Model construction for improving seismic data density, method for improving seismic data density

Abstract

The present invention discloses a model for improving seismic data density and a method for improving seismic data density. The method comprises: obtaining data from multiple seismic profiles of a three-dimensional seismic data volume of a target layer; respectively obtaining corresponding profile matrices, inserting several columns into each profile matrix, and setting the data in the matrix after the columns are inserted. The original data matrix, the to-be-encrypted data matrix, the random data matrix, and the identification matrix corresponding to each profile matrix are generated as training samples. Each training sample is input into a pre-built deep learning model for training, and a loss function of the deep learning model is calculated. Based on the result of the loss function, a first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained seismic data density improvement model; and the seismic data density improvement model outputs the seismic data matrix after the density is improved. The resolution of seismic data can be improved accurately and efficiently at low cost.

Description

Technical Field

[0001] The present invention relates to the field of oil and gas exploration, and in particular to a model construction for improving seismic data density and a method for improving seismic data density. Background Art

[0002] With the continuous advancement of oil and gas exploration, exploration targets are becoming increasingly complex. This, in particular, places higher demands on the resolution of seismic data for the identification and prediction of thin layers, small faults, and fractures. Improving the resolution of seismic data has always been a challenge for geophysicists. Currently, there are two approaches to improving seismic data resolution: increasing the frequency of seismic data and increasing its density. Methods for increasing the frequency of seismic data are relatively mature, including methods such as enhancing the effective signal through seismic data processing, expanding the effective frequency band of seismic data, and obtaining high-frequency components through frequency division. A number of commercial software programs have been developed and have achieved good results in practical applications. However, existing technologies for increasing the density of seismic data primarily focus on acquiring high-density seismic data. Specifically, increasing the spatial sampling rate to achieve alias-free sampling of the seismic wavefield can comprehensively and accurately record both the useful signal and noise, facilitating subsequent effective signal recovery and noise removal. However, due to its reliance on seismic data acquisition, these methods are often costly and time-consuming.

[0003] In recent years, with the rapid development of computer technology, artificial intelligence, represented by deep learning, has been widely applied to all aspects of oil and gas exploration and development. Statistical deep learning effectively increases the density and resolution of seismic data, which is of great significance to oil and gas exploration. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a method for constructing a model for improving seismic data density and improving seismic data density, which overcomes the above problems or at least partially solves the above problems.

[0005] In a first aspect, an embodiment of the present invention provides a model construction method for improving seismic data density, comprising:

[0006] Acquire a three-dimensional seismic data volume of a target layer segment, and extract data of multiple seismic sections from the three-dimensional seismic data volume; obtain a corresponding section matrix based on the data of each seismic section, wherein the columns in the section matrix represent the original seismic traces in the corresponding seismic section, and the data in each column represent the actual seismic data in the corresponding seismic section;

[0007] Inserting a number of columns into each of the profile matrices, and setting the data in the profile matrices after the insertion of the columns, respectively generating an original data matrix, a data matrix to be encrypted, a random data matrix, and an identification matrix corresponding to each profile matrix as training samples, wherein the inserted columns represent encrypted seismic traces, and the data on the columns represent encrypted seismic data;

[0008] Each of the training samples is input into a pre-built deep learning model for training. During the iterative process of training, the loss function of the deep learning model is calculated. According to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained seismic data density improvement model; the seismic data density improvement model outputs a seismic data matrix with improved density.

[0009] In one embodiment, a plurality of columns are inserted into the profile matrix, and data in the profile matrix after the plurality of columns are inserted is set to generate an original data matrix, a data matrix to be encrypted, a random data matrix, and an identification matrix corresponding to the profile matrix as training samples, including:

[0010] Inserting a preset number of columns between every two adjacent columns in the profile matrix to obtain an extended matrix, wherein the extended matrix includes encrypted columns and original columns, the encrypted columns are inserted columns, and the original columns are non-inserted columns;

[0011] Copying the extended matrix to obtain a first copied matrix, and setting the data on the encrypted columns in the first copied matrix to zero, while leaving the data on the original columns unchanged, to obtain a matrix of data to be encrypted;

[0012] Copying the extended matrix to obtain a second copied matrix, and setting the data on the encrypted columns in the second copied matrix to random numbers and the data on the original columns to 0 to obtain a random data matrix;

[0013] Copying the extended matrix to obtain a third copied matrix, and setting the data on the encrypted columns in the third copied matrix to 0 and the data on the original columns to 1 to obtain an identification matrix;

[0014] The data on the encrypted column in the matrix after the column insertion is set to be empty, and the data on the original column remains unchanged, to obtain the original data matrix;

[0015] The original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix are used as training samples corresponding to the profile matrix.

[0016] In one embodiment, the pre-built deep learning model includes a generator and a discriminator;

[0017] Input the training samples into a pre-built deep learning model for training, including:

[0018] Input the data matrix to be encrypted, the random data matrix and the identification matrix in the training sample into the generator for encryption to obtain the encrypted data matrix corresponding to the training sample;

[0019] The original data matrix, identification matrix and encrypted data matrix of the training sample are input into the discriminator to obtain the probability matrix output by the discriminator. The elements in the probability matrix represent the probability that the data corresponding to the element position in the encrypted data matrix is ​​real earthquake data.

[0020] In one embodiment, the generator includes an encoder and a decoder;

[0021] The step of inputting the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the generator for encryption to obtain the encrypted data matrix corresponding to the training sample includes:

[0022] Inputting the data matrix to be encrypted, the random data matrix and the identification matrix in the training sample into the encoder for feature extraction to obtain a first vector;

[0023] Using a semantic coding algorithm to reduce the dimension of the first vector to obtain a second vector;

[0024] The second vector is input into the long short-term memory recurrent neural network module of the decoder to obtain an encrypted data matrix corresponding to the training sample, where the encrypted data matrix has the same dimension as any matrix in the training sample.

[0025] In one embodiment, the encoder includes a convolutional module and a bidirectional long short-term memory recurrent neural network module;

[0026] The method of inputting the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the encoder for feature extraction to obtain the first vector includes:

[0027] The data matrix to be encrypted, the random data matrix and the identification matrix in the training sample are input into the convolution module to perform feature extraction and feature fusion to obtain a fused feature vector;

[0028] The fused feature vector is input into the bidirectional long short-term memory recurrent neural network module for feature extraction to obtain a first vector.

[0029] In one embodiment, the original data matrix of the training sample, the identification matrix, and the obtained encrypted data matrix are input into the discriminator to obtain the probability matrix output by the discriminator, including:

[0030] The original data matrix of the training sample and the obtained encrypted data matrix are input into the long short-term memory recurrent neural network module of the discriminator for feature extraction to obtain the third vector and the fourth vector respectively;

[0031] A probability matrix is ​​obtained according to the third vector and the fourth vector; the identification matrix is ​​used to indicate whether the data corresponding to each element position in the probability matrix is ​​real seismic data or encrypted seismic data.

[0032] In one embodiment, the loss function of the deep learning model includes a generator loss function and a discriminator loss function;

[0033] During the iterative training process, a loss function of the deep learning model is calculated, and according to a result of the loss function, a value of a first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained model for improving seismic data density, including:

[0034] During the training iteration process, the value of the discriminator loss function is calculated until the discriminator loss function reaches a preset convergence condition, and the discriminator is fixed; during the training iteration process after the discriminator is fixed, the value of the generator loss function is calculated until the generator loss function reaches a preset convergence condition, and then the iteration is stopped to obtain a trained seismic data density improvement model; or

[0035] During the training iteration process, the value of the generator loss function is calculated until the generator loss function reaches the preset convergence condition, and the generator is fixed; during the training iteration process after the generator is fixed, the value of the discriminator loss function is calculated until the discriminator loss function reaches the preset convergence condition, and then the iteration is stopped to obtain a trained model for improving seismic data density.

[0036] In one embodiment, the method further comprises:

[0037] A test sample is input into the improved seismic data density model completed in each iteration for testing, and the root mean square error value of the improved seismic data density model is calculated. According to the value of the root mean square error, the value of the second parameter of the improved seismic data density model is adjusted until the value of the root mean square error is less than or equal to a preset root mean square error threshold. The test sample is determined based on data of other seismic profiles of the three-dimensional seismic data volume.

[0038] In a second aspect, an embodiment of the present invention provides a method for increasing the density of seismic data, comprising:

[0039] Acquire data of a to-be-densified seismic section of a three-dimensional seismic data volume whose density is to be increased; and acquire a corresponding to-be-densified section matrix based on the data of the to-be-densified seismic section;

[0040] Inserting a plurality of columns into the section matrix to be encrypted to set the data in the matrix after the inserted columns, and generating a data matrix to be encrypted, a random data matrix and an identification matrix corresponding to the section matrix to be encrypted;

[0041] Inputting the data matrix to be encrypted, the random data matrix and the identification matrix into a seismic data density improvement model to obtain a seismic data matrix with improved density;

[0042] The model for improving seismic data density is obtained through the aforementioned model construction method for improving seismic data density.

[0043] In one embodiment, the method further comprises: converting the obtained density-enhanced seismic data matrix into a seismic profile.

[0044] In a third aspect, an embodiment of the present invention provides a model building device for improving seismic data density, comprising:

[0045] The first acquisition module is configured to acquire a three-dimensional seismic data volume of a target layer segment, extract data of multiple seismic sections from the three-dimensional seismic data volume, and acquire a corresponding section matrix based on the data of each seismic section, wherein the columns in the section matrix represent the original seismic traces in the corresponding seismic section, and the data in each column represent the actual seismic data in the corresponding seismic section;

[0046] A first generating module is configured to insert a plurality of columns into each of the profile matrices, and to set the data in the profile matrices after the plurality of columns are inserted, and to generate an original data matrix, a data matrix to be encrypted, a random data matrix, and an identification matrix corresponding to each profile matrix as training samples, wherein the inserted plurality of columns represent encrypted seismic traces, and the data on the plurality of columns represent encrypted seismic data;

[0047] A training module is used to input each of the training samples into a pre-built deep learning model for training. During the iterative process of training, the loss function of the deep learning model is calculated. According to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained seismic data density improvement model; the seismic data density improvement model outputs a seismic data matrix with improved density.

[0048] In a fourth aspect, an embodiment of the present invention provides a device for increasing seismic data density, comprising:

[0049] The second acquisition module is used to acquire data of a to-be-densified seismic section of the three-dimensional seismic data volume to be densified; and acquire a corresponding to-be-densified section matrix based on the data of the to-be-densified seismic section;

[0050] The second generating module is used to insert a plurality of columns of data in the matrix after the inserted columns into the section matrix to be encrypted, and generate a data matrix to be encrypted, a random data matrix and an identification matrix corresponding to the section matrix to be encrypted;

[0051] A density improvement module, configured to input the to-be-encrypted data matrix, the random data matrix, and the identification matrix into a seismic data density improvement model to obtain a seismic data matrix after density improvement;

[0052] The model for improving seismic data density is obtained through the aforementioned model construction method for improving seismic data density.

[0053] In a fifth aspect, an embodiment of the present invention provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the aforementioned model construction method for improving seismic data density or the aforementioned method for improving seismic data density is implemented.

[0054] In the sixth aspect, an embodiment of the present invention provides a terminal device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, it implements the aforementioned model construction method for improving the density of seismic data or the aforementioned method for improving the density of seismic data.

[0055] The beneficial effects of the above technical solutions provided by the embodiments of the present invention include at least:

[0056] The model construction method for improving the density of seismic data provided by the embodiment of the present invention extracts data of multiple seismic profiles from the three-dimensional seismic data volume of the target layer segment, obtains the profile matrix corresponding to each seismic profile according to the data of each seismic profile, each column in the profile matrix represents the original seismic trace in the corresponding seismic profile, and accordingly, the data in each column of the profile matrix represents the real seismic data in the corresponding seismic profile, inserts several columns into each profile matrix, and sets the data in the profile matrix after inserting several columns, the inserted several columns represent the encrypted seismic trace, and the data on the inserted several columns represent the encrypted seismic data, and obtains the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to each profile matrix, and the corresponding data matrix of each profile matrix is ​​converted into the encrypted seismic trace matrix. The original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix are used as training samples; each training sample is input into a pre-built deep learning model for training, and in the iterative process of training, the loss function of the deep learning model is calculated, and according to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches the preset convergence condition, thereby obtaining a trained model for improving the density of seismic data; based on the deep learning model, the internal relationship and overall characteristics between the seismic data are found, and a nonlinear model of the real seismic data and the encryption target is established, and then the seismic data is reconstructed. The density of the reconstructed seismic data is greater than that of the original seismic data, and the seismic data is encrypted efficiently and accurately, thereby improving the resolution of the seismic data.

[0057] Furthermore, the encoder component of the generator in the seismic data density model was improved, employing a convolutional structure to fully extract the spatial characteristics of the seismic data, ensuring that the lateral data features are fully utilized. Furthermore, a bidirectional long-short-term memory recurrent neural network structure was combined to extract the temporal characteristics of the seismic data, ensuring that the longitudinal and global data features are fully utilized, thereby improving the accuracy of the reconstructed seismic data.

[0058] 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 purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0059] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] 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:

[0061] Figure 1This is a flow chart of a model building method for improving seismic data density in an embodiment of the present invention;

[0062] Figure 2 Flowchart of the training method of the model in the embodiment of the present invention;

[0063] Figure 3 Flowchart of a method for generating an encrypted data matrix in an embodiment of the present invention;

[0064] Figure 4 is a flowchart of a method for obtaining a first vector in an embodiment of the present invention;

[0065] Figure 5 A flow chart of a method for increasing seismic data density according to an embodiment of the present invention;

[0066] Figure 6 Schematic diagram of the structure of a model building device for improving seismic data density in an embodiment of the present invention;

[0067] Figure 7 Schematic diagram of the structure of a device for improving seismic data density according to an embodiment of the present invention;

[0068] Figure 8 Schematic diagram of the model structure for improving seismic data density in an embodiment of the present invention

[0069] Figure 9 This is a comparison chart of the relationship between deep learning parameters and RMSE in an embodiment of the present invention;

[0070] Figure 10 This is a schematic diagram of an original seismic section before density enhancement in an embodiment of the present invention;

[0071] Figure 11 Schematic diagram of the original seismic profile after density improvement in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] 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.

[0073] By using deep learning of artificial intelligence, we can find the internal relationships and overall characteristics between seismic data, establish a nonlinear model of data and targets, and then reconstruct the seismic data. The reconstructed seismic data is denser than the original seismic data, which can effectively encrypt seismic data and improve the resolution of seismic data, which is of great significance for oil and gas exploration.

[0074] In order to solve the problems existing in the prior art, embodiments of the present invention provide a method for constructing a model for improving seismic data density and improving seismic data density.

[0075] In order to facilitate the description of a model construction for improving seismic data density and a method for improving seismic data density provided by an embodiment of the present invention, the process of improving density is also referred to as an encryption process in this embodiment.

[0076] Example

[0077] The first embodiment of the present invention provides a model construction method for improving the density of seismic data, the process of which is as follows: Figure 1 As shown, the following steps are included:

[0078] Step S1: Acquire a 3D seismic data volume of the target layer segment, and extract data of multiple seismic sections from the 3D seismic data volume; obtain a corresponding section matrix based on the data of each seismic section, wherein the columns in the section matrix represent the original seismic traces in the corresponding seismic section, and the data in each column represent the actual seismic data in the corresponding seismic section;

[0079] Step S2: inserting several columns into each profile matrix and setting the data in the profile matrix after inserting several columns, respectively generating the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to each profile matrix as training samples, the inserted several columns characterizing the encrypted seismic traces, and the data on the several columns representing the encrypted seismic data;

[0080] Step S3: Input each training sample into a pre-built deep learning model for training. During the iterative process of training, calculate the loss function of the deep learning model. According to the result of the loss function, adjust the first parameter of the deep learning model until the loss function reaches the preset convergence condition, and obtain the trained seismic data density improvement model; the seismic data density improvement model outputs the seismic data matrix with increased density.

[0081] Generally, there are singular values ​​such as maximum and minimum in the data of seismic profiles. The existence of singular values ​​will cause the data value range to span a large range, making it difficult for deep learning algorithm models to converge quickly. Therefore, it is necessary to standardize the data of each seismic profile to eliminate singular values ​​such as maximum and minimum, so that the seismic data density model can converge quickly.

[0082] In some optional embodiments, the seismic traces in the seismic profile data are used as the original seismic traces. In the above step S1, the seismic profile is converted into a corresponding matrix according to the original seismic traces in the seismic profile and the sampling interval time in the seismic profile. Referring to Table 1 below, the profile matrix is ​​obtained. In Table 1 below, x 11 represents the amplitude of the first sampling point of the first original seismic trace, xtn represents the amplitude of the t-th sampling point of the n-th original seismic trace,

[0083] Table 1:

[0084]

[0085] In some optional embodiments, the above step S2 inserts several columns into the profile matrix, and sets the data in the matrix after the columns are inserted, to generate the original data matrix, the data matrix to be encrypted, the random data matrix, and the identification matrix corresponding to the profile matrix as training samples. Specifically, this can be achieved in the following manner:

[0086] (1) inserting a preset number of columns between every two adjacent columns in the profile matrix to obtain an extended matrix, wherein the extended matrix includes encrypted columns and original columns, wherein the encrypted columns are inserted columns and the original columns are non-inserted columns;

[0087] Specifically, a column is inserted between every two adjacent columns in the profile matrix to obtain an extended matrix. Referring to Table 2 below, there are n columns of data in Table 2, representing n seismic traces respectively. The first and third columns correspond to the first column (x 11 The column where it is located) and the second column (x 12 The column where the original column is located is called the original column, which represents the original seismic trace. The second and fourth columns in Table 2, that is, the inserted columns, are called the encrypted columns, which represent the encrypted seismic trace.

[0088] Table 2:

[0089]

[0090] (2) Copy the expanded matrix to obtain a first copy matrix, and set the data on the encrypted columns in the first copy matrix to zero, while leaving the data on the original columns unchanged, to obtain the data matrix to be encrypted;

[0091] Specifically, the extended matrix shown in Table 2 is copied to obtain the first copied matrix. The data on the encrypted column in the first copied matrix is ​​set to 0, and the data on the original column remains unchanged to obtain the matrix of data to be encrypted, which is recorded as Refer to Table 3 below:

[0092] Table 3:

[0093]

[0094] (3) Copy the expanded matrix to obtain a second copied matrix, and set the data on the encrypted columns in the second copied matrix to random numbers and the data on the original columns to 0 to obtain a random data matrix;

[0095] Specifically, the extended matrix shown in Table 2 is copied to obtain a second copied matrix. The data on the encrypted columns in the second copied matrix are supplemented with random numbers z, and the data on the original columns are set to 0 to obtain a random data matrix, denoted as Z, as shown in Table 4 below:

[0096] Table 4:

[0097]

[0098]

[0099] (4) Copy the expanded matrix to obtain a third copied matrix, and set the data on the encrypted columns in the third copied matrix to 0 and the data on the original columns to 1 to obtain an identification matrix;

[0100] Specifically, the extended matrix shown in Table 2 is copied to obtain a third copied matrix. The data in the encrypted columns of the third copied matrix are set to 0, and the data in the original columns are set to 1 to obtain an identification matrix, denoted as M. Referring to Table 5 below, the identification matrix is ​​used to record the location of missing data. 1 indicates that the data at that location exists, and 0 indicates that the data at that location is missing:

[0101] Table 5:

[0102]

[0103] (5) Set the data on the encrypted column in the matrix after the insertion column to be empty, and the data on the original column remains unchanged, and obtain the original data matrix, which is recorded as X. Specifically, refer to Table 6,

[0104] Table 6:

[0105]

[0106] Where n is the number of seismic traces, t is the sampling time, then x tn is the sampling point data of the nth channel when the sampling time is t, where N represents the total number of traces in a profile, and T represents the total number of sampling points in each seismic trace.

[0107] (6) The original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix are used as training samples corresponding to the profile matrix.

[0108] In some optional embodiments, the pre-built deep learning model includes a generator and a discriminator. In step S3, the training sample is input into the pre-built deep learning model for training. Specifically, Figure 2 As shown, this can be achieved by:

[0109] Step S21: Input the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the generator for encryption to obtain the encrypted data matrix corresponding to the training sample;

[0110] Step S22: The original data matrix, identification matrix and encrypted data matrix of the training sample are input into the discriminator to obtain the probability matrix output by the discriminator. The elements in the probability matrix represent the probability that the data corresponding to the element position in the encrypted data matrix is ​​the real earthquake data.

[0111] In the above step S21, the data matrix to be encrypted The random data matrix Z and the identification matrix M are input to the generator, and the output result is the earthquake estimation matrix, that is, the encrypted data matrix, denoted as X′, as shown in Table 7 below:

[0112] Table 7:

[0113]

[0114] In some optional embodiments, the generator includes an encoder and a decoder, and the above step S21, referring to Figure 3 As shown, this can be achieved by:

[0115] Step S31: inputting the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the encoder for feature extraction to obtain a first vector;

[0116] Step S32: using a semantic coding algorithm to reduce the dimension of the first vector to obtain a second vector;

[0117] Step S33: Input the second vector into the long short-term memory recurrent neural network module of the decoder to obtain an encrypted data matrix corresponding to the training sample. The encrypted data matrix has the same dimension as any matrix in the training sample.

[0118] The encoder and decoder are connected by a low-dimensional semantic coding vector. Specifically, the output data of the generator (first vector) is mapped into a low-dimensional semantic coding vector (second vector) through the semantic encoder, and the task of the decoder is to reconstruct the low-dimensional semantic coding vector into complete seismic data.

[0119] In some optional embodiments, the encoder includes a convolution module and a bidirectional long short-term memory recurrent neural network module. In the above step S31, specifically, refer to Figure 4 As shown, this can be achieved by:

[0120] Step S41: Input the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the convolution module to perform feature extraction and feature fusion to obtain a fused feature vector;

[0121] The data matrix to be encrypted, the random data matrix and the identification matrix are input into the convolution module. The embodiment of the present invention adopts a multi-scale convolution structure. In the multi-scale convolution, the embodiment of the present invention adopts a three-channel mode: the first channel has only one convolution layer, the convolution kernel size is 1×1, the padding is 0, and the step size is 1; the second channel contains two convolution layers, the first convolution kernel size is 1×1, the padding is 1, and the step size is 1; the second convolution kernel size is 3×3, the padding is 1, the step size is 1, and the activation function is the ReLU function; the third channel contains three convolution layers, the first convolution kernel size is 1×1, the padding is 1, and the step size is 1, the second convolution kernel size is 3×3, the padding is 1, the step size is 1, and the third convolution kernel size is 5×5, and two 3×3 convolutions are used instead. The features extracted by the three channels are fused with the input layer after a 1×1 convolution. The spatial features of the seismic data are fully extracted.

[0122] Specifically, feature extraction is achieved through the following formula:

[0123]

[0124] In formula (1), conv represents the convolution operation.

[0125] The features extracted by convolution at different scales are fused using the following formula:

[0126]

[0127] Step S42: Input the fused feature vector into the bidirectional long short-term memory recurrent neural network module for feature extraction to obtain a first vector.

[0128] Specifically, the fused features are input into BiLSTM, and the first vector is obtained using the following formula:

[0129]

[0130] In some optional embodiments, the decoder is composed of 64 LSTMs to restore the data dimension, specifically, using the following formula:

[0131]

[0132] In formula (4), C is the encoder operation, and X′ is the encrypted seismic data and identification matrix output by the encoder G.

[0133] In some optional embodiments, the above step S22 can be implemented in the following manner:

[0134] (1) Inputting the original data matrix of the training sample and the obtained encrypted data matrix into the long short-term memory recurrent neural network module of the discriminator for feature extraction, and obtaining the third vector and the fourth vector respectively;

[0135] (2) A probability matrix is ​​obtained based on the third vector and the fourth vector; the identification matrix is ​​used to characterize whether the data corresponding to each element position in the probability matrix is ​​real earthquake data or encrypted earthquake data.

[0136] The discriminator consists of a long short-term memory network (LSTM) module and a fully connected layer (FC) module. The LSTM module extracts the eigenvectors of the encrypted seismic data X' and the original seismic data X, and the identification matrix M locates the position. The fully connected layer FC reduces the eigenvector from high dimension to one dimension, and then calculates the probability of each encrypted data element being true or false through the activation function (Sigomid function). Finally, the obtained probability data is located by the identification matrix to form a probability matrix

[0137] In some optional embodiments, the loss function of the deep learning model includes a generator loss function and a discriminator loss function. The above step S3 can be implemented as follows:

[0138] During the training iteration process, the value of the discriminator loss function is calculated until the discriminator loss function reaches a preset convergence condition, and the discriminator is fixed; during the training iteration process after the discriminator is fixed, the value of the generator loss function is calculated until the generator loss function reaches a preset convergence condition, and then the iteration is stopped to obtain a trained seismic data density improvement model; or

[0139] During the training iteration process, the value of the generator loss function is calculated until the generator loss function reaches the preset convergence condition, and the generator is fixed; during the training iteration process after the generator is fixed, the value of the discriminator loss function is calculated until the discriminator loss function reaches the preset convergence condition, and then the iteration is stopped to obtain the trained seismic data density improvement model.

[0140] The error between the encrypted data matrix generated by the generator and the original data matrix is ​​used as the discriminant error to construct the discriminator loss function; the error between the data matrix to be encrypted and the encrypted data matrix is ​​used as the reconstruction error to construct the generator loss function. The generator loss function and the discriminator loss function are combined to form the loss function of the seismic data density model through the following formula:

[0141] MGB LOSS =-(1-λ)D(X+Z)+λ‖M⊙(G(X+Z)-X)‖2 (5)

[0142] In formula (5), λ represents the number of generator loss functions, D represents the discriminator operation, G represents the generator operation, M represents the identification matrix, X represents the original data matrix, Z represents the random matrix, and ‖…‖ represents the l2 norm.

[0143] The generator generates an encrypted data matrix for training the discriminator, and updates the parameters of the discriminator until the discriminator loss function reaches the preset convergence point. The discriminator is fixed, and the parameters of the generator are updated according to the output of the discriminator. The iteration is repeated until the loss function expressed by formula (5) reaches the convergence condition. The iteration is stopped to obtain a trained seismic data density improvement model.

[0144] In some optional embodiments, a test sample is input into the improved seismic data density model completed in each iteration for testing, and the root mean square error value of the improved seismic data density model is calculated. According to the value of the root mean square error, the value of the second parameter of the improved seismic data density model is adjusted until the value of the root mean square error is less than or equal to a preset root mean square error threshold. The test sample is determined based on data of other seismic profiles of the three-dimensional seismic data body.

[0145] In order to make the seismic data density improvement model converge faster, the seismic data density improvement model can be trained once per iteration and tested with test samples. Specifically, the test samples are generated by generating training samples. The generation method has been described above and will not be repeated here in the embodiment of the present invention. The test samples are input into the seismic data density improvement model, and the root mean square error of the model is calculated. The root mean square error is used as an evaluation indicator, as shown in the following formula:

[0146]

[0147] In formula (6), y t The real earthquake data of the sampling point at time t in the encrypted earthquake data; y t ′ represents the seismic data generated by the sampling point at time t in the encrypted seismic data,

[0148] If the loss function of the seismic data density model has converged and the root mean square error value is greater than the preset root mean square error threshold, the value of the second parameter in the seismic data density model is adjusted, for example, the second parameter is the learning rate, the number of hidden layer nodes, the parameter λ, the number of iterations and other parameters, until the loss function of the seismic data density model reaches the preset convergence condition and the root mean square error value is less than the preset root mean square error threshold.

[0149] Based on the aforementioned model building method for improving seismic data density, the embodiment of the present invention further provides a method for improving seismic data density, the process of which is as follows: Figure 5 As shown, the following steps are included:

[0150] Step S61: obtaining data of a to-be-densified seismic section of a three-dimensional seismic data volume to be densified; obtaining a corresponding to-be-densified section matrix based on the data of the to-be-densified seismic section;

[0151] Step S62: inserting a number of columns into the section matrix to be encrypted and setting the data in the matrix after the inserted columns, generating a data matrix to be encrypted, a random data matrix and an identification matrix corresponding to the section matrix to be encrypted;

[0152] Step S63: inputting the data matrix to be encrypted, the random data matrix and the identification matrix into the seismic data density improvement model to obtain the seismic data matrix after density improvement;

[0153] The model for improving the density of seismic data is obtained through the aforementioned model construction method for improving the density of seismic data.

[0154] In some optional embodiments, after obtaining the density-enhanced seismic data matrix in step S13, the density-enhanced seismic data matrix can be converted into a seismic profile.

[0155] Based on the same inventive concept, the embodiment of the present invention further provides a model building device for improving the density of seismic data. The structure of the device is as follows: Figure 6 As shown, including:

[0156] The first acquisition module 71 is used to acquire a 3D seismic data volume of the target layer segment and extract data of multiple seismic sections from the 3D seismic data volume; based on the data of each seismic section, a corresponding section matrix is ​​obtained, where the columns in the section matrix represent the original seismic traces and the data in each column represent the actual seismic data;

[0157] The first generation module 72 is used to insert a number of columns into each profile matrix and set the data in the matrix after the columns are inserted, and respectively generate the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to each profile matrix as training samples, wherein the inserted columns represent the encrypted seismic traces, and the data on the columns represent the encrypted seismic data;

[0158] The training module 73 is used to input each training sample into a pre-built deep learning model for training. During the iterative process of training, the loss function of the deep learning model is calculated. According to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches the preset convergence condition, thereby obtaining a trained seismic data density improvement model; the seismic data density improvement model outputs the seismic data matrix with increased density.

[0159] Regarding the device for increasing the density of seismic data in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method and will not be elaborated on here.

[0160] Based on the same inventive concept, the embodiment of the present invention further provides a device for improving the density of seismic data. The structure of the device is as follows: Figure 7 As shown, including:

[0161] The second acquisition module 81 is used to acquire data of the seismic section to be encrypted of the three-dimensional seismic data volume to be increased in density; and acquire the corresponding section matrix to be encrypted based on the data of the seismic section to be encrypted;

[0162] The second generating module 82 is used to insert a number of columns into the section matrix to be encrypted and set the data in the matrix after the inserted columns, to generate the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to the section matrix to be encrypted;

[0163] The density improvement module 83 is used to input the data matrix to be encrypted, the random data matrix and the identification matrix into the seismic data density improvement model to obtain the seismic data matrix after density improvement;

[0164] The model for improving the density of seismic data is obtained through the aforementioned model construction method for improving the density of seismic data.

[0165] Regarding the device for increasing the density of seismic data in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method and will not be elaborated on here.

[0166] The above-mentioned model construction method and method for improving seismic data density are described using a specific example. Taking the Cambrian platform margin belt in the Gucheng area of ​​the Tarim Basin in northwestern China as an example, the specific steps are as follows:

[0167] 1. Data preprocessing

[0168] The 3D seismic data volume of the target layer in the study area was preprocessed. Seismic data from 1,000 sections were extracted with equal spacing (1×1) between the main and tie lines. 800 of these sections were used as training data and 200 as test data. Each seismic section contained 100 seismic traces, each containing 500 data points, with a sampling interval of 4 ms. Each seismic section in the 800 training data sets was then converted into an original data matrix, a matrix of data to be encrypted, a random matrix, and an identification matrix.

[0169] 2. Build the model generator

[0170] The model's generator consists of an encoder, a decoder, and a low-dimensional semantic encoder. The encoder consists of multi-scale convolutions and BiLSTMs. The encoder and decoder are connected by a low-dimensional semantic encoding vector. The encoder maps the generator input to a low-dimensional semantic encoding vector, while the decoder reconstructs the vector into complete seismic data. The decoder, consisting of 64 LSTMs, restores the data dimension and generates the input for the discriminator.

[0171] 3. Build the discriminator of the model

[0172] The discriminator of the model is mainly composed of LSTM modules and FC modules. Among them, the LSTM module consists of 64 LSTM units, which is responsible for processing the generated seismic data and obtaining the historical memory vector from the encrypted data matrix and the original data matrix. The fully connected layer maps the historical memory vector into a probability output, that is, using the Sigmoid function to calculate the probability that the input data is true. The training goal of the discriminator is to identify real samples as true and generated samples as false as much as possible. The generator and discriminator form a model to improve the density of seismic data. Figure 8 As shown;

[0173] 4. Input the training samples into the generator, which generates encrypted seismic data for training the discriminator and updating its parameters. The discriminator is fixed, and the generator parameters are updated based on its results. Repeat this process until the MC-GAN-BiLSTM model reaches the optimal value.

[0174] 5. Use the test samples to test the seismic data density improvement model. By continuously adjusting the learning rate, number of hidden layer nodes, parameter λ, number of iterations and other parameters in deep learning, and monitoring the changes in the evaluation index RMSE in real time, until the RMSE value is less than or equal to the preset root mean square error threshold, the trained seismic data density improvement model is obtained. Figure 9 As shown, Figure 9 (a) is the learning rate curve; (b) is the hidden layer node number curve; (c) is the parameter λ curve; and (d) is the number of iterations. When the learning rate is 0.001 and the number of hidden layer nodes is 16, the RMSE is the lowest and the model performance is the best. Therefore, the number of hidden layer nodes and the learning rate of the BiLSTM layer are set to 16 and 0.001, respectively, and then the model's λ parameter is adjusted. When the parameter λ is close to 1 or 0, the encryption effect is poor. This means that the two parts of the generator loss function are not performing as expected. When the parameter λ is set to 0.3, the RMSE is the lowest and the performance is the best. When the number of iterations is 1000, the RMSE is basically stable. Therefore, the maximum number of iterations is set to 1000. The learning rate, number of hidden layer nodes, parameter λ, and number of iterations are set to 0.001, 16, 0.3, and 1000, respectively.

[0175] 6. Use the trained seismic data density improvement model to encrypt the seismic data to be encrypted, such as Figure 10 As shown in the figure, it is the original seismic section before encryption. Figure 11 As shown in the figure, the seismic profile after density improvement, the original seismic data trace spacing is 25 meters, and the trace spacing after doubling the seismic data density is equivalent to 12.5 meters. The improved seismic profile has a clearer structure of the Cambrian platform margin mound and shoal body progradation in the target area, a higher resolution, and can better reflect the distribution of multiple periods of platform margin mound and shoal.

[0176] Due to the development of carbonate platform margin mound-shoal deposits in the Cambrian in the Gucheng area, multiple wells have revealed the development of multi-stage mound-shoal bodies in the platform margin. Through well-seismic calibration, the multi-stage mound-shoal bodies appear as superimposed reflection features of progradational mounds on the seismic profile. However, due to the large burial depth of the target area and the low main frequency of the seismic data, the low-resolution seismic data does not clearly show the progradational reflection features of the mound-shoal bodies. How to identify the multi-stage mound-shoal bodies in the platform margin through seismic data is a difficult problem in the target area. However, the conventional interpretation of seismic data such as frequency extension or frequency division to increase the frequency of seismic data to identify mound-shoal bodies often creates many false impressions and is not effective. The present invention improves the density of seismic data and thus the resolution of seismic data to identify Cambrian mound-shoal bodies in the Gucheng area, achieving relatively good results.

[0177] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the aforementioned model construction method for improving seismic data density or the aforementioned method for improving seismic data density is implemented.

[0178] Based on the same inventive concept, an embodiment of the present invention also provides a terminal 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 aforementioned model construction method for improving seismic data density or the aforementioned method for improving seismic data density.

[0179] Unless otherwise specifically stated, terms such as process, calculate, compute, determine, display, and the like may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, that manipulate and convert data represented as physical (e.g., electronic) quantities within registers or memories of a processing system into other data similarly represented as physical quantities within the memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals may be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0180] It should be understood that the specific order or hierarchy of steps in the disclosed processes is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The accompanying method claims present elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0181] In the foregoing detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that embodiments of the claimed subject matter require more features than are expressly recited in each claim. On the contrary, as reflected in the appended claims, the invention comprises less than all the features of any individual disclosed embodiment. The appended claims are therefore hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0182] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein may be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described around their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person may implement the described functions in an adaptable manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of this disclosure.

[0183] The steps of the methods or algorithms described in conjunction with the embodiments herein may be directly embodied as hardware, software modules executed by a processor, or a combination thereof. The software module may be located in a RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and storage medium may also be present in a user terminal as discrete components.

[0184] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or external to the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well known in the art.

[0185] The foregoing description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purposes of describing the above embodiments, but one of ordinary skill in the art will recognize that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to encompass all such changes, modifications and variations that fall within the scope of the appended claims. Furthermore, to the extent the term "comprising" is used in the specification or claims, the term is intended to be encompassed in a manner similar to the term "including," as explained in terms of "including," used as a transitional word in the claims. Furthermore, any use of the term "or" in the specification of the claims is intended to mean a "non-exclusive or."

Claims

1. A model construction method for improving seismic data density, characterized in that: include: Acquire a three-dimensional seismic data volume of a target layer segment, and extract data of multiple seismic sections from the three-dimensional seismic data volume; According to the data of each seismic profile, a corresponding profile matrix is ​​obtained, wherein the columns in the profile matrix represent the original seismic traces in the corresponding seismic profile, and the data in each column represent the real seismic data in the corresponding seismic profile; Inserting several columns into each of the profile matrices, and setting the data in the profile matrix after inserting the several columns, respectively generating the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to each profile matrix as training samples, the inserted several columns characterizing the encrypted seismic traces, and the data on the several columns characterizing the encrypted seismic data; inserting several columns into the profile matrix, and setting the data in the profile matrix after inserting the several columns, generating the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to the profile matrix as training samples, including: inserting a preset number of columns between every two adjacent columns in the profile matrix to obtain an extended matrix, the extended matrix including encrypted columns and original columns, the encrypted columns being inserted columns, and the original columns being non-inserted columns; copying The expanded matrix is ​​used to obtain a first copy matrix, and the data on the encrypted columns in the first copy matrix are set to zero, while the data on the original columns remain unchanged, to obtain a data matrix to be encrypted; the expanded matrix is ​​copied to obtain a second copy matrix, and the data on the encrypted columns in the second copy matrix are set to random numbers, while the data on the original columns are set to 0, to obtain a random data matrix; the expanded matrix is ​​copied to obtain a third copy matrix, and the data on the encrypted columns in the third copy matrix are set to 0, while the data on the original columns are set to 1, to obtain an identification matrix; the data on the encrypted columns in the matrix after the inserted columns are set to empty, while the data on the original columns remain unchanged, to obtain an original data matrix; the original data matrix, the data matrix to be encrypted, the random data matrix, and the identification matrix are used as training samples corresponding to the profile matrix; Each of the training samples is input into a pre-built deep learning model for training. During the iterative process of training, the loss function of the deep learning model is calculated. According to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained seismic data density improvement model; the seismic data density improvement model outputs a seismic data matrix with improved density.

2. The method according to claim 1, wherein The pre-built deep learning model includes a generator and a discriminator; Input the training samples into a pre-built deep learning model for training, including: Input the data matrix to be encrypted, the random data matrix and the identification matrix in the training sample into the generator for encryption to obtain the encrypted data matrix corresponding to the training sample; The original data matrix, identification matrix and encrypted data matrix of the training sample are input into the discriminator to obtain the probability matrix output by the discriminator. The elements in the probability matrix represent the probability that the data corresponding to the element position in the encrypted data matrix is ​​real earthquake data.

3. The method according to claim 2, wherein The generator includes an encoder and a decoder; The step of inputting the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the generator for encryption to obtain the encrypted data matrix corresponding to the training sample includes: Inputting the data matrix to be encrypted, the random data matrix and the identification matrix in the training sample into the encoder for feature extraction to obtain a first vector; Using a semantic coding algorithm to reduce the dimension of the first vector to obtain a second vector; The second vector is input into the long short-term memory recurrent neural network module of the decoder to obtain an encrypted data matrix corresponding to the training sample, where the encrypted data matrix has the same dimension as any matrix in the training sample.

4. The method according to claim 3, wherein The encoder includes a convolution module and a bidirectional long short-term memory recurrent neural network module; The method of inputting the data matrix to be encrypted, the random data matrix, and the identification matrix in the training sample into the encoder for feature extraction to obtain the first vector includes: The data matrix to be encrypted, the random data matrix and the identification matrix in the training sample are input into the convolution module to perform feature extraction and feature fusion to obtain a fused feature vector; The fused feature vector is input into the bidirectional long short-term memory recurrent neural network module for feature extraction to obtain a first vector.

5. The method according to claim 2, wherein The original data matrix, the identification matrix and the obtained encrypted data matrix of the training sample are input into the discriminator to obtain the probability matrix output by the discriminator, including: The original data matrix of the training sample and the obtained encrypted data matrix are input into the long short-term memory recurrent neural network module of the discriminator for feature extraction to obtain the third vector and the fourth vector respectively; A probability matrix is ​​obtained according to the third vector and the fourth vector; the identification matrix is ​​used to indicate whether the data corresponding to each element position in the probability matrix is ​​real seismic data or encrypted seismic data.

6. The method according to claim 2, wherein The loss function of the deep learning model includes a generator loss function and a discriminator loss function; During the iterative training process, a loss function of the deep learning model is calculated, and according to a result of the loss function, a value of a first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained model for improving seismic data density, including: During the training iteration process, the value of the discriminator loss function is calculated until the discriminator loss function reaches a preset convergence condition, and the discriminator is fixed; during the training iteration process after the discriminator is fixed, the value of the generator loss function is calculated until the generator loss function reaches a preset convergence condition, and then the iteration is stopped to obtain a trained seismic data density improvement model; or During the training iteration process, the value of the generator loss function is calculated until the generator loss function reaches the preset convergence condition, and the generator is fixed; during the training iteration process after the generator is fixed, the value of the discriminator loss function is calculated until the discriminator loss function reaches the preset convergence condition, and then the iteration is stopped to obtain a trained model for improving seismic data density.

7. The method according to claim 1, wherein The method further comprises: A test sample is input into the improved seismic data density model completed in each iteration for testing, and the root mean square error value of the improved seismic data density model is calculated. According to the value of the root mean square error, the value of the second parameter of the improved seismic data density model is adjusted until the value of the root mean square error is less than or equal to a preset root mean square error threshold. The test sample is determined based on data of other seismic profiles of the three-dimensional seismic data volume.

8. A method for increasing seismic data density, characterized in that: include: Acquire data of a to-be-densified seismic section of a three-dimensional seismic data volume whose density is to be increased; and acquire a corresponding to-be-densified section matrix based on the data of the to-be-densified seismic section; Inserting a plurality of columns into the section matrix to be encrypted to set the data in the matrix after the inserted columns, and generating a data matrix to be encrypted, a random data matrix and an identification matrix corresponding to the section matrix to be encrypted; Inputting the data matrix to be encrypted, the random data matrix and the identification matrix into a seismic data density improvement model to obtain a seismic data matrix with improved density; The model for improving seismic data density is obtained by the model construction method for improving seismic data density according to any one of claims 1 to 7.

9. The method according to claim 8, wherein The method further comprises: The obtained density-enhanced seismic data matrix is ​​converted into a seismic profile.

10. A model building device for improving seismic data density, characterized in that: include: A first acquisition module is used to acquire a three-dimensional seismic data volume of a target layer segment and extract data of multiple seismic sections from the three-dimensional seismic data volume; According to the data of each seismic profile, a corresponding profile matrix is ​​obtained, wherein the columns in the profile matrix represent the original seismic traces in the corresponding seismic profile, and the data in each column represent the real seismic data in the corresponding seismic profile; The first generating module is used to insert several columns into each of the profile matrices, and set the data in the profile matrix after the several columns are inserted, and respectively generate the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to each profile matrix as training samples, the inserted several columns characterize the encrypted seismic traces, and the data on the several columns characterize the encrypted seismic data; insert several columns into the profile matrix, and set the data in the profile matrix after the several columns are inserted, and generate the original data matrix, the data matrix to be encrypted, the random data matrix and the identification matrix corresponding to the profile matrix as training samples, including: inserting a preset number of columns between every two adjacent columns in the profile matrix to obtain an extended matrix, the extended matrix including encrypted columns and original columns, the encrypted columns are inserted columns, and the original columns are non-inserted columns ; copy the extended matrix to obtain a first copied matrix, and set the data on the encrypted columns in the first copied matrix to zero, while the data on the original columns remain unchanged, to obtain a data matrix to be encrypted; copy the extended matrix to obtain a second copied matrix, and set the data on the encrypted columns in the second copied matrix to random numbers, while the data on the original columns are set to 0, to obtain a random data matrix; copy the extended matrix to obtain a third copied matrix, and set the data on the encrypted columns in the third copied matrix to 0, while the data on the original columns are set to 1, to obtain an identification matrix; set the data on the encrypted columns in the matrix after the inserted columns to empty, while the data on the original columns remain unchanged, to obtain an original data matrix; use the original data matrix, the data matrix to be encrypted, the random data matrix, and the identification matrix as training samples corresponding to the profile matrix; A training module is used to input each of the training samples into a pre-built deep learning model for training. During the iterative process of training, the loss function of the deep learning model is calculated. According to the result of the loss function, the first parameter of the deep learning model is adjusted until the loss function reaches a preset convergence condition, thereby obtaining a trained seismic data density improvement model; the seismic data density improvement model outputs a seismic data matrix with improved density.

11. A device for increasing seismic data density, characterized in that: include: The second acquisition module is used to acquire data of the seismic profile to be densified of the three-dimensional seismic data volume to be densified; According to the data of the seismic section to be encrypted, a corresponding section matrix to be encrypted is obtained; The second generating module is used to insert a plurality of columns of data in the matrix after the inserted columns into the section matrix to be encrypted, and generate a data matrix to be encrypted, a random data matrix and an identification matrix corresponding to the section matrix to be encrypted; A density improvement module, configured to input the to-be-encrypted data matrix, the random data matrix, and the identification matrix into a seismic data density improvement model to obtain a seismic data matrix after density improvement; The model for improving seismic data density is obtained by the model construction method for improving seismic data density according to any one of claims 1 to 7.

12. A computer storage medium, characterized in that The computer storage medium stores computer executable instructions, which, when executed by a processor, implement the model construction method for improving seismic data density according to any one of claims 1 to 7 or the method for improving seismic data density according to claim 8.

13. A terminal device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the model construction method for improving the density of seismic data as described in any one of claims 1 to 7 or the method for improving the density of seismic data as described in claim 8 is implemented.

Images