CN113722893A

CN113722893A - Seismic record inversion method, device, equipment and storage medium

Info

Publication number: CN113722893A
Application number: CN202110925389.0A
Authority: CN
Inventors: 王梓旭
Original assignee: China University of Petroleum Beijing
Current assignee: China University of Petroleum Beijing
Priority date: 2021-08-12
Filing date: 2021-08-12
Publication date: 2021-11-30
Anticipated expiration: 2041-08-12
Also published as: CN113722893B

Abstract

The embodiment of the invention provides a seismic record inversion method, a device, equipment and a storage medium, wherein the method comprises the following steps: the method comprises the steps of obtaining small sample seismic records, corresponding small sample wave impedance and large sample unlabeled seismic records, constructing a dual WGAN model based on a neural network model, wherein the dual WGAN model comprises an inversion WGAN and a forward WGAN, optimizing the dual WGAN model through the small sample local seismic records, the small sample wave impedance and the large sample unlabeled seismic records, inputting the large sample unlabeled seismic records into an optimized inversion generator to generate corresponding large sample wave impedance, constructing the dual WGAN model, realizing high-precision inversion of the small sample local seismic records based on the cycle consistency of dual learning, replacing a cross entropy loss function in the traditional GAN with Wassertein distance, and improving the inversion stability of the seismic records.

Description

Seismic record inversion method, device, equipment and storage medium

Technical Field

The invention relates to the field of energy development and exploration, in particular to a seismic record inversion method, a device, equipment and a storage medium.

Background

Wave impedance inversion is a post-stack inversion technique that estimates elastic parameters of the subsurface medium. According to the convolution formula, the amplitude of the same reflection point of the stacked seismic gather is related to the elastic properties of the upper medium and the lower medium of the reflection interface, the inversion of the stacked seismic records is based on the relation, and the elastic parameters of the underground medium, namely wave impedance, are calculated according to the stacked seismic data by using an inversion method.

The existing inversion method is a new method for directly establishing a velocity model from an original seismic record based on a supervised depth full convolution neural network by applying the convolution neural network to wave impedance inversion. In the training phase, the network establishes a nonlinear projection from the multi-shot seismic data to the corresponding velocity model. In the prediction phase, the trained network may be used to estimate a velocity model for the new input seismic data.

However, when facing a large number of seismic records and a small number of corresponding wave impedances, the method is difficult to train and optimize a large number of unlabeled seismic records so as to achieve the accuracy required by inversion.

Disclosure of Invention

The embodiment of the invention provides a seismic record inversion method, a device, equipment and a storage medium, which are used for improving the inversion accuracy of a large number of unlabeled seismic records under the condition of a small sample.

In a first aspect, an embodiment of the present invention provides a seismic record inversion method, where the method includes:

acquiring a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record and a large sample unlabeled seismic record;

constructing a dual WGAN model based on a neural network model, wherein the dual WGAN model comprises an inversion WGAN and a forward WGAN, the inversion WGAN comprises an inversion generator and an inversion discriminator, and the forward WGAN comprises a forward generator and a forward discriminator;

optimizing the dual WGAN model through the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record;

and inputting the large sample label-free seismic record into an optimized inversion generator to generate corresponding large sample wave impedance.

Optionally, the inversion generator is a one-dimensional U-net generator which is input as two channels under the constraint of condition information, and the inversion discriminator is a one-dimensional Markovian discriminator which is input as two channels under the constraint of condition information;

the forward generator is a one-dimensional U-net generator, and the forward discriminator is a one-dimensional AlexNet discriminator.

Optionally, the optimizing the dual WGAN model by the small sample local seismic record, the small sample wave impedance, and the large sample unlabeled seismic record includes:

determining an inversion discriminator loss function, a forward discriminator loss function and a generator loss function according to the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record;

optimizing the inversion discriminator, the forward discriminator and the generator according to the inversion discriminator loss function, the forward discriminator loss function and the generator loss function;

wherein the generator comprises the forward generator and the inversion generator.

Optionally, determining an inversion discriminator loss function, a forward discriminator loss function and a generator loss function according to the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record, including:

inputting a low-frequency model, the small sample wave impedance and a seismic record into the inversion generator to obtain a wave impedance predicted value, wherein the seismic record comprises a small sample local seismic record and a large sample non-label seismic record;

inputting the wave impedance predicted value and the small sample wave impedance into the inversion discriminator to obtain an inversion discriminator loss function;

inputting the wave impedance predicted value into the positive generator to obtain a reconstructed seismic record;

inputting the small sample wave impedance into the positive generator to obtain a seismic record predicted value;

inputting the seismic record predicted value and the seismic record into the forward direction discriminator to obtain a loss function of the forward direction discriminator;

inputting the seismic record predicted value into the inversion generator to obtain reconstructed wave impedance;

determining the generator loss function according to the wave impedance predicted value, the seismic record predicted value, the reconstructed seismic record and the reconstructed wave impedance.

Optionally, the inversion discriminator loss function is:

wherein D is_inverInversion discriminator for inverting WGAN, G_inverFor inverting the inversion generator of WGAN, AI is the small sample wave impedance, S is the small sample local seismic record corresponding to AI, and (A | B) is expressed as A input network, gp under the constraint of B condition₁Is a gradient penalty term, lambda is a gradient penalty term coefficient, and m is the number of small samples for training;

the forward arbiter penalty function is:

wherein D is_forwardInversion discriminator for inverting WGAN, G_forwardInversion Generator, gp, for inverting WGAN₂Is a gradient penalty term.

Optionally, the generator loss function includes a loss function generating a countermeasure portion and a loss function generating a cyclic consistency portion; the generating a loss function for the countering portion comprises generating a loss function for the countering inversion portion and generating a loss function for the countering forward portion; the loss function of the cyclic consistency part comprises a loss function of a cyclic consistency open-loop part and a loss function of a cyclic consistency closed-loop part;

the loss function for generating the antagonistic inversion part is:

wherein S^*For large sample unlabeled seismic records, AI_lowIs a wave impedance low-frequency model;

the loss function for generating the forward part of the countermeasure is:

the loss function of the cyclic consistency open loop portion is:

the loss function of the loop consistency closed-loop part is as follows:

the generator loss function is:

g_loss＝g_inver_loss+g_forward_loss+α×(loss_3+loss_4+loss_5)+β×(loss_1+loss_2)

wherein, alpha and beta are respectively the constraint coefficients contributed by the control open loop part and the control closed loop part, and M is the number of large samples for training.

In a second aspect, an embodiment of the present invention provides a seismic recording inversion apparatus, including:

the acquisition module is used for acquiring a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record and a large sample unlabeled seismic record;

the system comprises a building module, a processing module and a processing module, wherein the building module is used for building a dual WGAN model based on a neural network model, the dual WGAN model comprises an inversion WGAN and a forward WGAN, the inversion WGAN comprises an inversion generator and an inversion discriminator, and the forward WGAN comprises a forward generator and a forward discriminator;

the optimization module is used for optimizing the dual WGAN model through the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record;

and the generating module is used for inputting the large sample label-free seismic record into the optimized inversion generator to generate corresponding large sample wave impedance.

In a third aspect, an embodiment of the present invention provides a seismic recording inversion apparatus, including:

a memory and at least one processor;

the memory stores computer-executable instructions;

the at least one processor executing the memory stored computer-executable instructions causes the at least one processor to perform a seismic recording inversion method as described in any one of the first aspects above.

In a fourth aspect, embodiments of the present invention provide a computer-readable storage medium having stored therein computer-executable instructions, which, when executed by a processor, are configured to implement a seismic record inversion method as described in any one of the first aspects above.

In a fifth aspect, embodiments of the invention provide a computer program product comprising a computer program that, when executed by a processor, implements a method of seismic record inversion as described in any one of the first aspects above.

The seismic record inversion method, the device, the equipment and the storage medium provided by the embodiment of the invention construct a dual WGAN model based on a neural network model by obtaining a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record and a large sample unlabeled seismic record, wherein the dual WGAN model comprises an inversion WGAN and a forward WGAN, the inversion WGAN comprises an inversion generator and an inversion discriminator, the forward WGAN comprises a forward generator and a forward discriminator, the dual WGAN model is optimized through the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record, the large sample unlabeled seismic record is input into the optimized inversion generator to generate a corresponding large sample wave impedance, a dual WGAN model is constructed, and the high-precision inversion of the small sample local seismic record can be realized based on the cycle consistency of dual learning, the Wasserstein distance is used for replacing a cross entropy loss function in the traditional GAN, so that the problems of gradient loss and mode collapse in the traditional GAN inversion process are solved, and the stability of seismic record inversion is improved.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a schematic view of an application scenario provided in an embodiment of the present invention;

FIG. 2 is a schematic flow chart of a seismic record inversion method according to an embodiment of the present invention;

fig. 3 is a schematic diagram of a network structure of an inversion generator according to an embodiment of the present invention;

fig. 4 is a schematic diagram of a network structure of an inversion discriminator according to an embodiment of the present invention;

fig. 5 is a schematic diagram of a network structure of a forward generator according to an embodiment of the present invention;

fig. 6 is a schematic diagram of a network structure of a forward arbiter according to an embodiment of the present invention;

FIG. 7 is a block diagram of a framework for a dual WGAN model according to an embodiment of the invention;

FIG. 8 is a schematic flow chart of another seismic recording inversion method provided by an embodiment of the invention;

fig. 9 is a schematic diagram of a Marmousi2 wave impedance model according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of wavelets for a synthetic seismic record according to an embodiment of the present invention;

FIG. 11 is a cross-sectional view of a synthetic seismic record provided by an embodiment of the invention;

FIG. 12 is a graph comparing results of a single inversion and actual results provided by the present invention;

FIG. 13A is a two-dimensional cross-sectional view of a real impedance value provided by an embodiment of the present invention;

FIG. 13B is a two-dimensional cross-sectional view of a low frequency model according to an embodiment of the invention;

FIG. 13C is a two-dimensional cross-sectional view of a dual WGAN based inversion result according to an embodiment of the invention;

FIG. 14 is a schematic structural diagram of a seismic logging inversion apparatus according to an embodiment of the present invention;

fig. 15 is a schematic structural diagram of a seismic recording inversion apparatus according to an embodiment of the present invention.

With the above figures, certain embodiments of the invention have been illustrated and described in more detail below. The drawings and the description are not intended to limit the scope of the inventive concept in any way, but rather to illustrate it by those skilled in the art with reference to specific embodiments.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.

The following describes the technical solution of the present invention and how to solve the above technical problems with specific examples. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments.

The following explains an application scenario provided by an embodiment of the present invention: the scheme provided by the embodiment of the invention relates to inversion of large sample unlabeled seismic records to obtain corresponding wave impedance. In actual seismic exploration inversion, the problem that a large number of seismic records are available but the corresponding wave impedance is low exists, so that inversion of large-sample unlabeled seismic records needs to be realized through small-sample seismic records, and the wave impedance corresponding to large-sample unlabeled seismic records is obtained.

In some technologies, linear approximation is performed on the line-branching mapping from the seismic record to the wave impedance based on a generalized linear inversion method, so that generalized linear inversion of the post-stack seismic record is realized through the linear inversion method, but the method may have the problem of poor inversion accuracy.

In other technologies, an algorithm of a self-adaptive damping term can be provided according to the sparse feature of the strong reflection coefficient and two different regularization criteria, so that the constraint sparse pulse inversion of the wave impedance is realized. However, the seismic recording bandwidth recorded according to the convolution model is often limited, and the bandwidth of the inversion target is wide, which leads to underdefining the seismic inversion problem.

In other techniques, a wide-band seismic inversion method may be implemented by sparse constraint methods as well constraint methods. And the inversion method based on the geology statistics provides that the seismic parameters are used as random variables, and the inversion of the seismic data is carried out based on a Bayes frame, so that the optimal solution of the statistics is obtained. Assuming that all uncertainties can be described by multi-dimensional Gaussian probability density, the environmental noise, the discretization error, the theoretical error and the prior information of the underground medium model are brought into the calculation information, and the inversion of the surface seismic data is realized through the Bayesian framework, so that the method has the problem of low inversion accuracy.

In other technologies, as the development of deep learning technology has been successfully applied to various fields, inversion methods based on deep learning have received extensive attention and research in recent years. The method can directly establish a velocity model from an original seismic record based on a supervised depth full convolution neural network, in a training stage, the network establishes a nonlinear projection to multi-shot seismic data to a corresponding velocity model, in a prediction stage, the trained network can be used for estimating the new velocity model of input seismic data, but the method has no clear physical significance.

In other technologies, a method for wave impedance inversion based on a closed-loop convolution neural network under the constraint of logging data is provided for a small sample data learning method in seismic inversion, wherein back-CNN and forward-CNN in the closed-loop convolution neural network are forward modeling methods for realizing seismic inversion from seismic records to wave impedance and seismic records from wave impedance to seismic records respectively, but the accuracy required by inversion is still difficult to achieve.

Therefore, an embodiment of the present invention provides an earthquake record inversion method, which includes constructing a dual WGAN (dynamic adaptive network) model based on a neural network model, where the dual WGAN model includes an inversion WGAN and a forward WGAN, inputting the obtained small sample local earthquake record, the corresponding small sample wave impedance, and the large sample local earthquake record into a network model to perform optimization training, inputting the large sample unlabeled record into an optimized inversion generator to generate the corresponding large sample wave impedance, and improving the inversion accuracy of the large sample unlabeled earthquake record under the condition of the small sample local earthquake record by using the dual learning concept in the deep learning through the dual WGAN model.

Fig. 1 is a schematic view of an application scenario provided in an embodiment of the present invention. As shown in fig. 1, the small sample wave impedance, the small sample local seismic record and the large sample unlabeled seismic record are input into a dual WGAN model, and the model is optimally trained.

Some embodiments of the invention are described in detail below with reference to the accompanying drawings. The features of the embodiments and examples described below may be combined with each other without conflict between the embodiments.

Fig. 2 is a schematic flow chart of a seismic record inversion method according to an embodiment of the present invention. As shown in fig. 2, the method in this embodiment may include:

step 201, obtaining a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record, and a large sample unlabeled seismic record.

The seismic record can also be called post-stack seismic record, and the small sample wave impedance can be called logging data and also called elastic parameters of the underground medium. The small sample wave impedance and the post-stack seismic recording are corresponding. A large sample unlabeled seismic record may be a large number of unlabeled datasets.

Wherein a small sample means a relatively small number. Large samples means that the number is large. Unlabeled seismic records refer to seismic records that do not have a corresponding wave impedance.

The seismic record may be two-dimensional data, and the small sample wave impedance may be corresponding one-dimensional data in the two-dimensional small sample seismic record.

Alternatively, the seismic record may be approximated by a convolution model, specifically, the reflection coefficient is calculated by a known impedance model:

wherein Z is₂Is the lower layer wave impedance, Z₁Is the upper wave impedance and R is the reflection coefficient. And solving the seismic record according to a convolution formula:

S(t)＝W(t)*R(t)

where S is the seismic record, R is the reflection coefficient, and W is the wavelet.

Optionally, the small sample wave impedance may be an approximate wave impedance corresponding to the small sample local seismic record.

Step 202, constructing a dual WGAN model based on a neural network model, wherein the dual WGAN model includes an inversion WGAN and a forward WGAN, the inversion WGAN includes an inversion generator and an inversion discriminator, and the forward WGAN includes a forward generator and a forward discriminator.

Wherein, the Dual WGAN model can be called Dual-WGAN-c model. The dual WGAN model consists of two WGANs, the inversion process from seismic record to wave impedance is realized by inverting the WGANs, and the inversion process from experimental wave impedance of the WGANs to the seismic record is forward calculated.

The inversion generator obtains a corresponding wave impedance predicted value according to the seismic record, the forward generator obtains a corresponding seismic record predicted value according to the wave impedance, the inversion discriminator is used for discriminating the performance of the inversion generator, and the forward discriminator is used for discriminating the performance of the forward generator.

And 203, optimizing the dual WGAN model through the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record.

Optionally, after the dual WGAN model is constructed, the inversion discriminator, inversion generator, forward discriminator and forward generator in the dual WGAN model are optimized through the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record, so that the dual WGAN model is optimized.

And 204, inputting the large sample label-free seismic record into an optimized inversion generator to generate corresponding large sample wave impedance.

Optionally, after the dual WGAN model is optimized, the large sample unlabeled seismic record is input into the optimized inversion generator to generate a corresponding large sample wave impedance.

The seismic record inversion method provided in this embodiment obtains a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record, and a large sample unlabeled seismic record, constructs a dual WGAN model based on a neural network model, where the dual WGAN model includes an inversion WGAN and a forward WGAN, where the inversion WGAN includes an inversion generator and an inversion discriminator, and optimizes the dual WGAN model by using the small sample local seismic record, the small sample wave impedance, and the large sample unlabeled seismic record, inputs the large sample unlabeled seismic record into the optimized inversion generator, generates a corresponding large sample wave impedance, constructs a dual WGAN model, and can implement high-precision inversion of the small sample local seismic record based on the cycle consistency of dual learning, the Wasserstein distance is used for replacing a cross entropy loss function in the traditional GAN, so that the problems of gradient loss and mode collapse in the traditional GAN inversion process are solved, and the stability of seismic record inversion is improved.

On the basis of the technical scheme provided by the embodiment, optionally, the inversion generator is a one-dimensional U-net generator which is input as two channels under the constraint of condition information, and the inversion discriminator is a one-dimensional Markovian discriminator which is input as two channels under the constraint of condition information; the forward generator is a one-dimensional U-net generator, and the forward discriminator is a one-dimensional AlexNet discriminator.

Fig. 3 is a schematic diagram of a network structure of an inversion generator according to an embodiment of the present invention. As shown in fig. 3, the inversion generator is a one-dimensional U-net generator that is input as a two-channel (channel 2) under the constraint of conditional information. Specifically, a pair of small sample local seismic records and small sample wave impedance, and a pair of large sample unlabeled seismic records and a wave impedance low-frequency model are input, and two pairs of data can be input circularly in the input process. In the figure, the rectangle marked with the number 2 is shown as a double channel, and the rest is similar. The arrow to the right represents convolution + bn (batch normalization) + activation function, the arrow to the bottom represents pooling layer, the arrow to the top represents transposed convolution, [ a,1] represents the vector length of the input data, 1 represents dimension, a/2 represents the vector length of the input data as original 1/2, and the rest is similar.

Fig. 4 is a schematic diagram of a network structure of an inversion discriminator according to an embodiment of the present invention. As shown in fig. 4, the inversion discriminator is a one-dimensional Markovian discriminator that is input as a two-channel under the constraint of conditional information. In the figure, black arrows indicate convolution, dashed arrows indicate activation function + convolution + BN, where a in [ a,1] indicates the vector length of the input data, 1 indicates the dimension, and a/2 indicates that the vector length of the input data is 1/2 as it is, and the rest is similar.

Fig. 5 is a schematic diagram of a network structure of a forward generator according to an embodiment of the present invention. As shown in fig. 5, the forward generator is a one-dimensional U-net generator. Specifically, the input may be a small sample wave impedance. In the figure, the rectangle marked with the number 1 is shown as a single channel, and the rest is similar. The arrow to the right represents the convolution + BN + activation function, the arrow down represents the pooling layer, the arrow up represents the transposed convolution, a in [ a,1] represents the vector length of the input data, 1 represents the dimension, a/2 represents the vector length of the input data as original 1/2, and the rest is similar.

Fig. 6 is a schematic network structure diagram of a forward arbiter according to an embodiment of the present invention. As shown in fig. 6, the forward arbiter is a one-dimensional AlexNet arbiter. Wherein, the starting dotted arrow represents convolution, the middle solid arrow represents activation function + convolution + BN, the last two arrows represent full connection, a in [ a,1] represents the vector length of the input data, 1 represents dimension, a/2 represents that the vector length of the input data is 1/2 as before, and the rest is similar.

The convolutional layer in the neural network architecture realizes weight sharing and characteristic extraction, the pooling layer realizes down-sampling, the activation function realizes algorithm nonlinearity, and the BN layer prevents over-training fitting.

Low-frequency information constraint is added in the deep learning inversion process through Conditional GAN, and the black box type neural network training method is endowed with the geophysical significance, namely, the numerical values obtained in the model training process have certain physical meanings and are not only simple numerical values.

determining an inversion discriminator loss function, a forward discriminator loss function and a generator loss function according to the small sample local seismic record, the small sample wave impedance and the large sample unlabeled seismic record; optimizing the inversion discriminator, the forward discriminator and the generator according to the inversion discriminator loss function, the forward discriminator loss function and the generator loss function; wherein the generator comprises the forward generator and the inversion generator.

Optionally, the local seismic record of the small sample, the wave impedance of the small sample and the unlabeled seismic record of the large sample are input into the dual WGAN model, so that an inversion discrimination loss function, a forward discrimination loss function and a generator loss function can be obtained, when the value of the loss function of the inversion discriminator reaches the minimum value, the optimization of the inversion discriminator can be realized, when the value of the loss function of the forward discriminator reaches the minimum value, the optimization of the forward discriminator can be realized, and when the value of the loss function of the generator reaches the minimum value, the optimization of the generator can be realized.

In this embodiment, the optimization of the network model is realized by optimizing the generator and the arbiter in the dual WGAN model, and the network optimization efficiency can be improved.

Fig. 7 is a schematic diagram of a framework passing through a dual WGAN model according to an embodiment of the present invention, and as shown in fig. 7, determining an inversion discriminator loss function, a forward discriminator loss function, and a generator loss function through the small sample local seismic record, the small sample wave impedance, and the large sample unlabeled seismic record includes:

inputting a low-frequency model, the small sample wave impedance and a seismic record into the inversion generator to obtain a wave impedance predicted value, wherein the seismic record comprises a small sample local seismic record and a large sample non-label seismic record; inputting the wave impedance predicted value and the small sample wave impedance into the inversion discriminator to obtain an inversion discriminator loss function; inputting the wave impedance predicted value into the positive generator to obtain a reconstructed seismic record; inputting the small sample wave impedance into the positive generator to obtain a seismic record predicted value; inputting the seismic record predicted value and the seismic record into the forward direction discriminator to obtain a forward direction discriminator loss function; inputting the seismic record predicted value into the inversion generator to obtain reconstructed wave impedance; determining the generator loss function according to the wave impedance predicted value, the seismic record predicted value, the reconstructed seismic record and the reconstructed wave impedance.

The wave impedance predicted value comprises a small sample wave impedance predicted value and a large sample wave impedance predicted value, and the reconstructed seismic record comprises a small sample reconstructed seismic record and a large sample reconstructed seismic record.

Optionally, the pair of small sample local seismic records and the corresponding small sample wave impedance are input into the inversion generator to obtain a small sample wave impedance predicted value, and the pair of large sample local seismic records and the corresponding wave impedance low-frequency model are input into the inversion generator to obtain a large sample wave impedance predicted value. And inputting the small sample wave impedance predicted value and the wave impedance into an inversion discriminator to obtain an inversion discriminator loss function.

And inputting the small sample wave impedance predicted value into a positive generator to obtain a small sample reconstructed seismic record, and inputting the large sample wave impedance predicted value into the positive generator to obtain a large sample reconstructed seismic record.

And inputting the small sample wave impedance into a positive generator to obtain a small sample local seismic record predicted value. And inputting the seismic record predicted value and the small sample local seismic record into a forward arbiter to obtain a forward arbiter loss function. And inputting the seismic record predicted value into an inversion generator to obtain reconstructed wave impedance.

And determining a generator loss function according to the wave impedance predicted value, the seismic record predicted value, the reconstructed seismic record and the reconstructed wave impedance.

In the embodiment, based on the cycle consistency of dual learning, the problem of low inversion accuracy caused by insufficient sample data in seismic exploration is solved, and the inversion accuracy of local seismic records of small samples is realized.

Optionally, the inversion discriminator loss function is:

wherein D is_inverInversion discriminator for inverting WGAN, G_inverFor inverting the inversion generator of WGAN, AI is the small sample wave impedance, S is the small sample local seismic record corresponding to AI, (A | B) represents the input network of A under the constraint of B condition, gp₁Is a gradient penalty term, lambda is a gradient penalty term coefficient, and m is the number of small samples for training.

Wherein m is specifically represented as a small sample local seismograph for trainingThe number of records. D_inver(AI | AI) is expressed as AI input network under AI condition constraint, and real small sample wave impedance AI, D is obtained through inversion discriminator action_inver(G_inver(S | AI) | AI) is expressed as S input network under the constraint of AI condition, and corresponding G is generated through the action of an inversion generator_inver(S | AI), and then G is added under the constraint of AI condition_inverAnd (S | AI) is input into the network, and a small sample wave impedance predicted value is generated through the action of an inversion discriminator.

Optionally, a minimum value between the true small sample wave impedance and the generated small sample wave impedance predicted value is calculated in the inversion discriminator, so that the loss function of the inversion discriminator is optimized, and the inversion discriminator is optimized.

The forward arbiter penalty function is:

And S is a real sample local seismic record, and AI is a label corresponding to the sample local seismic record.

Optionally, the minimum value between the real small sample local seismic record and the generated small sample local seismic record prediction value is calculated in the inversion discriminator, so that the loss function of the forward discriminator is optimized, and the forward discriminator is optimized.

In this embodiment, the minimum value between the real small sample wave impedance and the generated small sample wave impedance predicted value is calculated, and the minimum value between the real small sample local seismic record and the generated small sample local seismic record predicted value is calculated, so that the inversion discriminator and the forward discriminator are optimized, and the accuracy of the generated wave impedance can be improved.

the loss function for generating the antagonistic inversion part is:

wherein S^*For large sample unlabeled seismic records, AI_lowA low frequency model of wave impedance obtained from well log data;

the loss function for generating the forward part of the countermeasure is:

the loss function of the cyclic consistency open loop portion is:

the loss function of the loop consistency closed-loop part is as follows:

the generator loss function is:

wherein α and β are constraint coefficients respectively contributing to control of the open-loop part and the closed-loop part, M is the number of large samples to be trained, and β > α > 1, α is 10, β is 1000, and M is specifically represented as the number of large samples to be trained without tag seismic records.

Optionally, the generator loss function may be determined by the predicted value of wave impedance, the predicted value of seismic record, the reconstructed seismic record, and the reconstructed wave impedance obtained in fig. 7.

In this embodiment, the generator loss function is used to find the minimum value, so that the generator can be optimized in the case of the above embodiment, and the network model can be optimized accordingly.

Fig. 8 is a schematic flow chart of another seismic recording inversion method according to an embodiment of the present invention. On the basis of the foregoing embodiment, the dual WGAN model is optimized specifically with a hyper-parameter and optimization method, as shown in fig. 8, the method includes:

step 801, acquiring seismic records and small sample wave impedance (logging data), and labeling the label data.

The seismic records comprise small sample local seismic records and large sample unlabeled seismic records.

Optionally, label data labeling is performed on the small sample local seismic record and the small sample wave impedance, so that a small sample local seismic record and small sample wave impedance label data pair is obtained, and a large sample unlabeled seismic record without label data labeling is obtained.

And step 802, constructing a dual WGAN model.

Optionally, a dual WGAN model may be constructed based on a deep neural network building method.

Step 803, set the penalty function.

Optionally, after the network model is built, a proper loss function is selected to calculate an error, and when the error reaches a minimum value, the obtained network model is an optimal network model.

And step 804, adjusting the hyper-parameters and selecting an optimization method.

Optionally, the loss function is subjected to hyper-parameter adjustment, and an optimal loss function value can be obtained through calculation. The adjusted hyper-parameters are 300 iteration (epoch), 0.001 learning rate (learning rate), and 33 batch size (batch size).

Optionally, an optimization algorithm (Adam) is selected through the determined hyper-parameters, the loss function is optimized, and the minimum value of the loss function is determined, so that the aim of training the dual WGAN model is fulfilled.

And 805, performing inversion through the optimized network.

Optionally, the large sample unlabeled seismic record is input into the trained dual WGAN model, and the corresponding large sample wave impedance is obtained through seismic record inversion.

In this embodiment, the dual WGAN model is optimized by combining the hyper-parameters and the optimization method, so that the optimization of the network model can be improved, and the inversion accuracy of the network model can be improved.

Fig. 9 is a schematic diagram of a Marmousi2 wave impedance model according to an embodiment of the present invention. As shown in FIG. 9, the abscissa is x, the left ordinate axis is time in ms, and the right ordinate axis is the impedance value in g/cm³*m/s。

FIG. 10 is a schematic diagram of wavelets for a synthetic seismic record according to an embodiment of the present invention. As shown in fig. 10, the abscissa is time in ms, and the ordinate is amplitude.

FIG. 11 is a cross-sectional view of a synthetic seismic record provided by an embodiment of the invention. As shown in fig. 11, the abscissa axis is x and the ordinate axis is time in ms.

FIG. 12 is a graph comparing results of a single pass inversion with actual results provided by the practice of the present invention. As shown in fig. 12, the abscissa is time in ms, and the ordinate is the impedance value.

FIG. 13A is a schematic representation of the present inventionThe two-dimensional section of a real impedance value is provided. As shown in FIG. 13A, the abscissa is x, the left ordinate axis is time in ms, and the right ordinate axis is the impedance value in g/cm³*m/s。

Fig. 13B is a two-dimensional cross-sectional view of a low-frequency model according to an embodiment of the invention. As shown in FIG. 13B, the abscissa is x, the left ordinate axis is time in ms, and the right ordinate axis is the impedance value in g/cm³*m/s。

Fig. 13C is a two-dimensional cross-sectional view of a dual WGAN inversion result according to an embodiment of the present invention. As shown in FIG. 13C, the abscissa is x, the left ordinate axis is time in ms, and the right ordinate axis is the impedance value in g/cm³*m/s。

In an actual numerical simulation experiment, the inversion accuracy can be improved. The Marmousi2 wave impedance model shown in fig. 9 and the 20Hz rake wavelet shown in fig. 10 were used to generate the seismic record shown in fig. 11 by convolution formula, and 33 traces were extracted as small sample seismic records, and the remaining traces were extracted as large sample unlabeled seismic records. Inversion is carried out by the seismic record inversion method, and the inversion result is compared with the one-dimensional single channel of the real model and the low-frequency model, so that the inversion result shown in figure 12 is very close to the real result. And comparing the inversion result of fig. 13C with the two-dimensional profiles of the real model of fig. 13A and the low-frequency model of fig. 13B, wherein the inversion result is very close to the real value, and thus, the method has very high precision.

Fig. 14 is a schematic structural diagram of a seismic logging inversion apparatus according to an embodiment of the present invention. As shown in fig. 14, the seismic recording inversion apparatus provided in this embodiment may include:

an obtaining module 1401, configured to obtain a small sample local seismic record, a small sample wave impedance corresponding to the small sample local seismic record, and a large sample unlabeled seismic record;

a building module 1402, configured to build a dual WGAN model based on a neural network model, where the dual WGAN model includes an inversion WGAN and a forward WGAN, where the inversion WGAN includes an inversion generator and an inversion discriminator, and the forward WGAN includes a forward generator and a forward discriminator;

an optimization module 1403, configured to optimize the dual WGAN model through the small sample local seismic record, the small sample wave impedance, and the large sample unlabeled seismic record;

a generating module 1404, configured to input the large sample unlabeled seismic record into the optimized inversion generator, and generate a corresponding large sample wave impedance.

Optionally, the optimization module 1403 is specifically configured to:

Optionally, the optimization module 1403, when determining the inversion discriminator loss function, the forward discriminator loss function, and the generator loss function through the small sample local seismic record, the small sample wave impedance, and the large sample unlabeled seismic record, is specifically configured to:

Optionally, the inversion discriminator loss function is:

the forward arbiter penalty function is:

the loss function for generating the antagonistic inversion part is:

the loss function for generating the forward part of the countermeasure is:

the loss function of the cyclic consistency open loop portion is:

the loss function of the loop consistency closed-loop part is as follows:

the generator loss function is:

The apparatus provided in this embodiment may implement the technical solutions of the method embodiments shown in fig. 1 to fig. 13C, and the implementation principles and technical effects are similar, which are not described herein again.

Fig. 15 is a schematic structural diagram of a seismic recording inversion apparatus according to an embodiment of the present invention. As shown in fig. 15, the apparatus provided in this embodiment may include: a memory 152 and at least one processor 151;

the memory 152 stores computer-executable instructions;

the at least one processor 151 executes computer-executable instructions stored by the memory 152 to cause the at least one processor 151 to perform a method as in any of the embodiments described above.

Wherein the memory 152 and the processor 151 may be connected by a bus 153.

For specific implementation principles and effects of the device provided in this embodiment, reference may be made to relevant descriptions and effects corresponding to the embodiments shown in fig. 1 to fig. 13C, which are not described herein again.

Embodiments of the invention also provide a computer-readable storage medium having a computer program stored thereon, where the computer program is executed by a processor to implement the seismic record inversion method provided in any embodiment of the invention.

Embodiments of the invention also provide a computer program product comprising a computer program which, when executed by a processor, implements a method of seismic record inversion as described in any of the embodiments of the invention.

In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of modules is merely a division of logical functions, and an actual implementation may have another division, for example, a plurality of modules 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 through some interfaces, devices or modules, and may be in an electrical, mechanical or other form.

The modules described as separate parts may or may not be physically separate, and parts displayed as modules 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 modules may be selected according to actual needs to implement the solution of the present embodiment.

In addition, functional modules in the embodiments of the present invention may be integrated into one processing unit, or each module may exist alone physically, or two or more modules are integrated into one unit. The unit formed by the modules can be realized in a hardware form, and can also be realized in a form of hardware and a software functional unit.

The integrated module implemented in the form of a software functional module may be stored in a computer-readable storage medium. The software functional module is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) or a processor to execute some steps of the methods according to the embodiments of the present invention.

It should be understood that the Processor may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present invention may be embodied directly in a hardware processor, or in a combination of the hardware and software modules within the processor.

The memory may comprise a high-speed RAM memory, and may further comprise a non-volatile storage NVM, such as at least one disk memory, and may also be a usb disk, a removable hard disk, a read-only memory, a magnetic or optical disk, etc.

The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended ISA (Extended Industry Standard Architecture) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, the buses in the figures of the present invention are not limited to only one bus or one type of bus.

The storage medium may be implemented by any type or combination of volatile or non-volatile memory devices, such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.

An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may reside in an Application Specific Integrated Circuits (ASIC). Of course, the processor and the storage medium may reside as discrete components in an electronic device or host device.

Those of ordinary skill in the art will understand that: all or a portion of the steps of implementing the above-described method embodiments may be performed by hardware associated with program instructions. The program may be stored in a computer-readable storage medium. When executed, the program performs steps comprising the method embodiments described above; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

It will be understood that the invention is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims

1. A seismic record inversion method, comprising:

2. The method according to claim 1, wherein the inversion generator is a one-dimensional U-net generator which is input as two channels under the constraint of condition information, and the inversion discriminator is a one-dimensional Markovian discriminator which is input as two channels under the constraint of condition information;

3. The method of claim 1, wherein optimizing the dual WGAN model by the small sample seismographic record, the small sample wave impedance, and the large sample unlabeled seismographic record comprises:

4. The method of claim 3, wherein determining an inversion discriminator loss function, a forward discriminator loss function, and a generator loss function from the small sample local seismic records, the small sample wave impedances, and the large sample unlabeled seismic records comprises:

5. The method of claim 3 or 4, wherein the inverse discriminator loss function is:

the forward arbiter penalty function is:

6. The method of claim 3 or 4, wherein the generator loss function comprises a loss function generating a countermeasure portion and a loss function generating a cyclic consistency portion; the generating a loss function for the countering portion comprises generating a loss function for the countering inversion portion and generating a loss function for the countering forward portion; the loss function of the cyclic consistency part comprises a loss function of a cyclic consistency open-loop part and a loss function of a cyclic consistency closed-loop part;

the loss function for generating the antagonistic inversion part is:

the loss function for generating the forward part of the countermeasure is:

the loss function of the cyclic consistency open loop portion is:

the loss function of the loop consistency closed-loop part is as follows:

the generator loss function is:

g_loss＝g_inver_loss+g_forward_loss +α×(loss_3+loss_4+loss_5) +β×(loss_1+loss_2)

7. A seismic-record inversion apparatus, comprising:

the device comprises a building module, a processing module and a processing module, wherein the building module is used for building a dual WGAN model based on a neural network model, the dual WGAN model comprises an inversion WGAN and a forward WGAN, the inversion WGAN comprises an inversion generator and an inversion discriminator, and the forward WGAN comprises a forward generator and a forward discriminator;

8. A seismic recording inversion apparatus, comprising: a memory and at least one processor;

the memory stores computer-executable instructions;

the at least one processor executing the memory-stored computer-executable instructions cause the at least one processor to perform the seismic recording inversion method of any of claims 1-6.

9. A computer readable storage medium having computer executable instructions stored thereon which when executed by a processor are configured to implement a seismic record inversion method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the seismic recording inversion method of any of claims 1-6.