1 / 22

Data-driven estimates of reservoir properties from 3D/4D seismic A brown field study

Data-driven estimates of reservoir properties from 3D/4D seismic A brown field study. Evgeny Tolstukhin*, Reidar Midtun, Pål Navestad, ConocoPhillips Norway Tetyana Kholodna, CapGemini Norway Evgeny.Tolstukhin@conocophillips.com.

pnutter
Télécharger la présentation

Data-driven estimates of reservoir properties from 3D/4D seismic A brown field study

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Data-driven estimates of reservoir properties from 3D/4D seismicA brown field study Evgeny Tolstukhin*, Reidar Midtun, Pål Navestad, ConocoPhillips Norway Tetyana Kholodna, CapGemini Norway Evgeny.Tolstukhin@conocophillips.com

  2. Project: Properties from SeismicMotivation and Business Case • Idea / Innovation: • Use Machine Learning to predict reservoir properties directly from 3D seismic and well logs data • Motivation: • Increase value of G&G data through data-driven approach • Build a Machine Learning and AI platform for Subsurface Domain • Project scope: • Duration: 3 months • Prove the concept • Evaluate business impact • Business impact: • Drill less water-wet wells and side-tracks • Better understanding of reservoir properties and mechanisms

  3. DigitizationofSubsurface: seismic and wells data Survey 2016 Survey 2010 A09 A18 A27 ElasticImpedance Poro well log Swewell log RFT pressure Voxels Table

  4. Illustrationofresolution and sampling effects Sample bias Swe • 100 ftaverage • - Observed Well log vs. Seismic Pressure Swat > 0.5 - Observed - Polynom - Linear - Swe Fly by Pluto with the New Horizons probe | New Scientist

  5. Data available: poro, swe, pressure, A09, A18,A27 and ratios EI A09 EI A09 Poro Swe Swe EI A09 Poro EI A09/A27

  6. Causation vs Correlation: poro, swe, pressure, A09, A18,A27 and ratios Seismic, available in 3D volume Swe Properties, available only from well logs EI A09 Poro EI A09/A27

  7. CausationvsCorrelation Swe EI A09 Color is Pressure Poro EI A09/A27

  8. CausationvsCorrelation Swe EI A09 Color is Pressure LowSwe High Poro Poro EI A09/A27

  9. CausationvsCorrelation Swe EI A09 High Swe Low Poro Color is Pressure Poro EI A09/A27

  10. CausationvsCorrelation: NEXT LEVEL, divisionintogroups or clusters Color is Formation Pore intervals • Scope: • Try alternative clustering methods: • DensityDbscan • K-means • Normal Mixture • Hierarchical • Dimesionalityreduction: • Multi-DimensionalScaling • Principal Component Analysis • Try alternative ML methods: • Multi-Adaptive Regression • Neural Networks • DecisionTrees • Support Vector Machine • Random Forest • etc. PressureIntervals A09 Swat 0-1

  11. Conceptillustration: clustering and prediction SeismicType 1 Cluster 1 Category H RockType 1 Cluster 1 Category H RockType 1 SeismicType 1 RockType 2 Onlyseismic Use ML model from wells A09, A18, A27 Cluster 2 Category L RockType 2 SeismicType 2 SeismicType 2 Category H Cluster 2 Category L Cluster 1 Cluster 2 Category L RockType 1 Poro, Press, Swe RockType 2

  12. Software architecture SAS JMP / R-scripts Select Transform Filter Interpolate CARET PredictedFactor: Category Well Master table K-means Clustering 3 Seismic 3 props per formation • Random Forest • Predict Cluster • With Factor: Water and Factor: Category • PredictSwe, Poro, Press within Cluster using Random Forest Clusters Software used in the project: •SAS Enterprise Guide 7.1 •JMP 14.2.0 •JMP.R version 14.0 Distributions Within Clusters Factor: Category High, Medium, Low Scoring in 3D

  13. Agenda • Introduction • Methodologyreview • Results • Validation • Summary

  14. Resultsofnew Clustering at welllevel Swe Clusters Poro Pressure Seismic A09

  15. Resultsofnew Clustering: zoom Swe Clusters Poro ed ed ed ed Pressure Seismic A09

  16. BLIND TEST Prediction stageValidation at well level: future wells drilled in 2017-2018 Well 1 Well 4 Well 2 Well 5 Well 3 Well 6

  17. Effect of faults and fault shadows: Well 4 example (prediction validation) Well 4 water saturation Well 4 Faultedzone Well 4 water saturation Faultedzone

  18. Summaryofprediction at welllevel: wellsdrilled in 2017-2018 • What data do we compare to: • «Blind test» or validation wells drilled in 2017-2018 • Observed properties from well logs (lumped into «voxels») • What data do we have: • ML models trained on wells drilled in 2010-2016 • Reservoir properties scored in 3D using 2016 seismic

  19. 3D Comparison with simulation model: prediction of 3D Water Saturation, formation average Water saturation 0-1 Polygons are manual interpetations of water fronts based on well and production data

  20. Agenda • Introduction • Methodologyreview • Results • Validation • Summary

  21. Summary • The methodology allowed to predict 3D volumes of porosity, water saturation and pressure: • Predictions show good results at wells and in 3D • This data-driven model can be further utilized for: • Further quantitative analysis • Reservoir characterization • Multi-disciplinary communication • Key learnings from the project: • Agile project management • Collaboration between data and geoscientists • «Test fast, fail fast, adjust fast» • Quick feed-in of more data from Subsurface Data Lake: • New wells, new seismic, new simulation models, other observations • Consider addition of tracers, pressure, temperature and other data • Strength of the methodology: • Quick to run and update • Overall 10 min from training at well level to scoring in 3D

  22. Acknowledgements

More Related