CN110361782B - Seismic waveform clustering method and device - Google Patents
Seismic waveform clustering method and device Download PDFInfo
- Publication number
- CN110361782B CN110361782B CN201810313025.5A CN201810313025A CN110361782B CN 110361782 B CN110361782 B CN 110361782B CN 201810313025 A CN201810313025 A CN 201810313025A CN 110361782 B CN110361782 B CN 110361782B
- Authority
- CN
- China
- Prior art keywords
- frequency band
- band range
- seismic
- seismic data
- inherent modal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 230000001360 synchronised effect Effects 0.000 claims abstract description 15
- 230000006870 function Effects 0.000 claims description 45
- 238000010586 diagram Methods 0.000 claims description 11
- 238000010183 spectrum analysis Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000010587 phase diagram Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 abstract description 18
- 238000001125 extrusion Methods 0.000 abstract description 13
- 238000000354 decomposition reaction Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 abstract description 7
- 238000004422 calculation algorithm Methods 0.000 abstract description 6
- 238000012544 monitoring process Methods 0.000 abstract description 6
- 238000001228 spectrum Methods 0.000 abstract description 3
- 230000006872 improvement Effects 0.000 abstract description 2
- 238000005728 strengthening Methods 0.000 abstract 1
- 238000011426 transformation method Methods 0.000 abstract 1
- 239000013598 vector Substances 0.000 description 19
- 208000035126 Facies Diseases 0.000 description 18
- 238000004458 analytical method Methods 0.000 description 11
- 230000009466 transformation Effects 0.000 description 5
- 230000036039 immunity Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000007621 cluster analysis Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 235000006810 Caesalpinia ciliata Nutrition 0.000 description 1
- 241000059739 Caesalpinia ciliata Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 210000000987 immune system Anatomy 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/30—Analysis
- G01V1/306—Analysis for determining physical properties of the subsurface, e.g. impedance, porosity or attenuation profiles
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/62—Physical property of subsurface
- G01V2210/624—Reservoir parameters
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention relates to a seismic waveform clustering method and device, and belongs to the field of seismic exploration foundation application. The method makes full use of a synchronous extrusion transformation method, and self-defines the number and frequency band range of the inherent modal components aiming at different seismic data and the frequency spectrum characteristics of different intervals in the process of decomposing the seismic data into the inherent modal function components; and (3) optimizing the inherent modal component, and denoising the high-frequency component with high noise without directly abandoning the high-frequency component. The improvement of the decomposition algorithm and the component processing thought is helpful for better retaining or strengthening the critical information and removing the noise and the redundant information. Meanwhile, effective quality monitoring can be provided, the screened/discarded components and the reconstructed seismic data volume in the reconstruction process can be mapped to perform quality monitoring, whether the removed components are noise or redundant information or not is checked, and whether main frequency band information is reserved or not is checked, so that the reliability of waveform clustering is greatly improved.
Description
Technical Field
The invention belongs to the field of seismic exploration basic application, and particularly relates to a seismic waveform clustering method and device.
Background
The seismic waveform is a comprehensive attribute parameter, contains information such as amplitude, frequency, phase, geometric shape and the like, and can more reliably reflect real underground geological conditions when used for cluster analysis; but the seismic waveform also contains a lot of noise and redundant information, so that the clustering result is seriously interfered by the noise and has lower resolution.
Unsupervised seismic waveform clustering is a widely used seismic facies analysis method, and plays an important role in seismic sedimentary facies analysis and reservoir prediction, and a common method is to input seismic waveforms into a Self-Organizing neural network (SOM) for Learning (Learning) and then clustering.
An unsupervised seismic facies analysis method based on wavelet transformation was proposed by the authors Matos et al (geopysic, 2007, vol.72, No.1, P: 9-21). The method applies a time-frequency technology to a pattern recognition system for the first time, represents singular points of seismic signals by using WTIMLA (wavelet Transform Module) Line amplifiers) conversion, and performs SOM waveform clustering by using the whole WTIMLA curve as input data so as to improve the interpretation fault tolerance of a clustering result. The method shows higher capability of tolerating interpretation errors in model experiments, but because WTIMMALA curves are used for replacing seismic data, the original information of the seismic data is greatly reduced, the reliability of clustering results is poor, and the application effect in actual seismic data is not obvious.
Authors Saraswat and Sen propose a seismic facies analysis method (geophhysics, 2012, vol.77, No.4, P: 45-53) based on AIS, which uses AIS (artificial Immune system) to greatly reduce seismic data, removes noise and redundant information of the seismic data, uses the reduced seismic data for cluster analysis, and makes the clustering result not easily affected by the noise and redundant data. The method shows better noise immunity in the clustering of actual seismic data (Dutch F3 block), but AIS reduces the information of the seismic data by 50%, and noise and redundant data are difficult to be removed without quality control, and some main frequency band information can be removed, so that the reliability is poor.
An EMD-based seismic facies analysis method (Journal of applied geophilics, 2015, No.112, P: 52-61) is proposed by an author, Duhakun et al. According to the method, an EMD technology is utilized to decompose seismic signals into finite Intrinsic Mode Functions (IMFs) with different frequency band ranges, then components with less display noise and higher correlation with original seismic signals are screened out to be reconstructed, and finally, the reconstructed data is utilized to perform seismic waveform clustering. The method has a good anti-noise effect, but in the process of decomposing the seismic data, the number of components and the frequency band range are determined by the seismic data and the convergence condition of an algorithm, generally 7-10 components, so that the frequency band range of a single component is too narrow, the difference with an original profile is large, and an effective judgment criterion is lacked in component screening and processing; on the other hand, when the components are filtered, the components with more noise are directly discarded, and the components may contain some valid information. Therefore, the method has limitations on the decomposition algorithm and the component processing idea, and the reliability of the clustering result is influenced.
In summary, how to improve the noise immunity and resolution of waveform clustering while ensuring reliability is a subject of research in recent years for cluster analysis; the application of the time-frequency analysis method to unsupervised seismic waveform clustering is one of the most potential research directions, but the currently used time-frequency analysis method has respective limitations and still needs to be improved aiming at the characteristics of seismic signals.
Disclosure of Invention
The invention aims to provide a seismic waveform clustering method and a device, which can strengthen quality monitoring, optimize component screening thought, remove strong noise points, ensure the noise immunity and reliability of clustering results and solve the problem that the existing seismic waveform clustering method cannot simultaneously ensure the clustering reliability and noise immunity when seismic data are preprocessed.
In order to solve the technical problem, the invention provides a seismic waveform clustering method, which comprises the following solutions:
the first method scheme comprises the following steps:
1) acquiring a frequency band range of seismic channel signals in target layer seismic data, and dividing the frequency band range to at least comprise a first set frequency band range and a second set frequency band range; the first set frequency band range is a main frequency band range of the seismic trace signal, and the main frequency band range is a frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the second set frequency band range is the frequency band range except the first set frequency band range, and the target layer seismic data keeps the layered spread characteristic on the section;
2) decomposing the target layer seismic data to generate at least two natural modal function components, wherein the at least two natural modal function components comprise a first natural modal component and a second natural modal component, the frequency band range of the first natural modal component is the first set frequency band range, and the frequency band range of the second natural modal component is the second set frequency band range;
3) and denoising the second inherent modal component, superposing and reconstructing the second inherent modal component and the first inherent modal component after the second inherent modal component is processed to obtain a reconstructed seismic data volume, and learning and clustering the reconstructed seismic data volume to obtain a waveform seismic phase diagram.
According to the second method scheme, on the basis of the first method scheme, the first set frequency band range is obtained after the target layer seismic data are subjected to spectrum analysis; and the upper limit value of the second set frequency band range is the cut-off frequency of the seismic channel signal.
And a third method scheme and a fourth method scheme, wherein on the basis of the first method scheme and the second method scheme, the first set frequency band range is the main frequency band range of the seismic channel signals determined after synchronous extrusion wavelet transformation is carried out on the seismic data of the target layer and the time-frequency graph of the seismic channel signals is obtained.
And a fifth method scheme, which is to decompose the seismic data of the target layer by adopting synchronous extrusion wavelet transform on the basis of the first method scheme.
And a sixth method, on the basis of the first method, learning is self-organizing neural network learning.
In order to solve the above technical problem, the present invention further provides a seismic waveform clustering apparatus, including the following solutions:
apparatus aspect one, comprising a processor configured to execute instructions that implement a method comprising:
1) acquiring a frequency band range of seismic channel signals in target layer seismic data, and dividing the frequency band range to at least comprise a first set frequency band range and a second set frequency band range; the first set frequency band range is a main frequency band range of the seismic trace signal, and the main frequency band range is a frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the second set frequency band range is the frequency band range except the first set frequency band range, and the target layer seismic data keeps the layered spread characteristic on the section;
2) decomposing the target layer seismic data to generate at least two natural modal function components, wherein the at least two natural modal function components comprise a first natural modal component and a second natural modal component, the frequency band range of the first natural modal component is the first set frequency band range, and the frequency band range of the second natural modal component is the second set frequency band range;
3) and denoising the second inherent modal component, superposing and reconstructing the second inherent modal component and the first inherent modal component after the second inherent modal component is processed to obtain a reconstructed seismic data volume, and learning and clustering the reconstructed seismic data volume to obtain a waveform seismic phase diagram.
According to the second device scheme, on the basis of the first device scheme, the first set frequency band range is a main frequency band range of a target layer after the target layer seismic data are subjected to spectrum analysis; and the upper limit value of the second set frequency band range is the cut-off frequency of the seismic channel signal.
And the first set frequency band range is the main frequency band range of the seismic channel signals determined after synchronous extrusion wavelet transformation is carried out on the seismic data of the target layer and a time-frequency graph of the seismic channel signals is obtained on the basis of the first and second device schemes.
And in the fifth device scheme, the target layer seismic data are decomposed by adopting synchronous extrusion wavelet transform on the basis of the first device scheme.
And a sixth device scheme, wherein on the basis of the first device scheme, the learning is self-organizing neural network learning.
The invention has the beneficial effects that: according to the method, original seismic data are decomposed and reconstructed to obtain a seismic facies map which at least comprises a first modal function component and a second modal function component, the second modal function component is subjected to denoising processing and then is superposed and reconstructed with the first modal function component, and a reconstructed data body is subjected to learning and clustering to obtain the seismic facies map. The method can set the number and the frequency band range of the modal function components in a user-defined manner, retain the first modal function component corresponding to the main frequency band range, and superpose the second modal function component which has more noise and still shows the layered spread characteristic in the second set frequency band range with the first modal function component after denoising treatment to obtain the reconstructed seismic data body for generating the seismic facies diagram, thereby simultaneously ensuring the reliability and the antinoise performance of seismic waveform clustering and effectively removing the noise and redundant information.
Furthermore, in order to obtain the accurate main frequency band range of the seismic channels, the method utilizes synchronous extrusion wavelet transformation to obtain a time-frequency graph of each seismic channel signal, then carries out time-frequency analysis on each seismic channel respectively, determines the main frequency band range and the number of modal function components of each seismic channel, and then reconstructs the modal function components channel by channel. However, when the synchronous extrusion wavelet transform processing is performed on the actual seismic data, the seismic data often include hundreds of thousands of seismic channels, and in order to reduce the calculation amount and improve the data processing efficiency, the main frequency band range of all seismic channel signals adopts the main frequency band range of the target interval, and the main frequency band range of the target interval is determined by performing the spectrum analysis on the seismic data of the target interval.
Further, the first set frequency band range set by the invention is 4-39Hz, and the first mode function component corresponding to the frequency band range occupies the main energy information of the seismic signal, and reserves and does not take any measures. And setting a second set frequency band range to be 40-80Hz, wherein a second modal function component corresponding to the frequency band range is represented as a layered spread characteristic, useful information in the second modal function component is reserved after denoising, and other modal function components which are larger than 80Hz and cannot keep the layered spread characteristic are discarded. The reflection of the main energy information of the seismic trace signals is highlighted while the information such as noise is suppressed.
Furthermore, before the seismic data body is reconstructed for learning, the size of a time window which can just contain the information of the target interval is set, the waveform information in the time window is not influenced by noise, and the reliability of the clustering result is not influenced because the information of the non-target interval is not contained. And the cluster number which is larger than the number of sedimentary facies is also set, so that sedimentary facies information is fully and accurately reflected.
Drawings
FIG. 1 is a flow chart of a seismic waveform clustering method;
FIG. 2 is a graph of spectral analysis of a target interval in the West Chuanxi region, wherein the main frequency band of the target interval is 4-39 Hz;
FIG. 3 is a single-pass time-frequency analysis plot of a target interval generated by SST;
FIG. 4 is a diagram of IMF component 1 of SST decomposition, with a frequency band in the range of 4-39 Hz;
FIG. 5 is a diagram of IMF component 2 of SST decomposition, with a frequency band in the range of 40-80 Hz;
FIG. 6 is a diagram of the IMF component 3 of SST decomposition with a frequency band range greater than 80 Hz;
FIG. 7 is a schematic diagram of IMF component 2 after denoising;
FIG. 8 is a reconstructed seismic profile of IMF components 2 after SST processing and IMF component 1 stacking denoising;
FIG. 9 is an original seismic section;
FIG. 10 is a graph of SST-based unsupervised seismic waveform clustering of the present invention;
FIG. 11 is a graph of EMD-based unsupervised seismic waveform clusters.
Detailed Description
The invention is further described below by taking seismic waveform clustering of karst reservoirs at the tops of the Leiko slope groups in the West-Chuancheng area as an example and combining the accompanying drawings. The implementation method of the invention is shown in figure 1, the target layer section is the top of the mine mouth slope group, and the implementation method comprises the following steps:
1. and carrying out spectrum analysis on the seismic data volume of the target interval. As can be seen from the analysis result of FIG. 2, the main frequency band of the seismic data body at the top of the Raido slope group in the area is 4-39Hz, and the judgment standard is the frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the center frequency was 21.5 Hz.
2. And performing synchronous extrusion wavelet Transform (SST) on the seismic channel signals at the top of the lightning hole slope group of the target layer section in the work area channel by channel, analyzing a time-frequency diagram generated by each channel, and reconstructing to generate an IMF component in a user-defined frequency band range.
And 2.1, synchronously extruding wavelet transform (SST) channel by channel for the seismic channel signals of the target layer section in the work area, and analyzing the time-frequency diagram generated by each channel. The synchronous squeeze wavelet transform is an adaptive nonlinear non-stationary signal decomposition algorithm combining wavelet transform and frequency redistribution algorithm proposed by an author Daubechies et al in 2011.
Performing Continuous Wavelet Transform (CWT) on the seismic channel signals s (t):
where a is a scale factor and b is a time shift factor, #*Is a complex conjugate of a source wavelet, WsThe (a, b) coefficients represent the concentrated time-frequency image used to extract the instantaneous frequency.
According to Plancherel's theorem, equation (1) can be written as:
in the formula, ξ represents the angular frequency,
and
representing the fourier transforms of ψ (t) and s (t), respectively,
meaning that the spectrum of psi (t) is conjugated,
considering a fourier pair of a single harmonic signal s (t) ═ Acos (ω t), can be expressed as:
equation (2) can be transformed into:
because of the wavelets
At a central frequency ω0So W is extracteds(a, b) will be at horizontal line a ═ ω0Extracting near/omega, a is omega0And/ω represents the ratio of the center frequency of the wavelet to the center frequency of the signal. But actually Ws(a, b) always lie around this horizontal line, producing a blurred projection in the time scale representation. This blurring occurs mainly on a scale along the time shift factor b. This ambiguity is rarely found along the dimension b of the scale axis. It can be shown that the instantaneous frequency ω is negligible when the ambiguity in the b dimension is negligibles(a, b) can be calculated using the following equation:
for arbitrary points (a, b), Ws(a, b) ≠ 0, maps information from the time-scale plane to the time-frequency plane, and transforms each point (b, a) to (b, ω)s(a, b)), this step is called extrusion synchronization.
Since a and b are discrete values, the scaling step Δ ak＝ak-1-akWill encounter any akThe above. Then the time-frequency distribution of synchronous extrusion transformation can be only in the frequency band range [ omega ]l-Δω/2,ωl+Δω/2]Center of (a) (-)lDetermining:
wherein Δ ω ═ ωl-ωl-1，Ts(ωlB) is at the center frequency ωlSST time-frequency distribution.
And 2.2, analyzing the main energy distribution range of each seismic signal according to the SST time-frequency diagram. As shown in fig. 3, it is a time-frequency diagram of a single seismic signal in a target interval in a work area, and it is preferable that the frequency band range of the main energy of the seismic signal is 6-42 Hz. To reduce the effects of noise and redundant information, three Intrinsic Mode Function (IMF) components are generated here for reconstruction of the IMF generated after SST decomposition of the seismic trace signal. The IMF component 1 is determined as a main frequency band range and is a main component of the seismic trace signal energy; the IMF component 2 is determined as a high-frequency component from the main frequency band of the seismic trace signal to the cutoff frequency of the main energy; the part larger than the cutoff frequency is the IMF component 3. As another embodiment, the seismic trace signal SST may be decomposed to generate an IMF component 1 and an IMF component 2, and the IMF component 3 may not be generated.
The method is an accurate method for determining the frequency band range, and performs time-frequency analysis on each channel to obtain main frequency band information and then reconstructs IMF channel by channel. However, when SST processing is performed on actual seismic data, the seismic data often contains hundreds of thousands of seismic traces. In order to reduce the amount of calculation and improve the data processing efficiency, as another embodiment, the main frequency band range determined in step 1 is 4-39Hz for all seismic trace signals, and is used as the main frequency band range uniformly for the whole seismic data. FIG. 4 shows IMF component 1, with a frequency band range of 4-39 Hz; FIG. 5 shows IMF component 2, in the band range of 40-80 Hz; fig. 6 shows the IMF component 3 over a frequency band greater than 80 Hz.
Equation (6) shows that the new time-frequency representation of the signal is synchronized only along the frequency axis. The coefficients of the CWT are redistributed by the SST to obtain an image on the time-frequency plane. The instantaneous frequency is also extracted from the new time-frequency representation. IMF component skConvertible by discrete simultaneous extrusion
And (3) carrying out reconstruction:
in the formula (I), the compound is shown in the specification,
is a constant dependent on the selected wavelet, Re (-) is the real part of the representation;
is Ts(ωlDiscrete versions of b)
tmIs a discrete time, tm＝t0+ m Δ t, m ═ 0,1,. ·, n-1; n is a discrete signal
Total number of samples in (1); Δ t is the sampling rate.
3. Processing the IMF component obtained in the step 2: retaining all information of the IMF component 1; denoising the IMF component 2; the IMF component 3 is removed. And adding the processed IMF component 1 and the IMF component 2 to obtain a reconstructed seismic data volume.
3.1, the IMF component 2 is a high-frequency component, has a frequency band range of 40-80Hz and has more noise, but still shows the layered spread characteristic, which indicates that the layered spread characteristic contains available information, and the judgment standard of the available information is whether the layered spread characteristic can be maintained on the section. Therefore, the IMF component 2 is not directly discarded, but is retained after denoising, and fig. 7 shows the IMF component 2 after denoising; the IMF component 3, greater than 80Hz, exhibits a profile essentially consisting of noise components, which cannot maintain the laminar spreading characteristics and is therefore directly removed; the IMF component 1(4-39Hz) occupies the bulk energy information of the seismic signal, and no action is taken and retained. Such processing is intended to highlight reflection of the energy information of the main body of the seismic-trace signal while suppressing information such as noise.
The component processing mainly reserves the main frequency band part and carries out denoising processing on the high-frequency component with more noise, and the main frequency band range and the high-frequency component frequency band range are different according to actual data.
And 3.2, adding the IMF component 2 obtained in the step 3.1 after denoising with the IMF component 1 to obtain a reconstructed seismic data volume processed by SST, wherein the images of the reconstructed seismic section and the original seismic section are compared in the steps of FIG. 8 and FIG. 9, so that the reconstructed seismic section retains the main characteristic information of the original section, the noise is reduced, and the reflection of the main body energy information is highlighted.
4. And (3) selecting a proper time window and clustering number for the top of the lightning interface slope group of the target layer section, performing SOM learning on the reconstructed seismic signals obtained in the step (3), and performing K-means clustering after learning is completed to obtain a seismic waveform clustering chart.
And 4.1, selecting a proper time window for the target interval. The time window selection directly influences the result of seismic waveform clustering, and an overlarge time window contains information of a non-target interval and influences the clustering result; a too small time window contains insufficient waveform information and is susceptible to noise. Ideally, the selected time window can just contain the information of the target interval; the karst reservoir at the top of the Leikou slope group generally develops within 90m below the unconformity weathering crust, and the time window is selected within 30ms below the top (unconformity surface) of the Leikou slope group according to the calculation of 6000 m/s.
And 4.2, selecting a proper clustering number. The seismic facies information reflects sedimentary facies information, so that the number of the selected seismic facies is larger than that of the sedimentary facies, and on one hand, the sedimentary facies information is reflected more fully and accurately; on one hand, noise or redundant information occupies one or more seismic facies, so that the number of clusters is preferably large, but not small, and is generally 5-7. And determining the waveform clustering number as 7 types according to the sedimentary facies and seismic facies characteristics of the top of the mine opening slope group in the work area.
And 4.3, performing SOM unsupervised learning on the reconstructed seismic data body which is obtained in the step 3.2 and is subjected to SST processing to obtain a prototype vector after learning.
Converting seismic data x to [ x ]1,x2,...xn]Expressed by an N-dimensional space vector and input into the two-dimensional grid to start learning. This two-dimensional grid is composed of Prototype vectors (Prototype vectors). The arrangement mode of the prototype vector is rectangular or hexagonal; the number of the devices can be freely set; the dimensions of which are the same as those of the seismic signals.
Defining the number of prototype vectors in the two-dimensional grid as Y, b ═ b[b1,b2,...bi]I is 1,2, …, Y, and all input prototype vectors are connected to each other in a hexagonal arrangement. The prototype vector with the minimum distance to the input vector x is denoted as bmThis prototype vector, called Best Match Unit (Best Match Unit), can be used to find b using equation 8m：
||x-bm||＝min||x-bi|| (8)
Updating the prototype vector corresponding to the best matching unit and the adjacent prototype vectors thereof, moving towards the winning vector belonging to the space, and the updating rule is as follows:
bi(t+1)＝bi(t)+λ(t)hbi(t)[x-bi(t)](9)
λ (t) is a learning parameter, hbi(t) is a neighborhood function of the best matching unit, expressed as:
σ (t) is the effective width of the neighboring vector, | | rb-ri||2Is the winning vector bmAnd its neighboring vector biThe lateral distance therebetween.
These steps will be iterated until the learning is finished.
And 4.4, inputting the learned prototype vector acquired in the step 4.3 into seismic data for classification to obtain an unsupervised seismic waveform cluster map, namely a waveform seismic phase map. K mean clustering is generally used, K cluster points are defined, and an objective function satisfies:
where n represents the input data point and the dimension of the cluster-like point.
Finally, the result of the SST-based unsupervised seismic waveform clustering is shown in FIG. 10, and FIG. 11 is a diagram of the result of the EMD-based unsupervised seismic waveform clustering. Comparing fig. 10 and fig. 11, it can be seen that the seismic waveform clustering result obtained by the invention has less noise, and the detail depiction of fault and the like is more accurate, and simultaneously, the clustering result has high conformity with the original seismic data and good reliability. The method fully utilizes the advantages of the SST, autonomously defines the number and the frequency band range of IMF components, processes the characteristics of each component, simultaneously keeps good quality monitoring, and has better reliability and noise resistance of the seismic waveform clustering result.
The invention fully utilizes the advantages of the synchronous extrusion wavelet transform method, and can automatically define the number and the frequency band range of the components aiming at different seismic data and the frequency spectrum characteristics of different intervals in the process of decomposing the seismic channel signals into the inherent mode function components. Meanwhile, the processing of the inherent modal component is optimized, and the high-frequency component with much noise is not directly abandoned but is subjected to denoising processing.
The decomposition algorithm (synchronous extrusion wavelet transform method) and the improvement of the component processing idea are beneficial to better retaining or highlighting the information of the main body energy and removing noise and redundant information. Meanwhile, effective quality monitoring can be provided, the screened/discarded components and the reconstructed seismic data volume in the reconstruction process can be mapped to perform quality monitoring, whether the removed components are noise or redundant information or not is checked, and whether main frequency band information is reserved or not is checked, so that the reliability of waveform clustering is greatly improved. Finally, the reconstructed seismic data volume is converted into a two-dimensional potential space grid consisting of prototype vectors for learning; and clustering and grouping the learned prototype vectors by using K-means clustering to obtain a clustering distribution map.
The invention also provides a seismic waveform clustering device, which comprises a processor, wherein the processor is used for executing instructions for realizing the following method:
1) acquiring a frequency band range of seismic channel signals in target layer seismic data, and dividing the frequency band range to at least comprise a first set frequency band range and a second set frequency band range; the first set frequency band range is a main frequency band range of the seismic trace signal, and the main frequency band range is a frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the second set frequency band range is the frequency band range except the first set frequency band range, and the target layer seismic data keeps the layered spread characteristic on the section;
2) decomposing the target layer seismic data to generate at least two natural modal function components, wherein the at least two natural modal function components comprise a first natural modal component and a second natural modal component, the frequency band range of the first natural modal component is the first set frequency band range, and the frequency band range of the second natural modal component is the second set frequency band range;
3) and denoising the second inherent modal component, superposing and reconstructing the second inherent modal component and the first inherent modal component after the second inherent modal component is processed to obtain a reconstructed seismic data volume, and learning and clustering the reconstructed seismic data volume to obtain a waveform seismic phase diagram.
The seismic waveform clustering device is a computer solution based on the method flow, namely a software framework, and can be applied to a processor, and the device is a processing process corresponding to the method flow. The above-described method will not be described in detail since it is sufficiently clear and complete.
Claims (10)
1. A seismic waveform clustering method is characterized by comprising the following steps:
1) acquiring a frequency band range of seismic channel signals in target layer seismic data, and dividing the frequency band range to at least comprise a first set frequency band range and a second set frequency band range; the first set frequency band range is a main frequency band range of the seismic trace signal, and the main frequency band range is a frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the second set frequency band range is the frequency band range except the first set frequency band range, and the target layer seismic data keeps the layered spread characteristic on the section;
2) decomposing the target layer seismic data to generate at least two inherent modal function components, wherein the at least two inherent modal function components comprise a first inherent modal function component and a second inherent modal function component, the frequency band range of the first inherent modal function component is the first set frequency band range, and the frequency band range of the second inherent modal function component is the second set frequency band range;
3) and denoising the second inherent modal function component, superposing and reconstructing the second inherent modal function component and the first inherent modal function component after the second inherent modal function component is processed to obtain a reconstructed seismic data volume, and learning and clustering the reconstructed seismic data volume to obtain a waveform seismic phase diagram.
2. The seismic waveform clustering method according to claim 1, wherein the first set frequency band range is a main frequency band range of a target layer obtained after spectral analysis is performed on seismic data of the target layer; and the upper limit value of the second set frequency band range is the cut-off frequency of the seismic channel signal.
3. The seismic waveform clustering method according to claim 1 or 2, wherein the first set frequency band range is a main frequency band range of the seismic trace signals determined after performing synchronous squeeze wavelet transform on the seismic data of the target layer and acquiring a time-frequency graph of the seismic trace signals.
4. The seismic waveform clustering method of claim 1, wherein the target interval seismic data is decomposed using a simultaneous squeeze wavelet transform.
5. The seismic waveform clustering method of claim 1, wherein the learning is a self-organizing neural network learning.
6. A seismic waveform clustering apparatus comprising a processor for executing instructions that implement a method of:
1) acquiring a frequency band range of seismic channel signals in target layer seismic data, and dividing the frequency band range to at least comprise a first set frequency band range and a second set frequency band range; the first set frequency band range is a main frequency band range of the seismic trace signal, and the main frequency band range is a frequency band range when the amplitude energy intensity is reduced to half of the maximum value; the second set frequency band range is the frequency band range except the first set frequency band range, and the target layer seismic data keeps the layered spread characteristic on the section;
2) decomposing the target layer seismic data to generate at least two inherent modal function components, wherein the at least two inherent modal function components comprise a first inherent modal function component and a second inherent modal function component, the frequency band range of the first inherent modal function component is the first set frequency band range, and the frequency band range of the second inherent modal function component is the second set frequency band range;
3) and denoising the second inherent modal function component, superposing and reconstructing the second inherent modal function component and the first inherent modal function component after the second inherent modal function component is processed to obtain a reconstructed seismic data volume, and learning and clustering the reconstructed seismic data volume to obtain a waveform seismic phase diagram.
7. The seismic waveform clustering device according to claim 6, wherein the first set frequency band range is a subject frequency band range of a target layer obtained after spectral analysis is performed on seismic data of the target layer; and the upper limit value of the second set frequency band range is the cut-off frequency of the seismic channel signal.
8. The seismic waveform clustering device according to claim 6 or 7, wherein the first set frequency band range is a main frequency band range of the seismic trace signals determined after performing synchronous squeeze wavelet transform on the seismic data of the target layer and acquiring a time-frequency diagram of the seismic trace signals.
9. The seismic waveform clustering device of claim 6, wherein the target layer seismic data is decomposed using a simultaneous squeeze wavelet transform.
10. The seismic waveform clustering device of claim 6, wherein the learning is a self-organizing neural network learning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810313025.5A CN110361782B (en) | 2018-04-09 | 2018-04-09 | Seismic waveform clustering method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810313025.5A CN110361782B (en) | 2018-04-09 | 2018-04-09 | Seismic waveform clustering method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110361782A CN110361782A (en) | 2019-10-22 |
| CN110361782B true CN110361782B (en) | 2020-10-30 |
Family
ID=68212693
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810313025.5A Active CN110361782B (en) | 2018-04-09 | 2018-04-09 | Seismic waveform clustering method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110361782B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112711070B (en) * | 2019-10-24 | 2024-02-20 | 中国石油化工股份有限公司 | Oil gas detection method and device based on seismic signal decomposition |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1314997B1 (en) * | 2001-11-24 | 2005-01-12 | EADS Deutschland Gmbh | Method for HPRF-radar measurement |
| US6691039B1 (en) * | 2002-08-30 | 2004-02-10 | John M. Robinson | Removal of noise from seismic data using improved radon transformations |
| CN103698808B (en) * | 2012-09-28 | 2016-08-03 | 中国石油天然气集团公司 | Earthquake is separated and wave reconstruction method by one with log data waveform extremal features point |
| CN104280771A (en) * | 2014-10-27 | 2015-01-14 | 中国石油集团川庆钻探工程有限公司地球物理勘探公司 | Three-dimensional seismic data waveform semi-supervised clustering method based on EM algorithm |
| US10578758B2 (en) * | 2015-03-19 | 2020-03-03 | Exxonmobil Upstream Research Company | Sequence pattern characterization |
| CN105243388B (en) * | 2015-09-09 | 2018-12-04 | 电子科技大学 | Waveform classification based on dynamic time warping and partitioning algorithm |
| CN107831549A (en) * | 2017-11-20 | 2018-03-23 | 中国地质大学（武汉） | A kind of NMP cepstrum SST Time-frequency methods of ENPEMF signals |
-
2018
- 2018-04-09 CN CN201810313025.5A patent/CN110361782B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110361782A (en) | 2019-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Seismic signal denoising and decomposition using deep neural networks | |
| CN110361778B (en) | Seismic data reconstruction method based on generation countermeasure network | |
| CN107817527B (en) | Seismic exploration in desert stochastic noise suppression method based on the sparse compressed sensing of block | |
| CN103279957B (en) | A kind of remote sensing images area-of-interest exacting method based on multi-scale feature fusion | |
| CN113723171B (en) | Electroencephalogram signal denoising method based on residual error generation countermeasure network | |
| CN104809357B (en) | The analysis method and device of a kind of multiple dimensioned touching act of three-dimensional rough surface | |
| CN107144879A (en) | A kind of seismic wave noise-reduction method combined based on adaptive-filtering with wavelet transformation | |
| Kumar et al. | Biosignal denoising via wavelet thresholds | |
| CN114492538A (en) | Local discharge signal denoising method for urban medium-voltage distribution cable | |
| CN116153329A (en) | CWT-LBP-based sound signal time-frequency texture feature extraction method | |
| CN109389553A (en) | Meteorological causes isopleth interpolation method based on T batten | |
| CN110361782B (en) | Seismic waveform clustering method and device | |
| CN104143031A (en) | Vegetation index time series data reconstruction method based on wavelet multi-scale decomposition | |
| CN113568058B (en) | Magnetotelluric signal-noise separation method and system based on multi-resolution singular value decomposition | |
| Wang et al. | Desert seismic noise suppression based on multimodal residual convolutional neural network | |
| CN108226996B (en) | Self-adaptive anisotropic frequency division partition filtering method based on energy frequency band distribution | |
| CN102509268B (en) | Immune-clonal-selection-based nonsubsampled contourlet domain image denoising method | |
| Dong et al. | A potential solution to insufficient target-domain noise data: Transfer learning and noise modeling | |
| CN112817056B (en) | Magnetotelluric signal denoising method and system | |
| CN114139586A (en) | Electric energy quality composite disturbance identification method based on multi-scale analysis and two-dimensional feature extraction | |
| CN107590788A (en) | A kind of remote sensing image processing method | |
| CN115236744A (en) | Method for strong shielding stripping and weak signal energy compensation of igneous rock | |
| CN104865603A (en) | SVD filtering method and device for inclined layer | |
| Oroumchian | ECEI Spectrogram Denoising and Feature Extraction Using a Multi-Wavelet Convolutional Neural Network Model | |
| CN111142134A (en) | Coordinate time series processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |