A method, device and related equipment for processing waveform of seismic data

Abstract

The present invention discloses a waveform processing method, device and related equipment for seismic data. The method may include: using adaptive local averaging sampling processing on seismic data to extract single-channel waveform data of the target layer; inputting the single-channel waveform data of the target layer into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain single-channel waveform data after dimensionality reduction; inputting the single-channel waveform data after dimensionality reduction into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data. This method can more conveniently extract the difference characteristics between different seismic waveforms. Using seismic waveforms to better generate seismic phase maps effectively helps reservoir distribution prediction work and provides reliable support for subsequent seismic phase analysis, reservoir parameter prediction, fluid identification, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data processing, and in particular to a seismic data waveform processing method, device and related equipment. Background Art

[0002] Seismic waveforms reflect changes in underground geological conditions, with different waveforms corresponding to different types of geological features. Seismic waveform classification, which uses cluster analysis of seismic waveforms, is an effective and rapid method for predicting the spatial distribution of reservoirs. Traditional seismic waveform classification primarily uses pattern recognition methods to classify seismic waveforms. By classifying typical seismic model trace shapes and iteratively updating neural network algorithms, all seismic traces in the actual work area are ultimately classified. These model traces represent the diversity of seismic trace shapes within the target layer of the entire region. The primary purpose of seismic waveform classification is to establish a correspondence between the overall changes in seismic signals and the distribution patterns of these changes and the distribution of oil and gas reservoirs or fluids. Summary of the Invention

[0003] The inventors discovered that in early algorithms, seismic phase analysis was performed using unsupervised clustering techniques. Waveform features within a time window above and below the target layer were typically used as sample features. Traditional methods such as K-means or self-organizing maps were then used to automatically classify reservoir data based on waveform sample point features. However, typical waveform features contain a significant amount of information redundancy. Using principal component analysis (PCA) to extract (compress) waveform data and then perform cluster analysis presents the problem of difficulty determining the number of features to extract.

[0004] In view of the above problems, the present invention is proposed to provide a waveform processing method, apparatus and related equipment for seismic data that overcome the above problems or at least partially solve the above problems.

[0005] In a first aspect, an embodiment of the present invention provides a waveform processing method for seismic data, which may include:

[0006] Adaptive local averaging sampling is used to process seismic data to extract single-channel waveform data of the target layer;

[0007] The single-channel waveform data of the target layer is input into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain the single-channel waveform data after dimensionality reduction; the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

[0008] Optionally, the process of using adaptive local averaging sampling to process the seismic data to extract single-channel waveform data of the target layer may include:

[0009] Extracting seismic waveform data of unequal time window lengths between two target layers in the seismic data;

[0010] Counting the waveform length of each piece of seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data;

[0011] The seismic waveform data of all seismic channels are interpolated using the waveform length with the largest length in the histogram to obtain single-channel waveform data with the same maximum waveform length.

[0012] Optionally, the interpolation processing is a nearest neighbor interpolation processing method.

[0013] Optionally, the deep neural network model is pre-trained through the following steps:

[0014] Acquire a training sample set, wherein each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference features thereof;

[0015] The training samples in the training sample set are used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for classification parameter estimation.

[0016] In a second aspect, an embodiment of the present invention provides a method for training a machine learning model, which may include:

[0017] Acquire a training sample set, wherein each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference characteristics thereof;

[0018] The training samples in the training sample set are used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for parameter estimation of clustering.

[0019] Optionally, obtaining a training sample set may specifically include:

[0020] Adaptive local averaging sampling is used to process seismic data to extract single-channel waveform data of the target layer;

[0021] Analyzing the single-channel waveform data based on principal component analysis to obtain single-channel waveform data after dimensionality reduction;

[0022] Performing cluster analysis on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data;

[0023] The single-channel waveform data, the single-channel waveform data after dimensionality reduction and the waveform difference characteristics thereof are determined as a training sample, and a training sample set consisting of the training samples is obtained.

[0024] Optionally, the process of using adaptive local averaging sampling to process the seismic data to extract single-channel waveform data of the target layer may include:

[0025] Extracting seismic waveform data of unequal time window lengths between two target layers in seismic data;

[0026] Counting the waveform length of each piece of seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data;

[0027] The seismic waveform data of all seismic channels are interpolated using the waveform length with the largest length in the histogram to obtain single-channel waveform data with the same maximum waveform length.

[0028] Optionally, the interpolation processing is a nearest neighbor interpolation processing method.

[0029] Optionally, the convolutional neural network may include: a downsampling network and an upsampling network;

[0030] The downsampling network includes, in sequence: a first convolutional layer, a first batch of regularization layers, a first maximum pooling layer, a first parameter optimization layer, a second convolutional layer, a second batch of regularization layers, a second maximum pooling layer, and a third convolutional layer;

[0031] The upsampling network includes, in sequence: a first upsampling layer, a first fusion interface, a third parameter optimization layer, a fourth convolutional layer, a second upsampling layer, a second fusion interface, a fourth parameter optimization layer, and a fifth convolutional layer;

[0032] The first upsampling layer processes upsampling based on bilinear interpolation;

[0033] The first fusion interface is used to connect the sampling result of the first upsampling layer with the result of the second convolutional layer;

[0034] The second fusion interface is used to connect the sampling result of the second upsampling layer with the result of the first convolutional layer.

[0035] Optionally, the center point parameters of the unsupervised cluster analysis network are:

[0036] Among them, K * is the number of clusters; m r is the center of each cluster subclass; ∑ r is the covariance matrix of each cluster subclass; r is the subscript index.

[0037] In a third aspect, an embodiment of the present invention provides an application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification;

[0038] The waveform difference characteristics in the seismic data are obtained according to the waveform processing method of seismic data described in the first aspect.

[0039] In a fourth aspect, an embodiment of the present invention provides a waveform processing device for seismic data, which may include:

[0040] An extraction module is used to process seismic data using adaptive local averaging sampling to extract single-channel waveform data of a target layer;

[0041] A determination module is used to input the single-channel waveform data of the target layer into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain reduced-dimensional single-channel waveform data; and the reduced-dimensional single-channel waveform data is input into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

[0042] In a fifth aspect, an embodiment of the present invention provides a training device for a machine learning model, which may include:

[0043] An acquisition module is used to acquire a training sample set, wherein each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference characteristics thereof;

[0044] A training module is used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network using the training samples in the training sample set, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for parameter estimation of clustering.

[0045] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, it implements the waveform processing method of seismic data described in the first aspect, or implements the training method of the machine learning model as described in the second aspect, or implements the application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification as described in the third aspect.

[0046] In the seventh aspect, an embodiment of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the waveform processing method for seismic data as described in the first aspect is implemented, or the training method for the machine learning model as described in the second aspect is implemented, or the application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction, and / or fluid identification as described in the third aspect is implemented.

[0047] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0048] An embodiment of the present invention provides a waveform processing method, apparatus and related equipment for seismic data, which may include: using adaptive local averaging sampling processing on seismic data to extract single-channel waveform data of a target layer; inputting the single-channel waveform data of the target layer into a pre-trained deep neural network model comprising a convolutional neural network and an unsupervised clustering analysis network, wherein the single-channel waveform data of the target layer passes through a feature extraction layer in the convolutional neural network to obtain single-channel waveform data after dimensionality reduction; inputting the single-channel waveform data after dimensionality reduction into an unsupervised clustering analysis network to determine waveform difference characteristics in the seismic data.

[0049] The above method provided by the embodiment of the present invention has better explored and applied the automatic extraction of information from seismic waveform features based on the combination of deep learning autoencoder convolutional network and cluster analysis method, which can more conveniently extract the difference features between different seismic waveforms. The seismic waveform is used to better generate seismic phase maps, which effectively assists in reservoir distribution prediction. In order to avoid the problems existing in traditional cluster analysis methods and their impact on oil and gas prediction, the accuracy of oil and gas prediction is ultimately improved, the drilling risk of the research target is reduced, and reliable data is provided for the efficient exploration and development of oil and gas fields. Furthermore, the seismic data is compressed and feature extracted through a pre-trained deep neural network model to extract the maximum difference features of the waveform features, providing reliable support for subsequent seismic phase analysis, reservoir parameter prediction, fluid identification, etc.

[0050] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0051] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0053] Figure 1 Schematic diagram of the process of training the machine learning model provided in Example 1 of the present invention;

[0054] Figure 2 This is a schematic diagram of the process of extracting single-channel waveform data of the target layer provided in Example 1 of the present invention;

[0055] Figure 3 This is a specific flow chart of step S21 provided in Example 1 of the present invention;

[0056] Figure 4 This is a schematic diagram of a deep neural network model provided in Example 1 of the present invention;

[0057] Figure 5 This is a schematic diagram of a specific process for training the machine learning model provided in Example 1 of the present invention;

[0058] Figure 6 This is a schematic diagram of the structure of the training device for the machine learning model provided in Example 1 of the present invention;

[0059] Figure 7 Schematic diagram of the flow of the waveform processing method for seismic data provided in Example 2 of the present invention;

[0060] Figure 8 This is an example of the river channel effect obtained by automatic feature extraction and automatic waveform clustering provided in Example 2 of the present invention;

[0061] Figure 9 This is a schematic structural diagram of the waveform processing device for seismic data provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0062] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0063] Example 1

[0064] In embodiment 1 of the present invention, a method for training a machine learning model is provided, referring to Figure 1 As shown, the method may include the following steps:

[0065] Step S11: obtaining a training sample set, wherein each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference features thereof.

[0066] Step S12: Use the training samples in the training sample set to train the deep neural network model including the convolutional neural network and the unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for parameter estimation of clustering.

[0067] It should be noted that the above-mentioned convolutional neural network in the embodiment of the present invention is a U-net type autoencoding convolutional neural network.

[0068] The training method of the above-mentioned machine learning model provided in the embodiment of the present invention is achieved by combining an autoencoding convolutional neural network and an unsupervised cluster analysis network. That is, after the autoencoding convolutional neural network compresses the seismic data, the reduced-dimensional seismic data is subjected to feature extraction by an unsupervised cluster analysis network to extract the maximum difference features of the waveform characteristics, thereby providing reliable support for subsequent seismic phase analysis, reservoir parameter prediction, fluid identification, etc.

[0069] In the above step S11, refer to Figure 2 As shown, obtaining a training sample set may specifically include the following steps:

[0070] Step S21: Apply adaptive local averaging sampling to the seismic data to extract single-channel waveform data of the target layer.

[0071] This step extracts single-channel waveform data from the target layer of seismic data. This extraction process involves creating time windows of equal length Δt above and below the target layer l0, pre-determined by the interpreter. This allows for the extraction of single-channel waveform data with identical sampling points based on the time window length. The inventors have discovered that single-channel waveform data extracted in this manner does not fully conform to geological sedimentary patterns. If only windows of equal length are selected above and below a given layer, the environment along that layer may be assumed to be uniform. This results in an inability to accurately predict the stratigraphic orientation of geological deposits when changes in geological sedimentation occur.

[0072] Therefore, the inventors of this application innovatively proposed the following adaptive local averaging sampling processing method, referring to Figure 3 As shown, the following steps may be specifically included:

[0073] Step S211: extract seismic waveform data of unequal time window lengths between two target layers in the seismic data.

[0074] In this step, along the given upper and lower target layers l1 and l2, the unequal time window lengths between layers (set as Δt up and Δt down ) earthquake waveform data T is the total number of seismic traces, and i is the seismic trace index variable.

[0075] Step S212: Count the waveform length of each piece of seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data.

[0076] In this step, for each seismic waveform data s i The waveform length len_s i Perform statistics to obtain earthquake waveform data s i Waveform length len_s i Distribution histogram map S .

[0077] Step S213: interpolating the seismic waveform data of all seismic channels using the waveform length with the largest length in the histogram to obtain single-channel waveform data with the same maximum waveform length.

[0078] In this step, from the above histogram map S Find the length len_s corresponding to the distribution with the largest length * . All s i According to len_s * The length is interpolated by the nearest interpolation method (the value of the interpolation point is replaced by the value of the point closest to it), and finally the data set is Convert to have the same length len_s *Data collection

[0079] The single-channel waveform data of the target layer obtained through the above steps is more consistent with the geological sedimentary law and is more accurate in extracting differential features.

[0080] Step S22: Analyze the single-channel waveform data based on principal component analysis to obtain single-channel waveform data after dimensionality reduction. In this step, when obtaining training samples, principal component analysis (PCA) can be used to analyze the single-channel waveform data to eliminate a large amount of redundant information in the waveform features, thereby obtaining single-channel waveform data after dimensionality reduction.

[0081] Step S23: cluster analysis is performed on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data. In this step, when obtaining training samples, existing cluster analysis methods can be used to obtain waveform difference characteristics of the single-channel waveform data.

[0082] Step S24 : determining the single-channel waveform data, the single-channel waveform data after dimensionality reduction, and the waveform difference features thereof as a training sample, and obtaining a training sample set consisting of the training samples.

[0083] The acquisition process of the above-mentioned training sample set in the embodiment of the present invention includes extracting the single-channel waveform data of the target layer, the dimensionality-reduced single-channel waveform data after compression of the single-channel waveform data, and the waveform difference features extracted by clustering based on the dimensionality-reduced single-channel waveform data, which are used as a training sample to constitute a training sample set.

[0084] In another optional embodiment, in order to extract seismic waveform features, it is first necessary to define the fuzzy degradation attributes of each input With the corresponding output The correspondence between them can be simply expressed as the objective function:

[0085]

[0086] Among them, the function and are operators that need to be simulated. In this embodiment, the inventor uses a neural network model for simulation. is the set of single-channel waveform data samples obtained above. In order to make the function and It is not a simple identity transformation. In the initial stage of the algorithm, random noise needs to be added to initialize the network model. Dropout operation needs to be added during the algorithm process to make the algorithm not equal to the identity transformation, which can be expressed as follows:

[0087]

[0088] in, and are the random noise perturbation operations corresponding to the network.

[0089] In another optional embodiment, the above convolutional neural network (U-net network) net auto It is mainly composed of convolutional layers, pooling layers, batch normalization layers, upsampling layers and skip layers, among which the skip layer can merge multi-scale features. The specific structure is: conv1 (including three convolution kernels, using 3*3 convolution kernels, same scale convolution), norm1 (batch normalization), maxpool1 (using maximum pooling), dropout1 (parameter optimization layer), conv2 (including three convolution kernels, using 3*3 convolution kernels, same scale convolution), norm2 (batch normalization), maxpool2 (using maximum pooling), conv3 (including three convolution kernels, using 3*3 convolution kernels, same scale convolution), dropout3 (parameter optimization layer), uppool1 (using bilinear interpolation), Value upsampling), fuse1 (connecting the uppool upsampling result with the result of conv2 using a side-by-side concat operation), conv4 (4 convolution kernels, using a 3*3 convolution kernel, same scale convolution), uppool2 (using bilinear interpolation upsampling), fuse2 (connecting the uppool2 upsampling result with the result of conv1 using a side-by-side concat operation), conv5 (5 convolution kernels, using a 3*3 convolution kernel, same scale convolution), using 1*1*c (c is the number of channels) full convolution to the final fusion result

[0090] The network mainly consists of two parts, and its architecture can include: downsampling network and upsampling network; that is, conv1+norm1+maxpool1+dropout1+conv2+norm2+maxpool2+conv3 downsampling network net down and uppool1+fuse1+dropout3+conv4+uppool2+fuse2+dropout4+conv5 upsampling network net up The downsampling network includes: the first convolution layer, the first batch of regularization layers, the first maximum pooling layer, the first parameter optimization layer, the second convolution layer, the second batch of regularization layers, the second maximum pooling layer, and the third convolution layer; the upsampling network includes: the first upsampling layer, the first fusion interface, the third parameter optimization layer, the fourth convolution layer, the second upsampling layer, the second fusion interface, the fourth parameter optimization layer, and the fifth convolution layer.

[0091] The above network structure is similar to the end-to-end self-organizing type network structure, and the input is seismic waveform data Its output is also seismic waveform data net dow n and net up Together they form the entire network auto , the network is a self-input and self-output Through network learning, On the net down The network is reduced in dimension (feature extraction), which can be achieved through the net up Restored. Network net down The calculated output result is the extracted sample data set Each Characteristic information after waveform compression This method establishes a feature extraction process based on its own waveform characteristics for the subsequent unsupervised cluster analysis. All by a network net down Output compression features and upsampling network net up The output of Corresponding to it.

[0092] In another optional embodiment, the center point parameters of the unsupervised cluster analysis network are:

[0093] Among them, K * is the number of clusters; m r is the center of each cluster subclass; ∑ r is the covariance matrix of each cluster subclass; r is the subscript index.

[0094] 1) The embodiment of the present invention uses an adaptive K-means unsupervised clustering network

[0095] The inventors found that the main disadvantage of the existing K-means unsupervised clustering network is the number of clusters K * It must be given before the algorithm is run, and the number of clusters K cannot be changed during the algorithm. * In order to make the K-means algorithm suitable for the deep learning process, we set the center point parameter of K-means to Among them, m r is the center of each cluster subclass, ∑ r Optimize the covariance matrix of each cluster subclass, r is the subscript index. While optimizing, the number of clusters K * Also optimized, and each time within a certain range [Ks ,K e ]Inside, K s and K e They are given in advance by the interpreter according to the actual working area conditions, and they represent the upper and lower bounds of the number of subclasses in the clustering process. * Intra-class dispersion after clustering and inter-class dispersion choose The biggest K * The corresponding error Merging networks for backpropagation.

[0096] Finally, based on the above convolutional neural network and unsupervised cluster analysis network, a cluster analysis and feature extraction combined network, namely a deep neural network model, is established.

[0097] The role of the merge network is to extract each optimized earthquake features, and combine them for cluster analysis to generate the final classification results The network model is shown in the figure below: Figure 4 As shown in the figure, its biggest innovation is to use unsupervised clustering analysis network to analyze the output of convolutional neural network. Perform clustering and calculate the clustering error The output of the above autoencoder network With data collection The error between And let the total error function be: The parameters α and β are weight parameters of the total error accounted for by the errors generated by the two steps, respectively, and can be set in advance according to the actual work area.

[0098] Combine Figure 5 As shown, the embodiment of the present invention obtains seismic waveforms of the entire work area and performs denoising preprocessing on the seismic data using the existing median filtering algorithm; then, local mean decomposition is used to extract features from each seismic waveform; after the above-mentioned single waveform data is placed in the autoencoder network for training, the trained network model is used to output the encoder results for cluster analysis, and the weights are adjusted and updated based on the classification results; until the algorithm is stable, the compressed feature data is obtained; and the trained network model is used to classify all seismic waveforms.

[0099] Based on the same inventive concept, an embodiment of the present invention further provides a training device for a machine learning model, referring to Figure 6 As shown, the device may include: an acquisition module 61 and a training module 62, and its working principle is as follows:

[0100] The acquisition module 61 is used to acquire a training sample set, where each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference characteristics thereof;

[0101] The training module 62 is used to train a deep neural network model including a convolutional neural network and an unsupervised clustering analysis network using the training samples in the training sample set, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised clustering analysis network for parameter estimation of clustering.

[0102] In an optional embodiment, the acquisition module 61 is specifically configured to:

[0103] Adaptive local averaging sampling is used to process seismic data to extract single-channel waveform data of the target layer;

[0104] Analyzing the single-channel waveform data based on principal component analysis to obtain single-channel waveform data after dimensionality reduction;

[0105] Performing cluster analysis on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data;

[0106] The single-channel waveform data, the single-channel waveform data after dimensionality reduction and the waveform difference characteristics thereof are determined as a training sample, and a training sample set consisting of the training samples is obtained.

[0107] In another optional embodiment, when extracting the single-channel waveform data of the target layer, the acquisition module 61 is specifically configured to:

[0108] Extracting seismic waveform data of unequal time window lengths between two target layers in seismic data;

[0109] Counting the waveform length of each piece of seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data;

[0110] The seismic waveform data of all seismic channels are interpolated using the longest waveform length in the histogram to obtain single-channel waveform data with the same maximum waveform length. The interpolation process is a nearest neighbor interpolation process.

[0111] In another optional embodiment, the convolutional neural network in the training module 62 may include: a downsampling network and an upsampling network;

[0112] The downsampling network includes, in sequence: a first convolutional layer, a first batch of regularization layers, a first maximum pooling layer, a first parameter optimization layer, a second convolutional layer, a second batch of regularization layers, a second maximum pooling layer, and a third convolutional layer;

[0113] The upsampling network includes, in sequence: a first upsampling layer, a first fusion interface, a third parameter optimization layer, a fourth convolutional layer, a second upsampling layer, a second fusion interface, a fourth parameter optimization layer, and a fifth convolutional layer;

[0114] The first upsampling layer processes upsampling based on bilinear interpolation;

[0115] The first fusion interface is used to connect the sampling result of the first upsampling layer with the result of the second convolutional layer;

[0116] The second fusion interface is used to connect the sampling result of the second upsampling layer with the result of the first convolutional layer.

[0117] In another optional embodiment, the center point parameters of the unsupervised cluster analysis network in the training module 62 are:

[0118] Among them, K * is the number of clusters; m r is the center of each cluster subclass; ∑ r is the covariance matrix of each cluster subclass; r is the subscript index.

[0119] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, which implements the training method of the above-mentioned machine learning model when executed by a processor.

[0120] Based on the same inventive concept, an embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein the processor implements the above-mentioned machine learning model training method when executing the program.

[0121] The principles of the problems solved by the above-mentioned devices, media, and related equipment in the embodiments of the present invention are similar to those of the above-mentioned methods. Therefore, their implementation can refer to the implementation of the above-mentioned methods, and the repeated parts will not be repeated.

[0122] Example 2

[0123] In embodiment 2 of the present invention, a waveform processing method for seismic data is provided, referring to Figure 7 As shown,

[0124] Step S71: Use adaptive local averaging sampling processing on seismic data to extract single-channel waveform data of the target layer.

[0125] Step S72: input the single-channel waveform data of the target layer into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain the single-channel waveform data after dimensionality reduction; and input the single-channel waveform data after dimensionality reduction into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

[0126] It should be noted that the above-mentioned pre-trained deep neural network model including the convolutional neural network and the unsupervised clustering analysis network in the embodiment of the present invention can be trained in advance according to the method in Example 1, or can be trained by other means, and the embodiment of the present invention does not make specific limitations on this.

[0127] The above method provided by the embodiment of the present invention has better explored and applied the automatic extraction of information from seismic waveform features based on the combination of deep learning autoencoder convolutional network and cluster analysis method, which can more conveniently extract the difference features between different seismic waveforms. The seismic waveform is used to better generate seismic phase maps, which effectively assists in reservoir distribution prediction. In order to avoid the problems existing in traditional cluster analysis methods and their impact on oil and gas prediction, the accuracy of oil and gas prediction is ultimately improved, the drilling risk of the research target is reduced, and reliable data is provided for the efficient exploration and development of oil and gas fields. Furthermore, the seismic data is compressed and feature extracted through a pre-trained deep neural network model to extract the maximum difference features of the waveform features, providing reliable support for subsequent seismic phase analysis, reservoir parameter prediction, fluid identification, etc.

[0128] Taking the actual 3D seismic data of a certain work area as an example, refer to Figure 8 As shown, the river channel effect obtained by automatic feature extraction and automatic waveform clustering of this solution can be seen. It can be seen that the above method provided by the embodiment of the present invention has achieved good results in extraction.

[0129] In an optional embodiment, adaptive local averaging sampling is used on seismic data to extract single-channel waveform data of the target layer, including:

[0130] Extracting seismic waveform data of unequal time window lengths between two target layers in seismic data;

[0131] Counting the waveform length of each seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data;

[0132] The seismic waveform data of all seismic channels are interpolated using the waveform length with the largest length in the histogram to obtain single-channel waveform data with the same maximum waveform length.

[0133] In another optional embodiment, the interpolation process is a nearest neighbor interpolation process method.

[0134] In another optional embodiment, the above-mentioned deep neural network model is pre-trained through the following steps:

[0135] Obtaining a training sample set, where each training sample in the sample set includes a single-channel waveform data set, single-channel waveform data after dimensionality reduction, and waveform difference features thereof;

[0136] The training samples in the training sample set are used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network. The single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for classification parameter estimation.

[0137] Based on the same inventive concept, an embodiment of the present invention further provides a waveform processing device for seismic data, referring to Figure 9 As shown, the device may include: an extraction module 91 and a determination module 92, and its working principle is as follows:

[0138] The extraction module 91 is used to process the seismic data using adaptive local averaging sampling to extract single-channel waveform data of the target layer;

[0139] The determination module 92 is used to input the single-channel waveform data of the target layer into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain the single-channel waveform data after dimensionality reduction; and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

[0140] Based on the same inventive concept, an embodiment of the present invention further provides an application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification;

[0141] The waveform difference characteristics in the seismic data are obtained according to the above-mentioned waveform processing method of seismic data.

[0142] Based on the same inventive concept, an embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it implements the above-mentioned waveform processing method of seismic data, or implements the above-mentioned application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification.

[0143] Based on the same inventive concept, an embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, it implements the above-mentioned waveform processing method for seismic data, or implements the above-mentioned application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification.

[0144] The principles of the problems solved by the above-mentioned devices, media, and related equipment in the embodiments of the present invention are similar to those of the above-mentioned methods. Therefore, their implementation can refer to the implementation of the above-mentioned methods, and the repeated parts will not be repeated.

[0145] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0146] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0149] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A waveform processing method for seismic data, characterized in that: include: Extracting seismic waveform data of unequal time window lengths between two target layers in seismic data; Counting the waveform length of each piece of seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data; Interpolating the seismic waveform data of all seismic channels using the waveform length corresponding to the maximum distribution of waveform lengths in the histogram to extract single-channel waveform data of the target layer with the same maximum distribution waveform length; The single-channel waveform data of the target layer is input into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain the single-channel waveform data after dimensionality reduction; the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

2. The method according to claim 1, characterized in that The interpolation process is a nearest neighbor interpolation process method.

3. The method according to claim 1 or 2, characterized in that The deep neural network model is pre-trained through the following steps: Acquire a training sample set, extract seismic waveform data of unequal time window lengths between two target layers in the seismic data; perform statistics on the waveform length of each seismic waveform data to determine a waveform length distribution histogram of the seismic waveform data; perform interpolation processing on the seismic waveform data of all seismic channels using the waveform length corresponding to the maximum distribution of waveform length in the histogram to extract single-channel waveform data of the target layer with the same maximum distribution waveform length; analyze the single-channel waveform data based on the principal component analysis method to obtain single-channel waveform data after dimensionality reduction; Performing cluster analysis on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data; The single-channel waveform data, the single-channel waveform data after dimensionality reduction, and the waveform difference characteristics thereof are determined as a training sample, and a training sample set consisting of the training samples is obtained; The training samples in the training sample set are used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for classification parameter estimation.

4. A method for training a machine learning model, characterized in that: include: Obtain a training sample set and extract seismic waveform data of unequal time window lengths between two target layers in the seismic data; The waveform length of each seismic waveform data is counted to determine a waveform length distribution histogram of the seismic waveform data; the seismic waveform data of all seismic channels are interpolated using the waveform length corresponding to the maximum distribution of waveform length in the histogram to extract single-channel waveform data of the target layer with the same maximum distribution waveform length; the single-channel waveform data is analyzed based on the principal component analysis method to obtain single-channel waveform data after dimensionality reduction; Performing cluster analysis on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data; The single-channel waveform data, the single-channel waveform data after dimensionality reduction, and the waveform difference characteristics thereof are determined as a training sample, and a training sample set consisting of the training samples is obtained; The training samples in the training sample set are used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for parameter estimation of clustering.

5. The method according to claim 4, characterized in that The interpolation process is a nearest neighbor interpolation process method.

6. The method according to claim 4 or 5, characterized in that The convolutional neural network includes: a downsampling network and an upsampling network; The downsampling network includes, in sequence: a first convolutional layer, a first batch of regularization layers, a first maximum pooling layer, a first parameter optimization layer, a second convolutional layer, a second batch of regularization layers, a second maximum pooling layer, and a third convolutional layer; The upsampling network includes, in sequence: a first upsampling layer, a first fusion interface, a third parameter optimization layer, a fourth convolutional layer, a second upsampling layer, a second fusion interface, a fourth parameter optimization layer, and a fifth convolutional layer; The first upsampling layer processes upsampling based on bilinear interpolation; The first fusion interface is used to connect the sampling result of the first upsampling layer with the result of the second convolutional layer; The second fusion interface is used to connect the sampling result of the second upsampling layer with the result of the first convolutional layer.

7. The method according to claim 4 or 5, characterized in that The center point parameters of the unsupervised cluster analysis network are: Among them, K * is the number of clusters; m r is the center of each cluster subclass; ∑ r is the covariance matrix of each cluster subclass; r is the subscript index.

8. An application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification; in, The waveform difference feature in the seismic data is obtained by the waveform processing method for seismic data according to any one of claims 1 to 3.

9. A waveform processing device for seismic data, characterized in that: include: An extraction module is used to extract seismic waveform data of unequal time window lengths between two target layers in the seismic data; The waveform length of each seismic waveform data is counted to determine a waveform length distribution histogram of the seismic waveform data; the seismic waveform data of all seismic channels are interpolated using the waveform length corresponding to the maximum distribution of waveform length in the histogram to extract single-channel waveform data of the target layer with the same maximum distribution waveform length; A determination module is used to input the single-channel waveform data of the target layer into a pre-trained deep neural network model including a convolutional neural network and an unsupervised cluster analysis network, wherein the single-channel waveform data of the target layer passes through the feature extraction layer in the convolutional neural network to obtain reduced-dimensional single-channel waveform data; and the reduced-dimensional single-channel waveform data is input into the unsupervised cluster analysis network to determine the waveform difference characteristics in the seismic data.

10. A training device for a machine learning model, characterized in that: include: An acquisition module is used to acquire a training sample set and extract seismic waveform data of unequal time window lengths between two target layers in the seismic data; The waveform length of each seismic waveform data is counted to determine a waveform length distribution histogram of the seismic waveform data; the seismic waveform data of all seismic channels are interpolated using the waveform length corresponding to the maximum distribution of waveform length in the histogram to extract single-channel waveform data of the target layer with the same maximum distribution waveform length; the single-channel waveform data is analyzed based on the principal component analysis method to obtain single-channel waveform data after dimensionality reduction; Performing cluster analysis on the single-channel waveform data after dimensionality reduction to determine waveform difference characteristics of the single-channel waveform data; The single-channel waveform data, the single-channel waveform data after dimensionality reduction, and the waveform difference characteristics thereof are determined as a training sample, and a training sample set consisting of the training samples is obtained; A training module is used to train a deep neural network model including a convolutional neural network and an unsupervised cluster analysis network using the training samples in the training sample set, wherein the single-channel waveform data of the target layer is input into the convolutional neural network for parameter estimation of feature extraction, and the single-channel waveform data after dimensionality reduction is input into the unsupervised cluster analysis network for parameter estimation of clustering.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the waveform processing method of seismic data as described in any one of claims 1 to 3, or implements the training method of the machine learning model as described in any one of claims 4 to 7, or implements the application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification as described in claim 8.

12. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the waveform processing method of seismic data as described in any one of claims 1 to 3, or implements the training method of the machine learning model as described in any one of claims 4 to 7, or implements the application of waveform difference characteristics in seismic data in seismic phase analysis, storage parameter prediction and / or fluid identification as described in claim 8.

Images