A Q value estimation method and device based on a deep neural network, and an electronic device

Abstract

The application relates to the technical field of seismic data interpretation, and discloses a Q value estimation method and device based on a deep neural network and electronic equipment. The method comprises the following steps: training a deep neural network according to a generated steady-state seismic signal, obtaining a trained deep neural network, retraining the trained deep neural network according to to-be-estimated seismic data, obtaining an optimized deep neural network, and predicting the Q value of the to-be-estimated seismic data through the optimized deep neural network. Based on phase correction and multiple characteristics of seismic signals, the method based on phase correction has the advantages of good anti-interference performance to noise and independence from information of a seismic wavelet, and does not need to eliminate the influence of effective frequency band information of the wavelet on a Q value estimation result, so that the method has strong adaptability to actual seismic data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of seismic data interpretation, in particular to a Q value estimation method and device based on a deep neural network, a storage medium and an electronic device. BACKGROUND

[0002] Due to the serious signal attenuation of fluid in the fracture-vug, accurate estimation of the attenuation information or Q value in the fracture-vug is the key to fine characterization of the fracture-vug and fine identification of the fluid in the fracture-vug.

[0003] Conventional Q value estimation methods include the spectral ratio method and the rise time method. Due to the difficulty in obtaining the Q value, the Gabor transform can be used to estimate the amplitude of the attenuation function and the wavelet amplitude in the Gabor domain. Under the assumption of minimum phase, the Hilbert transform is used to obtain the attenuation function and the minimum phase wavelet, and then non-stationary deconvolution is performed to improve the resolution of the seismic record, thereby obtaining high-resolution seismic data. However, the assumption is that the reflection coefficient is white noise and the attenuation function and the wavelet are minimum phase. Wang rearranges the time-frequency spectrum, uses the attenuation function or the compensation function for fitting, and performs Q value analysis to obtain relatively good analysis results.

[0004] The methods described above are relatively sensitive to noise and have certain assumptions, and the effect is not ideal in actual application. SUMMARY

[0005] To solve the above problems, the present application provides a Q value estimation method and device based on a deep neural network, a storage medium and an electronic device. Based on the phase correction and multiple characteristics of the seismic signal, the method based on the phase correction has the advantage of better noise immunity, and does not depend on the information of the seismic wavelet, and does not need to eliminate the influence of the effective frequency band information of the wavelet on the Q value estimation result, so it has strong adaptability to actual seismic data.

[0006] In a first aspect, the present application provides a Q value estimation method based on a deep neural network, which comprises:

[0007] training the deep neural network according to the generated steady-state seismic signal to obtain a trained deep neural network;

[0008] re-training the trained deep neural network according to the seismic data to be estimated to obtain an optimized deep neural network;

[0009] predicting the Q value of the seismic data to be estimated by the optimized deep neural network.

[0010] Further, the training of the deep neural network according to the generated steady-state seismic signal to obtain the trained deep neural network comprises:

[0011] generating a one-dimensional steady-state seismic signal according to a preset minimum phase wavelet and a preset random sparse reflection coefficient sequence;

[0012] generating a first Q value sequence in a preset range, and respectively attenuating a one-dimensional steady-state seismic signal according to different Q values in the first Q value sequence to obtain a non-steady-state seismic signal;

[0013] generating a second Q value sequence for attenuation compensation, and respectively performing attenuation compensation processing on the non-steady-state seismic signal according to Q values in the second Q value sequence to obtain a compensated seismic signal;

[0014] extracting a plurality of different features from the compensated seismic signal to construct a feature matrix;

[0015] normalizing the feature matrix to obtain training data;

[0016] using a preset real Q value as label data, training a deep neural network using the training data and the label data to obtain the trained deep neural network.

[0017] Further, before the trained deep neural network is obtained, the method further comprises:

[0018] modifying the one-dimensional steady-state seismic signal, generating new training data according to the modified one-dimensional steady-state seismic signal, and retraining the deep neural network one or more times through the new training data.

[0019] Further, in the process of performing attenuation processing, the attenuation model comprises a Kolsky-Futterman model.

[0020] Further, under the Kolsky-Futterman model, the attenuation formula of the seismic signal comprises:

[0021]

[0022] wherein U is a seismic signal, f is a frequency, t is a time, t0 is a reference time, f r is a reference frequency, Q is a quality factor, and i is an imaginary unit.

[0023] Further, the attenuation compensation processing is performed by phase correction, wherein the phase correction is performed by the following formula:

[0024]

[0025] wherein U is a seismic signal, τ is a time, ω is a circular frequency, ω γwhere f is the reference circular frequency, g is the attenuation factor, and i is the imaginary unit.

[0026] Further, in extracting a plurality of different features from the compensated seismic signal, the feature value extraction manner comprises acquiring non-Gaussianity of the compensated seismic signal by using different non-Gaussianity criteria.

[0027] In a second aspect, the application provides a Q value estimation device based on a deep neural network, comprising:

[0028] a training module configured to train the deep neural network according to the generated steady-state seismic signal, and obtain a trained deep neural network;

[0029] an optimization training module configured to retrain the trained deep neural network according to the seismic data to be estimated, and obtain an optimized deep neural network;

[0030] a prediction module configured to predict the Q value of the seismic data to be estimated by using the optimized deep neural network.

[0031] In a third aspect, the application provides a computer readable storage medium storing a computer program, which can be executed by one or more processors to implement the method described above.

[0032] In a fourth aspect, the application provides an electronic device comprising a memory and one or more processors, wherein the memory stores a computer program, and the memory and the one or more processors are communicatively connected, and the computer program is executed by the one or more processors to implement the method described above.

[0033] Compared with the prior art, the technical scheme of the application has the following advantages or beneficial effects:

[0034] Based on the phase correction and a plurality of features of the seismic signal, the method based on the phase correction has the advantages of better anti-interference to noise and independence from information of the seismic wavelet, and does not need to eliminate the influence of the effective frequency band information of the wavelet on the Q value estimation result, thus having strong adaptability to actual seismic data. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical scheme in the embodiments of the application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only embodiments of the application, and those skilled in the art can obtain other drawings according to the provided drawings without creative labor.

[0036] The drawings constituting part of the present application are intended to provide a further understanding of the present application. The illustrative embodiments and descriptions thereof in the present application are intended to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0037] Figure 1 A flowchart of a Q-value estimation method based on a deep neural network provided in an embodiment of the present application;

[0038] Figure 2 A flowchart of preparing a Q-value prediction network training sample provided in an embodiment of the present application;

[0039] Figure 3 A flowchart of a Q-value estimation based on a deep neural network provided in an embodiment of the present application;

[0040] Figure 4 A two-dimensional profile of actual seismic data provided in an embodiment of the present application;

[0041] Figure 5 A schematic diagram of a training sample provided in an embodiment of the present application;

[0042] Figure 6 A method for Figure 3 Schematic diagram of the Q value profile obtained after prediction of seismic data;

[0043] Figure 7 A schematic diagram of the structure of a device provided in an embodiment of the present application;

[0044] Figure 8 A connection block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following will describe the implementation methods of this application in detail with reference to the accompanying drawings and examples, so that the application can fully understand how technical means are used to solve technical problems and achieve corresponding technical effects, and implement them accordingly. The embodiments of this application and the various features therein can be combined with each other without conflict, and the resulting technical solutions are all within the scope of protection of this application.

[0046] It should also be noted that, for the convenience of description, only the parts related to the present disclosure are shown in the drawings.

[0047] Example 1

[0048] This embodiment discloses a Q-value estimation method based on a deep neural network.

[0049] Due to the underground propagation medium, i.e. the inhomogeneous elastic medium of the formation, the seismic wave will produce attenuation in the propagation process, including the attenuation of amplitude and phase, and the higher the frequency component, the more serious the attenuation in the propagation process. The attenuation process of seismic data seriously affects the resolution of seismic data, and then affects the subsequent seismic data processing and interpretation work, causing the decline of seismic imaging accuracy. In order to better describe the attenuation process of seismic data, the quality factor Q is introduced to quantify the attenuation process of seismic data. The estimation of Q value is mainly divided into two categories. One is to estimate the Q value according to the centroid frequency shift of the attenuation signal, and the other is to use different Q values to perform sparse inversion on the seismic signal, and compare the sparsity of the inversion coefficient to obtain the approximate equivalent Q value. Using the centroid frequency shift of the attenuation signal to estimate the Q value requires higher resolution of the data, and needs to know the accurate wavelet. When processing actual seismic data, the estimated wavelet is not accurate, and the distribution of the formation is irregular, which causes the irregularity of the centroid frequency shift corresponding to different Q values, and then affects the accuracy of the Q value estimation. The other method uses different Q values to perform sparse inversion on the seismic signal to obtain the reflection coefficient sequence, and compares the sparsity of the reflection coefficient to obtain the estimated Q value. This method also needs to know the wavelet, and due to the existence of noise, the sparsity of the reflection coefficient is affected, causing the instability of the amplitude compensation, so it is necessary to select a sparsity criterion to balance the influence of the existence of weak noise and the change of amplitude on the calculation of the sparsity of the reflection coefficient. The commonly used sparsity criterion is the l p norm criterion, usually taking p = 0.5. Due to the influence of noise level on sparsity and compensation amplitude, the l p norm criterion shows instability. Accurate Q value estimation not only plays a significant role in effectively improving the resolution of seismic data and then improving the accuracy of seismic imaging, but also helps oil and gas prediction. A Q value estimation method that is more accurate, stable and has good extrapolation performance is urgently needed.

[0050] To solve the above problems, the application provides a Q value estimation method based on a deep neural network, which can obtain relatively stable and accurate Q value estimation results on synthetic seismic data and be applied to actual seismic data. Since the amplitude of actual seismic data will be affected in the process of noise suppression, the method is based on phase correction and multiple features of seismic signals. The advantage of using the phase correction based method is that it has good anti-interference performance to noise, and does not depend on the information of the seismic wavelet, so it does not need to eliminate the influence of the effective frequency band information of the wavelet on the Q value estimation result, and therefore has strong adaptability to actual seismic data.

[0051] Figure 1 A flow chart of a Q value estimation method based on a deep neural network provided by the embodiment of the application is shown in Figure 1 The method disclosed by the embodiment comprises the following steps:

[0052] Step 110, training the deep neural network according to the generated steady-state seismic signal to obtain a trained deep neural network.

[0053] In some embodiments, the training the deep neural network according to the generated steady-state seismic signal to obtain a trained deep neural network comprises:

[0054] generating a one-dimensional steady-state seismic signal according to a preset minimum phase wavelet and a preset random sparse reflection coefficient sequence;

[0055] generating a first Q value sequence in a preset range, and respectively performing attenuation processing on the one-dimensional steady-state seismic signal according to different Q values in the first Q value sequence to obtain a non-steady-state seismic signal;

[0056] generating a second Q value sequence for attenuation compensation, and respectively performing attenuation compensation processing on the non-steady-state seismic signal according to Q values in the second Q value sequence to obtain a compensated seismic signal;

[0057] extracting a plurality of different features from the compensated seismic signal to construct a feature matrix;

[0058] performing normalization processing on the feature matrix to obtain training data;

[0059] using a preset real Q value as label data, training a deep neural network using the training data and the label data to obtain the trained deep neural network.

[0060] The preset minimum phase wavelet comprises a minimum phase wavelet, and the preset random sparse reflection coefficient comprises a random sparse reflection coefficient satisfying a Bernoulli-Gaussian distribution.

[0061] Optionally, the minimum phase wavelet is generated first, and then the random sparse reflection coefficient sequence satisfying the Bernoulli-Gaussian distribution is generated, and the one-dimensional steady-state seismic signal is synthesized using the wavelet and the reflection coefficient sequence.

[0062] In some embodiments, before the obtaining a trained deep neural network, the method further comprises:

[0063] In some embodiments, in the process of performing the attenuation processing, the attenuation model comprises a Kolsky-Futterman model.

[0064] In some embodiments, the attenuation formula of the seismic signal under the Kolsky-Futterman model comprises:

[0065]

[0066] wherein U is a seismic signal, f is a frequency, t is a time, t0 is a reference time, f r is a reference frequency, Q is a quality factor, and i is an imaginary unit.

[0067] In some possible cases, the reference time and the reference frequency can be selected and determined according to actual requirements.

[0068] Optionally, the preset range includes a reasonable range (which can be set according to actual requirements). A Q value sequence in the reasonable range is constructed, and the steady-state seismic signal is attenuated by using different Q values in the sequence, to obtain a non-steady-state seismic signal. The attenuation model uses the most commonly used Kolsky-Futterman model.

[0069] In some embodiments, the attenuation compensation processing is performed in a phase correction manner, wherein the phase correction is performed by the following formula:

[0070]

[0071] wherein U is a seismic signal, τ is a time, ω is a circular frequency, ω γ is a reference circular frequency, γ is an attenuation factor, and i is an imaginary unit.

[0072] In some possible cases, the reference circular frequency can be selected and determined according to actual requirements.

[0073] Optionally, a Q value sequence for attenuation compensation is constructed, and each Q value in the sequence is traversed, and the non-steady-state seismic signal is phase-corrected by using the current Q value. Since the amplitude correction process is prone to instability and is sensitive to noise, the pure phase correction method is used for attenuation compensation of the non-steady-state seismic signal. The phase correction method is described as follows:

[0074]

[0075] wherein, ω r is a reference frequency.

[0076] In some embodiments, when a plurality of different features are extracted from the compensated seismic signal, the feature value extraction manner includes obtaining non-Gaussianity of the compensated seismic signal by using different non-Gaussian criteria.

[0077] Further, a plurality of different features are extracted from the phase-corrected seismic signal, and the feature value extraction manner is to obtain non-Gaussianity of the compensated seismic signal by using different non-Gaussian criteria, which are l1-norm, l p-norm, p = 0.5, 0.2, 0.1, 0.05, and Hoyer criterion, thereby constituting a feature matrix, each column of the matrix corresponding to a different feature extracted after compensating the seismic signal using a separate Q value.

[0078] The feature matrix cannot be directly trained as a training sample. To ensure the uniformity of the influence of each criterion in the training process, each row of the feature matrix is normalized, thereby obtaining the training sample of the neural network. The preset true Q value includes the true Q value, and the neural network is trained using the true Q value as a label.

[0079] The one-dimensional steady-state seismic signal is modified, and new training data is generated according to the modified one-dimensional steady-state seismic signal. The deep neural network is trained again one or more times through the new training data.

[0080] Further, the wavelet used to synthesize the steady-state seismic signal, the sparsity of the reflection coefficient sequence, and the degree of added Gaussian noise are changed respectively, new test data is regenerated, and the neural network is tested and trained again one or more times.

[0081] Step 120: The trained deep neural network is trained again according to the seismic data to be estimated, and an optimized deep neural network is obtained.

[0082] Optionally, for actual seismic data to be estimated, the foregoing training steps are re-executed to train the trained deep neural network again, and an optimized deep neural network is obtained.

[0083] Step 130: The Q value of the seismic data to be estimated is predicted by the optimized deep neural network.

[0084] Optionally, the Q value is estimated by the optimized deep neural network, and then the actual seismic data is compensated for attenuation using the estimated Q value.

[0085] To facilitate understanding of the technical solutions of the present application, reference can be made to Figure 2 and Figure 3 wherein, Figure 2 is a flowchart of a Q value prediction network training sample preparation provided by an embodiment of the present application; Figure 3 is a flowchart of a deep neural network-based Q value estimation provided by an embodiment of the present application.

[0086] The Q value estimation method based on the deep neural network provided in the embodiment comprises: training a deep neural network according to a generated steady-state seismic signal, obtaining a trained deep neural network; re-training the trained deep neural network according to to-be-estimated seismic data, obtaining an optimized deep neural network; and predicting a Q value of the to-be-estimated seismic data through the optimized deep neural network. Based on the phase correction and multiple characteristics of the seismic signal, the method based on the phase correction has the advantages of better anti-interference to noise and independence from information of a seismic wavelet, and does not need to eliminate the influence of effective frequency band information of the wavelet on the Q value estimation result, and therefore has strong adaptability to actual seismic data.

[0087] Embodiment two

[0088] The embodiment provides a specific example based on the embodiment one.

[0089] In the embodiment, an actual seismic data with an obvious attenuation reservoir is used as an example, as shown in Figure 4 .

[0090] As an example, the method can specifically comprise the following processing procedure:

[0091] (1) First, a minimum phase wavelet is generated, and then a random sparse reflection coefficient sequence satisfying a Bernoulli-Gaussian distribution is generated, and a one-dimensional steady-state seismic signal is synthesized by using the wavelet and the reflection coefficient sequence.

[0092] (2) Then, a Q value sequence in a reasonable range is constructed, and the steady-state seismic signal is attenuated by using different Q values in the sequence, to obtain a non-steady-state seismic signal. The attenuation model can use the most commonly used Kolsky-Futterman model, and the attenuation formula of the seismic signal in the model can be expressed as:

[0093]

[0094] (3) A Q value sequence for attenuation compensation is constructed, each Q value in the sequence is traversed, and the seismic signal obtained in the step (2) is phase-corrected by using the current Q value. Since an unstable situation is prone to occur in the amplitude correction process and the amplitude correction is sensitive to noise, a pure phase correction method is used when the non-steady-state seismic signal is attenuated and compensated, and the phase correction can be performed through the following formula:

[0095]

[0096] wherein, ω r is a reference frequency.

[0097] A plurality of different features are extracted from the phase-corrected seismic signal, and the feature value extraction method is to calculate the non-Gaussianity of the compensated seismic signal using different non-Gaussianity criteria, which are l1-norm, l p -norm, p = 0.5, 0.2, 0.1, 0.05, and Hoyer criterion, thereby forming a feature matrix, each column of the matrix corresponding to a different feature extracted after compensating the seismic signal using a single Q value.

[0098] The feature matrix cannot be directly used as a training sample for training. To ensure the uniformity of the influence of each criterion in the training process, each row of the feature matrix is normalized to obtain the training sample of the neural network; the real Q value is used as the label, and the neural network is trained.

[0099] (4) Change the wavelet used to synthesize the steady-state seismic signal, the sparsity of the reflection coefficient sequence, and the degree of added Gaussian noise, respectively, repeat steps (1), (2), and (3), generate test data, and test the neural network trained in step (3).

[0100] (5) Repeat steps (2) and (3) for actual seismic data, and use the neural network trained in step (3) to estimate the Q value, and use the estimated Q value to compensate the actual seismic data for attenuation.

[0101] In this embodiment, the label of the training sample is given by the sequence number of the column in the feature matrix corresponding to the real Q value used when attenuating the seismic signal, that is, the training samples are classified according to the position of the real Q in the Q value sequence, and the number of categories is the same as the length of the Q value sequence.

[0102] Among them, the training sample can refer to Figure 5 .

[0103] In this embodiment, the feature matrix is generated using actual two-dimensional seismic data, and the feature matrix set corresponding to the two-dimensional seismic data is constrained in the horizontal direction, that is, the feature matrix corresponding to each single seismic trace is obtained by averaging the feature matrices of the left and right adjacent seismic traces, thereby adding spatial constraint information to the feature matrix and further improving the horizontal continuity of the Q value estimation result.

[0104] The feature matrix of the two-dimensional seismic data is used to estimate the Q value of the seismic data, and the predicted Q value is as shown in Figure 6 .

[0105] Since the fluid in the fracture-vug has a serious signal attenuation, the application can accurately estimate the attenuation information or Q value in the fracture-vug, which is the key to finely characterize the fracture-vug and finely identify the fluid in the fracture-vug. A large number of label samples available for training of the deep neural network are obtained by using the synthetic seismic data, the deep neural network trained by using the label samples is applied to the actual seismic data, has a good anti-interference performance to noise, and does not depend on the information of the seismic wavelet, and has a strong self-adaptability to the actual seismic data.

[0106] Embodiment three

[0107] The embodiment of the device can be used to execute the method embodiments of the application. For details not disclosed in the device embodiment, please refer to the method embodiments of the application. Figure 7 A structural schematic diagram of a device provided by the embodiment of the application is shown in Figure 7 The device 700 disclosed by the embodiment includes:

[0108] The training module 701 is configured to train the deep neural network according to the generated steady-state seismic signal, and obtain a trained deep neural network.

[0109] The optimization training module 702 is configured to retrain the trained deep neural network according to the to-be-estimated seismic data, and obtain an optimized deep neural network.

[0110] The prediction module 703 is configured to predict the Q value of the to-be-estimated seismic data by using the optimized deep neural network.

[0111] In some embodiments, the training module 701 includes a steady-state seismic signal generation unit, a non-steady-state seismic signal generation unit, a compensated seismic signal determination unit, an extraction unit, a normalization unit, and a training unit.

[0112] The steady-state seismic signal generation unit is configured to generate a one-dimensional steady-state seismic signal according to a preset minimum phase wavelet and a preset random sparse reflection coefficient sequence.

[0113] The non-steady-state seismic signal generation unit is configured to generate a first Q value sequence in a preset range, and perform attenuation processing on the one-dimensional steady-state seismic signal according to different Q values in the first Q value sequence, to obtain a non-steady-state seismic signal.

[0114] The compensated seismic signal determination unit is configured to generate a second Q value sequence for attenuation compensation, and perform attenuation compensation processing on the non-steady-state seismic signal according to Q values in the second Q value sequence, to obtain a compensated seismic signal.

[0115] The extraction unit is configured to extract a plurality of different features from the compensated seismic signal to construct a feature matrix.

[0116] a normalization unit, configured to perform normalization on the feature matrix to obtain training data;

[0117] a training unit, configured to use a preset real Q value as label data, and train a deep neural network using the training data and the label data to obtain the trained deep neural network.

[0118] In some embodiments, the retraining unit further includes, before the trained deep neural network is obtained, modifying the one-dimensional steady-state seismic signal, generating new training data according to the modified one-dimensional steady-state seismic signal, and retraining the deep neural network one or more times using the new training data.

[0119] In some embodiments, in the process of performing the attenuation processing, the attenuation model includes a Kolsky-Futterman model.

[0120] In some embodiments, the attenuation formula of the seismic signal under the Kolsky-Futterman model includes:

[0121]

[0122] wherein U is the seismic signal, f is the frequency, t is the time, t0 is the reference time, f r is the reference frequency, Q is the quality factor, and i is the imaginary unit.

[0123] In some embodiments, the attenuation compensation processing is performed by phase correction, wherein the phase correction is performed by the following formula:

[0124]

[0125] wherein U is the seismic signal, τ is the time, ω is the circular frequency, ω γ is the reference circular frequency, γ is the attenuation factor, and i is the imaginary unit.

[0126] In some embodiments, in the process of extracting a plurality of different features from the compensated seismic signal, the feature value extraction manner includes obtaining the non-Gaussianity of the compensated seismic signal using different non-Gaussianity criteria.

[0127] Those skilled in the art can understand that, Figure 7 The structure shown in the above embodiments is not a limitation on the device of the present application, and can include more or less modules / cells than the figure, or combine certain modules / cells, or different module / cell arrangement.

[0128] Those skilled in the art should understand that the modules or steps of the present application described above can be realized by general computing devices, which can be centralized on a single computing device or distributed on a network composed of multiple computing devices. Alternatively, they can be realized by program codes executable by computing devices, so that they can be stored in storage devices and executed by computing devices, and in some cases, the steps shown or described herein can be executed in different order, or they can be made into individual integrated circuit modules, or multiple modules or steps thereof can be made into a single integrated circuit module.

[0129] The device provided by the embodiment comprises: a training module 701, configured to train a deep neural network according to a generated steady-state seismic signal, and obtain a trained deep neural network; an optimization training module 702, configured to retrain the trained deep neural network according to to-be-estimated seismic data, and obtain an optimized deep neural network; and a prediction module 703, configured to predict a Q value of the to-be-estimated seismic data by using the optimized deep neural network. Based on phase correction and multiple characteristics of seismic signals, the method based on phase correction has the advantage of better anti-interference to noise, and does not depend on information of a seismic wavelet, and does not need to eliminate the influence of effective frequency band information of the wavelet on a Q value estimation result, so that the method has strong adaptability to actual seismic data.

[0130] Embodiment Four

[0131] The embodiment further provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the method steps in the foregoing method embodiments, which will not be repeated here.

[0132] The computer-readable storage medium can also include, or be, alone, a computer program, a data file, a data structure, etc., or a combination thereof. The computer-readable storage medium or the computer program can be specifically designed and understood by those skilled in the computer software field, or can be known and available to those skilled in the computer software field. Examples of the computer-readable storage medium include: magnetic media, such as a hard disk, a floppy disk, and a magnetic tape; optical media, such as a CD ROM disk and a DVD; a magneto-optical medium, such as an optical disk; and a hardware device specifically configured to store and execute a computer program, such as a read-only memory (ROM), a random access memory (RAM), a flash memory; or a server, an app application store, etc. Examples of the computer program include machine code (e.g., code generated by a compiler) and a file containing high-level code that can be executed by a computer by using an interpreter. The described hardware device can be configured to function as one or more software modules to perform the above-described operations and methods, and vice versa. In addition, the computer-readable storage medium can be distributed in a networked computer system, and the program code or computer program can be stored and executed in a distributed manner.

[0133] Embodiment Five

[0134] Figure 8 A connection block diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 8. The electronic device 800 can include one or more processors 801, a memory 802, a multimedia component 803, an input / output (I / O) interface 804, and a communication component 805. Figure 8

[0135] The one or more processors 801 are configured to perform all or part of the steps in the above-described method embodiments. The memory 502 is configured to store various types of data, which can include, for example, instructions of any application program or method in the electronic device, and application-related data.

[0136] ​The one or more processors 801 can be implemented with an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a controller, a microcontroller, a microprocessor or other electronic devices, for performing the methods as described in the aforementioned method embodiments.

[0137] The memory 802 can be implemented with any type of volatile or nonvolatile storage devices or a combination thereof, such as a Static Random Access Memory (SRAM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), an Erasable Programmable Read-Only Memory (EPROM), a Programmable Read-Only Memory (PROM), a Read-Only Memory (ROM), a magnetic storage, a flash memory, a magnetic disk, or a compact disk.

[0138] The multimedia component 803 can include a screen, which can be a touch screen, and audio components for outputting and / or inputting audio signals. For example, the audio components can include a microphone for receiving external audio signals. The received audio signals can be further stored in the memory or transmitted through the communication component. The audio components also include at least one speaker for outputting audio signals.

[0139] The I / O interface 804 provides an interface between the one or more processors 801 and other interface modules, which can be a keyboard, a mouse, a button, and the like. The buttons can be virtual buttons or physical buttons.

[0140] The communication component 805 is configured to perform wired or wireless communication between the electronic device 800 and other devices. The wired communication includes communication through a network port, a serial port, etc., and the wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, 5G, or a combination of one or more of them. Therefore, the corresponding communication component 805 can include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0141] In summary, the application provides a Q value estimation method and device based on a deep neural network, a storage medium, and an electronic device. The method includes training a deep neural network according to a generated steady-state seismic signal to obtain a trained deep neural network, retraining the trained deep neural network according to to-be-estimated seismic data to obtain an optimized deep neural network, and predicting a Q value of the to-be-estimated seismic data through the optimized deep neural network. Based on phase correction and multiple characteristics of seismic signals, the method based on phase correction has better anti-interference performance to noise and does not depend on information of a seismic wavelet, and does not need to eliminate the influence of effective frequency band information of the wavelet on a Q value estimation result, thus having strong adaptability to actual seismic data.

[0142] It should also be understood that the methods and systems disclosed in the embodiments of the application can also be implemented in other ways. The above-described method and system embodiments are only illustrative, for example, the flowchart and block diagram in the drawings show the possible implementation architecture, function and operation of the method and device according to the embodiments of the application. In this regard, each block in the flowchart or block diagram can represent a module, a computer program segment or a part of a computer program, which includes one or more computer programs for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the block can occur in different orders from those noted in the drawings, and can actually be executed in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer programs.

[0143] In this application, the terms "comprise", "contain", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, the element defined by the phrase "comprises a" does not exclude the presence of additional identical elements in the process, method, device, or apparatus that includes the element; if there is a description of "first", "second", etc., it is only for the purpose of description and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the sequence of indicated technical features; in the description of the present application, unless otherwise specified, the term "a plurality of" or "a plurality" means at least two; if there is a description of a server, it should be noted that the server can be a stand-alone physical server or terminal, or a server cluster composed of multiple physical servers, or a cloud server capable of providing cloud server, cloud database, cloud storage and CDN and other basic cloud computing services; if there is a description of a smart terminal or a mobile device in the present application, it should be noted that the smart terminal or mobile device can be a mobile phone, a tablet computer, a smart watch, a netbook, a wearable electronic device, a personal digital assistant (PDA), an augmented reality technology device (AR), a virtual reality device (VR), a smart television, a smart sound, a personal computer (PC), etc., but is not limited thereto, and the specific form of the smart terminal or mobile device is not specially limited in the present application.

[0144] Finally, it should be noted that in the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "one example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0145] Although the embodiments of the present application have been shown and described above, it is understood that all the above-described embodiments are exemplary only, the contents described are merely adopted for the purpose of facilitating the understanding of the present application, and are not intended to limit the present application. Any person skilled in the art to which the present application belongs can make any modification and change in the form and details without departing from the spirit and scope of the present application, but the protection scope of the present application shall be subject to the scope defined by the appended claims.

Claims

1. A Q-value estimation method based on deep neural network, characterized in that: The method comprises: The deep neural network is trained according to the generated steady-state seismic signal to obtain a trained deep neural network; including: generating a one-dimensional steady-state seismic signal according to a preset minimum phase wavelet and a preset random sparse reflection coefficient sequence; generating a first Q value sequence within a preset range, and performing attenuation processing on the one-dimensional steady-state seismic signal according to different Q values ​​in the first Q value sequence to obtain a non-steady-state seismic signal; generating a second Q value sequence for attenuation compensation, and performing attenuation compensation processing on the non-stationary seismic signal according to the Q values ​​in the second Q value sequence to obtain a compensated seismic signal; extracting a plurality of different features from the compensated seismic signal to construct a feature matrix; Normalizing the feature matrix to obtain training data; Using the preset true Q value as label data, and using the training data and the label data to train the deep neural network to obtain the trained deep neural network; Retraining the trained deep neural network according to the earthquake data to be estimated to obtain an optimized deep neural network; The Q value of the seismic data to be estimated is predicted by the optimized deep neural network.

2. The method according to claim 1, characterized in that Before obtaining the trained deep neural network, the method further includes: The one-dimensional steady-state seismic signal is modified, new training data is generated according to the modified one-dimensional steady-state seismic signal, and the deep neural network is trained again one or more times using the new training data.

3. The method according to claim 1, characterized in that During the attenuation process, the attenuation model includes the Kolsky-Futterman model.

4. The method according to claim 3, characterized in that The attenuation formula of the seismic signal under the Kolsky-Futterman model includes: Where U is the earthquake signal, f is the frequency, t For time, is the reference time, is the reference frequency, is the quality factor, Is an imaginary unit.

5. The method according to claim 1, wherein Attenuation compensation is performed by phase correction, where phase correction is performed using the following formula: Where U is the earthquake signal, For time, is the circular frequency, is the reference circular frequency, is the attenuation factor, Is an imaginary unit.

6. The method according to claim 1, characterized in that When extracting a plurality of different features from the compensated seismic signal, the feature value extraction method includes obtaining the non-Gaussianity of the compensated seismic signal using different non-Gaussianity criteria.

7. A Q-value estimation device based on a deep neural network, characterized in that: include: A training module, used to train the deep neural network according to the generated steady-state seismic signal to obtain a trained deep neural network; The training module includes: a steady-state seismic signal generating unit, a non-steady-state seismic signal generating unit, a compensated seismic signal determining unit, an extraction unit, a normalization unit and a training unit; wherein: The steady-state seismic signal generating unit is configured to generate a one-dimensional steady-state seismic signal according to a preset minimum phase wavelet and a preset random sparse reflection coefficient sequence; The non-steady-state seismic signal generating unit is configured to generate a first Q value sequence within a preset range, and perform attenuation processing on the one-dimensional steady-state seismic signal according to different Q values ​​in the first Q value sequence to obtain a non-steady-state seismic signal; The compensated seismic signal determining unit is configured to generate a second Q value sequence for attenuation compensation, and perform attenuation compensation processing on the non-stationary seismic signal according to the Q values ​​in the second Q value sequence to obtain a compensated seismic signal; The extraction unit is used to extract a plurality of different features from the compensated seismic signal to construct a feature matrix; The normalization unit is used to perform normalization processing on the feature matrix to obtain training data; A training unit, configured to use a preset true Q value as label data, and train a deep neural network using the training data and the label data to obtain the trained deep neural network; An optimization training module is used to train the trained deep neural network again according to the seismic data to be estimated to obtain an optimized deep neural network; A prediction module is used to predict the Q value of the seismic data to be estimated through the optimized deep neural network.

8. A computer-readable storage medium, characterized in that The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements the method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: The method comprises a memory and one or more processors, wherein a computer program is stored in the memory, and the memory and the one or more processors are communicatively connected to each other, and when the computer program is executed by the one or more processors, the method according to any one of claims 1 to 6 is executed.

Images