SEG D segmentation and classification hybrid compression method for massive seismic source data
Through the SEG D segmented and classified hybrid compression method, the problem that seismic data compression in the existing technology is difficult to meet the on-site quality monitoring needs, and efficient and accurate data compression is achieved, which meets the data compression requirements of massive seismic exploration and acquisition projects.
Patent Information
- Application Number
- CN202011114683.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-10-16
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-10-16
AI Technical Summary
Existing seismic data compression technology is difficult to meet the needs of on-site quality monitoring of massive seismic source data, especially the problem of difficulty in achieving a balance between noise-signal ratio, distortion rate, compression ratio and compression time.
Using massive seismic source data SEG D segmented and classified hybrid compression method, the SEG D data file is decompiled, and the relevant parameters of quality control are obtained for Hoffman encoding, null irrelevant data, time difference correction and coefficient of variation are calculated, and Daubechies 9 wavelet basis is used. Functions and multi-level tree set splitting algorithms perform wavelet transformation and quantization, and combine Hoffman encoding to form a compressed character stream.
It realizes that on the basis of ensuring data quality and compression efficiency, the compression ratio and compression time of seismic data are improved, and the requirements for efficient quality monitoring and analysis of on-site seismic data are met, and a practical engineering data compression solution is provided.
Smart Images

Figure CN114371501B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of oil and gas seismic exploration and development, and in particular to a segmentation classification and mixed compression method for massive seismic source data SEG D. Background Art
[0002] Seismic data compression methods have gradually emerged and developed with the increase in the amount of oil and gas seismic exploration data. Around the beginning of this century, there were sporadic research reports and inventions on seismic acquisition instrument data encoding and acquisition system transmission data compression technology. Later, seismic data compression research with wavelet transform as the main content has been long-lasting. In recent years, with the continuous deepening of single-point high-density (two widths and one height) seismic exploration theory and practice and compressed sensing technology research, seismic data compression technology has made great progress.
[0003] At present, there are mainly the following ways to compress seismic data: starting from the seismic acquisition signal, the seismic signal is appropriately transformed, such as wavelet transform, and the acquisition system group code and formatted data are based on the transformed value; starting from the seismic acquisition system, the channel data stream of the seismic acquisition equipment is compressed to improve the transmission efficiency; extracting the information representing its attributes from the general dumped seismic data SEG Y file, realizing random access to data indoors and saving storage space; data compression methods dedicated to data processing and maintaining spatial attribute information are usually used for data processing. Through searching, there are few reports on the compression of seismic source data at the massive seismic acquisition site (the seismic source data here refers to the seismic data directly generated by the field seismic acquisition system, generally referring to the SEG D standard, and some equipment adopts the SEG Y standard); data compression methods used for indoor data processing, especially data compression and reconstruction methods driven by compressed sensing technology.
[0004] From the compression technology used, seismic data compression mainly adopts the following methods: Wavelet transform method has been used for seismic data compression for a long time. From the early days to the present, a large number of compression technologies have emerged. The ideas of using wavelet transform to achieve seismic data compression are similar. Using different wavelet bases, group coding methods and encoding algorithms, the data compression ratio and fidelity vary greatly; compressed sensing is a new technology that has emerged with the advancement of seismic exploration technology. Although it has not been long, the research results are rich. The amount of literature on compressed sensing reconstruction of massive seismic data is increasing. The research hotspot is based on the methods and technologies in different fields to reconstruct compressed sensing of seismic data. These field methods include wavelet transform, Fourier discrete cosine transform, etc. At present, it is a major trend to reconstruct data by combining compressed sensing with wavelet transform; treating seismic single shot records as images and compressing seismic single shot records using image compression technology is also an idea. JPEG 2000 and its improved algorithm, multi-level tree set splitting algorithm (SPIHT) are quite representative algorithms; in addition, there are also studies on seismic data compression through typing technology and chaos theory, as well as seismic data lossless compression methods, but the application effect is not obvious. In summary, there are many achievements in the massive seismic data compression methods with wavelet transform and compressed sensing as typical features. They have made many effective explorations in ensuring the signal-to-noise ratio of data and improving the compression ratio, and have been tested in engineering practice.
[0005] There are many existing methods and technologies for seismic data compression, but they lack specificity in data compression for field seismic data monitoring, and their achievements are rare. In addition, some commonly used methods, such as the error of reconstructing seismic data based on predictive filtering methods, have large errors, and the accuracy of seismic data reconstruction based on certain mathematical transformations is low; in the field of compressed sensing, the goal of data reconstruction is to obtain better imaging seismic data under the premise of reducing the amount of data. Therefore, its starting point and end point are very different from the compression of seismic source data for field quality monitoring. Due to the differences in goals and means, compressed sensing seismic data compression is not suitable for field quality monitoring of seismic source data; wavelet transform uses different wavelet bases for data compression to achieve different application effects, but the quality control requirements based on seismic source data are still quite different from these methods. Field quality control must take into account both compression ratio and compression and decompression time; analogy to seismic image data compression, especially the use of JPEG 2000 technology for data compression, has achieved certain technical results, but it is far from enough to treat seismic data as ordinary image data. The simple image compression method drowns out the information characteristics reflected by seismic data.
[0006] In the Chinese patent application with application number: CN201310570935.9, a method for parallel compression of massive seismic data is involved, which belongs to the technical field of petroleum seismic exploration data processing. The method is characterized in that three-dimensional seismic data is converted by using a CPU to obtain seismic data volume data, and the seismic data volume data is decomposed and re-decomposed to obtain seismic data sub-blocks and seismic data blocks, and then the seismic data blocks are transmitted to the GPU, and a visual perception model is used for parallel compression, and the compression results of all seismic data blocks are merged to obtain the compression result of the entire seismic data. The visual perception model adopted by the patent simulates the intuitive understanding of seismic data by the naked eye. Therefore, the patent first converts the seismic data into voxel data, takes the voxel as the compression object, divides the seismic data into strong edge blocks, detail blocks and smooth sliders according to the entropy and variance of the data blocks, assigns them different visual weights, and uses wavelet transform for compression to ensure that the most important visual coefficients are given priority. This is a typical application of image compression in seismic data. Image compression masks the frequency, signal-to-noise ratio and other physical characteristics of the seismic data itself. Therefore, this method is not universal. In addition, seismic acquisition is currently widely distributed in various complex surfaces and geological conditions. The entropy and variance calculated by the fixed-size data sub-blocks and block division blur the boundary between the large local visual variation within the block and the large overall visual difference of the block. The visual weight established in this way will greatly affect the final compression effect.
[0007] In the Chinese patent application with application number: CN201310495175.X, a method and device for compressing massive seismic data that maintains spatial attribute information is involved, which is applied to the technical field of seismic data compression processing. The method includes: determining seismic data that meets preset compression conditions as seismic data to be compressed; determining the compression ratio according to observation system parameters, common center point CMP gather surface element attributes, underground interface morphology and seismic wave field characteristics; determining the direction in which the seismic wave field is most densely distributed in space; gridding the seismic data according to spatial position based on the data attribute distribution; selecting a compression function based on the seismic data distribution and / or compression requirements; and using the compression function and the compression ratio to compress the seismic data to be compressed in each grid along the direction in which the seismic wave field is most densely distributed in space. Usually, seismic data compression is the compression of all sampling data of seismic channels. Different from the compression method in the general sense, the essence of this patent is the sparseness of seismic channels. The compression formula is only a linear processing of the overall numerical value of the seismic channel sampling data. The goal of reducing the amount of seismic data is achieved by deleting a large number of seismic channels that have little influence on spatial attributes. Therefore, the compressed data cannot be reconstructed at all, and because it completely changes the relationship between the gun and the inspection, it has a serious impact on the later seismic data processing and is only effective for specific target processing.
[0008] In the Chinese patent application with application number: CN201810010947.9, a method for compressing and reconstructing seismic data is involved, which includes the following steps: first, wavelet transform the seismic data to increase the compressibility of the seismic data; then, construct a hardware-implementable measurement matrix based on the chaotic sequence, and use the chaotic measurement matrix to compress and observe the seismic data after wavelet transform; finally, improve the Bayesian wavelet tree structure compressed sensing reconstruction algorithm, and use the improved BTSWCS-vb algorithm to restore the complete seismic data. This patent is a data compression that relies on hardware for seismic sampling data streams. The measurement matrix used for compression is implemented by hardware and adopts a compression method while collecting. Therefore, in order to ensure the reconstruction accuracy, its compression ratio is relatively low, with a maximum of less than 1:5, which is difficult to meet the on-site quality control requirements of seismic source data.
[0009] To this end, we invented a new SEG D segmentation classification hybrid compression method for massive seismic source data to solve the above technical problems. Summary of the invention
[0010] The purpose of the present invention is to provide a massive seismic source data SEG D segment classification hybrid compression method which provides an engineering practical data compression solution for on-site quality monitoring and analysis of massive seismic acquisition.
[0011] The object of the present invention can be achieved by the following technical measures: a massive seismic source data SEG D segmented classification hybrid compression method, the massive seismic source data SEG D segmented classification hybrid compression method comprises: step 1: decode the SEG D data file, obtain all quality control related parameters from various header blocks and track headers, and perform Huffman coding on them; step 2: for auxiliary track data, record the jump time and amplitude information within a specific time window; step 3: according to the exploration target requirements of the work area, eliminate the sampling data irrelevant to quality control in the arrangement according to the time and space range; step 4: by obtaining the time difference correction factor of the seismic track in the arrangement, perform time difference correction on the seismic track, after the time difference correction of all tracks in the arrangement, calculate the coefficient of variation of all samples of the arrangement and the previous arrangement; step 5: if the coefficient of variation is greater than the set threshold, perform integer normalization processing on the arrangement samples; step 6: adopt Daubechies 9 wavelet basis functions perform integer wavelet transform on the normalized data; Step 7: Calculate the dynamic threshold from the wavelet coefficients, and use the multi-level tree set splitting algorithm to quantize the wavelet coefficients, and the accuracy control is limited by the controlled factor; Step 8: Perform arithmetic coding on the quantized data, and count the coding summary information; Step 9: If the coefficient of variation obtained in step 4 is not greater than the set threshold, perform differential coding on the difference between the arrangement and the previous arrangement sampling and normalize it; Step 10: Perform Huffman coding on the normalized data again, and count the coding summary information; Step 11: Group the above segmented and classified data according to certain rules to form a character stream.
[0012] The purpose of the present invention can also be achieved by the following technical measures:
[0013] In step 1, first decode the total header block, calculate the channel group position and data channel position, identify the auxiliary channel and data channel, and then decode various header blocks and track headers, remove the marker track header word and seismic track index information and other parameters not related to field data analysis and quality control, and then perform Huffman coding on the quality control parameters.
[0014] In step 2, the auxiliary channel data is scanned to record the typical data of the reference signal and the verification signal within a specific time window, including the start time and amplitude, which are directly used for the group code.
[0015] In step 3, the data in the time and space range are eliminated based on the monitoring target, and the eliminated content is the far-offset arrangement or seismic trace and deep seismic data.
[0016] In step 4, the time difference correction factor is used for the time difference correction of the arranged seismic trace data, and its calculation formula is:
[0017]
[0018] Where: Δt s,i is the time difference correction factor of the i-th channel in the current arrangement s, in units of s, p x 、p y is the horizontal and vertical coordinates of the shot point, in m, p s,i (x), p s,i (y) is the horizontal and vertical coordinates of the ith track in the current arrangement s, in m, and v is the speed of the low-speed reduction belt, in m / s;
[0019] The coefficient of variation reflects the similarity between two adjacent seismic trace samples, and its calculation formula is:
[0020]
[0021] Where: cv is the coefficient of variation, M is the number of seismic channels in the arrangement, and σ is the standard deviation, which is defined as follows:
[0022]
[0023] Where: T is the maximum sampling number of the seismic trace after the arrangement data is removed, x s,i (t j ), x s-1,i (t j ) are the sampling values of the jth discrete point in the i-th channel in the current arrangement s and the previous arrangement s-1 respectively.
[0024] In step 5, the setting of the threshold is closely related to the earthquake source type, surface and underground geological conditions, and is determined by the analysis of single shot test data before construction; the normalization calculation formula is:
[0025]
[0026] Where: z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit, which is a positive integer value not greater than 32, generally 8, 16 or 24; x s,i (t j ) is the jth discrete point sampling value in the i-th channel of the current arrangement s; min(x s,i )、max(x s,i ) are the minimum sampling value and the maximum sampling value of the i-th channel in the current arrangement s, respectively.
[0027] In step 6, the integer wavelet transform uses Daubechies 9 wavelet basis functions, and the number of decomposition layers is 3 to 5.
[0028] In step 7, the dynamic threshold of the multi-level tree set splitting algorithm is calculated by wavelet coefficients. The calculation formula of the dynamic threshold is:
[0029]
[0030] Where: T p is the dynamic threshold, p is the number of cycles, p = 0, 1, 2, ..., k-1; is the floor function; lg(·) is the logarithmic function with base 10; max(·) is the maximum value function; W f (a, b) represents the wavelet coefficients, a is the time scale, and b is the frequency scale.
[0031] In step 7, the controlled factor is obtained by the mean square error of the compressed bit, compressed time and data before and after compression, and its calculation formula is as follows:
[0032]
[0033] Where: CF is the controlled factor, unit: db; ξ is the compression time factor, which is used to adjust the compression time and distortion, and takes 0<ξ<100, k is still the compression position, and MSE is the mean square error, which is defined as follows:
[0034]
[0035] Where: M is the number of seismic traces after the current arrangement data is removed, T is the maximum sampling number of seismic traces after the current arrangement data is removed, z s,i(t j ), They are the data before and after compression of the jth sampling point of the i-th channel in the current arrangement s.
[0036] In step 8, the statistical summary information includes the spatial and temporal ranges M and T of data removal, the compression bit k, the compression time factor ξ, the normalization factor, the seismic trace sampling extreme value min(x s,i ) and max(x s,i ).
[0037] In step 9, if the coefficient of variation obtained in step 4 is not greater than the set threshold, the difference between the arrangement and the previous arrangement sampling is differentially encoded and integer normalized. The normalization calculation formula is:
[0038]
[0039] Where: z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit; is the difference between the jth sample value of the ith channel in the current arrangement s and the corresponding sample of the previous arrangement; is the minimum and maximum value of the difference of the ith track in the current arrangement s.
[0040] In step 10, Huffman coding is performed on the normalized data, and the coding summary information is statistically calculated, including the threshold, compression bit, and minimum value of the sampling difference. and maximum value
[0041] In step 11, the group code is organized in the order of the data in step 1, step 2, step 8, and step 10, wherein corresponding serialization rules are formulated for each type of information.
[0042] The mass seismic source data SEG D segmentation classification hybrid compression method of the present invention is oriented to seismic source data and has strong pertinence; the compression target is used for efficient quality monitoring and data analysis of on-site seismic data, and a balance needs to be achieved between noise-to-signal ratio, distortion rate, compression ratio and compression time; when compressing seismic samples, the similarity of adjacent arrangement samples is fully considered, that is, the more similar the adjacent arrangement samples are, the higher the compression ratio is, and the shorter the compression time is. Therefore, the purpose of the present invention is to provide an engineering practical data compression solution for on-site quality monitoring and analysis of large-scale single-shot data acquisition for the field of massive seismic exploration and acquisition engineering, which not only meets the on-site timeliness requirements, but also does not cause serious data distortion due to compression. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1It is a flow chart of a specific embodiment of the method for segmentation classification and hybrid compression of massive seismic source data SEG D of the present invention;
[0044] Figure 2 This is a comparison diagram of single shot records before and after data compression using a compression bit k of 8 and a controlled factor CF of 90 in a specific implementation of the present invention. DETAILED DESCRIPTION
[0045] In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.
[0046] like Figure 1 As shown, Figure 1 This is a flow chart of the segmented classification and mixed compression method of massive seismic source data SEG D of the present invention.
[0047] In step 1, the SEG D data file is decoded, all quality control related parameters are obtained from various header blocks and trace headers, and Huffman encoding is performed on them;
[0048] Eliminate the marker trace header and seismic trace index information and other parameters irrelevant to field data analysis and quality control, especially the redundant data in the standard, and then perform Huffman coding on the quality control parameters.
[0049] In step 2, for the auxiliary channel data, the take-off time and amplitude information within a specific time window are recorded;
[0050] The typical data of the reference signal and verification signal within a specific time window in the auxiliary channel, including the start time and amplitude, are recorded and directly used for the group code.
[0051] In step 3, according to the exploration target requirements of the work area, the sampling data irrelevant to quality control in the arrangement are eliminated according to the time and space range;
[0052] According to the exploration target requirements of the work area, sampling data irrelevant to quality control are eliminated in the arrangement according to the time and space range. Generally, the content eliminated is the far-deviated arrangement or seismic trace and deep seismic data.
[0053] In step 4, the time difference correction factor of the seismic trace in the arrangement is obtained to correct the time difference of the trace. After the time difference correction of all the traces in the arrangement, the coefficient of variation of all samples of the arrangement and the previous arrangement is calculated;
[0054] Obtain the time difference correction factors of the seismic channels in the current arrangement in turn, and make time difference correction for each seismic channel. The calculation formula is:
[0055]
[0056] Where: Δt s,iis the time difference correction factor of the i-th channel in the current arrangement s, in units of s, p x 、p y is the horizontal and vertical coordinates of the shot point, in m, p s,i (x), p s,i (y) is the horizontal and vertical coordinates of the ith track in the current arrangement s, in m, and v is the speed of the low deceleration zone, in m / s.
[0057] Then, the coefficient of variation of all samples of the last arranged seismic trace is obtained, and the calculation formula is:
[0058]
[0059] Where: cv is the coefficient of variation, M is the number of seismic channels in the arrangement, and σ is the standard deviation, which is defined as follows:
[0060]
[0061] Where: T is the maximum sampling number of the seismic trace after the s data of the arrangement is removed, x s,i (t j ), x s-1,i (t j ) is the sampling value of the jth discrete point in the i-th channel in the current arrangement s and the previous arrangement s-1.
[0062] In step 5, if the coefficient of variation is greater than the set threshold, integer normalization is performed on the arrangement sampling, and the normalization calculation formula is:
[0063]
[0064] Where: z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; the compression bit k is a positive integer value not greater than 32, generally 8, 16 or 24; x s,i (t j ) is the jth discrete point sampling value in the i-th channel of the current arrangement s; min(x s,i )、max(x s,i ) are the minimum and maximum sampling values of the i-th channel in the current arrangement s.
[0065] In step 6, integer wavelet transform is performed on the normalized data using Daubechies 9 wavelet basis functions;
[0066] Integer wavelet transform uses Daubechies 9 wavelet basis functions, and the number of wavelet decomposition layers is preferably 3 to 5.
[0067] In step 7, the dynamic threshold is calculated from the wavelet coefficients, and the wavelet coefficients are quantized using the multi-level tree set splitting (SPIHT) algorithm, and the precision control is limited by the controlled factor;
[0068] The dynamic threshold is calculated based on the wavelet coefficients, and the multi-level tree set splitting (SPIHT) algorithm is used to quantize the wavelet coefficients. The calculation formula of the dynamic threshold is:
[0069]
[0070] Where: T p is the dynamic threshold, p is the number of cycles, p = 0, 1, 2, ..., k-1; is the floor function; lg(·) is the logarithmic function with base 10; max(·) is the maximum value function; W f (a, b) represents the wavelet coefficients, a is the time scale, and b is the frequency scale.
[0071] The precision control is determined by the controlled factor, which is limited by the compression bit, compression time and the mean square error of the data before and after compression. The calculation formula is as follows:
[0072]
[0073] Where: CF is the controlled factor (unit: db), ξ is the compression time factor, which mainly controls the compression time and distortion, generally 0<ξ<100, k is still the compression bit, and MSE is the mean square error, which is defined as follows:
[0074]
[0075] Where: M is the number of seismic traces after the data of the arrangement s is removed, T is the maximum sampling number of seismic traces after the data of the arrangement is removed, z s,i (t j ), They are the data before and after compression of the jth sampling point of the i-th channel in the current arrangement s.
[0076] In step 8, arithmetic coding is performed on the quantized data, and coding summary information is statistically calculated;
[0077] The quantized data is arithmetically coded again, and the coding summary information is statistically analyzed, including the spatial and temporal range M and T of data removal, compression bit k, compression time factor ξ, normalization factor, sampling extreme value min(x s,i ) and max(x s,i ).
[0078] In step 9, if the coefficient of variation obtained in step 4 is not greater than the set threshold, the difference between the arrangement and the previous arrangement sampling is differentially encoded and integer normalized. The normalization calculation formula is:
[0079]
[0080] Where: z s,i (t j ) is the normalized data of the jth sampling difference of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit; is the difference between the jth sample value of the ith channel in the current arrangement s and the corresponding sample of the previous arrangement s-1; is the minimum and maximum value of the difference of the ith track in the current arrangement s.
[0081] In step 10, Huffman coding is performed on the normalized data, and the coding summary information is counted, including the compression bit and the minimum value of the sampling difference in the current arrangement s. and
[0082] In step 11, after all the arrangements of steps 3 to 10 are completed, the data obtained in steps 1, 2, 8 and 10 are grouped according to certain rules to form a character stream.
[0083] The group code is organized according to the data sequence in step 1, step 2, step 8, and step 10, wherein corresponding serialization rules are formulated for each type of information.
[0084] In a specific embodiment of the present invention, the following steps are included:
[0085] (1) Input the original seismic record SEG D file, distinguish between versions 1.0, 2.1 and 3.0 according to the SEG D related standards, decode the file, calculate the channel group position and data track position, identify the auxiliary track and data track, and then decode various track headers. Through these processes, identify the file header block, track header and sampling data, including data index and redundant information, and especially distinguish quality control parameters from non-quality control parameters;
[0086] (2) Keyword screening: select keyword data related to quality control for Huffman coding, including shot point location and sampling rate;
[0087] (3) For auxiliary channel data, identify and extract reference signal and verification signal data within a specific time window. For vibroseis acquisition, it also includes scan type data, which are directly used for group coding;
[0088] (4) Process the data in order according to the arrangement, set the monitoring range according to the exploration target requirements, and eliminate the far-deviated arrangement or seismic traces and deep seismic data;
[0089] (5) According to the speed of the low-speed reduction zone, the time difference correction factor of each seismic channel in the current arrangement is obtained, and the time difference correction is performed on each seismic channel. The calculation formula is: In the formula, Δt s,i is the time difference correction factor of the i-th channel in the current arrangement s, in units of s, p x 、p y is the horizontal and vertical coordinates of the shot point, in m, p s,i (x), p s,i (y) is the horizontal and vertical coordinates of the ith track in the current arrangement s, in m, and v is the speed of the low-speed reduction belt, in m / s;
[0090] (6) Obtain the coefficient of variation of all seismic trace samples of the current arrangement and the previous arrangement, and the calculation formula is: Where cv is the coefficient of variation, M is the number of seismic traces in the current arrangement s, and σ is the standard deviation, which is defined as Where T is the maximum sampling number of the seismic trace after the current arrangement s data is removed, x s,i (t j ), x s-1,i (t j ) is the jth discrete point sampling value in the i-th channel in the current arrangement s and the previous arrangement s-1;
[0091] (7) If the coefficient of variation is greater than the set threshold, the array sampling is normalized by integer. The threshold here is related to the earthquake excitation factor, surface and geological conditions, environmental noise, array length and recording time. This value can be easily obtained through test analysis before formal production. The normalization calculation formula is: In the formula, z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit, which is a positive integer value not greater than 32, generally 8, 16 or 24; x s,i (t j ) is the jth discrete point sampling value in the i-th channel of the current arrangement s; min(x s,i )、max(x s,i ) are the minimum and maximum sampling values of the ith channel in the current arrangement s;
[0092] (8) The Daubechies 9 wavelet basis function is used to perform integer wavelet transform on the normalized data, and the number of wavelet decomposition layers is 3;
[0093] (9) For coefficients of different frequencies, the multi-level tree set splitting (SPIHT) algorithm is used to quantize the wavelet coefficients, and its dynamic threshold is defined as Where: T p is the dynamic threshold, p is the number of cycles, p = 0, 1, 2, ..., k-1; is the floor function; lg(·) is the logarithmic function with base 10; max(·) is the maximum value function; W f (a, b) represents the wavelet coefficients, a is the time scale, b is the frequency scale, where p = 0 is the initial threshold;
[0094] (10) The controlled factor controls the data compression accuracy, which is defined by the compression ratio, compression quality and compression time. Its calculation formula is: Where CF is the controlled factor, unit is db, ξ is the compression time factor, which mainly controls the compression time and distortion, generally takes 0<ξ<100, k is still the compression position, and MSE is the mean square error, defined as Where M is the number of seismic traces after the current arrangement s data is removed, T is the maximum sampling number of seismic traces after the current arrangement s data is removed, and z s,i (t j ), They are the data before and after compression of the jth sampling point of the i-th channel in the current arrangement s;
[0095] (11) Encode the obtained quantized sampled data using arithmetic coding, and count the relevant summary information of the arrangement in the above process, including the arrangement number, the time and space range M and T, the compression bit k, the compression time factor ξ, the normalization factor, the i-th sampling extreme value min(x s,i ) and max(x s,i );
[0096] (12) If the coefficient of variation in (6) is not greater than the set threshold, the difference between the sampling of the arrangement and the previous arrangement is differentially encoded and integer normalized. The normalization formula is: In the formula, z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit; is the difference between the jth sample value of the ith channel in the current arrangement s and the corresponding sample of the previous arrangement s-1; is the minimum and maximum value of the difference of the ith track in the current arrangement s;
[0097] (13) Perform Huffman coding on the normalized data and calculate the coding summary information, including the compression bit and the minimum value of the sampling difference in the current arrangement s. and maximum value
[0098] (14) The data in (2), (3), (11) and (13) above are grouped according to certain formatting rules. These rules include: the content is organized in the order of file header, auxiliary channel information, summary information of each arrangement code, seismic channel header summary information and the channel code data. The sizes of the above items are: file header is 256B, auxiliary channel is 64B, arrangement is 128B, seismic channel header is 32B and seismic channel sampling data is unlimited. The size of these data can be adjusted appropriately and finally a compressed file is output.
[0099] The character stream file formed by the above can be directly used for seismic source data SEG D compression, providing an efficient and relatively fidelity compression technology for on-site seismic data monitoring and analysis.
[0100] In a specific embodiment of the present invention, Figure 2 For the comparison of seismic data after compression by this method with compression bit of 8 and controlled factor of 90, the upper figure is the original seismic single shot (partial), the middle figure is the seismic record (partial) corresponding to the left figure after compression and decompression, and the lower figure is the difference between the compressed and original seismic records. Table 1 is a parameter comparison of seismic data compression using 8-bit compression and different controlled factors using multi-threaded parallelism.
[0101] Table 1 Parameters obtained by 8-bit compression and compression with different controlled factors
[0102]
[0103]
[0104] Experimental tests in multiple 3D seismic acquisition areas in my country's eastern old oil fields and western desert areas show that when the compression bit is 8 and the controlled factor is 60 decibels for data compression, the compression ratio is about 16.32, the signal-to-noise ratio is only about 0.09%, and the average time required to compress 10,000 seismic data is 2.71 seconds, which fully meets the quality control, analysis and data dump requirements of massive on-site seismic source data.
[0105] The present invention discloses a segmented classification hybrid joint compression method for massive seismic source data SEG D for field applications, which provides effective support for the analysis and quality monitoring of massive seismic acquisition field data, and greatly improves production efficiency. At present, the main targets of oil and gas exploration and development in my country are lithological and concealed oil and gas reservoirs, as well as complex areas such as complex fault blocks, small faults and micro-structures, which require high-quality seismic data as support. With the advancement of oil and gas seismic exploration technology, seismic equipment, computers and network technology, the large amount of data in the acquisition area and the instantaneous single-shot large data flow have become the main characteristics of massive seismic data acquisition. The traditional seismic data monitoring mode encounters challenges in such massive acquisition, and it is necessary to accelerate the field data analysis and quality control processing rate through data compression. The seismic source data SEG D is compressed according to the header and the sampling data respectively and then grouped, especially considering the similarity of adjacent arranged seismic data, which can quickly and efficiently realize SEG D data compression and meet the requirements of near real-time quality control on site. First, obtain the quality control related parameters from the SEG D file decoding and perform Huffman coding; record the key parameters for the seismic auxiliary channel data; cut the sampled data in the time and space dimensions to obtain the quality control range data; obtain the time difference correction factor of the current arrangement, perform time difference correction on all seismic channels, and then calculate the coefficient of variation between this arrangement and the previous arrangement; if the coefficient is greater than the set threshold, then for each arrangement data, the wavelet coefficient is calculated by integer wavelet transform, and the wavelet coefficient is quantized with the help of dynamic threshold. Its accuracy is controlled by the controlled factors limited by the compression bit, compression time and mean square error of the data before and after compression, and the quantized data is arithmetic coded; if the coefficient of variation is less than the set threshold, the difference between the sampling of this arrangement and the previous arrangement is integer normalized, differentially coded first, and then Huffman coded, and the relevant information of the process is recorded; by merging the different codes of keywords, auxiliary channels and seismic sampling data and their related information according to certain rules, a compressed character stream is formed for field data analysis and quality control.
[0106] The above-described embodiment is only one of the more preferred specific implementations of the present invention. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.
Claims
1. A hybrid compression method for massive seismic source data SEG D segmentation classification, characterized in that: The SEGD segmentation classification hybrid compression method for massive seismic source data includes: Step 1: Decode the SEG D data file, obtain all quality control related parameters from various header blocks and trace headers, and perform Huffman coding on them; Step 2: For the auxiliary channel data, record the take-off time and amplitude information within a specific time window; Step 3: According to the exploration target requirements of the work area, eliminate the sampling data that are not related to quality control in the arrangement according to the time and space range; Step 4: By obtaining the time difference correction factor of the seismic traces in the array, the time difference correction is performed on the seismic traces. After the time difference correction of all traces in the array, the coefficient of variation of all samples of the array and the previous array is calculated; Step 5: If the coefficient of variation is greater than the set threshold, integer normalization is performed on the arrangement sampling; Step 6: Use Daubechies 9 wavelet basis function to perform integer wavelet transform on the normalized data; Step 7: Calculate the dynamic threshold value from the wavelet coefficients, and use the multi-level tree set splitting algorithm to quantize the wavelet coefficients. The accuracy control is limited by the controlled factor. Step 8: Perform arithmetic coding on the quantized data and calculate the coding summary information; Step 9: If the coefficient of variation obtained in step 4 is not greater than the set threshold, the difference between the arrangement and the previous arrangement is differentially encoded and normalized; Step 10: Perform Huffman coding on the normalized data and count the coding summary information; Step 11: Group the data processed by the above segmentation and classification according to certain rules to form a character stream; In step 1, first decode the total header block, calculate the channel group position and data channel position, identify the auxiliary channel and data channel, then decode various header blocks and channel headers, remove the marker channel header word and seismic channel index information and other parameters irrelevant to field data analysis and quality control, and then perform Huffman coding on the quality control parameters; In step 2, the auxiliary channel data is scanned to record the typical data of the reference signal and the verification signal within a specific time window, including the jump time and amplitude, which are directly used for the group code; In step 3, the data elimination in the time and space range is determined based on the monitoring target, and the eliminated content is the far deviation arrangement or seismic trace and deep seismic data; In step 4, the time difference correction factor is used for the time difference correction of the arranged seismic trace data, and its calculation formula is: Where: Δt s,i is the time difference correction factor of the i-th channel in the current arrangement s, in units of s, p x 、p y is the horizontal and vertical coordinates of the shot point, in m, p s,i (x), p s.i (y) is the horizontal and vertical coordinates of the ith track in the current arrangement s, in m, and v is the speed of the low-speed reduction belt, in m / s; The coefficient of variation reflects the similarity between two adjacent seismic trace samples, and its calculation formula is: Where: cv is the coefficient of variation, M is the number of seismic channels in the arrangement, and σ is the standard deviation, which is defined as follows: Where: T is the maximum sampling number of the seismic trace after the arrangement data is removed, x s,i (t j ), x s-1,i (t j ) are the jth discrete point sampling values in the i-th channel in the current arrangement s and the previous arrangement s-1 respectively; In step 5, the setting of the threshold is closely related to the earthquake source type, surface and underground geological conditions, and is determined by the analysis of single shot test data before construction; the normalization calculation formula is: Where: z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit, which is a positive integer value not greater than 32, generally 8, 16 or 24; x s,i (t j ) is the jth discrete point sampling value in the i-th channel of the current arrangement s; min(x s,i )、max(x s,i ) are respectively the minimum sampling value and the maximum sampling value of the ith channel in the current arrangement s; In step 6, the integer wavelet transform uses Daubechies 9 wavelet basis function, and the number of decomposition layers is 3-5; In step 7, the dynamic threshold of the multi-level tree set splitting algorithm is calculated by wavelet coefficients. The calculation formula of the dynamic threshold is: Where: T p is the dynamic threshold, p is the number of cycles, p = 0, 1, 2, ..., k-1; is the floor function; lg(·) is the logarithmic function with base 10; max(·) is the maximum value function; W f (a,b) represents the wavelet coefficients, a is the time scale, b is the frequency scale; In step 7, the controlled factor is obtained by the mean square error of the compressed bit, compressed time and data before and after compression, and its calculation formula is as follows: Where: CF is the controlled factor, unit: db; ξ is the compression time factor, which is used to adjust the compression time and distortion, and takes 0<ξ<100, k is still the compression position, and MSE is the mean square error, which is defined as follows: Where: M is the number of seismic traces after the current arrangement data is removed, T is the maximum sampling number of seismic traces after the current arrangement data is removed, z s,i (t j ), They are the data before and after compression of the jth sampling point of the i-th channel of the current arrangement s; In step 8, the statistical summary information includes the spatial and temporal ranges M and T of data removal, the compression bit k, the compression time factor ξ, the normalization factor, the seismic trace sampling extreme value min(x s,i ) and max(x s,i ); In step 9, the difference between the arrangement and the previous arrangement is differentially encoded and integer normalized. The normalization calculation formula is: Where: z s,i (t j ) is the normalized data of the jth sample of the ith channel in the current arrangement s; int[·] is the integer function; k is the compression bit; is the difference between the jth sample value of the ith channel in the current arrangement s and the corresponding sample of the previous arrangement; is the minimum and maximum value of the difference of the ith track in the current arrangement s; In step 10, Huffman coding is performed on the normalized data, and the coding summary information is statistically calculated, including the threshold, compression bit, and minimum value of the sampling difference. and maximum value In step 11, the group code is organized in the order of the data in step 1, step 2, step 8, and step 10, wherein corresponding serialization rules are formulated for each type of information.
