Risk Reduction through Seismic Data Mining The Ibhubesi Field, Orange River Basin Republic of South Africa © 1998-2003 Rock Solid Images, all rights reserved
Seismic Data Mining “Data mining is a process that uses a variety of data analysis tools to discover patterns and relationships in data that may be used to make valid predictions”
Seismic Attributes
Two Crows Corporation.
PHI, Sw, Vcl
Progressive information extraction and knowledge enrichment from seismic and well data © 1998-2003 Rock Solid Images, all rights reserved
Project Goals Determine whether seismic data contains any information that might be utilized to reduce exploration risk in the future
© 1998-2003 Rock Solid Images, all rights reserved
Topography & Bathymetry
© 1998-2003 Rock Solid Images, all rights reserved
Prospect Leads
© 1998-2003 Rock Solid Images, all rights reserved
Orange River Basin Architecture
© 1998-2003 Rock Solid Images, all rights reserved
Project Workflow Precondition amplitude volume for attribute analysis via spectral extension and edge-preserving smoothing. “Mine” the 3D offset volumes using multiple seismic attributes. Determine which attributes are the best discriminators of lithology, fluid and geometry. Combine this attribute suite in an artificial neural network – Kohonen Self-Organizing Map Produce a volume that is calibrated to rock properties. © 1998-2003 Rock Solid Images, all rights reserved
Preconditioning of Seismic Data Amplitude Full-stack
Spectral balancing Increases frequency spectrum
© 1998-2003 Rock Solid Images, all rights reserved
Preconditioning of Seismic Data Spectral extension Increases frequency spectrum
Smoothing Suppresses random noise
© 1998-2003 Rock Solid Images, all rights reserved
Preconditioning of Seismic Data Input Data - Amplitude Full-stack
Output Data - Spectral extension + Smoothing
© 1998-2003 Rock Solid Images, all rights reserved
Establishing a Structural Framework Similarity
Spectrally decomposed Amplitude centered at 8 Hertz
© 1998-2003 Rock Solid Images, all rights reserved
Visual Data Mining Opacity Sculpting
© 1998-2003 Rock Solid Images, all rights reserved
Spectral Decomposition Volume centered at 8 hz – frontal view
Arrows delineate fault trace
Interpreted fault plane
© 1998-2003 Rock Solid Images, all rights reserved
Spectral Decomposition Volume centered at 8 hz – reverse view
Interpreted fault plane
Arrows delineate fault trace
© 1998-2003 Rock Solid Images, all rights reserved
Structural Context
4 wells upthrown, 1 well downthrown © 1998-2003 Rock Solid Images, all rights reserved
Amplitude Data in Structural Context
Isolated amplitudes generally do not conform to structural grain © 1998-2003 Rock Solid Images, all rights reserved
Similarity
Meander cutoff © 1998-2003 Rock Solid Images, all rights reserved
Similarity & Relative Acoustic Impedance
Meander lobes © 1998-2003 Rock Solid Images, all rights reserved
Relative Acoustic Impedance
Sand bodies correspond to volumes low in relative AI
© 1998-2003 Rock Solid Images, all rights reserved
Voxel based seed picking
© 1998-2003 Rock Solid Images, all rights reserved
Connectivity Analysis
© 1998-2003 Rock Solid Images, all rights reserved
Project Information
3 gas wells (AK-1, AK-2 and AV-1) drilled on bright amplitudes.
AW-1 well drilled on brightest amplitude in volume – tested wet.
Amplitude & AI good lithology discriminator – poor fluid discriminator
© 1998-2003 Rock Solid Images, all rights reserved
A-W1 well (wet) Sw
VSH
PR
AI PR INCREASE
A-Y1 well (gas) Sw
VSH
PR
AI
PR DECREASE
© 1998-2003 Rock Solid Images, all rights reserved
AVO Class III Anomalies Near-angle stack AK-2 AY-1
Brightening occurs over offset at top sand
AK-2 AY-1
Mid-angle stack
© 1998-2003 Rock Solid Images, all rights reserved
Elastic Attributes 2 key elastic rock properties: shear modulus aka rigidity aka μ, μ is insensitive to fluid change
bulk modulus aka incompressibility aka K K is sensitive to fluid change
Lamé’s coefficient λ = K – 2/3 μ and, λρ = Ip2 – 2Is2 In presence of gas – term 1 decreases & term 2 increases © 1998-2003 Rock Solid Images, all rights reserved
Lambda-Rho
AV-1 AK-2 AK-1
Elastic constants
© 1998-2003 Rock Solid Images, all rights reserved
Fluid Factor
AK-1
Red-green couplets indicate Vp decrease due to gas presence
© 1998-2003 Rock Solid Images, all rights reserved
Kohonen Self-Organizing Map (SOM) Artificial neural network featuring unsupervised learning N-dimensional clustering technology Non-linear mapping of multi-attribute signature to seismic lithofacies Volumetric classification
© 1998-2003 Rock Solid Images, all rights reserved
Mapped Final Calibrated Volume
Gas/Water Contact
wet sand
shale
gas sand
© 1998-2003 Rock Solid Images, all rights reserved
Final Calibrated Volume – AY-1 (gas)
wet sand
shale
gas sand
© 1998-2003 Rock Solid Images, all rights reserved
Final Calibrated Volume – AK-1&2 (gas)
wet sand
shale
gas sand
© 1998-2003 Rock Solid Images, all rights reserved
Final Calibrated Volume – AV-1 (gas)
wet sand
shale
gas sand
© 1998-2003 Rock Solid Images, all rights reserved
Final Calibrated Volume – AW-1 (wet)
wet sand
shale
gas sand
© 1998-2003 Rock Solid Images, all rights reserved
Conclusions
Attribute Analysis reveals that multiple attributes can be combined in a non-linear fashion via an ANN featuring unsupervised learning – Kohonen’s Self-Organizing Map – to effectively discriminate between lithologies and fluids. Seismic data mining mitigates drilling risk
© 1998-2003 Rock Solid Images, all rights reserved
Data Mining Summary Data mining provides knowledge discovery while reducing the volume being mined. Latest workflows being developed encompass this concept.
© 1998-2003 Rock Solid Images, all rights reserved
Acknowledgements Forest Oil Tim Berge
Anschutz Corp Kevin Corbett
Soekor Eric Jungschlager
Rock Solid Images Matt Carr, Maggie Smith, Tury Taner, Gareth Taylor, Uwe Strecker
© 1998-2003 Rock Solid Images, all rights reserved
alternatives
© 1998-2003 Rock Solid Images, all rights reserved