Patent Application
20100228486
1. Field of the Invention
This invention relates to the field of seismic data processing. In particular, the invention relates to seismic data pre-processing for extrema classification and Extrema Sequence classification.
2. Background
Seismic data acquisition and processing are key components in geophysical exploration. Seismic data are collected in order to analyze the subsurface of the Earth, in particular for hydrocarbon exploration. The seismic signals released from a seismic source, which may comprise explosives or an impulse of compressed air. The signals reflected by the various layers beneath the surface of the Earth are known as traces and are sensed by a larger number, typically hundreds of sensors such as geophones. The reflected signals are recorded and the results are analyzed to derive an indication of the layer formations beneath the subsurface. In a seismic survey, formation layers, lithological boundaries, sedimentary bedding, etc. can be defined through the interface between different formation layers, which produces seismic reflections due to impedance contrast.
Seismic reflections are referred as seismic horizons, which is important in structural characterization of 3D seismic data. Seismic horizons are commonly interpreted as being located along minimum, maximum, or zero crossing value in a seismic volume. Various seismic data processing methods, including manual interpretation and automatic extraction of the seismic horizon have been developed. U.S. Pat. No. 7,248,539, issued to Borgos et al., discloses a method of automatic seismic reflector interpretation and fault displacement calculations. The method provides a sophisticated automatic approach for extracting interpretation primitives using a sub-sample precision algorithm that also computes attributes for each primitive. Thus, fast and accurately merging the interpretation primitives into geological structures. According to Borgos' method, generating extrema cubes is a basic and one of the most important part of the process. The term “extrema” as used throughout this patent application includes any characters or seismic traces that can be used to track the position of seismic horizons, such as minimum values, maximum values, zero-crossing values and midpoints between zero-crossing values and maximum or minimum values, etc. According to Borgos', high accuracy positions of extrema points can be calculated by orthogonal polynomial reconstruction of seismic traces. However, for some reasons, the seismic measurement data may contain distortions and the seismic traces are not always continuous and valid. Thus, the positions and the amplitudes of the extrema points derived from the distortions are useless, which causes faults in the resulting extrema cubes and seismic interpretation faults.
Another important feature of Borgos' method is to classify seismic waveforms around the reflectors, thus, provided an improved automatic interpretation. However, such classification relies on the user to decide the right number of classes to produce. It is always hard for the user to view and decide, because the samples to be classified are typically 3 dimensional. Thus, users have to guess a number and take hours to produce the result, and maybe another couple of hours for another number of classes.
In one aspect, the present invention relates to methods for data processing, particularly seismic data represented in three dimensions (3D). A method in accordance with one embodiment of the invention includes identifying extrema points from the 3D seismic data set; removing artificial distortion from 3D seismic data set; generating extrema cubes and derivatives along extrema points; estimating a class number associated with the extrema points; and determining number of classes and dynamically classifying extrema points.
In another aspect, the present invention relates to methods for reducing artificial distortion from 3D seismic data set. In accordance with one embodiment of the invention, a method for processing 3D seismic data set comprises selecting a seismic trace from the 3D seismic data set; removing dead trace data from the seismic trace; extracting extrema points from the seismic trace; removing extrema points by comparing distance between adjacent extrema points and characteristics of the adjacent extrema points; and outputting remaining extrema points.
In another aspect, the present invention relates to methods for auto-estimating class number for extrema classification or extrema sequence classification. In accordance with one embodiment of the invention, a method for processing 3D seismic data set comprises extracting extrema points from the 3D seismic data set; K-mean clustering the extracted extrema points into N clusters; calculating Gaussian parameters of each cluster; grouping clusters according to the Gaussian parameters of all clusters; and outputting amount of groups as class number for extrema classification or extrema sequence classification.
U.S. Pat. No. 7,248,539, issued to Borgos et al., the disclosures of which are incorporated here by reference, discloses a method of Extrema Classification, which discloses generating extrema points by using Taylor expansion method. However, the generated points can only be called extrema points in theory, because the raw seismic measurement data always include noise. These extrema points derived from raw measurement data may result in unacceptable seismic interpretations. Normally, the user has to manually identify the interpretation faults and remove those. Since most of these interpretation faults originate from the fake extrema points generated in the previous detection step, better interpretation results could be achieved by removing the artificial distortions before the step of extrema point detection.
According to an embodiment of the invention, the distortions and noise of the seismic database can be categorized in three main types that are useless dithering, clamped values and dead traces.
1. Useless dithering:
2. Clamped value: The clamped values are actually distortions caused by the data processing of data acquisition tools. Some acquisition tools do rough pre-processing that replaces data having amplitude more than threshold amplitude with the threshold. As shown in
3. Dead trace: Dead traces are caused by the rough data processing by data acquisition tools. Seismic data are reflections captured by geophones, and the captured reflections are time-indexed. In order to demonstrate the subsurface formation, the received reflections need to be transformed from time domain to depth domain. This transformation could also cause distortions. For instance, a seismic wave propagates faster in rigid formation layers than in loose or porous layers, and the depth interval for each sampling in rigid layers will be longer. That could result in the depth errors of the seismic traces. In order to compensate the depth errors, acquisition tools normally scale down a portion of seismic trace for a loose formation layer and fill the break with a straight line of zero amplitude. This straight line with zero amplitude does not include any meaningful data. In extrema extraction and classification process, the breaks of the seismic data traces should not be considered. Thus, each individual trace needs to be check and meaningless samples need to be removed from the trace.
According to an embodiment of the invention, as shown in
According to an embodiment of the present invention, generating extrema cubes 105 involves reconstructing the seismic traces by using orthogonal polynomials, which is described in significantly more detailed in U.S. Pat. No. 6,240,370, issued May 29, 2001 to Las Sonneland et al. Orthogonal polynomial reconstruction provides an analytic representation of the seismic signal from which high accuracy positions of extrema points can be calculated. The resulting extrema points may be represented through two sparse 3D volumes. One of the extrema cubes may contain the amplitudes of the extrema points, 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.
After the extrema cubes being generated 105, derivatives along extrema points will be generated 106. According to one embodiment of the invention, the derivatives/coefficients characterize the seismic data in the vicinity of the extrema positions. The derivatives can be calculated by reconstructing seismic signals or the entire seismic trace in vicinity of the extrema point, which is described in significantly more detailed in U.S. Pat. No. 6,240,370, issued May 29, 2001 to Las Sonneland et al.
Then, auto-estimating class number 107. The step of auto-estimating class number will be more specifically explained in the following paragraphs by reference of
According to an embodiment of the invention, the step of artificial distortion reduction starts with determining whether all traces in the cube have been processed 201. As shown in
According to an embodiment of the invention, determining whether there are valid changes in the related sample values includes comparing the characteristics of every two contiguous extrema points. The characteristics include, for example, whether an extrema point has a maximum value, minimum value or zero crossing value. As shown in
Then, determining whether all seismic samples in the trace have been handled 210, if so, go to step 201 to determining whether all traces in the cube have been processed, otherwise, go back to step 204 to extract extrema points from the trace.
According to an embodiment of the invention, also as shown in
According to an embodiment of the invention, auto-estimating class number includes using random sample coordinates as seeds for K-mean clustering.
According to an embodiment of the invention, for a one-attribute space of single dimension d, the similarity of two Gaussian distributed clusters in d dimension is presented as follow:
l=modulus(μi−μj)
d=normalize(abs(μi−μj))
t=c·modulus(d·(σi+σj))
closeness=l−t(the lower the value is, the closer the clusters will be)
Some embodiments of the invention relate to systems that implement the above described methods. A system of the invention may include a processor and a memory that store a program having instructions for causing the processor to perform the steps of a method of the invention. Such systems may be implemented on any computer (such as a personal computer or workstation) or any computing unit known in the art. Some embodiments of the invention relate to computer readable media, which store a program having instructions for causing the processor to perform the steps of a method of the invention.
Advantages of the invention may include one or more of the following. Methods of the invention use dynamic filtering of a large collection of geometric primitives to quickly isolate a desired subset of available geometric primitives. The filtering may be based on proximity to a selected point in 3D as well as a selected property of the object. This will facilitate analysis of complex data set to afford quick identification of useful information.
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this disclosure, will appreciate that other embodiments can be envisioned that do not depart from the scope of the invention as disclosed herein. Accordingly, the scope of the invention shall be limited only by the attached claims.
