Seismic data frequency expanding method based on combination of least square deconvolution and deep learning

Abstract

The embodiment of the invention discloses a seismic data frequency-expanding method based on combination of least square deconvolution and deep learning, and the method comprises the steps: carrying out the preliminary frequency-expanding processing of seismic data through least square deconvolution, and obtaining preliminary frequency-expanded seismic data; taking the seismic data as sample data, taking the preliminary frequency-broadening seismic data as target data corresponding to the sample data, and constructing a sample data set; training a pre-constructed convolutional neural network model by using the sample data set to obtain a seismic data frequency-expanding model; and inputting the to-be-processed preliminary frequency-broadening seismic data into the seismic data frequency-broadening model, and performing secondary frequency-broadening processing on the to-be-processed preliminary frequency-broadening seismic data through the seismic data frequency-broadening model to obtain target seismic data. The seismic data quality is greatly improved, the resolution of the seismic data is effectively improved, and a seismic result profile with higher resolution can be obtained.

Description

Technical Field

[0001] This application relates to the field of marine oil geophysical exploration technology, specifically to a seismic data overlay method based on a combination of least squares deconvolution and deep learning. Background Technology

[0002] In the processing of seismic data from geophysical exploration, expanding the seismic data bandwidth is a crucial step, as the bandwidth significantly impacts the quality of seismic imaging results. Obtaining broadband seismic profiles with high-resolution seismic data has always been a key focus and challenge in seismic data processing. Current techniques typically employ conventional geophysical methods alone for seismic data upscaling; however, the results achieved through this method are limited. Summary of the Invention

[0003] In view of the above problems, this application proposes a seismic data upscaling method, apparatus, computing device and computer storage medium based on the combination of least square deconvolution and deep learning, to solve the following problem: the seismic data upscaling processing effect obtained by existing seismic data upscaling methods is limited.

[0004] According to one aspect of the embodiments of this application, a seismic data overlay method based on a combination of least squares deconvolution and deep learning is provided, comprising:

[0005] Preliminary frequency-spreading processing of seismic data is performed using least squares deconvolution to obtain preliminary frequency-spreading seismic data;

[0006] Seismic data was used as sample data, and preliminary frequency-spreading seismic data was used as the target data corresponding to the sample data to construct a sample dataset;

[0007] Using the sample dataset, a pre-built convolutional neural network model is trained to obtain a seismic data overlay model;

[0008] The preliminary frequency-spreading seismic data to be processed is input into the seismic data frequency-spreading model. The seismic data frequency-spreading model performs secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed, and the target seismic data is obtained.

[0009] Furthermore, using least squares deconvolution, preliminary frequency upsampling processing is performed on the seismic data to obtain preliminary frequency upsampling seismic data, which further includes:

[0010] Calculate the autocorrelation value of seismic data and the cross-correlation value between seismic data and desired target data; where the desired target data is the stretched portion of the seismic data.

[0011] Based on the autocorrelation and cross-correlation values, solve for the least square deconvolution filter operator;

[0012] The first frequency domain data corresponding to the desired target data is calculated using the least squares deconvolution filter operator.

[0013] Perform an inverse Fourier transform on the data in the first frequency domain to obtain the desired target data;

[0014] Based on the seismic data and the desired target data, the seismic data after least square deconvolution filtering is calculated and used as the seismic data in the preliminary topology seismic data.

[0015] Furthermore, calculating the autocorrelation value of seismic data and the cross-correlation value between seismic data and the desired target data further includes:

[0016] Perform a Fourier transform on the seismic data to obtain the second frequency domain data corresponding to the seismic data;

[0017] Using the real and imaginary parts of the second frequency domain data, calculate the third frequency domain data corresponding to the autocorrelation value of the seismic data;

[0018] The autocorrelation value of the seismic data in the seismic data is obtained by performing an inverse Fourier transform on the third frequency domain data.

[0019] Based on the autocorrelation value and the step size of the least square deconvolution, the cross-correlation value between the seismic data and the desired target data is estimated.

[0020] Furthermore, the calculation of the first frequency domain data corresponding to the desired target data using the least squares deconvolution filter operator further includes:

[0021] The least squares deconvolution filter operator is transformed into the frequency domain to obtain the fourth frequency domain data corresponding to the least squares deconvolution filter operator;

[0022] The first frequency domain data is calculated using the fourth frequency domain data and the second frequency domain data.

[0023] Furthermore, the convolutional neural network model includes: an encoding structure, a decoding structure, a bottom structure, and a connection structure;

[0024] The encoding structure is used to extract features from the input sample data;

[0025] The decoding structure is used to reconstruct 3D seismic data based on features.

[0026] Furthermore, the encoding structure includes: multiple convolutional layers, multiple max pooling layers, and multiple dropout layers;

[0027] The decoding structure includes: multiple upsampling layers and multiple convolutional layers;

[0028] The bottom structure includes: multiple convolutional layers;

[0029] The connection structure includes: multiple connection layers; different connection layers are used to connect different upsampling layers in the decoding structure and different convolutional layers in the encoding structure.

[0030] According to another aspect of the embodiments of this application, a seismic data overlay device based on a combination of least squares deconvolution and deep learning is provided, comprising:

[0031] The preliminary frequency extension module is suitable for performing preliminary frequency extension processing on seismic data using least squares deconvolution to obtain preliminary frequency extension seismic data;

[0032] The module is suitable for using seismic data as sample data and preliminary frequency-spreading seismic data as the target data corresponding to the sample data to construct a sample dataset.

[0033] The training module is suitable for training a pre-built convolutional neural network model using a sample dataset to obtain a seismic data overlay model.

[0034] The application module is suitable for inputting the preliminary frequency-spreading seismic data to be processed into the seismic data frequency-spreading model, and then performing secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed through the seismic data frequency-spreading model to obtain the target seismic data.

[0035] According to another aspect of the embodiments of this application, a computing device is provided, including: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0036] The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning.

[0037] According to another aspect of the embodiments of this application, a computer storage medium is provided, wherein at least one executable instruction is stored in the storage medium, the executable instruction causing a processor to perform the operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning described above.

[0038] According to another aspect of the embodiments of this application, a computer program product is provided, including at least one executable instruction that causes a processor to perform operations corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning described above.

[0039] According to the technical solution provided in this application, preliminary frequency extension processing of seismic data is performed using least squares deconvolution, thus achieving preliminary frequency extension of seismic data. Furthermore, a convolutional neural network model suitable for processing geophysical exploration seismic data is constructed, combining geophysical methods with deep learning artificial intelligence. The model is trained using the obtained preliminary frequency extension seismic data as target data for deep learning. This allows the seismic data to be trained on the seismic data after least squares deconvolution filtering, resulting in a trained seismic data frequency extension model based on deep learning artificial intelligence. This seismic data frequency extension model can be used for effective frequency extension. This method expands the frequency band of seismic data. A trained seismic data frequency expansion model is applied to the initially frequency-expanded seismic data after least-squares deconvolution filtering. The trained model is then used again to perform frequency expansion processing on the initially frequency-expanded seismic data. This achieves frequency expansion processing of seismic data using a self-supervised artificial intelligence deep learning method, thereby further improving the effect of seismic data frequency expansion processing. This scheme realizes the combined least-squares deconvolution and deep learning to expand the frequency band of seismic data, greatly improving the quality of seismic data, effectively increasing the resolution of seismic data, obtaining higher-resolution seismic profiles, and solving the problem of seismic data bandwidth.

[0040] The above description is merely an overview of the technical solutions of the embodiments of this application. In order to better understand the technical means of the embodiments of this application and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of this application more obvious and understandable, specific implementation methods of the embodiments of this application are described below. Attached Figure Description

[0041] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0042] Figure 1 A flowchart illustrating a seismic data overlay method based on a combination of least squares deconvolution and deep learning according to an embodiment of this application is shown.

[0043] Figure 2 This diagram illustrates a structural schematic of a convolutional neural network model.

[0044] Figure 3 This diagram illustrates the effect of the original data from the forward model.

[0045] Figure 4 This diagram illustrates the effect of forward modeling data after least squares deconvolution.

[0046] Figure 5 This diagram illustrates the effect of forward modeling data after combined least-squares deconvolution and deep learning for extended seismic data bandwidth processing.

[0047] Figure 6 This shows a schematic diagram illustrating the effect of actual seismic profile data;

[0048] Figure 7 This diagram illustrates the effect of the actual seismic profile data after least square deconvolution processing.

[0049] Figure 8 This diagram illustrates the effect of extended seismic profile data after combined least-squares deconvolution and deep learning.

[0050] Figure 9 A structural block diagram of a seismic data overlay device based on a combination of least squares deconvolution and deep learning according to an embodiment of this application is shown.

[0051] Figure 10 A schematic diagram of the structure of a computing device according to an embodiment of this application is shown. Detailed Implementation

[0052] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0053] Figure 1 A flowchart illustrating a seismic data overlay method based on a combination of least squares deconvolution and deep learning, according to an embodiment of this application, is shown below. Figure 1 As shown, the method includes the following steps:

[0054] Step S101: Use least squares deconvolution to perform preliminary frequency upsampling processing on the seismic data to obtain preliminary frequency upsampling seismic data.

[0055] This application proposes a method for effectively expanding the frequency band of seismic data profiles. It utilizes least squares deconvolution to perform preliminary frequency extension processing on the seismic data. Then, it combines geophysical methods with deep learning artificial intelligence, using the obtained preliminary frequency extension seismic data as target data for deep learning model training. This results in a trained seismic data frequency extension model based on deep learning artificial intelligence. Finally, the seismic data frequency extension model is used to perform further frequency extension processing on the preliminary frequency extension seismic data, thereby further improving the seismic data frequency extension processing effect.

[0056] In step S101, the seismic data in the seismic data can be processed by least squares deconvolution to achieve preliminary frequency extension. Specifically, by calculating the autocorrelation value of the seismic data and estimating the cross-correlation value between the seismic data and the desired target data, the seismic data after least squares deconvolution filtering is obtained, which is the seismic data in the preliminary frequency extension seismic data.

[0057] Specifically, the autocorrelation value of the seismic data and the cross-correlation value between the seismic data and the desired target data are calculated, where the desired target data is the stretched portion of the seismic data. Based on the autocorrelation and cross-correlation values, the least squares deconvolution filter operator is solved. Using the least squares deconvolution filter operator, the first frequency domain data corresponding to the desired target data is calculated. The first frequency domain data is subjected to an inverse Fourier transform to obtain the desired target data. Based on the seismic data and the desired target data, the seismic data after least squares deconvolution filtering is calculated and used as the seismic data in the preliminary frequency-spreading seismic data.

[0058] The method for obtaining the autocorrelation value of seismic data is as follows: perform a Fourier transform on the seismic data to obtain the second frequency domain data corresponding to the seismic data; use the real and imaginary parts of the second frequency domain data to calculate the third frequency domain data corresponding to the autocorrelation value of the seismic data; and perform an inverse Fourier transform on the third frequency domain data to obtain the autocorrelation value of the seismic data in the seismic data.

[0059] Specifically, a Fourier transform is performed on the seismic data d(t) to obtain the second frequency domain data D corresponding to the seismic data d(t), that is, the data form of the seismic data d(t) in the frequency domain. The second frequency domain data D is a complex number, and its expression is:

[0060]

[0061] Where D(ω) represents the second frequency domain data corresponding to the seismic data d(t); ω is the angular frequency; N is the number of frequency domain samples; t is the time index, representing the position of the data point in the time series, with values ​​ranging from 1 to N-1; d(t) is the seismic data; and j is the imaginary unit, satisfying j 2=-1.

[0062] Wherein, the real part RD(ω) of D(ω) is:

[0063] The imaginary part ID(ω) of D(ω) is:

[0064] Find the third frequency domain data A(ω) corresponding to its autocorrelation value a(t), that is, the data form of the autocorrelation value a(t) in the frequency domain.

[0065] The real part RA(ω) of A(ω) is: RA(ω) = RD(ω) 2 +ID(ω) 2 Formula 4;

[0066] The imaginary part IA(ω) of A(ω) is: IA(ω)=0.

[0067] The third frequency domain data A(ω) corresponding to the obtained autocorrelation value a(t) of the seismic data d(t) is subjected to an inverse Fourier transform to obtain the autocorrelation value a(t) of the seismic data d(t). The expression for a(t) is:

[0068]

[0069] Where a(t) is the autocorrelation value of the seismic data d(t); N is the number of frequency domain samples; and A(ω) is the third frequency domain data corresponding to the autocorrelation value a(t) of the seismic data d(t).

[0070] The cross-correlation value between seismic data and the desired target data is obtained by estimating the cross-correlation value between the seismic data and the desired target data based on the autocorrelation value and the step size of the least square deconvolution.

[0071] Specifically, the cross-correlation value between the seismic data d(t) and the desired target data can be estimated using b(t). The expression for b(t) is:

[0072] b(t) = a(t + gap) (Formula 6)

[0073] Where gap is the step size of least squares deconvolution. The desired target data is the stretched portion of the seismic data d(t).

[0074] The calculation method for the first frequency domain data corresponding to the desired target data is as follows: the least squares deconvolution filter operator is transformed into the frequency domain to obtain the fourth frequency domain data corresponding to the least squares deconvolution filter operator; the first frequency domain data is calculated using the fourth frequency domain data and the second frequency domain data.

[0075] Based on the autocorrelation value a(t) and the cross-correlation value b(t), the least squares deconvolution filter operator f1 is solved. Specifically, the least squares deconvolution filter operator f1 can be solved using the following formula:

[0076]

[0077] Where t = 0 to M-1, a(-t) = a(t), and M is the total length of the autocorrelation value a(t).

[0078] Transforming the least squares deconvolution filter operator (or filter) f1 into the frequency domain yields the fourth frequency domain data F1, which is the frequency domain data form of the least squares deconvolution filter operator f1. The fourth frequency domain data F1 is a complex number, and its expression is:

[0079]

[0080] Where F1(ω) is the fourth frequency domain data corresponding to the least square deconvolution filter operator f1(t); ω is the angular frequency; and N is the number of frequency domain samples.

[0081] The real part RF1(ω) of F1(ω) is:

[0082] The imaginary part IF1(ω) of F1(ω) is:

[0083] Assume the desired target data is dd1(t), and its corresponding first frequency domain data (i.e., the data form of the desired target data in the frequency domain) is DD(ω), with the real part being RDD(ω) and the imaginary part being IDD(ω).

[0084] DD(ω)=RDD(ω)+i*IDD(ω) Formula 11

[0085] Where RDD(ω) = RF1(ω) * RD(ω) - IF1(ω) * ID(ω) (Formula 12)

[0086] IDD(ω)=RF1(ω)*ID(ω)+IF1(ω)*RD(ω) Formula 13

[0087] Perform an inverse Fourier transform on the obtained first frequency domain data DD(ω) to obtain the desired target data dd1(t);

[0088]

[0089] Then, based on the seismic data d(t) and the desired target data dd1(t), the seismic data dd(t) after least square deconvolution filtering is calculated. The seismic data dd(t) after least square deconvolution filtering is the seismic data dd(t) in the preliminary topology seismic data.

[0090] Where, dd(t) = d(t) - dd1(t) (Formula 15)

[0091] Step S102: Using seismic data as sample data and preliminary frequency-spreading seismic data as the target data corresponding to the sample data, construct a sample dataset.

[0092] In this embodiment, seismic data is used as sample data, and preliminary upsampling seismic data is used as the target data corresponding to the sample data. In other words, preliminary upsampling seismic data is used as the target data (i.e., label data) for deep learning. Therefore, there is no need to create additional label data, and the sample dataset can be easily constructed, which greatly improves the efficiency of sample dataset construction.

[0093] Step S103: Using the sample dataset, train the pre-built convolutional neural network model to obtain the seismic data overlay model.

[0094] After obtaining the sample dataset, a pre-built Convolutional Neural Network (CNN) model can be trained using the sample dataset to obtain the seismic data extension model. Specifically, the seismic data d(t) is fed into the seismic data dd(t) after least-squares deconvolution filtering for deep learning training, thereby obtaining the trained seismic data extension model. The CNN model is used for deep learning training.

[0095] Convolutional neural network models are trained using operations such as convolution, pooling, dropout, and upsampling. A convolutional neural network model can include an encoding structure, a decoding structure, a bottom layer structure, and connection structures.

[0096] The encoding structure is used to extract features from the input sample data, specifically from the input seismic data d(t). The encoding structure may include multiple convolutional layers, multiple max-pooling layers, and multiple dropout layers. For example, the encoding structure may include 6 convolutional layers, 3 max-pooling layers, and 3 dropout layers, with the number of convolutional kernels in the convolutional layers remaining consistent, typically 96.

[0097] The decoding structure is used to reconstruct 3D seismic data based on the extracted features. The decoding structure may include multiple upsampling layers and multiple convolutional layers. For example, a decoding structure may include 3 upsampling layers and 6 convolutional layers, with the number of convolutional kernels in the convolutional layers remaining consistent, typically 96.

[0098] The bottom structure may include multiple convolutional layers. For example, the bottom structure may include two convolutional layers with a large number of convolutional kernels, typically 160.

[0099] The connection structure may include multiple connection layers. Different connection layers are used to connect different upsampling layers in the decoding structure and different convolutional layers in the encoding structure. For example, the connection structure includes three connection layers: the first connection layer connects the first upsampling layer of the decoding structure and the sixth convolutional layer of the encoding structure; the second connection layer connects the second upsampling layer of the decoding structure and the fourth convolutional layer of the encoding structure; and the third connection layer connects the third upsampling layer of the decoding structure and the second convolutional layer of the encoding structure.

[0100] Figure 2 This shows a schematic diagram of a convolutional neural network model, such as... Figure 2 As shown, layers 1 to 12 belong to the encoding structure; layers 13 and 14 belong to the bottom structure; layers 15 to 24 belong to the decoding structure; the connection layer in the connection structure is used to connect layer 15 to layer 10, layer 18 to layer 6, and layer 21 to layer 2.

[0101] Layer 0 is the input layer.

[0102] Encoding structure:

[0103] The first layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0104] The second layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. The third layer is a max pooling layer with a pooling window of 2×2.

[0105] The fourth layer is the Dropout layer, with a drop rate of 0.2%.

[0106] Layer 5 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. Layer 6 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. Layer 7 is a max pooling layer with a pooling window of 2×2.

[0107] The 8th layer is the Dropout layer, with a drop rate of 0.2%.

[0108] Layer 9 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. Layer 10 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0109] The 11th layer is the max pooling layer, with a pooling window of 2×2.

[0110] The 12th layer is the Dropout layer, with a drop rate of 0.2%.

[0111] Bottom structure:

[0112] Layer 13 is a convolutional layer: 160 3×3 convolutional kernels with a stride of 1. Layer 14 is a convolutional layer: 160 3×3 convolutional kernels with a stride of 1. Decoding structure:

[0113] The 15th layer is an upsampling layer with a size of 2×2.

[0114] Layer 16 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. Layer 17 is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1. Layer 18 is an upsampling layer with a size of 2×2.

[0115] The 19th layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0116] The 20th layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0117] The 21st layer is an upsampling layer with a size of 2×2.

[0118] The 22nd layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0119] The 23rd layer is a convolutional layer: 96 3×3 convolutional kernels with a stride of 1.

[0120] Layer 24 is a convolutional layer: a single 3×3 convolutional kernel with a stride of 1. Connection structure:

[0121] Connection layer:

[0122] The 15th floor is connected to the 10th floor.

[0124] The 18th floor is connected to the 6th floor.

[0125] The 21st floor is connected to the 2nd floor.

[0126] Step S104: Input the preliminary frequency-spreading seismic data to be processed into the seismic data frequency-spreading model, and perform secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed through the seismic data frequency-spreading model to obtain the target seismic data.

[0127] After training the seismic data extension model, it can be applied to further improve the seismic data extension processing effect, realizing the use of self-supervised artificial intelligence deep learning methods for seismic data extension processing. Specifically, the input data of the model are: seismic data in the preliminary extension seismic data to be processed, i.e., seismic data dd(t) after least square deconvolution filtering; the loaded model is the seismic data extension model trained in step S103; the seismic data extension model predicts the input data to achieve secondary extension processing, and the output data is the seismic data in the target seismic data, i.e., the frequency band extended and optimized seismic data ddd(t). The seismic data ddd(t) is the result of the seismic data d(t) after joint extension of the seismic data frequency band by least square deconvolution and deep learning.

[0128] The following diagram illustrates the effect of seismic data overlay processing using the method provided in this application.

[0129] Figure 3 This shows a schematic diagram illustrating the effect of the original data from the forward model. Figure 4 Shown Figure 3 This is a schematic diagram illustrating the effect of forward modeling data after least squares deconvolution. Figure 4 and Figure 3 A comparison clearly shows that the resolution has been effectively improved. Figure 5 This diagram illustrates the effect of forward modeling data after combined least-squares deconvolution and deep learning to extend the frequency bands of seismic data. In other words, Figure 5 for Figure 4 The forward modeling data, after being augmented and optimized through deep learning, is then... Figure 5 and Figure 4 A comparison clearly shows that the resolution has been further improved. Figure 5 The data shown is the extended seismic data bandwidth processed by a combination of least squares deconvolution and deep learning. Figure 3 Compared to the data shown, Figure 3 Previously unrecognizable in-phase axes are now clearly visible. Among them, Figures 3 to 5 The horizontal axis represents the number of tracks; the vertical axis represents time, in seconds (s).

[0130] Figure 6 This shows a schematic diagram illustrating the effect of actual seismic profile data. Figure 7 Shown Figure 6 A schematic diagram illustrating the effect of actual seismic profile data after least square deconvolution processing, by... Figure 7 and Figure 6 A comparison clearly shows that the resolution has been effectively improved. Figure 8 This diagram illustrates the effect of extended seismic profile data after combined least-squares deconvolution and deep learning processing. In other words, Figure 8 for Figure 7 The actual seismic profile data, after being extended and optimized by deep learning, is obtained by... Figure 8 and Figure 7 A comparison clearly shows that the resolution has been further improved. Figure 8 The actual seismic profile data shown is processed by a combination of least squares deconvolution and deep learning, resulting in extended seismic data frequency bands. Figure 6 Compared to the data shown, Figure 6 Previously unrecognizable in-phase axes are now clearly visible. Among them, Figures 6 to 8 The horizontal axis represents the number of tracks; the vertical axis represents time, in seconds (s).

[0131] According to the seismic data upscaling method based on least squares deconvolution and deep learning provided in this application embodiment, the least squares deconvolution is used to perform preliminary upscaling processing on seismic data, achieving preliminary upscaling of seismic data; furthermore, a convolutional neural network model suitable for processing geophysical exploration seismic data is constructed, combining geophysical methods with deep learning artificial intelligence. The obtained preliminary upscaled seismic data is used as target data for deep learning model training, allowing deep learning training on seismic data after least squares deconvolution filtering, thereby obtaining a trained seismic data upscaling model based on deep learning artificial intelligence. The frequency extension model can be used to effectively extend the frequency band of seismic data. The trained seismic data frequency extension model is applied to the initially extended seismic data after least squares deconvolution filtering. The trained model is then used again to extend the frequency band of the initially extended seismic data, realizing the use of self-supervised artificial intelligence deep learning methods for seismic data frequency extension processing, thereby further improving the effect of seismic data frequency extension processing. This scheme achieves the combined least squares deconvolution and deep learning for extending the frequency band of seismic data, greatly improving the quality of seismic data, effectively increasing the resolution of seismic data, obtaining higher resolution seismic profiles, and solving the problem of seismic data bandwidth.

[0132] Figure 9 A structural block diagram of a seismic data overlay device based on a combination of least squares deconvolution and deep learning according to an embodiment of this application is shown, as follows: Figure 9 As shown, the device includes: a preliminary frequency extension module 910, a construction module 920, a training module 930, and an application module 940.

[0133] The preliminary frequency extension module 910 is suitable for: using least squares deconvolution to perform preliminary frequency extension processing on seismic data to obtain preliminary frequency extension seismic data.

[0134] The construction module 920 is suitable for: using seismic data as sample data and preliminary frequency-adjusted seismic data as the target data corresponding to the sample data to construct a sample dataset.

[0135] Training module 930 is suitable for: training a pre-built convolutional neural network model using a sample dataset to obtain a seismic data overlay model.

[0136] Application module 940 is suitable for: inputting the preliminary frequency-spreading seismic data to be processed into the seismic data frequency-spreading model, and performing secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed through the seismic data frequency-spreading model to obtain the target seismic data.

[0137] Optionally, the preliminary frequency extension module 910 is further adapted to: calculate the autocorrelation value of the seismic data and the cross-correlation value between the seismic data and the desired target data; wherein the desired target data is the stretched portion of the seismic data; solve the least square deconvolution filter operator based on the autocorrelation value and the cross-correlation value; calculate the first frequency domain data corresponding to the desired target data using the least square deconvolution filter operator; perform an inverse Fourier transform on the first frequency domain data to obtain the desired target data; and calculate the seismic data after least square deconvolution filtering based on the seismic data and the desired target data, as the seismic data in the preliminary frequency extension seismic data.

[0138] Optionally, the preliminary frequency extension module 910 is further adapted to: perform a Fourier transform on the seismic data to obtain second frequency domain data corresponding to the seismic data; use the real and imaginary parts in the second frequency domain data to calculate third frequency domain data corresponding to the autocorrelation value of the seismic data; perform an inverse Fourier transform on the third frequency domain data to obtain the autocorrelation value of the seismic data in the seismic data; and estimate the cross-correlation value between the seismic data and the desired target data based on the autocorrelation value and the step size of the least squares deconvolution.

[0139] Optionally, the preliminary frequency extension module 910 is further adapted to: convert the least squares deconvolution filter operator to the frequency domain to obtain the fourth frequency domain data corresponding to the least squares deconvolution filter operator; and use the fourth frequency domain data and the second frequency domain data to calculate the first frequency domain data.

[0140] Optionally, the convolutional neural network model includes: an encoding structure, a decoding structure, a bottom structure, and a connection structure; wherein the encoding structure is used to extract features from the input sample data; and the decoding structure is used to reconstruct the three-dimensional seismic data based on the features.

[0141] Optionally, the encoding structure includes: multiple convolutional layers, multiple max pooling layers, and multiple dropout layers; the decoding structure includes: multiple upsampling layers and multiple convolutional layers; the bottom structure includes: multiple convolutional layers; the connection structure includes: multiple connection layers; different connection layers are used to connect different upsampling layers in the decoding structure and different convolutional layers in the encoding structure.

[0142] The descriptions of the above modules refer to the corresponding descriptions in the method embodiments, and will not be repeated here.

[0143] According to the seismic data upscaling device based on least squares deconvolution and deep learning provided in this application embodiment, the seismic data is initially upscaled using least squares deconvolution. Furthermore, a convolutional neural network model suitable for processing geophysical exploration seismic data is constructed, combining geophysical methods with deep learning artificial intelligence. The model is trained using the obtained initially upscaled seismic data as target data for deep learning. This allows the seismic data to be trained on the seismic data after least squares deconvolution filtering, thereby obtaining a trained seismic data upscaling model based on deep learning artificial intelligence. The frequency extension model can be used to effectively extend the frequency band of seismic data. The trained seismic data frequency extension model is applied to the initially extended seismic data after least squares deconvolution filtering. The trained model is then used again to extend the frequency band of the initially extended seismic data, realizing the use of self-supervised artificial intelligence deep learning methods for seismic data frequency extension processing, thereby further improving the effect of seismic data frequency extension processing. This scheme achieves the combined least squares deconvolution and deep learning for extending the frequency band of seismic data, greatly improving the quality of seismic data, effectively increasing the resolution of seismic data, obtaining higher resolution seismic profiles, and solving the problem of seismic data bandwidth.

[0144] This application provides a non-volatile computer storage medium storing at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning in any of the above method embodiments.

[0145] This application provides a computer program product, which includes at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning in any of the above method embodiments.

[0146] Figure 10The diagram shows a structural schematic of a computing device according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the computing device.

[0147] like Figure 10 As shown, the computing device may include: a processor 1002, a communications interface 1004, a memory 1006, and a communications bus 1008.

[0148] The processor 1002, communication interface 1004, and memory 1006 communicate with each other via communication bus 1008. Communication interface 1004 is used to communicate with other network elements, such as clients or other servers. Processor 1002 executes program 1010, specifically performing the relevant steps in the embodiment of the seismic data overlay method based on least square deconvolution and deep learning for computing devices described above.

[0149] Specifically, program 1010 may include program code that includes computer operation instructions.

[0150] The processor 1002 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0151] Memory 1006 is used to store program 1010. Memory 1006 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0152] Specifically, program 1010 can be used to cause processor 1002 to execute the seismic data upscaling method based on the combination of least square deconvolution and deep learning in any of the above method embodiments. The specific implementation of each step in program 1010 can be found in the corresponding steps and units described in the above embodiments of seismic data upscaling based on the combination of least square deconvolution and deep learning, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0153] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of this application are not directed to any particular programming language. It should be understood that the contents of the embodiments of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best implementation of the embodiments of this application.

[0154] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0155] Similarly, it should be understood that, in order to simplify this disclosure and aid in understanding one or more of the various inventive aspects, in the foregoing description of exemplary embodiments of the present application, various features of the present application embodiments are sometimes grouped together into a single embodiment, figure, or description thereof. However, this approach to disclosure should not be construed as reflecting an intention that the claimed embodiments of the present application require more features than expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the present application.

[0156] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0157] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are meant to be within the scope of the embodiments of this application and form different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination.

[0158] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of this application. The embodiments of this application can also be implemented as device or apparatus programs (e.g., computer programs and computer program products) for performing part or all of the methods described herein. Such programs implementing the embodiments of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0159] It should be noted that the above embodiments are illustrative of the embodiments of this application and not limiting of the embodiments of this application, and those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. Embodiments of this application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

Claims

1. A seismic data overlay method based on a combination of least squares deconvolution and deep learning, characterized in that, include: Preliminary frequency-spreading processing of seismic data is performed using least squares deconvolution to obtain preliminary frequency-spreading seismic data; Using the earthquake data as sample data and the preliminary frequency-adjusted earthquake data as the target data corresponding to the sample data, a sample dataset is constructed. Using the aforementioned sample dataset, a pre-built convolutional neural network model is trained to obtain a seismic data overlay model; The preliminary frequency-spreading seismic data to be processed is input into the seismic data frequency-spreading model. The seismic data frequency-spreading model performs secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed, thereby obtaining the target seismic data.

2. The method according to claim 1, characterized in that, The preliminary frequency-upgrading processing of seismic data using least squares deconvolution to obtain preliminary frequency-upgraded seismic data further includes: Calculate the autocorrelation value of the seismic data in the seismic data and the cross-correlation value between the seismic data and the desired target data; wherein, the desired target data is the stretched portion of the seismic data; Based on the autocorrelation value and the cross-correlation value, solve for the least square deconvolution filter operator; Using the least squares deconvolution filter operator, calculate the first frequency domain data corresponding to the desired target data; Perform an inverse Fourier transform on the first frequency domain data to obtain the desired target data; Based on the earthquake data and the desired target data, the earthquake data after least square deconvolution filtering is calculated and used as the earthquake data in the preliminary frequency-spreading earthquake data.

3. The method according to claim 2, characterized in that, The calculation of the autocorrelation value of the seismic data in the seismic data and the cross-correlation value between the seismic data and the desired target data further includes: Perform a Fourier transform on the earthquake data to obtain second frequency domain data corresponding to the earthquake data; Using the real and imaginary parts of the second frequency domain data, calculate the third frequency domain data corresponding to the autocorrelation value of the seismic data; Perform an inverse Fourier transform on the third frequency domain data to obtain the autocorrelation value of the seismic data in the seismic data. Based on the autocorrelation value and the step size of the least square deconvolution, the cross-correlation value between the seismic data and the desired target data is estimated.

4. The method according to claim 3, characterized in that, The step of using the least squares deconvolution filter operator to calculate the first frequency domain data corresponding to the desired target data further includes: The least squares deconvolution filter operator is transformed into the frequency domain to obtain the fourth frequency domain data corresponding to the least squares deconvolution filter operator; The first frequency domain data is calculated using the fourth frequency domain data and the second frequency domain data.

5. The method according to any one of claims 1-4, characterized in that, The convolutional neural network model includes: an encoding structure, a decoding structure, a bottom structure, and a connection structure; The encoding structure is used to extract features from the input sample data; The decoding structure is used to reconstruct three-dimensional seismic data based on the features.

6. The method according to claim 5, characterized in that, The encoding structure includes: multiple convolutional layers, multiple max pooling layers, and multiple dropout layers; The decoding structure includes: multiple upsampling layers and multiple convolutional layers; The bottom structure includes: multiple convolutional layers; The connection structure includes: multiple connection layers; different connection layers are used to connect different upsampling layers in the decoding structure and different convolutional layers in the encoding structure.

7. A seismic data overlay device based on a combination of least squares deconvolution and deep learning, characterized in that, include: The preliminary frequency extension module is suitable for performing preliminary frequency extension processing on seismic data using least squares deconvolution to obtain preliminary frequency extension seismic data; The construction module is adapted to use the seismic data as sample data and the preliminary frequency-spreading seismic data as the target data corresponding to the sample data to construct a sample dataset; The training module is adapted to train a pre-built convolutional neural network model using the sample dataset to obtain a seismic data overlay model; The application module is adapted to input the preliminary frequency-spreading seismic data to be processed into the seismic data frequency-spreading model, and to perform secondary frequency-spreading processing on the preliminary frequency-spreading seismic data to be processed through the seismic data frequency-spreading model to obtain the target seismic data.

8. A computing device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning as described in any one of claims 1-6.

9. A computer storage medium storing at least one executable instruction that causes a processor to perform an operation corresponding to the seismic data overlay method based on the combination of least square deconvolution and deep learning as described in any one of claims 1-6.

10. A computer program product comprising at least one executable instruction that causes a processor to perform an operation corresponding to the seismic data overlay method based on the combination of least squares deconvolution and deep learning as described in any one of claims 1-6.

Images