Method and device for improving seismic data resolution

Abstract

The invention discloses a method and a device for improving the resolution of seismic data. The method comprises the following steps: selecting adjacent multi-channel seismic data from original seismic record data as training samples; using the training sample to train a seismic record deconvolution network model, and obtaining a corresponding loss function in each training; if the loss function satisfies a predetermined condition, taking the model parameter of the training as a trained seismic record deconvolution network model parameter; and inputting the original seismic record into the trained seismic record deconvolution network model parameters to obtain a seismic record after seismic data resolution is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data interpretation, and in particular to a method and device for improving the resolution of seismic data. Background Art

[0002] Seismic exploration refers to a geophysical exploration method that uses elastic waves caused by artificial excitation, utilizes the differences in density and velocity of underground media, and analyzes the propagation laws of seismic waves underground to infer the lithological characteristics of the formation. Improving the resolution of seismic data through signal processing methods is of great significance for the fine interpretation of stratigraphic structure. In particular, with the advancement of exploration technology and the complexity of exploration objects, higher requirements are placed on the high-resolution processing of seismic signals. At present, seismic exploration has shifted from oil and gas reservoirs with simple structures to oil and gas reservoirs with complex stratigraphic structures such as thin layers and pores, from medium and shallow layers to deep layers, and from interpreting stratigraphic thickness to predicting porosity, shale content, formation pressure, fluid properties and other more detailed reservoir parameters. The realization of these goals must rely on the acquisition of higher-resolution seismic data and more effective high-resolution processing technology for seismic data.

[0003] Therefore, it is urgent to propose a method to improve the resolution of seismic data in order to obtain higher resolution seismic data. Summary of the invention

[0004] The inventors of this application have discovered that: in recent years, the oil and gas exploration field has been using deep learning to perform high-resolution processing of seismic data, but most of the high-resolution processing methods for seismic data based on deep learning need to rely on the high-resolution results obtained by traditional methods as labels. In some technologies, U-net is used to achieve high-resolution processing of single-channel seismic data based on supervised learning, by learning the mapping from original seismic data to high-resolution seismic data, but can only perform high-frequency information recovery problems for single-channel seismic data 1. In some technologies, seismic record deconvolution is achieved through repeated iterations and simultaneous optimization of wavelets and reflection coefficients based on deep neural networks, but the network needs to be pre-trained. In addition, the supervised sample generation process in the above method requires manpower to repeatedly compare results and adjust parameters, which cannot meet the oil and gas industry's increasing demand for automation and intelligence of algorithms in seismic data processing and interpretation operations.

[0005] In view of the above problems, the present invention proposes a method for improving the resolution of seismic data based on unsupervised learning based on the principle of deep learning. This method can improve the resolution of seismic data without labeled samples, is more adaptable to different work areas and ground conditions than supervised learning methods, is easier to grasp essential features, and is easier to develop and improve than traditional algorithms.

[0006] The present application provides a method and device for improving the resolution of seismic data. The method selects adjacent multi-channel seismic data from the original seismic record profile to be processed as training samples; constructs a seismic wavelet estimation network, a deconvolution network and a convolution network to form a seismic record deconvolution network; constructs a loss function of the seismic wavelet estimation network and a loss function of the deconvolution network; uses the obtained training samples as inputs of the seismic wavelet estimation network and the deconvolution network respectively and trains the entire seismic record deconvolution network, and obtains a reflection coefficient sequence and a wavelet estimation result that satisfy the convolution relationship by repeatedly iteratively updating the wavelet estimation network and the deconvolution network parameters. Compared with traditional methods, this method can improve the resolution of seismic data without labeled samples, is more adaptable to different work areas and ground conditions than supervised learning methods, and is easier to grasp essential features.

[0007] In a first aspect, the present application provides a method for improving the resolution of seismic data, the method comprising:

[0008] Select adjacent multi-channel seismic data from the original seismic record data as training samples;

[0009] The training samples are used to train the seismic record deconvolution network model, and a corresponding loss function is obtained in each training;

[0010] If the loss function satisfies the predetermined conditions, the model parameters of this training are used as the trained seismic record deconvolution network model parameters;

[0011] The original seismic record is input into the trained seismic record deconvolution network model to obtain the seismic record with improved seismic data resolution.

[0012] In a second aspect, an embodiment of the present invention provides a device for improving the resolution of seismic data, the device comprising: a memory and a processor; the memory is used to store a program for improving the resolution of seismic data, and the processor is used to read and execute the program for improving the resolution of seismic data, and execute any one of the methods described in the above embodiments.

[0013] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a data processing program is stored, and the data processing program is executed by a processor to perform any of the methods for improving the resolution of seismic data.

[0014] Compared with the related art, a method and device for improving the resolution of seismic data, the method includes: selecting adjacent multi-channel seismic data from the original seismic record data as training samples; using the training samples to train the seismic record deconvolution network model, and obtaining a corresponding loss function for each training; if the loss function meets the predetermined conditions, the model parameters of this training are used as the parameters of the trained seismic record deconvolution network model; the original seismic record is input into the trained seismic record deconvolution network model to obtain a high-resolution seismic record. The present application implements a method for improving the resolution of seismic data based on unsupervised learning, which can improve the resolution of seismic data in the absence of labeled samples, is more adaptable to different work areas and terrain conditions than supervised learning methods, is easier to grasp the essential characteristics, and is easier to develop and improve than traditional algorithms. It is a potential solution to the digital transformation of oil and gas companies.

[0015] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by implementing the present application. Other advantages of the present application can be realized and obtained by the schemes described in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide an understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0017] Figure 1 is a flow chart of a method for improving the resolution of seismic data according to an embodiment of the present application;

[0018] Figure 2 is a schematic diagram of a seismic record deconvolution network model structure in some exemplary embodiments;

[0019] Figure 3 is a schematic diagram of the structure of a seismic wavelet estimation network module in some exemplary embodiments;

[0020] Figure 4 is a schematic diagram of the structure of a deconvolution network module in some exemplary embodiments;

[0021] Figure 5 is one of the actual original seismic records in some exemplary embodiments;

[0022] Figure 6 is the result of performing resolution enhancement processing on the actual original seismic record in some exemplary embodiments;

[0023] Figure 7 is a seismic wavelet obtained by increasing the resolution of an actual original seismic record in some exemplary embodiments;

[0024] Figure 8 is two of the actual original seismic records in some exemplary embodiments;

[0025] Fig. 9 is the result of performing resolution enhancement processing on the actual original seismic record in some exemplary embodiments;

[0026] Fig.10 is a seismic wavelet obtained by increasing the resolution of an actual original seismic record in some exemplary embodiments;

[0027] Fig.11 It is a schematic diagram of a device for improving the resolution of seismic data according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] The present application describes multiple embodiments, but the description is exemplary rather than restrictive, and it is obvious to those skilled in the art that there may be more embodiments and implementations within the scope of the embodiments described in the present application. Although many possible feature combinations are shown in the drawings and discussed in the specific embodiments, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.

[0029] The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features and elements disclosed in the present application may also be combined with any conventional features or elements to form a unique invention scheme defined by the claims. Any features or elements of any embodiment may also be combined with features or elements from other invention schemes to form another unique invention scheme defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in the present application may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the attached claims and their equivalents, the embodiments are not subject to other restrictions. In addition, various modifications and changes may be made within the scope of protection of the attached claims.

[0030] In addition, when describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps of the specific order described. As will be understood by those of ordinary skill in the art, other sequences of steps are also possible. Therefore, the specific sequence of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to the steps of performing them in the order written, and those skilled in the art can easily understand that these sequences can be changed and still remain within the spirit and scope of the embodiments of the present application.

[0031] The embodiment of the present invention provides a method for improving the resolution of seismic data, such as Figure 1 As shown, the method includes steps S100-S140, which are specifically as follows:

[0032] S100: Selecting adjacent multi-channel seismic data from the original seismic record data as training samples;

[0033] S110: using the training samples to train the seismic record deconvolution network model, and obtaining a corresponding loss function for each training;

[0034] S120: Determine whether the current loss function meets a predetermined condition;

[0035] S130: If the loss function satisfies a predetermined condition, the model parameters of this training are used as the trained seismic record deconvolution network model parameters, and step S140 is executed;

[0036] If the loss function does not meet the predetermined condition, iterate to perform the next model training and execute step S110;

[0037] S140: Inputting the original seismic record into the trained seismic record deconvolution network model to obtain the seismic record with improved seismic data resolution.

[0038] In an exemplary embodiment, within a predetermined time range, adjacent multi-channel seismic data are selected from the original seismic record data in a random uniform sampling manner as training samples. Assuming that the high-frequency components of the seismic wavelet decay rapidly during underground propagation, the seismic wavelet does not change much within a certain range, and the wavelet can be estimated within a time window. The time window for estimating the wavelet is generally selected to include a specific stratigraphic structure. Seismic wavelet estimation usually requires multi-channel seismic data to estimate its average amplitude spectrum, so several adjacent seismic data are input as multi-channel one-dimensional data, and their multi-channel average normalized amplitude spectrum is calculated as part of the loss function for wavelet estimation. For example: In sample selection, a random uniform sampling method is adopted, and within a selected time range, adjacent multi-channel seismic data with 256 sampling points and 256 channels are selected as training samples.

[0039] In an exemplary embodiment, the seismic record deconvolution network model includes: a seismic wavelet estimation network model, a deconvolution network model and a convolution network model.

[0040] In an exemplary embodiment, the seismic wavelet estimation network model includes: N network layers, Dense layers, Mean layers and Linear Unit layers; each network layer is composed of two convolution functions and a maximum pooling function; N is a positive integer. The seismic wavelet estimation network is equivalent to a parameter estimation network, which estimates the simulated wavelet, angular frequency and constant attenuation factor corresponding to the simulated wavelet from the original seismic record, and the estimated seismic wavelet can be obtained by substituting the following simulated wavelet formula.

[0041] The wavelet formula is:

[0042]

[0043] In the above formula, w(t) is the sub-wave, t is the time, m1 and m2 are the angular frequencies, and σ1 and σ2 are constant attenuation factors.

[0044] In order to more accurately estimate the angular frequency m and the constant attenuation factor σ, two parameters are Figure 3 As shown in the figure, two 10-layer parameter estimation subnetworks can be used to estimate m and σ respectively; the network structure design and parameter initialization method of the two parameter estimation subnetworks are the same, but the parameter updates are independent. The specific structure of the seismic wavelet estimation network is: N is 10 layers, each layer consists of two convolution functions and a maximum pooling function, and after the 10th layer, the Dense layer, Mean layer and Linear Unit layer are connected, that is, the "Dense+Mean+Linear Unit" operation is realized after the 10th layer.

[0045] In an exemplary embodiment, the deconvolution network model includes: multiple convolution layers, M basic block modules and upsampling operation modules; wherein each basic block module is composed of a convolution layer and a BN layer; and M is a positive integer. The deconvolution network includes three important structures: first, removing all pooling layers to avoid damage to effective information in the problem of improving resolution; second, using multiple residual structures and batch normalization layers (Batch Normalization, BN) to improve the convergence ability and convergence speed of the network; third, using a sub-pixel convolution layer for feature fusion, and then using a convolution layer with a step size of 2 to ensure that the input and output sizes are the same. Figure 4 The deconvolution network module shown in the figure has a structure of “convolution layer + 10 basic blocks + convolution layer + upsampling operation + 2 convolution layers” when M is 14 layers. The basic block is a basic residual unit composed of a convolution layer and a BN layer.

[0046] In an exemplary embodiment, a convolution network convolves the estimated seismic wavelet and reflection coefficient sequence in a certain manner to restore the original seismic record.

[0047] In an exemplary embodiment, a seismic record deconvolution network model is used to calculate the training samples to obtain a loss function, including:

[0048] The first step is to input the training samples into the seismic wavelet estimation network model to obtain the angular frequency and constant attenuation factor;

[0049] Step 2: According to the angular frequency and the constant attenuation factor, the seismic estimation wavelet is obtained by using the wavelet formula;

[0050] The third step is to input the training samples into the deconvolution network model to obtain the reflection coefficient corresponding to the sample data;

[0051] Step 4: Input the reflection coefficient and earthquake estimation wavelet into the convolution network model to obtain the processed earthquake record;

[0052] Step 5: Calculate based on the processed seismic records and sample data to obtain the loss function corresponding to the sample data.

[0053] In an exemplary embodiment, the loss function includes: a loss function of a seismic wavelet estimation network and a loss function of a deconvolution network.

[0054] In an exemplary embodiment, the loss function of the seismic wavelet estimation network is:

[0055]

[0056]

[0057]

[0058]

[0059] FWf j =|f(w(t))|

[0060] In the above formula, loss w is the loss function of the seismic wavelet estimation network, f(·) represents Fourier transform, N is the number of batches, M is the number of samples in a batch, i represents the i-th batch, and FW j represents the normalized amplitude spectrum of the wavelet, FD i,j represents the multi-channel average normalized amplitude spectrum of the seismic record, λ is a non-negative constant proportional coefficient, j=1,…,k0 represents the effective frequency band of the seismic signal, j=1,…,k1 represents the low frequency band of the seismic signal, Fd i,j represents the multi-channel average amplitude spectrum of the seismic record, d k (t) represents earthquake record, Fw j represents the wavelet amplitude spectrum, and w(t) represents the seismic wavelet.

[0061] From the above function formula, it can be seen that the loss function of the seismic wavelet estimation network is divided into two terms: the first term is the constraint term, which means that the multi-channel average normalized amplitude spectrum of the seismic record in the effective frequency band is similar to the normalized amplitude spectrum of the seismic wavelet; the second term uses the relu function as the penalty term, which means that the spectrum of the seismic wavelet in the effective low-frequency band is the envelope of the amplitude spectrum of the seismic record.

[0062] In an exemplary embodiment, the loss function of the deconvolution network is:

[0063]

[0064]

[0065]

[0066] In the above formula, N represents the batch number, M represents the number of samples in a batch, T represents the sampling point of each sample, and λ represents a non-negative constant adjustment factor. is the estimated result of the kth frequency band of the j'th sample in the i-th batch, is the high-resolution processing result of the kth frequency band of the j'th sample in the i-th batch, F(·) represents the multi-channel average amplitude spectrum, f(·) represents the Fourier transform, and F n (·) represents the multi-channel average normalized amplitude spectrum, r estRepresents the reflection coefficient output by the frequency domain deconvolution network module, r lable represents the initial reflection coefficient obtained by the frequency domain deconvolution method.

[0067] The frequency domain deconvolution method is a traditional deconvolution method. The input of the frequency domain deconvolution method is the original seismic record and the seismic wavelet, wherein the wavelet is the seismic wavelet estimated by the seismic wavelet estimation network module. The deconvolution module takes the original seismic record and the original estimated wavelet as input and outputs the corresponding reflection coefficient.

[0068] In an exemplary embodiment, the obtained training samples are used as inputs of the seismic wavelet estimation network and the deconvolution network respectively, and the entire seismic record deconvolution network is trained. The wavelet estimation network and the deconvolution network parameters are updated through repeated iterations. If the loss function meets the predetermined conditions, the model parameters of this training are used as the trained seismic record deconvolution network model parameters to obtain the reflection coefficient sequence and the wavelet estimation result that meet the convolution relationship. It can include: the seismic record deconvolution network, using the pre-trained seismic wavelet estimation network and the deconvolution network to estimate the wavelet and the reflection coefficient respectively, and then the convolution network synthesizes the original seismic record according to the seismic convolution principle. The convolution network does not participate in the training, and the final seismic record deconvolution result is obtained by repeatedly iteratively updating the wavelet estimation network and the deconvolution network parameters.

[0069] Compared with the traditional method for improving the resolution of seismic data, the method for improving the resolution of seismic data based on unsupervised learning implemented in this application has the following technical effects:

[0070] (1) In the absence of labeled samples, the resolution of seismic data can be improved;

[0071] (2) It is more adaptable to different work areas and terrain conditions than supervised learning methods and is easier to grasp essential features;

[0072] (3) Easier to develop and improve than traditional algorithms.

[0073] In the second aspect, an embodiment of the present invention provides a device for improving the resolution of seismic data, the device comprising: a memory S1101 and a processor S1102; the memory S1101 is used to store a program for improving the resolution of seismic data, and the processor S1102 is used to read and execute the program for improving the resolution of seismic data, and execute any one of the methods described in the above embodiments.

[0074] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a data processing program is stored, and the data processing program is executed by a processor to perform any of the methods for improving the resolution of seismic data.

[0075] Example 1

[0076] The present invention provides a method for improving the resolution of seismic data based on unsupervised learning, and the specific implementation process is as follows:

[0077] Step 1. Select adjacent multi-channel seismic data from the original seismic record section to be processed as training samples;

[0078] The adjacent multi-channel seismic data in the original seismic record section to be processed as training samples may include:

[0079] Assuming that the high-frequency components of seismic wavelets decay rapidly during underground propagation, and seismic wavelets do not change much within a certain range, the wavelets can be estimated within a time window. The time window for estimating wavelets is generally selected to include a specific stratigraphic structure. Seismic wavelet estimation usually requires multi-channel seismic data to estimate its average amplitude spectrum. Therefore, several adjacent seismic data are input as multi-channel one-dimensional data. In sample selection, random uniform sampling is adopted. Within the selected time range, adjacent multi-channel seismic data with 256 sampling points and 256 channels are selected as training samples.

[0080] Step 2. Construct a seismic wavelet estimation network, a deconvolution network and a convolution network to form a seismic record deconvolution network; Figure 2 Schematic diagram of the structure of the seismic record deconvolution network model shown.

[0081] Step 21. Construct seismic wavelet estimation network

[0082] The seismic wavelet estimation network model includes: N network layers, Dense layers, Mean layers and Linear Unit layers;

[0083] Each network layer consists of two convolution functions and one maximum pooling function; N is a positive integer. Figure 3 The seismic wavelet estimation network module shown in FIG. 1 has a specific structure as follows when N is 10: there are 10 layers in total, each layer consists of two convolutions and a maximum pooling function, and a "Dense+Mean+Linear Unit" operation is connected after the 10th layer.

[0084] The seismic wavelet estimation network model is equivalent to a parameter estimation network model, which estimates the angular frequency m and constant attenuation factor σ corresponding to the simulated wavelet from the original seismic record, and brings these two parameters into the simulated wavelet formula to obtain the estimated seismic wavelet;

[0085] The wavelet formula is:

[0086]

[0087] In the above formula, w(t) is the sub-wave, t is the time, m1 and m2 are the angular frequencies, and σ1 and σ2 are the attenuation factors.

[0088] Step 22. Construct a deconvolution network

[0089] The deconvolution network model includes: multiple convolution layers, M basic block modules and an upsampling operation module;

[0090] Among them, each basic block module is composed of a convolutional layer and a BN layer; M is a positive integer.

[0091] like Figure 4 The deconvolution network module shown in the figure has a structure of “convolution layer + 10 basic blocks + convolution layer + upsampling operation + 2 convolution layers” when M is 14 layers. The basic block is a basic residual unit composed of a convolution layer and a BN layer.

[0092] The deconvolution network includes three important structures: First, all pooling layers are removed to avoid damage to effective information when improving resolution; second, multiple residual structures and batch normalization layers (Batch Normalization, BN) are used to improve the convergence ability and speed of the network; third, sub-pixel convolution layers are used for feature fusion, and a convolution layer with a step size of 2 is used afterwards to ensure that the input and output sizes are the same.

[0093] Step 23. Construct a convolutional network model

[0094] The convolution network model is: convolute the estimated seismic wavelet and reflection coefficient sequence in a certain way to restore the original seismic record.

[0095] Step 3. Using the training samples to train the seismic record deconvolution network model;

[0096] Step 31. Using training samples to train the seismic wavelet estimation network model;

[0097] Step 311. Input the training sample into the seismic wavelet estimation network model to obtain the angular frequency m and the constant attenuation factor σ;

[0098] Step 312. Obtain the earthquake estimation wavelet using the wavelet formula according to the obtained angular frequency and constant attenuation factor;

[0099] The formula for estimating the wavelet is:

[0100]

[0101] Step 32: using the training samples to train the deconvolution network model;

[0102] Inputting the training sample into the deconvolution network model to obtain a reflection coefficient corresponding to the sample data;

[0103] Step 33. Input the reflection coefficient obtained in step 32 and the earthquake estimation wavelet obtained in step 31 into the convolution network model to obtain a processed earthquake record;

[0104] Step 4. Perform calculations based on the processed seismic records and the sample data to obtain a loss function corresponding to the sample data.

[0105] Step 41. Loss function of the seismic wavelet estimation network

[0106] The loss function of the seismic wavelet estimation network is divided into two terms: the first term is the constraint term, which means that the multi-channel average normalized amplitude spectrum of the seismic record in the effective frequency band is similar to the normalized amplitude spectrum of the seismic wavelet; the second term uses the relu function as the penalty term, which means that the spectrum of the seismic wavelet in the effective low-frequency band is the envelope of the amplitude spectrum of the seismic record.

[0107] The loss function of the seismic wavelet estimation network is:

[0108]

[0109]

[0110]

[0111]

[0112] f j =|f(w(t))|

[0113] In the above formula, loss w is the loss function of the seismic wavelet estimation network, f(·) represents Fourier transform, N is the number of batches, M is the number of samples in a batch, i represents the i-th batch, and FW j represents the normalized amplitude spectrum of the wavelet, FD i,j represents the multi-channel average normalized amplitude spectrum of the seismic record, λ is a non-negative constant proportional coefficient, j=1,…,k0 represents the effective frequency band of the seismic signal, j=1,…,k1 represents the low frequency band of the seismic signal, Fd i,j represents the multi-channel average amplitude spectrum of the seismic record, d k (t) represents earthquake record, Fw j represents the wavelet amplitude spectrum, and w(t) represents the seismic wavelet.

[0114] Step 42. Loss function of the deconvolution network

[0115] The loss function of the deconvolution network is divided into the following two parts, which respectively impose time domain and frequency domain constraints on the estimation results according to the frequency domain deconvolution formula:

[0116]

[0117]

[0118]

[0119] Where f(·) represents Fourier transform, F n (·) represents the multi-channel average normalized amplitude spectrum, r lable represents the high-resolution processing result of the frequency domain deconvolution method; i represents the batch number from 1 to N, j = 1, ..., k0 represents the effective frequency band of the seismic signal, j = 1, ..., k1 represents the low frequency band of the seismic signal; r est Represents the reflection coefficient output by the frequency domain deconvolution network module, r lable Represents the initial reflection coefficient obtained by the frequency domain deconvolution method; N represents the batch number, M represents the number of samples in a batch, T represents the sampling point of each sample, and λ represents a non-negative constant adjustment factor.

[0120] Step 5. Obtain a loss function that satisfies the predetermined conditions, and use the model parameters of this training as the trained seismic record deconvolution network model parameters;

[0121] Step 51. If the loss function meets the predetermined conditions, the model parameters of this training are used as the trained seismic record deconvolution network model parameters;

[0122] Step 52. If the loss function does not meet the predetermined conditions, the next model training is performed again to update the parameters of the seismic wavelet estimation network and the deconvolution network.

[0123] Step 6. Input the original seismic record into the trained seismic record deconvolution network model to obtain a high-resolution seismic record.

[0124] like Figure 5 Shown is one of the actual original earthquake records. Figure 6 For Figure 5 Schematic diagram of the resolution enhancement processing results corresponding to the actual original seismic records. Figure 7 For Figure 5 The seismic wavelet obtained after improving the resolution corresponding to the actual original seismic record in the image can be seen from this that the unclear event axis information in the original seismic data has been restored through the method of the present application, the resolution of the seismic record has been significantly improved, and it is effective for high-resolution processing of actual data.

[0125] like Figure 8 The following is the second actual original earthquake record. Fig. 9 For Figure 8 Schematic diagram of the resolution enhancement processing results corresponding to the actual original seismic records. Fig.10 For Figure 8 The seismic wavelet obtained after improving the resolution corresponding to the actual original seismic record in the image can be seen from this that the unclear event axis information in the original seismic data has been restored through the method of the present application, the resolution of the seismic record has been significantly improved, and it is effective for high-resolution processing of actual data.

[0126] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

Claims

1. A method for improving the resolution of seismic data, characterized in that: The method comprises: Select adjacent multi-channel seismic data from the original seismic record data as training samples; The training samples are used to train the seismic record deconvolution network model, and a corresponding loss function is obtained in each training; If the loss function satisfies the predetermined conditions, the model parameters of this training are used as the trained seismic record deconvolution network model parameters; The original seismic record is input into the trained seismic record deconvolution network model to obtain the seismic record with improved seismic data resolution.

2. The method for improving the resolution of seismic data according to claim 1, characterized in that: The step of selecting adjacent multi-channel seismic data from the original seismic record data as training samples includes: Within a predetermined time range, adjacent multi-channel seismic data are selected from the original seismic record data by random uniform sampling as training samples.

3. The method for improving the resolution of seismic data according to claim 1, characterized in that: The seismic record deconvolution network model includes: a seismic wavelet estimation network model, a deconvolution network model and a convolution network model.

4. The method for improving the resolution of seismic data according to claim 3, characterized in that: The seismic wavelet estimation network model includes: N network layers, a Dense layer, a Mean layer and a Linear Unit layer; Among them, each network layer is composed of two convolution functions and one maximum pooling function; N is a positive integer.

5. The method for improving the resolution of seismic data according to claim 3, characterized in that: The deconvolution network model includes: one or more convolution layers, M basic block modules and an upsampling operation module; Among them, each basic block module is composed of a convolutional layer and a BN layer; M is a positive integer.

6. The method for improving the resolution of seismic data according to claim 3, characterized in that: The seismic record deconvolution network model is trained using the training samples, and a corresponding loss function is obtained each time the training is performed, including: Inputting the training samples into the seismic wavelet estimation network model to obtain an angular frequency and a constant attenuation factor; According to the angular frequency and the constant attenuation factor, using a wavelet formula to obtain a seismic estimation wavelet; Inputting the training sample into the deconvolution network model to obtain a reflection coefficient corresponding to the sample data; Inputting the reflection coefficient and the earthquake estimation wavelet into the convolution network model to obtain a processed earthquake record; Calculation is performed based on the processed seismic records and the sample data to obtain a loss function corresponding to the sample data.

7. The method for improving the resolution of seismic data according to claim 6, characterized in that: The wavelet formula is: In the above formula, w(t) is the sub-wave, t is the time, m1 and m2 are the angular frequencies, and σ1 and σ2 are constant attenuation factors.

8. The method for improving the resolution of seismic data according to claim 6, characterized in that: The loss function includes: a loss function of a seismic wavelet estimation network and a loss function of a deconvolution network.

9. The method for improving the resolution of seismic data according to claim 8, characterized in that: The loss function of the seismic wavelet estimation network is: Fw j =|f(w(t))| In the above formula, loss w is the loss function of the seismic wavelet estimation network, f(·) represents Fourier transform, N is the number of batches, M is the number of samples in a batch, i represents the i-th batch, and FW j represents the normalized amplitude spectrum of the wavelet, FD ij represents the multi-channel average normalized amplitude spectrum of the seismic record, λ is a non-negative constant proportional coefficient, j=1,...,k0 represents the effective frequency band of the seismic signal, j=1,...,k1 represents the low frequency band of the seismic signal, Fd i，j represents the multi-channel average amplitude spectrum of the seismic record, d k (t) represents earthquake record, Fw j represents the wavelet amplitude spectrum, and w(t) represents the seismic wavelet.

10. The method for improving the resolution of seismic data according to claim 8, characterized in that: The loss function of the deconvolution network is: In the above formula, N represents the batch number, M represents the number of samples in a batch, T represents the sampling point of each sample, and λ represents a non-negative constant adjustment factor. is the estimated result of the kth frequency band of the j'th sample in the i-th batch, is the high-resolution processing result of the kth frequency band of the j'th sample in the i-th batch, F(·) represents the multi-channel average amplitude spectrum, f(·) represents the Fourier transform, and F n (·) represents the multi-channel average normalized amplitude spectrum, r est Represents the reflection coefficient obtained by the deconvolution network module, r lable represents the initial reflection coefficient obtained by the frequency domain deconvolution method.

11. A device for improving the resolution of seismic data, characterized in that: The device comprises: a memory and a processor; the memory is used to store a program for improving the resolution of seismic data, and the processor is used to read and execute the program for improving the resolution of seismic data, and execute the method described in any one of claims 1-10.

12. A computer-readable storage medium having a data processing program stored thereon, wherein the data processing program is executed by a processor to implement the method for improving the resolution of seismic data as described in any one of claims 1 to 10.

Images