This invention relates to the field of seismic reservoir characterisation, and in particular to the automated mapping of structural geometries, like seismic interfaces and fault displacements.
Formation layers, lithological boundaries, sedimentary bedding, etc. in the underground can be defined through the interfaces between different lithologies, which produce seismic reflections due to impedance contrasts. These seismic reflections are referred to as seismic horizons, and interpretation of seismic horizons plays an important role in the structural characterization of 3D seismic data. Seismic horizons are commonly interpreted as being located along minimum, maximum, or zero crossing values in a seismic volume, and interpretations can be obtained by manual or automatic extraction of such surfaces. Structurally complex regions are a challenge to existing interpretation procedures, both automated and manual. For the manual interpreter, structural complexity produces ambiguities, and the interpretation process can become very time consuming. The continuity constraint incorporated by most automated algorithms may fail to hold. In particular, automatic continuation of an interpretation across faults is a challenge.
The present invention addresses this problem by laterally detecting and combining horizon segments having similar seismic waveforms. The mapping is not restricted to spatial continuity in the position of the seismic horizon, but instead determines possible lateral continuations based on similarity of signal shape. The inventive method represents waveforms as coefficients, typically seismic derivatives, that can be used to reconstruct the seismic trace in the vicinity of the extrema positions, such as by utilizing Taylor series expansions. The term “extrema” as used throughout this patent application includes any characteristics of seismic traces that can be used to track the positions of seismic horizons, such as minimum values, maximum values, zero-crossing values, midpoints between zero-crossing values and maximum or minimum values, etc. Derivatives that can be used in the Taylor series reconstruction and sub-sample accuracy extrema positions can be calculated based on orthogonal polynomial spectral decompositions of the seismic traces, such as described in U.S. Pat. No. 6,240,370, issued May 29, 2001 to Lars Sønneland et al.
The lateral continuations of horizons can be exploited to automatically estimate fault displacement, with a spatially varying displacement offset along the fault plane. Fault displacement influences the connectivity of hydrocarbon bearing lithologies, and assessment of the displacement provides an improved description of a reservoir. The present invention can calculate the displacement along pre-interpreted fault planes, for example fault planes that have been extracted from the seismic data using the methodology described in U.K. patent GB 2,375,448 B issued Oct. 15, 2003 to Schlumberger Holdings Limited.
Accordingly, it is an object of the present invention to provide an improved method of processing and interpreting seismic data. In one embodiment, this involves extracting surface primitives, and grouping extrema positions where these surface primitives are both spatially continuous along the extrema of the seismic volume and are continuous in class index in the classification volume.
Classification tools are widely used in both 3D and 4D reservoir characterization, for example in the mapping of 3D structures, lithological properties and production effects. In this invention, we extend the application area of the classification methodology into automated interpretation of seismic reflectors and fault displacement calculations. By classifying the seismic waveform along reflectors, we gain an improved automated interpretation that also performs well in structurally complex regions. The seismic waveform around the extrema positions can be represented by a set of coefficients that can be input into the classification process. These coefficients can also be used to reconstruct the seismic data waveform shape in the vicinity of the extrema positions, such as by using Taylor series reconstruction. One-point support for the reconstruction is an important element in the classification of seismic reflectors, as it allows the classification to be performed only along extrema positions while utilizing information regarding the waveforms in intervals around the extrema positions. This substantially reduces the number of data points to be classified, allowing the 3D classification to be run on a sparse 3D volume. The workflow of this preferred embodiment of the method, referred to as extrema classification, is outlined in
For existing fault surfaces, the present invention can automatically extract and match pairs of horizons on opposite sides of each fault. The displacement can then be calculated as the offset of these horizons in their intersections with the fault plane. This constitutes a fully automatic procedure for fault displacement assessment, providing displacement estimates that vary spatially along the fault plane.
A flow diagram outlining the extrema classification methodology is shown in
Orthogonal polynomials may be used to reconstruct the seismic trace S(z) locally, such as by using the Volume Reflection Spectral decomposition technique (VRS) where orthogonal Chebyshev polynomials gi are used as the basis functions:
S(z)=b0g0(z)+b1g1(z)+ . . . +bngn(z) (1)
This method is described in significantly more detail in U.S. Pat. No. 6,240,370, issued May 29, 2001 to Lars Sønneland et al.
Orthogonal polynomial reconstruction provides an analytic representation of the seismic signal from which high-accuracy positions of extrema can be calculated. The resulting extrema points may be represented through two sparse 3D volumes. A first cube can contain the amplitudes at the extrema, positioned at the voxels vertically closest to the extrema points. If the types of extrema being identified are zero-crossings, then a non-zero placeholder value (such as 1.0) could be used to mark the identified voxels. Voxels between extrema points are generally assigned zero value. A second cube contains values describing the vertical distances between the voxel positions where the amplitudes are stored, and the exact positions of the extrema, i.e., the sub-sample precision. The set of voxels containing extrema data is the same for the two cubes, but they contain amplitude and sub-sample position values respectively.
The second step in the workflow of the current invention is to generate a set of coefficients that characterize the seismic data in the vicinity of the extrema positions, which we refer to as “waveform attributes”. The coefficients can be derivatives of the seismic signal at the single extrema data point and the seismic trace in the vicinity of the extrema point can be reconstructed using these coefficients in a Taylor series expansion:
S(z+h)≈S(z)+hS′(z)+h2S2(z)/2!+ . . . +hnSn(z)/n! (2)
This reconstruction gives a good fit to the seismic signal in a region around the point z, where the quality of the reconstruction depends on the number of coefficients/derivatives included in the Taylor series. Each trace in the 3D seismic volume is typically reconstructed using VRS decomposition, where the polynomial reconstruction (1) enables analytical calculations of the derivatives Sk(z) at each sample point along the trace. Although multiple data points are involved in calculating the derivative values, the seismic signal in a region can afterwards be reproduced based on a set of derivatives obtained for one single data point. In particular, if derivatives are obtained only at extrema points along the seismic trace, this is sufficient to reconstruct the full seismic trace. This process is shown in
Other types of reconstruction methods with one-point support can also be used in the inventive method. The one-point support of the reconstruction is an important feature, since this enables the classification to be limited to extrema points only, but still be related to the overall shape of the seismic signal in the vicinity of the extrema points.
The next step involves forming groups of extrema points that have similar coefficients. Four alternative approaches to classification of seismic extrema may be utilized: three aiming at producing surface primitives for horizon interpretation or volume building; and one for fault displacement estimation. Common to all approaches is an underlying statistical model, assuming the derivative attribute vectors ak of length p, where k is the voxel index, are Gaussian distributed within each class. The number of derivative attributes to include in the classification is typically selected by the user. Furthermore, the first derivative may be excluded from the classification, since they are zero, by definition, at maximum or minimum extrema points.
The mean vectors μ=(μ1, . . . , μn) and covariance matrices Σ=(Σ1, . . . ,Σn) of the Gaussian distributions describe respectively the position and shape in attribute space of classes 1, . . . , n. The probability density function (pdf) of the complete set of K attributes vectors a=(a1, . . . , aK) given Gaussian parameters and class indexes c=(c1, . . . , cK) of all voxels, is given as:
where the pdf of voxel k is given as:
The unknowns are the class indexes ck and the Gaussian parameters μ and Σ, which can be estimated using either supervised or unsupervised classification. The aim of the classification process is to subdivide the attribute vectors into n classes, providing a separation of the attribute space as illustrated for three attributes in
The volume of interest for performing the extrema classification can be chosen, for instance, as a vertical window with constant thickness or as the volume confined between two pre-interpreted seismic horizons. This enables, for example, the extrema classification to be run only within a reservoir formation.
In the first approach, supervised classification with picked seed points, the user picks seed points along reflectors of interest, each reflector representing a separate class. This is shown in
where Tc is the set of training data for class c and mc is the number of training data in Tc. Two additional background classes are included: positive extrema not being training data; and negative extrema not being training data. Each extrema not being a training point is then assigned the class index maximizing the pdf (4) in that voxel. This process is shown in
In the second approach, unsupervised classification with a specified number of classes, the number of classes is provided as a user input (shown in
In the third approach, unsupervised classification with an unspecified number of classes, the algorithm starts with selecting at random a seed point among all extrema, and extracts from the extrema cube a small, spatially continuous horizon segment locally around the seed point. This process is shown in
In a fourth approach, unsupervised classification for fault displacement estimation, fault surfaces are provided as input to the extrema classification algorithm. First, spatially continuous horizon segments are automatically extracted from one side of the fault, each assigned to a separate class index. This is shown in
As shown in
A class consistent horizon interpretation, resulting from one class only, consists of extrema points with a similar shape of the seismic waveform in a neighborhood around the extrema, which are likely to belong to the same reflector. The building of the final horizon interpretations typically requires some manual effort. The extrema classification provides a set of patches, or geometry primitives, forming the basis for the interpretation. The classification index guides the interpreter in structurally complex regions, reducing the risk of interpreting along the wrong events. Overall, the extrema classification method can substantially improve the manual horizon interpretation process, easing the job and reducing the time spent by the interpreter.
The inventive method may also be used to extract seismic volumes. When one or a few classes appear repeatedly vertically within a region, as illustrated in
The inventive method may also be used in the estimation of fault displacements. Fault displacement is defined as the distance a horizon has been offset along the fault surface, due to faulting. Starting with an existing fault surface [see, for instance, Randen, T., Pedersen, S. I., and Sønneland, L. (2001), Automatic extraction of fault surfaces from three-dimensional seismic data, In Expanded Abstr., Int. Mtg., Soc. Exploration Geophys., 551-554] (
Fault displacement calculations can be performed between the two intersection lines corresponding to a pair of horizons. The calculations are performed in the dip-slip direction, assuming no lateral movement along the fault surface. Displacements are calculated between pairs of intersection points, one from each of the two horizon segments, in a vertical plane placed perpendicular to the horizontal direction of the fault plane (
The sample units of a seismic volume typically differs vertically and laterally, e.g., milliseconds in the vertical direction and meters in the horizontal direction. To avoid confusion between different measurement units, the displacement may be decomposed into the vertical throw component and the horizontal heave component.
The primary advantage of the present invention over existing methodologies for automated seismic interpretation lies in its ability to track reflectors laterally in structurally complex regions, for example across faults.
In
The inventive method will typically be implemented on a computer system of the type shown in schematic format in
While the invention has been described herein with reference to certain examples and embodiments, it will be evident that various modifications and changes may be made to the embodiments described above without departing from the scope of the invention as set forth in the claims below.
This application claims the benefit of U.S. Provisional Application No. 60/461,782, filed Apr. 10, 2003, incorporated herein by reference.
Number | Name | Date | Kind |
---|---|---|---|
4672546 | Flinchbaugh | Jun 1987 | A |
4727488 | Flinchbaugh | Feb 1988 | A |
5570106 | Viswanathan | Oct 1996 | A |
5615171 | Hildebrand | Mar 1997 | A |
5966672 | Knupp | Oct 1999 | A |
6014344 | Baker, III | Jan 2000 | A |
6016287 | Klebba et al. | Jan 2000 | A |
6240370 | Sonneland et al. | May 2001 | B1 |
6278949 | Alam | Aug 2001 | B1 |
6388947 | Washbourne et al. | May 2002 | B1 |
6463387 | Runnestrand et al. | Oct 2002 | B1 |
20020126896 | Pedersen | Sep 2002 | A1 |
20030023383 | Stark | Jan 2003 | A1 |
Number | Date | Country |
---|---|---|
0 181 216 | Jul 1985 | EP |
407989 | Aug 1932 | GB |
2 375 448 | Oct 2003 | GB |
WO 9837437 | Aug 1998 | WO |
WO 0133255 | Oct 2000 | WO |
Number | Date | Country | |
---|---|---|---|
20040260476 A1 | Dec 2004 | US |
Number | Date | Country | |
---|---|---|---|
60461782 | Apr 2003 | US |