Title: What can classification of seismic wavelet information reveal with regard to reservoir size and qual
1What can classification of seismic wavelet
information reveal with regard to reservoir size
and quality? Presenter Adéle Groenewald
Authors J. Feddersen, A. Groenewald P. Young
2Seismic Classification
- Seismic attributes widely used over 3D seismic
surveys - Seismic classification vs individual attribute
maps - Enables interpreter to study
- Combination of attributes
- In a systematic way
- To identify similarities in lithology
- and
- To study spatial variation of seismic waveform
- Lateral variations in lithology, porosity, fluid
content
3Seismic Classification
- 2 case studies used to illustrate Seismic
Classification - Supervised classification of seismic attributes
- Oribi-Oryx field
- Unsupervised classification of seismic waveform
- E-BD2
4Oribi-Oryx field
- Low impedance oil bearing sandstone reservoir
- Encased in shale
Oryx
Oribi
5Supervised Classification
- When to use Supervised Classification?
- Reliable information available at wells
- A-priori information can supplement attribute
data - The Method
- Fischers Discriminant Cluster centers are
computed within a pre-classified set of training
and validation data
6Supervised Classification Oribi-Oryx field
- A combination of Seismic attributes extracted
along a horizon were used - Acoustic Impedance
- Amplitude
- Integrated Fluid Factor magnitude
- Isopach
- Instantaneous frequency
-
Lithology indicators
7Supervised Classification Oribi-Oryx
fieldCombination of Attributes
8Supervised Classification Training Phase
2-attribute example
Attribute 2
Attribute 1
From Hampson Russel manual
9Supervised Classification Result Oribi-Oryx 4
classes
10E-BD2 Amplitude map
- 2D seismic Amplitude anomaly
- High fluid factor response
- Water bearing low velocity sandstone
- High amplitude/Fluid factor anomaly attributed to
- Tuning
- Low velocity of sandstone
Oil bearing channel sandstone
Water bearing sandstone E-BD2
11Unsupervised Classification
- When to use Unsupervised Classification?
- No reliable a-priori information or
- Un-biased cluster analysis to verify existence of
natural clusters - The Method
- Competitive learning method Unsupervised
classification of seismic waveform (VRS)
12Volume Reflection Spectrum analysis (VRS)
Trace Reconstruction S(t)C0 C1 t C2 t2 C3
t3 .. Cn tn (min lt t lt max)
13Volume Reflection Spectrum analysis (VRS)
11 coefficients sufficient to exactly reconstruct
the trace
14Coefficient maps Volume Reflection Spectrum
(VRS)
15Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
First input vector
16Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
First input vector
17Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Second input vector
18Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Second input vector
19Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Third input vector
20Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Third input vector
21Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Fourth input vector
22Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Fourth input vector
23Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
First input vector
24Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
First input vector
25Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Second input vector
26Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Second input vector
27Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Third input vector
28Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Third input vector
29Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Fourth input vector
30Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
Fourth input vector
31Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
32Training phase Competitive Learning 2-attribute
graphical example
Euclidian Distance
33Classification phase Competitive
Learning 2-attribute graphical example
Training data Other data
34Unsupervised Classification
- Conscience factor
- One output vector could win all input vectors
- Reduces greediness from algorithm (during
training phase) - Output vector that wins often bad conscience,
withdraws from competition - Results in even spread of input vectors over
output classes
35Coefficient maps are used as the attribute maps
for classification
Now lets look at the real thing The
N-dimensional case
(X1, X2, .. XN) Represents one attribute vector
36Artificial Neural Network Determining the
cluster centers
X1
X2
X3
(X1, X2, .. XN) Represents one attribute vector
X4
X5
XN
Output Layer (Number of classes)
Input Layer (Number of attributes)
37Wavelet Classification Result E-BD2
Oil bearing channel sandstone
Water bearing sandstone E-BD2
38Conclusion What does classification of wavelet
information reveal about reservoir size and
quality?
39Conclusion What does classification of wavelet
information reveal about reservoir size and
quality?
40Thank you