阅读说明：本技术 地震数据的重建方法、装置、电子设备及可读存储介质 (Seismic data reconstruction method and device, electronic equipment and readable storage medium ) 是由 唐欢欢 毛伟建 于 2020-11-06 设计创作，主要内容包括：本申请提供了地震数据的重建方法、装置、电子设备及可读存储介质,所述重建方法包括：基于获取的待重建地震数据的地域范围的地质属性,构建深度学习网络；基于构建好的深度学习网络的网络结构,以及深度学习网络中的多个初始参数,构建归一化误差函数；基于获取到的多个样本地震数据和所述归一化误差函数,对所述深度学习网络进行训练,得到训练好的地震数据重建模型；将所述待重建地震数据输入至所述地震数据重建模型,得到重建后的地震数据。这样,通过构建地震数据重建模型,完成对地震数据的重建,有助于提高地震数据重建准确率和精度。(The application provides a reconstruction method, a reconstruction device, electronic equipment and a readable storage medium of seismic data, wherein the reconstruction method comprises the following steps: constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed; constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network; training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model; and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data. Therefore, the reconstruction of the seismic data is completed by constructing the seismic data reconstruction model, and the accuracy and precision of seismic data reconstruction are improved.)

1. A method of reconstruction of seismic data, the method comprising:

constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed;

constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network;

training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model;

and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

2. The reconstruction method according to claim 1, wherein the training the deep learning network based on the acquired plurality of sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model comprises:

inputting the acquired seismic data to be reconstructed into a constructed deep learning network to obtain initial reconstruction data;

detecting whether the difference value between each element in the initial reconstruction data and the corresponding label element is smaller than a preset error threshold value or not;

if the difference value between the elements and the corresponding label elements in the initial reconstruction data is not smaller than a preset error threshold value, performing iterative tuning training on the constructed deep learning network based on the normalized error function, and when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, determining that the tuning training of the constructed deep learning network is finished, so as to obtain a trained seismic data reconstruction model.

3. The reconstruction method according to claim 1, wherein before the training the deep learning network based on the acquired plurality of sample seismic data and the normalized error function to obtain the trained seismic data reconstruction model, the reconstruction method further comprises:

obtaining sample data to be reconstructed within a preset region range;

normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data;

and according to the size of a preset window, performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel to obtain divided sample seismic data.

4. The reconstruction method according to claim 1, wherein the constructing a deep learning network based on the obtained geological properties of the region range of the seismic data to be reconstructed comprises:

determining a deep learning network structure, wherein the deep learning network structure comprises a three-layer neural network;

and initializing a plurality of parameters and a plurality of transmission functions between every two layers of neural networks based on the obtained geological attributes of the region range of the seismic data to be reconstructed, and determining the constructed deep learning network.

5. The reconstruction method according to claim 2, wherein the constructed deep learning network is trained by the following steps:

determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element;

for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the normalized error function;

for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector;

and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the training of the constructed deep learning network is finished.

6. The reconstruction method according to claim 5, wherein the determining, for each parameter to be updated, a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and a corresponding gradient vector of the parameter comprises:

substituting the parameters to be updated as input parameters into the correction function to obtain iterative parameter gradient vectors;

updating the parameter to be updated based on the iterated parameter gradient vector to obtain an updated parameter;

and substituting the updated update parameter into the correction function again, and obtaining a target update parameter through iteration update of preset times.

7. The reconstruction method according to claim 1, wherein the tag element is determined by:

determining a spline function corresponding to each two adjacent elements in the seismic data to be reconstructed;

and determining an interpolation function corresponding to the seismic data to be reconstructed based on the determined spline functions, and determining a label element corresponding to each element based on the interpolation function.

8. An apparatus for reconstructing seismic data, the apparatus comprising:

the network construction module is used for constructing a deep learning network based on the geological attribute of the obtained region range of the seismic data to be reconstructed;

the function construction module is used for constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network;

the model determining module is used for training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model;

and the data reconstruction module is used for inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

9. An electronic device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is run, the machine-readable instructions when executed by the processor performing the steps of the method of seismic data reconstruction according to any of claims 1 to 7.

10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method of reconstruction of seismic data according to one of claims 1 to 7.

Technical Field

The present application relates to the field of data processing technologies, and in particular, to a method and an apparatus for reconstructing seismic data, an electronic device, and a readable storage medium.

Background

With the rapid development of scientific technology, more and more surveying techniques are applied to the process of predicting oil and gas, wherein geophysical exploration is one of the techniques, geological conditions such as stratum lithology, geological structure and the like are detected by researching and observing the change of various geophysical fields, so that geological properties are deduced, further, the oil gas is predicted, in geophysical exploration, seismic exploration is the most effective and important mode, but in the seismic data acquisition process of seismic exploration, data loss or data sparseness can be caused due to the limitation of forbidden areas, obstacles, acquisition cost and the like, incomplete seismic data can not only lead the overlapping and covering times of common central points to be uneven, but also cause space false frequency to influence the accuracy of surface multiple suppression and migration imaging results, therefore, how to reconstruct incomplete seismic data to obtain complete seismic data is an urgent problem to be solved.

At the present stage, with the continuous development of mathematical theories and computer technologies, seismic data reconstruction methods are also abundant and complete, mathematical methods such as Fourier transform, windowing technology, Radon transform and the like are widely applied to the seismic data reconstruction process, the mathematical methods all solve the problem of how to reconstruct data by using a linear inversion strategy, and require that seismic data have the same-phase axis linearity and the parabolic type, have sparsity in a transform domain, accurate underground medium velocity information and the like, but in the actual application process, the acquired seismic data hardly completely meet the conditions, and the accuracy and precision of seismic data reconstruction by using the methods are low.

Disclosure of Invention

In view of this, an object of the present application is to provide a seismic data reconstruction method, apparatus, electronic device and readable storage medium, which complete the reconstruction of seismic data by constructing a seismic data reconstruction model, and contribute to improving the accuracy and precision of seismic data reconstruction.

The embodiment of the application provides a reconstruction method of seismic data, which comprises the following steps:

constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed;

constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network;

training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model;

and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Further, the training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model includes:

inputting the acquired seismic data to be reconstructed into a constructed deep learning network to obtain initial reconstruction data;

detecting whether the difference value between each element in the initial reconstruction data and the corresponding label element is smaller than a preset error threshold value or not;

if the difference value between the elements and the corresponding label elements in the initial reconstruction data is not smaller than a preset error threshold value, performing iterative tuning training on the constructed deep learning network based on the normalized error function, and when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, determining that the tuning training of the constructed deep learning network is finished, so as to obtain a trained seismic data reconstruction model.

Further, before the training of the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model, the reconstruction method further includes:

obtaining sample data to be reconstructed within a preset region range;

normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data;

and according to the size of a preset window, performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel to obtain divided sample seismic data.

Further, the constructing a deep learning network based on the obtained geological attributes of the region range of the seismic data to be reconstructed includes:

determining a deep learning network structure, wherein the deep learning network structure comprises a three-layer neural network;

and initializing a plurality of parameters and a plurality of transmission functions between every two layers of neural networks based on the obtained geological attributes of the region range of the seismic data to be reconstructed, and determining the constructed deep learning network.

Further, the constructed deep learning network is trained through the following steps:

determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element;

for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the normalized error function;

for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector;

and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the training of the constructed deep learning network is finished.

Further, the determining, for each parameter to be updated, a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector includes:

substituting the parameters to be updated as input parameters into the correction function to obtain iterative parameter gradient vectors;

updating the parameter to be updated based on the iterated parameter gradient vector to obtain an updated parameter;

and substituting the updated update parameter into the correction function again, and obtaining a target update parameter through iteration update of preset times.

Further, the tag element is determined by:

determining a spline function corresponding to each two adjacent elements in the seismic data to be reconstructed;

and determining an interpolation function corresponding to the seismic data to be reconstructed based on the determined spline functions, and determining a label element corresponding to each element based on the interpolation function.

An embodiment of the present application further provides a seismic data reconstruction device, where the reconstruction device includes:

the network construction module is used for constructing a deep learning network based on the geological attribute of the obtained region range of the seismic data to be reconstructed;

the function construction module is used for constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network;

the model determining module is used for training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model;

and the data reconstruction module is used for inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Further, when the model determining module is configured to train the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model, the model determining module is configured to:

inputting the acquired seismic data to be reconstructed into a constructed deep learning network to obtain initial reconstruction data;

detecting whether the difference value between each element in the initial reconstruction data and the corresponding label element is smaller than a preset error threshold value or not;

if the difference value between the elements and the corresponding label elements in the initial reconstruction data is not smaller than a preset error threshold value, performing iterative tuning training on the constructed deep learning network based on the normalized error function, and when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, determining that the tuning training of the constructed deep learning network is finished, so as to obtain a trained seismic data reconstruction model.

Further, the reconstruction apparatus further includes a data processing module, and the data processing module is configured to:

obtaining sample data to be reconstructed within a preset region range;

normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data;

and according to the size of a preset window, performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel to obtain divided sample seismic data.

Further, when the network construction module is used for constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed, the network construction module is used for:

determining a deep learning network structure, wherein the deep learning network structure comprises a three-layer neural network;

and initializing a plurality of parameters and a plurality of transmission functions between every two layers of neural networks based on the obtained geological attributes of the region range of the seismic data to be reconstructed, and determining the constructed deep learning network.

Further, the model determining module is configured to train the constructed deep learning network by:

determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element;

for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the normalized error function;

for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector;

and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the training of the constructed deep learning network is finished.

Further, when the model determining module is configured to determine, for each parameter to be updated, a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector, the model determining module is configured to:

substituting the parameters to be updated as input parameters into the correction function to obtain iterative parameter gradient vectors;

updating the parameter to be updated based on the iterated parameter gradient vector to obtain an updated parameter;

and substituting the updated update parameter into the correction function again, and obtaining a target update parameter through iteration update of preset times.

Further, the reconstruction apparatus further includes a tag determination module, configured to:

determining a spline function corresponding to each two adjacent elements in the seismic data to be reconstructed;

and determining an interpolation function corresponding to the seismic data to be reconstructed based on the determined spline functions, and determining a label element corresponding to each element based on the interpolation function.

An embodiment of the present application further provides an electronic device, including: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when the electronic device is running, the machine-readable instructions when executed by the processor performing the steps of the method of reconstructing seismic data as described above.

Embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of the reconstruction method of seismic data as described above.

According to the seismic data reconstruction method, the seismic data reconstruction device, the electronic equipment and the readable storage medium, a deep learning network is constructed on the basis of the geological attributes of the obtained region range of the seismic data to be reconstructed; constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network; training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model; and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Thus, a deep learning network is constructed according to the geological attribute of the region range of the seismic data to be reconstructed, and a normalized error function is constructed according to the constructed network structure and a plurality of initial parameters in the deep learning network; training the constructed deep learning network based on a plurality of sample seismic data and a normalized error function to obtain a trained seismic data reconstruction model, inputting seismic data to be reconstructed into the seismic data reconstruction model to obtain reconstructed seismic data, and accordingly improving the accuracy and precision of seismic data reconstruction.

In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

FIG. 1 is a flow chart of a seismic data reconstruction method according to an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of a training process for a seismic data reconstruction model;

FIG. 3 is a flow chart of a method for reconstruction of seismic data according to another embodiment of the present application;

FIG. 4 is a BP neural network structure;

FIG. 5 is a plot of the mean square error training function training results;

FIG. 6 is a graph of normalized mean square error training function training results;

FIG. 7(a) is a schematic diagram of missing data;

FIG. 7(b) is a schematic cross-sectional view of the result of the BP model reconstruction using the mean square error training function;

FIG. 7(c) is a diagram illustrating a result of the BP model reconstruction using a normalized mean square error training function;

FIG. 7(d) is a diagram of the 98 th original data;

FIG. 7(e) is a schematic diagram of the mean square error training function reconstructing the 98 th data;

FIG. 7(f) is a schematic diagram of the normalized mean square error training function reconstructing the 98 th data;

FIG. 8 is a schematic structural diagram of an apparatus for reconstructing seismic data according to an embodiment of the present disclosure;

fig. 9 is a second schematic structural diagram of an apparatus for reconstructing seismic data according to an embodiment of the present application;

fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. Every other embodiment that can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present application falls within the protection scope of the present application.

First, an application scenario to which the present application is applicable will be described. The method can be applied to the technical field of data processing, along with the rapid development of scientific technology, more and more survey technologies are applied to the process of predicting oil and gas, geophysical exploration is one of the technologies, geological conditions such as formation lithology, geological structure and the like are detected by researching and observing the change of various geophysical fields, geological properties are deduced, and oil and gas are predicted, seismic exploration is the most effective and important mode in the geophysical exploration, but in the process of acquiring seismic data of seismic exploration, data loss or data sparseness can be caused by the limitation of forbidden areas, obstacles, acquisition cost and the like, incomplete seismic data can not only lead to uneven coverage times of common center stacking, but also cause space false frequency to influence the accuracy of surface multiple-wave suppression and migration imaging results, and how to reconstruct the incomplete seismic data, obtaining complete seismic data is an urgent problem to be solved.

Research shows that with the continuous development of mathematical theories and computer technologies, seismic data reconstruction methods are rich and complete, mathematical methods such as Fourier transformation, windowing technology, Radon transformation and the like are widely applied to seismic data reconstruction processes, the mathematical methods solve the problem of how to reconstruct data by using a linear inversion strategy, and require that seismic data have the same-phase axis linearity and the parabolic type, have sparsity in a transformation domain, accurate underground medium velocity information and the like, but in the practical application process, the acquired seismic data hardly completely meet the conditions, and the accuracy and precision of seismic data reconstruction by using the method are low.

Based on the above, the embodiment of the application provides a seismic data reconstruction method, which is beneficial to improving the seismic data reconstruction accuracy and precision.

Referring to fig. 1, fig. 1 is a flowchart illustrating a seismic data reconstruction method according to an embodiment of the present disclosure. As shown in fig. 1, a method for reconstructing seismic data provided by an embodiment of the present application includes:

s101, constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed.

In the step, a specific network structure of the deep learning network is determined according to the address attribute of the region range of the acquired seismic data, and then the deep learning network is constructed.

Here, the deep learning network may be a BP (back propagation) neural network, and in constructing the deep learning network, the requirement for the structure of the deep learning network is to determine the number of neural network layers of the BP neural network and the number of neurons included in each layer of the neural network.

Here, the geological properties of the region range may include a region in which the region range where the seismic data is acquired is located, a type of a terrain to which the region range belongs, and distribution of obstacles in the region range.

Here, the number of seismic data that may be acquired and the missing situation of the seismic data may be roughly determined according to the geological properties of the obtained region range of the seismic data to be reconstructed, so as to plan the network structure of the deep learning network, for example, in a deep learning network constructed in a region range where more seismic data may be obtained, the number of neurons included in each layer of neural network may be relatively large.

S102, constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network.

In the step, a normalized error function for adjusting the deep learning network is constructed according to the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network.

In the method, for tuning of the deep learning network, a normalized mean square error function is adopted, each dimension characteristic is considered by the normalized mean square error function, namely the difference between trained weight parameters is small, when data of a certain node in input changes greatly, the output change is small, the response of the network is ensured to be smoother, and over-adaptation is reduced, so that the generalization capability of the network is improved.

Wherein the normalized mean square error function can be determined by the following formula:

wherein, a_iFor the actual value of the ith output of the Kth iteration, t_iThe expected value (label element) of the ith output of the Kth iteration, M is the number of output nodes, gamma is the error performance adjustment rate of the number of the output nodes, and the value range is [0, 1]]And n is the number of parameters (weights and offsets) in the deep learning network.

S103, training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model.

In the step, a plurality of parameters of the deep learning network are adjusted according to the acquired plurality of sample seismic data and the normalized error function, and a trained seismic data reconstruction model is obtained.

Therefore, the seismic data reconstruction model with strong generalization capability is obtained through iterative training of the deep learning network, and the seismic data reconstruction model has higher reconstruction precision for various seismic data of different terrains.

And S104, inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

In the step, the acquired seismic data to be reconstructed is input into the seismic data reconstruction model trained in the step S103 to obtain target reconstruction data, and the reconstruction of the seismic data to be reconstructed is completed.

Therefore, the geological image of the corresponding region range can be constructed through the reconstructed target reconstruction data, and then geological analysis is carried out on the corresponding region range according to the obtained geological image.

Here, please refer to fig. 2, fig. 2 is a schematic diagram of a training process of a seismic data reconstruction model, as shown in fig. 2, inputting seismic data to be reconstructed, preprocessing the input seismic data to be reconstructed, inputting the preprocessed seismic data to be reconstructed into a constructed deep learning network, outputting a reconstruction result, detecting whether the output reconstruction result meets a reconstruction requirement (an error value is smaller than a preset error threshold or a signal-to-noise ratio is greater than a preset signal-to-noise ratio threshold), and if the reconstruction requirement is met, determining that the training of the seismic data reconstruction model is completed; and if the reconstruction requirement is not met, determining the label elements by utilizing cubic spline interpolation, and iteratively updating parameters (weight and bias) of the deep learning network by utilizing the label elements and the constructed error function until the reconstruction result meets the reconstruction requirement, and determining that the seismic data reconstruction model is trained completely.

According to the seismic data reconstruction method provided by the embodiment of the application, a deep learning network is constructed on the basis of the geological attributes of the obtained region range of the seismic data to be reconstructed; constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network; training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model; and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Thus, a deep learning network is constructed according to the geological attributes of the obtained region range of the seismic data to be reconstructed, and a normalized error function is constructed according to the constructed network structure and a plurality of initial parameters in the deep learning network; training the constructed deep learning network based on a plurality of sample seismic data and a normalized error function to obtain a trained seismic data reconstruction model, inputting seismic data to be reconstructed into the seismic data reconstruction model to obtain reconstructed seismic data, and accordingly improving the accuracy and precision of seismic data reconstruction.

Referring to fig. 3, fig. 3 is a flowchart of a seismic data reconstruction method according to another embodiment of the present application. As shown in fig. 3, a method for reconstructing seismic data provided by an embodiment of the present application includes:

s301, constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed.

S302, a normalization error function is constructed based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network.

And S303, obtaining sample data to be reconstructed within a preset region range.

In the step, sample data to be reconstructed within a preset region range is obtained.

Here, the acquired sample seismic data missing in the time-space domain is denoted as p (t)_m,x_n) Wherein N is 1, 2, 3, …, N; m is 1, 2, 3, …, M; n represents the number of receiving channels on each measuring line, and M represents the number of sampling points in the time direction, wherein the sampling points can be regarded as a two-dimensional matrix of time-space, namely, the obtained sample matrix to be reconstructed.

The seismic data mainly describes the amplitude, phase, vibration frequency, and the like of the vibration wave at a certain sampling time point or on a certain receiving channel.

S304, normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data.

In the step, each element included in the sample data to be reconstructed is subjected to normalization processing to obtain normalized seismic data.

Here, in the network training process, when the value ranges of the input matrix vector and the output matrix vector are [ -1, 1], the training effect is optimal, and in order to achieve a better training result, normalization processing of corresponding data (amplitude, etc.) is required.

S305, according to the size of a preset window, performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel to obtain divided sample seismic data.

In this step, according to the size of a preset window, performing space-time windowing on the normalized seismic data determined in step S304 according to the acquisition time and the receiving channel, that is, blocking the data to be reconstructed, to obtain divided sample seismic data.

Here, the normalized seismic data is partitioned to obtain local features of the data, the size of the partition window may be set to 11 × 11, and the sliding step of the window may be 5 in both directions on the sliding trajectory.

S306, training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model.

And S307, inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

The descriptions of S301, S302, S306 to S307 may refer to the descriptions of S101 to S104, and the same technical effect can be achieved, which is not described in detail herein.

Further, step S306 includes: inputting the acquired seismic data to be reconstructed into a constructed deep learning network to obtain initial reconstruction data; detecting whether the difference value between each element in the initial reconstruction data and the corresponding label element is smaller than a preset error threshold value or not; if the difference value between the elements and the corresponding label elements in the initial reconstruction data is not smaller than a preset error threshold value, performing iterative tuning training on the constructed deep learning network based on the normalized error function, and when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, determining that the tuning training of the constructed deep learning network is finished, so as to obtain a trained seismic data reconstruction model.

Inputting the acquired seismic data to be reconstructed, which are missing in a time-space domain, into a constructed deep learning network to obtain initial reconstruction data output by the deep learning network; for each element included in initial reconstruction data output through a deep learning network, detecting whether a difference value between the element and a corresponding tag element is smaller than a preset error threshold value; if the situation that the difference value between the element and the corresponding label element in the initial reconstruction data is not smaller than the preset error threshold value is determined, the deep learning network is determined not to reach the reconstruction requirement, the deep learning network needs to be adjusted again, the constructed deep learning network is subjected to repeated iterative tuning training through the constructed normalized mean square error function, a plurality of parameters of the deep learning network are adjusted, when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, the deep learning network is determined to reach the reconstruction requirement, the iterative tuning training process of the deep learning network is finished, and the trained seismic data reconstruction model is obtained.

Here, the acquired sample seismic data missing in the time-space domain is denoted as p (t)_m,x_n) Wherein N is 1, 2, 3, …, N; m is 1, 2, 3, …, M; n represents the number of receiving channels on each measuring line, M represents the number of sampling points in the time direction, the sampling points can be regarded as a two-dimensional matrix of time-space, and in order to obtain the local characteristics of the sample seismic data, the matrix corresponding to the sample seismic data is divided into a plurality of target sample matrixes to be reconstructed.

Here, in the sample data to be reconstructed for a plurality of targets input into the deep learning network, there may be missing seismic data, which may be missing seismic data on a plurality of receiving channels at a certain acquisition time point, or missing seismic data on an entire receiving channel, and the initial reconstruction data output via the deep learning network is data that is complemented with missing data and has no missing data.

Here, for the reconstruction accuracy of the deep learning network, whether the reconstruction requirement is met may be detected by determining an error value between each element and a tag element in the reconstructed initial reconstruction data, and determining the reconstruction effect of the model by comparing the error value between each element and the tag element in the initial reconstruction data with a preset error threshold.

Here, in other embodiments, it may also be determined whether the reconstruction of the seismic data by the deep learning network meets the reconstruction requirement according to the signal-to-noise ratio of the seismic data in the initial reconstruction data, and when the signal-to-noise ratio of the seismic data in the initial reconstruction data is greater than a preset signal-to-noise ratio threshold, it is determined that the reconstruction of the seismic data by the deep learning network meets the reconstruction requirement.

Therefore, the seismic data reconstruction model with strong generalization capability is obtained through iterative training of the deep learning network, and the seismic data reconstruction model has higher reconstruction precision for various seismic data of different terrains.

Further, step S301 includes: determining a deep learning network structure, wherein the deep learning network structure comprises a three-layer neural network; and initializing a plurality of parameters and a plurality of transmission functions between every two layers of neural networks based on the obtained geological attributes of the region range of the seismic data to be reconstructed, and determining the constructed deep learning network.

In the step, a deep learning network structure is determined, in the embodiment of the application, the deep learning network comprises three layers of neural networks, after the structure of the deep learning network is determined, a plurality of parameters and a plurality of transmission functions between every two layers of neural networks are initialized according to the obtained geological attributes of the region range of the seismic data to be reconstructed, and after the initialization is finished, the constructed deep learning network is determined.

Here, the deep learning network adopts a feedforward type three-layer BP neural network to realize the reconstruction of the missing seismic data, as shown in fig. 4, fig. 4 is a BP neural network structure, wherein the feedforward type three-layer BP neural network respectively includes an input layer, a hidden layer and an output layer.

The parameters between layers include weight values and bias values.

Here, a plurality of neurons are included in each layer of the BP neural network, and it is assumed that a vector of an input is p ═ p₁,p₂,...p_R]^T，w＝[w_i,1,w_i,2,...,w_i,R]The weight vector of the ith neuron connected with other neurons in the hidden layer; the output of the neuron can then be determined by the following formula:

where a is the output of the neuron, b is the bias of the ith neuron, and f (x) is the transfer function of the neuron.

The transmission function f (x) can adopt a common sigmoid function, can realize the nonlinear mapping relation between input and output, can limit the output of the hidden layer in a smaller range (0,1), and can output any value by adopting a pure linear function instead of the sigmoid function because the amplitude of the seismic data is divided into positive and negative.

Further, the constructed deep learning network is trained through the following steps: determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element; for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the normalized error function; for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector; and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the training of the constructed deep learning network is finished.

In the step, at least one parameter to be updated between every two layers of neural networks is determined based on the mean square error between each element and the corresponding label element in the initial reconstruction data obtained through the unoptimized neural networks, and a parameter gradient vector corresponding to the parameter to be updated is determined based on the constructed normalized error function for each parameter to be updated; for each parameter to be updated, determining a target updating parameter corresponding to the parameter to be updated through finite iterations according to the parameter to be updated and the corresponding parameter gradient vector; and after each parameter to be updated is updated to the corresponding target updating parameter, determining that the tuning training of the constructed deep learning network is finished.

Here, the parameter gradient vector corresponding to the parameter to be updated may be determined by the following formula:

wherein g (k) is a gradient vector of parameters to be updated of the mean square error of the kth iteration, and E (k) is an output error performance function of the neural network of the kth iteration.

Further, the step of determining a target update parameter corresponding to each parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector includes substituting the parameter to be updated as an input parameter into a correction function to obtain an iterated parameter gradient vector; updating the parameter to be updated based on the iterated parameter gradient vector to obtain an updated parameter; and substituting the updated update parameter into the correction function again, and obtaining a target update parameter through iteration update of preset times.

In the step, a parameter to be updated is input into a correction function as an input parameter to obtain an iterated parameter gradient vector, the parameter to be updated is updated according to the obtained iterated parameter gradient vector to obtain an updated update parameter, the updated update parameter is substituted into the correction function again, and the target update parameter is obtained through iteration update for a preset number of times.

Here, the expression of the correction function is:

x(k+1)＝x(k)-αg(k)；

wherein k is the number of iterations, x is the parameter to be updated, α is the learning rate, which is a constant with an empirical value of 0.01, the negative sign indicates the opposite direction of the gradient, and g (k) is the gradient vector of the parameter to be updated for the mean square error of the kth iteration.

Further, the tag element is determined by: determining a spline function corresponding to each two adjacent elements in the seismic data to be reconstructed; and determining an interpolation function corresponding to the seismic data to be reconstructed based on the determined spline functions, and determining a label element corresponding to each element based on the interpolation function.

In the step, spline functions between every two adjacent elements in the sample data to be reconstructed are determined, interpolation functions corresponding to the sample data to be reconstructed are determined by combining the determined spline functions, and label elements on each node in the sample data to be reconstructed are determined according to the interpolation functions.

In the application, a cubic spline interpolation method is adopted to obtain an interpolation function corresponding to a sample matrix to be reconstructed.

Further, please refer to fig. 5 and fig. 6, in which fig. 5 is a training result curve of the mean square error training function, fig. 6 is a training result curve of the normalized mean square error training function, and a simple sine function is fitted by the BP neural network as an example to illustrate the improvement of the network generalization capability by the normalized mean square error training function adopted in the present application. The input vector is P [ -1:0.05:1], the expected output sample data is t ═ sin (2 pi × P) +0.1randn (size (P)), 0.1 times of random noise is added to the true sine function, the training result is as shown in fig. 5 and fig. 6, it can be known from fig. 5 and fig. 6 that the output result of the mean square error training function is closer to the training sample data, and the output result of the normalized error training function is closer to the true sine function curve, so that the overfitting condition is improved, and the generalization capability of the network is improved.

To further illustrate the generalization capability of the BP neural network trained by the normalized mean square error function (msereg), we compare the reconstruction results of the BP neural network trained by the mean square error training function (mse) and the normalized mean square error function (msereg), and please refer to fig. 7(a) to 7(f), where each shot data has 100 receiving points, the inter-lane distance is 10m, each lane has 500 time samples, the time sample interval is 4ms, and the start time, form and structure of the event axes of different shots are different. Fig. 7(a) is a schematic diagram of missing data, fig. 7(b) is a schematic diagram of a cross-sectional view of a reconstruction result of a mean square error training function for a BP model, fig. 7(c) is a schematic diagram of a reconstruction result of a normalized mean square error training function for a BP model, fig. 7(d) is a schematic diagram of a 98 th original data, and fig. 7(e) is a schematic diagram of a reconstruction result of a 98 th data of a mean square error training function; fig. 7(f) is a schematic diagram of the 98 th data reconstructed by the normalized mean square error training function, which shows that the difference between the amplitude of the reconstructed result of the mean square error training function and the amplitude of the original data is large, while the amplitude of the reconstructed result of the normalized mean square error training function is better recovered and more consistent with the original data, and proves that the normalized mean square error function can improve the generalization capability of the BP network.

According to the seismic data reconstruction method provided by the embodiment of the application, a deep learning network is constructed on the basis of the geological attributes of the obtained region range of the seismic data to be reconstructed; constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network; obtaining sample data to be reconstructed within a preset region range; normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data; performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel according to the size of a preset window to obtain divided sample seismic data; training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model; and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Thus, a deep learning network is constructed according to the geological attributes of the region range of the acquired seismic data, and a normalized error function is constructed according to the constructed network structure and a plurality of initial parameters in the deep learning network; training the constructed deep learning network based on a plurality of sample seismic data and a normalized error function to obtain a trained seismic data reconstruction model, inputting seismic data to be reconstructed into the seismic data reconstruction model to obtain reconstructed seismic data, and accordingly improving the accuracy and precision of seismic data reconstruction.

Referring to fig. 8 and 9, fig. 8 is a first schematic structural diagram of a seismic data reconstruction apparatus according to an embodiment of the present disclosure, and fig. 9 is a second schematic structural diagram of a seismic data reconstruction apparatus according to an embodiment of the present disclosure. As shown in fig. 8, the reconstruction apparatus 800 includes:

the network construction module 810 is used for constructing a deep learning network based on the geological attributes of the obtained region range of the seismic data to be reconstructed;

a function constructing module 820, configured to construct a normalized error function based on the network structure of the constructed deep learning network and a plurality of initial parameters in the deep learning network;

the model determining module 830 is configured to train the deep learning network based on the acquired multiple sample seismic data and the normalized error function, so as to obtain a trained seismic data reconstruction model;

and the data reconstruction module 840 is used for inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Further, as shown in fig. 9, the reconstruction apparatus 800 further includes a data processing module 850, where the data processing module 850 is configured to:

obtaining sample data to be reconstructed within a preset region range;

normalizing each element included in the sample data to be reconstructed to obtain corresponding normalized seismic data;

and according to the size of a preset window, performing space-time windowing on the normalized seismic data according to the acquisition time and the receiving channel to obtain divided sample seismic data.

Further, as shown in fig. 9, the reconstruction apparatus 800 includes a tag determination module 860, where the tag determination module 860 is configured to:

determining a spline function corresponding to each two adjacent elements in the seismic data to be reconstructed;

and determining an interpolation function corresponding to the seismic data to be reconstructed based on the determined spline functions, and determining a label element corresponding to each element based on the interpolation function.

Further, when the model determining module 830 is configured to train the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model, the model determining module 830 is configured to:

inputting the acquired seismic data to be reconstructed into a constructed deep learning network to obtain initial reconstruction data;

detecting whether the difference value between each element in the initial reconstruction data and the corresponding label element is smaller than a preset error threshold value or not;

if the difference value between the elements and the corresponding label elements in the initial reconstruction data is not smaller than a preset error threshold value, performing iterative tuning training on the constructed deep learning network based on the normalized error function, and when the difference value between each element in the output reconstruction data and the corresponding label element is smaller than the preset error threshold value, determining that the tuning training of the constructed deep learning network is finished, so as to obtain a trained seismic data reconstruction model.

Further, when the network construction module 810 is configured to construct a deep learning network based on the obtained geological attributes of the region range of the seismic data to be reconstructed, the network construction module 810 is configured to:

determining a deep learning network structure, wherein the deep learning network structure comprises a three-layer neural network;

and initializing a plurality of parameters and a plurality of transmission functions between every two layers of neural networks based on the obtained geological attributes of the region range of the seismic data to be reconstructed, and determining the constructed deep learning network.

Further, the model determining module 830 is configured to perform iterative tuning training on the constructed deep learning network by:

determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element;

for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the constructed normalized error function;

for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector;

and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the well-constructed deep learning network tuning training is finished.

Further, the model determining module 830 is configured to train the constructed deep learning network by:

determining at least one parameter to be updated between each two layers of neural networks based on the mean square error between each element in the initial reconstruction data and the corresponding tag element;

for each parameter to be updated, determining a parameter gradient vector corresponding to the parameter to be updated based on the normalized error function;

for each parameter to be updated, determining a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector;

and when the target updating parameters corresponding to each parameter to be updated are determined to be finished, determining that the training of the constructed deep learning network is finished.

Further, when the model determining module 830 is configured to determine, for each parameter to be updated, a target update parameter corresponding to the parameter to be updated based on the parameter to be updated and the corresponding parameter gradient vector, the model determining module 830 is configured to:

substituting the parameters to be updated as input parameters into the correction function to obtain iterative parameter gradient vectors;

updating the parameter to be updated based on the iterated parameter gradient vector to obtain an updated parameter;

and substituting the updated update parameter into the correction function again, and obtaining a target update parameter through iteration update of preset times.

The seismic data reconstruction device provided by the embodiment of the application constructs a deep learning network based on the geological attribute of the obtained region range of the seismic data to be reconstructed; constructing a normalized error function based on the constructed network structure of the deep learning network and a plurality of initial parameters in the deep learning network; training the deep learning network based on the acquired multiple sample seismic data and the normalized error function to obtain a trained seismic data reconstruction model; and inputting the seismic data to be reconstructed into the seismic data reconstruction model to obtain the reconstructed seismic data.

Thus, a deep learning network is constructed according to the geological attributes of the region range of the acquired seismic data, and a normalized error function is constructed according to the constructed network structure and a plurality of initial parameters in the deep learning network; training the constructed deep learning network based on a plurality of sample seismic data and a normalized error function to obtain a trained seismic data reconstruction model, inputting seismic data to be reconstructed into the seismic data reconstruction model to obtain reconstructed seismic data, and accordingly improving the accuracy and precision of seismic data reconstruction.

Referring to fig. 10, fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. As shown in fig. 10, the electronic device 1000 includes a processor 1010, a memory 1020, and a bus 1030.

The memory 1020 stores machine-readable instructions executable by the processor 1010, when the electronic device 1000 runs, the processor 1010 and the memory 1020 communicate through the bus 1030, and when the machine-readable instructions are executed by the processor 1010, the steps of the method for reconstructing seismic data in the method embodiments shown in fig. 1 and fig. 3 may be executed.

An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method for reconstructing seismic data in the method embodiments shown in fig. 1 and fig. 3 may be executed.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the exemplary embodiments of the present application, and are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.