distance from the pressure transducer. Seismic exploration uses seismoacoustic waves over the entire fieldwork area to study geologic formations at depth. Seismic surveys can cover significant areas of space, from units to thousands of square kilometers. The area of investigation during well testing is determined by the spread of the pressure front in the pay zone within the radius of the well being tested. Depending on the time of investigation and

FAQs

How is the positive and negative class for training defined?

Positive class volumes contain gas-saturated sands identified by detailed seismic correlation of gas sandstone top and bottom, with a boundary corresponding to continuous radial gas filtration. Negative class volumes cover explored space near the wellbore where test results and log interpretation indicate futility within a one-hundred-meter radius.

What role does IARF play in the workflow?

IARF (radius of radial gas filtration) is estimated by interpreting gas well test data after deconvolution of test regimes, and it helps establish the spatial boundary used to label positive-class volumes.

How are the final predictions produced and validated?

Each point in class volumes is assigned a vector of seismic field attributes, then an ensemble of classification models via a metamodel-logistic regression predicts the probability of belonging to gas reservoirs. Forecast quality is evaluated on a delayed blind test sample, achieving an F1 score of 0.893846.

Aug 5, 2026

Upload by Hazel