Seismic waveform classification calculation method based on multi-frequency data volume

Abstract

The invention relates to a seismic waveform classification calculation method based on a multi-frequency data body, and belongs to the technical field of high-quality reservoir stratum prediction in the energy exploration fields of oil and gas field exploration, shale gas exploration, geothermal exploration and the like. A time-frequency analysis method is adopted to calculate three co-frequency data volumes such as a low frequency data volume, an intermediate frequency data volume and a high frequency data volume, then a waveform clustering analysis method based on neural network training is sequentially carried out on the three co-frequency data volumes to carry out binary analysis on seismic waveforms, and a research region is divided into eight sub-regions; and then performing waveform overall correlation analysis and combination of the 8 sub-regions by adopting an original data body, finally representing numerical values of the sub-regions by adopting different color codes, drawing, and representing a waveform classification result by adopting an image.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas field exploration, especially in the energy exploration fields such as oil and gas field exploration, shale gas exploration, geothermal exploration, etc., and specifically relates to a seismic waveform classification calculation method based on multi-frequency data volume. Background Art

[0002] Seismic facies is the comprehensive response feature of sedimentary facies on its seismic data. The seismic facies technology based on waveform classification classifies the waveform shapes of seismic traces in the target interval, and depicts the sedimentary facies belts in a larger area or the lithofacies zones in a smaller area based on this classification.

[0003] Seismic waveform classification is based on the relationship between parameters such as the frequency, amplitude, and phase of seismic signals and waveform characteristics, and combines technologies such as neural networks to process the waveform information of seismic waves to achieve the discrimination of the spatial similarity degree of seismic waveforms. Conventional seismic waveform classification algorithms mostly adopt waveform clustering analysis methods based on neural network training and waveform classification methods based on vector space models, such as neural networks, logistic regression, support vector machines, etc.

[0004] There are many classification methods for seismic waveforms, which are mainly divided into unsupervised classification methods and supervised classification methods. The method of this invention belongs to the unsupervised classification method. When the prior information is incomplete or does not exist, unsupervised learning classification is one of the most effective methods. Two common unsupervised learning methods are: ① Self-organizing neural network clustering, which uses unsupervised learning to train to generate a low-dimensional (usually two-dimensional), discretized training sample input space, called a map, and realizes the orderly mapping of high-dimensional distribution to a regular low-dimensional grid, which is a dimensionality reduction method; ② Gaussian mixture model is a parametric model of probability distribution, which provides greater flexibility and accuracy in modeling than traditional unsupervised clustering algorithms. Unsupervised classification algorithms are mainly applied to seismic facies analysis of post-stack seismic data because post-stack seismic data is simple and has a small amount of computation, and applying these algorithms can quickly obtain the necessary results.

[0005] In summary, the current research field of seismic waveform classification mainly focuses on the in-depth study of classification algorithms, while ignoring the in-depth excavation of the input data volume. Summary of the Invention

[0006] The present invention relates to a seismic waveform classification calculation method based on multi-frequency data volumes, belonging to the technical field of high-quality reservoir prediction in energy exploration fields such as oil and gas field exploration, shale gas exploration, and geothermal exploration. The method first obtains three co-frequency data volumes, namely low-frequency, medium-frequency, and high-frequency data volumes, based on the 3D seismic data of the target interval in the study area by using time-frequency analysis method. Then, for the three co-frequency data volumes, waveform dichotomy analysis based on the waveform clustering analysis method trained by neural network is carried out in sequence. That is, single waveform clustering analysis divides the waveforms in the target area into two categories, and the study area is divided into eight sub-regions. Then, the original data volume is used for overall waveform correlation analysis and merging of the eight sub-regions. Finally, different color scales are used to represent the values of the sub-regions, and a graph is drawn to represent the final waveform classification result by an image.

[0007] The specific steps of the present invention include:

[0008] (1) Input the 3D seismic data of the target interval in the study area, perform time-frequency analysis on the 3D data volume, and obtain three co-frequency data volumes corresponding to the original data volume according to the main frequency, approximately the low-frequency data volume, medium-frequency data volume, and high-frequency data volume;

[0009] (2) For the low-frequency data volume, first perform denoising processing based on the 3D edge-preserving denoising method, and then use the waveform clustering analysis method trained by neural network to carry out waveform clustering analysis and perform dichotomy analysis on the waveforms in the whole area. That is, by dividing the waveforms into two categories, the study area is divided into two sub-regions;

[0010] (3) For the medium-frequency data volume, use the waveform clustering analysis method trained by neural network to carry out waveform clustering analysis on the two sub-regions in step (2) respectively, and perform dichotomy analysis on the waveforms in the sub-regions to obtain four sub-regions. That is, after this step of processing, the study area is divided into four sub-regions;

[0011] (4) For the high-frequency data volume, use the waveform clustering analysis method trained by neural network to carry out waveform clustering analysis on the four sub-regions in step (3) respectively, and perform dichotomy analysis on the waveforms in the sub-regions to obtain eight sub-regions. That is, after this step of processing, the study area is divided into eight sub-regions;

[0012] (5) Use the original data volume (without frequency division processing) of the eight sub-regions divided in step (4) to perform correlation analysis and merging of the eight sub-regions;

[0013] (6) Use different color scales to represent the values of the sub-regions, draw a graph, and use an image to represent the final waveform classification result.

[0014] A waveform classification calculation method based on multi-frequency data volumes has the following characteristics, mainly manifested as:

[0015] (1) For the first time in the field of waveform classification, the present invention jointly uses three co-frequency data volumes, which approximately represent the low-frequency band data volume, the medium-frequency band data volume, the high-frequency band data volume, and performs waveform classification calculations on the original data volume (without frequency division processing).

[0016] (2) The present invention makes full use of the advantages that the low-frequency data volume represents macroscopic features and the high-frequency data volume represents detailed features. In the process design of waveform classification, it first conducts macroscopic (overall zoning), then microscopic (local zoning), and finally conducts sub-region merging analysis using the original data volume (respecting the waveform features of the original data volume). Description of the Drawings

[0017] Figure 1 is the technical flow chart of the present invention;

[0018] Figure 2 is the waveform classification plan view of the second member of the Qixia Formation in a certain area of Sichuan calculated by using the method of the present invention. Detailed Embodiments

[0019] Embodiment 1

[0020] A seismic waveform classification calculation method based on multi-frequency data volumes, the specific steps include:

[0021] Step 1, input the three-dimensional seismic data of the target interval in the study area, denoted as the array A(x, y, t);

[0022] Step 2, calculate the main frequency of the array A(x, y, t) using the Fourier transform analysis method, denoted as Fmain;

[0023] Step 3, perform time-frequency analysis on the array A(x, y, t) using the time-frequency wavelet transform method to obtain the time-frequency analysis array AW(x, y, t, f);

[0024] Step 4, extract three co-frequency data volumes with different frequencies,

[0025] Step 4-1, denote the frequency of the first co-frequency data volume as f1, denote the frequency of the second co-frequency data volume as f2, and denote the frequency of the third co-frequency data volume as f3,

[0026] f1 = int(Fmain - Fmain / 3), f2 = int(Fmain), f3 = int(Fmain + Fmain / 3)

[0027] Step 4-2: Extract three co-frequency data volumes, AW(x, y, t, f1), AW(x, y, t, f2), and AW(x, y, t, f3), from the time-frequency analysis array AW(x, y, t, f) in Step 3. Since f1, f2, and f3 are fixed values, the arrays AW(x, y, t, f1), AW(x, y, t, f2), and AW(x, y, t, f3) are three-dimensional arrays. Denote AW_low(x, y, t) = AW(x, y, t, f1), AW_med(x, y, t) = AW(x, y, t, f2), and AW_hig(x, y, t) = AW(x, y, t, f3).

[0028] Step 5: Conduct a binary analysis of the waveforms in the entire region using the array AW_low(x, y, t).

[0029] Step 5-1: Denoise the array AW_low(x, y, t) using a three-dimensional edge-preserving denoising method. Denote the denoised values as AW_low_n(x, y, t).

[0030] Step 5-2: For the array AW_low_n(x, y, t), in the time t direction, conduct waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and denote the classification result as Wave1(x, y). The array Wave1(x, y) uses two values, 1 or 2, representing two types of waveforms.

[0031] Step 6: Conduct a binary analysis of the waveforms in the sub-regions using the array AW_med(x, y, t).

[0032] Step 6-1: Denoise the array AW_med(x, y, t) using a three-dimensional edge-preserving denoising method. Denote the denoised values as AW_med_n(x, y, t).

[0033] Step 6-2: Extract the region where Wave1(x, y) = 1. For this region, use the array AW_med_n(x, y, t). In the time t direction, conduct waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and denote the classification result as Wave1_1(x, y). The array Wave1_1(x, y) uses two values, 11 or 12, representing two types of waveforms.

[0034] Step 6-3: Extract the region where Wave1(x,y) = 2. For this region, use the array AW_med_n(x,y,t) and, in the time t direction, perform waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and record the classification result as Wave1_2(x,y). The array Wave1_2 uses two values, 21 or 22, to represent two types of waveforms;

[0035] Step 7: Perform binary analysis of the waveforms in sub-regions using the array AW_hig(x,y,t).

[0036] Step 7-1: Denoise the array AW_hig(x,y,t) using a three-dimensional edge-preserving denoising method. Denote the denoised values as AW_hig_n(x,y,t).

[0037] Step 7-2: Extract the region where Wave1_1(x,y) = 11. For this region, use the array AW_hig_n(x,y,t) and, in the time t direction, perform waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and record the classification result as Wave1_1_1(x,y). The array Wave1_1_1 uses two values, 111 or 112, to represent two types of waveforms.

[0038] Step 7-3: Extract the region where Wave1_1(x,y) = 12. For this region, use the array AW_hig_n(x,y,t) and, in the time t direction, perform waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and record the classification result as Wave1_1_2(x,y). The array Wave1_1_2 uses two values, 121 or 122, to represent two types of waveforms.

[0039] Step 7-4: Extract the region where Wave1_2(x,y) = 21. For this region, use the array AW_hig_n(x,y,t) and, in the time t direction, perform waveform clustering analysis using a waveform clustering analysis method based on neural network training. Set the number of waveform classifications to 2, and record the classification result as Wave1_2_1(x,y). The array Wave1_2_1 uses two values, 211 or 212, to represent two types of waveforms.

[0040] Step 7-5: Extract the region where Wave1_2(x,y) = 22. For this region, use the array AW_hig _n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_2_2(x,y). The array Wave1_2_2 uses 2 values, 221 or 222, representing 2 types of waveforms.

[0041] Step 8: Merge the 8 regions obtained in Step 7, denoted as Wave0(x,y). The array Wave0(x,y) has 8 values 111, 112, 121, 122, 211, 212, 221, 222, corresponding to the 8 regions in Step 7 respectively.

[0042] Step 9: Based on A(x,y,t) in Step 1 and Wave0(x,y) in Step 8, divide A(x,y,t) into 8 sub-regions. Use the covariance matrix analysis method to calculate the correlation coefficients between the 8 sub-regions. Merge the sub-regions with correlation coefficients greater than 0.9 and update the values of Wave0(x,y). For example, if the overall correlation coefficient between the waveforms of the first sub-region and the second sub-region is greater than 0.9, modify the value of Wave0(x,y) in the second sub-region to the value of Wave0(x,y) in the first sub-region to achieve the merger of the first sub-region and the second sub-region.

[0043] Step 10: Output the Wave0(x,y) array, and use different color scales to represent the values of the array Wave0(x,y). Plot a graph to represent the final waveform classification result with an image.

[0044] Embodiment 2

[0045] Figure 2 It is a waveform classification plan view of the second member of the Qixia Formation in a certain research area in Sichuan. The 2nd, 3rd, and 4th type regions in the figure well represent the development areas of high-quality oil and gas reservoirs.

[0046] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. The specific steps of a seismic waveform classification calculation method based on multi-frequency data volume include: Step 1: Input the 3D seismic data of the target layer section in the study area, denoted as array A(x, y, t); Step 2: Calculate the main frequency of array A(x, y, t) using the Fourier transform analysis method, denoted as Fmain; Step 3: Perform time-frequency analysis on array A(x, y, t) using the time-frequency wavelet transform method to obtain the time-frequency analysis array AW(x, y, t, f); Step 4: Extract three co-frequency data volumes with different frequencies. Step 4-1: Denote the frequency of the first co-frequency data volume as f1, the frequency of the second co-frequency data volume as f2, and the frequency of the third co-frequency data volume as f3. f1 = int(Fmain - Fmain / 3), f2 = int(Fmain), f3 = int(Fmain + Fmain / 3) Step 4-2: Extract three co-frequency data volumes from the time-frequency analysis array AW(x, y, t, f) in Step 3, AW(x, y, t, f1), AW(x, y, t, f2), AW(x, y, t, f3). Since f1, f2, and f3 are fixed values, the arrays AW(x, y, t, f1), AW(x, y, t, f2), AW(x, y, t, f3) are 3D arrays. Denote AW_low(x, y, t) = AW(x, y, t, f1), AW_med(x, y, t) = AW(x, y, t, f2), AW_hig(x, y, t) = AW(x, y, t, f3); Step 5: Perform binary analysis on the waveforms in the entire region using array AW_low(x, y, t). Step 5-1: Denoise array AW_low(x, y, t) using the 3D edge-preserving denoising method. The denoised value is denoted as AW_low_n(x, y, t). Step 5-2: For array AW_low_n(x, y, t), in the time t direction, use the waveform clustering analysis method based on neural network training to perform waveform clustering analysis. The number of waveform classifications is set to 2, and the classification result is denoted as Wave1(x, y). Array Wave1(x, y) uses two values, 1 or 2, representing two types of waveforms. Step 6: Perform binary analysis on the waveforms in the sub-region using array AW_med(x, y, t). Step 6-1: Denoise array AW_med(x, y, t) using the 3D edge-preserving denoising method. The denoised value is denoted as AW_med_n(x, y, t). Step 6-2: Extract the region where Wave1(x, y) = 1. For this region, use array AW_med_n(x, y, t). In the time t direction, use the waveform clustering analysis method based on neural network training to perform waveform clustering analysis. The number of waveform classifications is set to 2, and the classification result is denoted as Wave1_1(x, y). Array Wave1_1 uses two values, 11 or 12, representing two types of waveforms. Step 6-3: Extract the region where Wave1(x,y) = 2. For this region, use the array AW_med_n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_2(x,y). The array Wave1_2 uses 2 values, 21 or 22, representing 2 types of waveforms; Step 7: Use the array AW_hig(x,y,t) to perform binary analysis of the waveforms in sub-regions. Step 7-1: Denoise the array AW_hig(x,y,t) using a three-dimensional edge-preserving denoising method. The denoised value is recorded as AW_hig_n(x,y,t). Step 7-2: Extract the region where Wave1_1(x,y) = 11. For this region, use the array AW_hig_n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_1_1(x,y). The array Wave1_1_1 uses 2 values, 111 or 112, representing 2 types of waveforms. Step 7-3: Extract the region where Wave1_1(x,y) = 12. For this region, use the array AW_hig_n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_1_2(x,y). The array Wave1_1_2 uses 2 values, 121 or 122, representing 2 types of waveforms. Step 7-5: Extract the region where Wave1_2(x,y) = 21. For this region, use the array AW_hig_n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_2_1(x,y). The array Wave1_2_1 uses 2 values, 211 or 212, representing 2 types of waveforms. Step 7-6: Extract the region where Wave1_2(x,y) = 22. For this region, use the array AW_hig_n(x,y,t). In the time t direction, adopt the waveform clustering analysis method based on neural network training to carry out waveform clustering analysis. Set the number of waveform classifications to 2, and record the classification result as Wave1_2_2(x,y). The array Wave1_2_2 uses 2 values, 221 or 222, representing 2 types of waveforms; Step 8: Merge the 8 regions obtained in Step 7, and record it as Wave0(x,y). The array Wave0(x,y) has 8 values 111, 112, 121, 122, 211, 212, 221, 222, corresponding to the 8 regions in Step 7 respectively; Step 9: Based on A(x, y, t) in Step 1 and Wave0(x, y) in Step 8, divide A(x, y, t) into 8 sub-regions, calculate the correlation coefficients between the 8 sub-regions using the covariance matrix analysis method, merge the sub-regions with a correlation coefficient greater than 0.9, and update the value of Wave0(x, y). For example, if the overall correlation coefficient between the waveforms of the first sub-region and the second sub-region is greater than 0.9, modify the value of Wave0(x, y) in the second sub-region to the value of Wave0(x, y) in the first sub-region to achieve the merger of the first sub-region and the second sub-region; Step 10: Output the Wave0(x, y) array, use different color scales to represent the values of the array Wave0(x, y), plot a graph, and use the image to represent the final waveform classification result.

Images