Seismic wave impedance high-resolution inversion method and device based on physical model constraint, equipment and storage medium

Abstract

The invention discloses a seismic wave impedance high-resolution inversion method and device based on physical model constraint, equipment and a storage medium. The method comprises the following steps: training a forward modeling network by using logging wave impedance data and corresponding seismic data; performing well logging constraint-based training on the inversion network by using the well logging data, the initial low-frequency wave impedance data and the seismic trace position text data; training the inversion network based on the cyclic consistency constraint by using the unlabeled data under the constraint of the forward modeling network; utilizing an inversion network to obtain wave impedance data and a reflection coefficient of initial prediction; and performing iterative training on the inversion network, training the inversion network by using logging constraint and loop consistency constraint in each round of iteration, using an inversion result of the previous round of iteration as input, and outputting a final inversion result when an iteration termination condition is satisfied. The forward modeling physical model and the inversion network model are combined to form a closed-loop network, label-free data are fully utilized, and the physical interpretability of an inversion result is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of geophysics, and in particular to a method, device, equipment and storage medium for high-resolution inversion of seismic wave impedance based on physical model constraints. Background Art

[0002] In geophysical exploration, high-resolution processing of seismic data plays an important role in clearly understanding the underground geological structure and accurately describing the oil and gas reservoir environment. In the entire process from the excitation to the reception of seismic waves, due to the heterogeneity and non-perfect elasticity of the real formation medium, when the seismic waves propagate in the formation medium, they will be affected by factors such as formation absorption and wavefront diffusion, causing the energy of the seismic waves to attenuate when propagating in the formation. The final result is that the frequency of the seismic waves is changed, especially the high-frequency band of the seismic waves is more susceptible to attenuation than the low-frequency band of the seismic waves. If the seismic data is represented as a linear time-invariant system of the convolution of a finite-band seismic wavelet and a broadband reflection coefficient, the goal of high-resolution seismic is to obtain the broadband reflection coefficient through the signal inverse problem solution method. Due to the band-limited nature of seismic data, the high-resolution seismic inversion problem faces multiple solutions and uncertainty.

[0003] At present, deconvolution methods such as optimal Wiener filtering, pulse deconvolution, predictive deconvolution, and homomorphic deconvolution are used to improve the resolution of seismic data. This type of method is the inverse process of the convolution process and has good physical interpretability, but it requires the estimation of seismic wavelets and assumes that the reflection coefficient sequence is Gaussian distributed. The processing results have poor amplitude preservation and are sensitive to noise. Seismic high-resolution processing methods based on time-frequency domain analysis, such as short-time Fourier transform, continuous wavelet transform, synchronous compression wavelet transform, empirical mode decomposition, etc., can characterize the frequency variation of non-stationary signals over time in the time-frequency spectrum, that is, estimate dynamic wavelets, but accurately estimating the time-frequency spectrum of seismic signals requires breaking through the contradictory relationship between time domain and frequency domain resolution. Inverse Q filtering can also estimate dynamic wavelets and directly compensate for frequency loss and amplitude attenuation, but the Q model is relatively complex, and the difficulty lies in the accurate estimation of the Q value. Recently, some high-resolution seismic data processing methods based on deep learning, such as end-to-end network-based, generative adversarial network-based, multi-task learning-based, and seismic phase-guided, are data-driven methods that achieve higher accuracy. However, these methods require a large number of labeled samples and lack physical interpretability.

[0004] Therefore, a high-resolution inversion method for seismic wave impedance is needed that can ensure physical interpretability and make full use of existing labeled data and unlabeled data. Summary of the invention

[0005] This paper proposes a high-resolution inversion scheme for seismic wave impedance constrained by a physical model. The scheme uses a forward physical model and an inversion network model to form a closed-loop network, introduces unlabeled data to participate in training, and physical information can improve the physical interpretability and controllability of high-resolution processing results of seismic data by optimizing wavelets, thereby reducing the multi-solution problem of high-resolution inversion.

[0006] According to an embodiment of the present disclosure, a high-resolution inversion method of seismic wave impedance based on physical model constraints is proposed, comprising:

[0007] The forward modeling network is trained using the well logging impedance data Z and the corresponding seismic data S.

[0008] Using the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L , seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints;

[0009] Forward Modeling Network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Conduct the first round of training based on cycle consistency constraints;

[0010] Using the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the corresponding reflection coefficient

[0011] The inversion network is iteratively trained. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on well logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

[0012] In some embodiments, the training forward network include:

[0013] Based on the forward physical constraint loss term L f (W F ) Train the forward network

[0014]

[0015] Among them, W F are the forward modeling network parameters.

[0016] In some embodiments, the inversion network The first round of training based on well logging constraints includes:

[0017] Based on the first well logging constraint loss term L b1 (W B ) for the inversion network Perform the first round of training based on well logging constraints:

[0018]

[0019] Among them, W B are the inversion network parameters.

[0020] In some embodiments, the inversion network The first round of training based on cycle consistency constraints includes:

[0021] Based on the first cycle consistency constraint L bf1 (W B ,W F ) for the inversion network Perform the first round of training based on cycle consistency constraints:

[0022]

[0023] Among them, W F is the forward modeling network parameter, W B are the inversion network parameters.

[0024] In some embodiments, the inversion network is trained in an iterative manner. Training based on well logging constraints includes:

[0025] Based on the second well logging constraint loss term L b2 (W B) for the inversion network Perform training based on well logging constraints:

[0026]

[0027] Among them, W B are the inversion network parameters.

[0028] In some embodiments, the inversion network is trained in an iterative manner. Training with cycle consistency constraints involves:

[0029] Based on the second cycle consistency constraint L bf2 (W B , W F ) for the inversion network Perform training based on cycle consistency constraints:

[0030]

[0031] Among them, W F is the forward modeling network parameter, W B are the inversion network parameters.

[0032] In some embodiments, during iterative training, the inversion network During the training based on the cycle consistency constraint, the forward modeling network also outputs the reflection coefficient, and the reflection coefficient output by the forward modeling network is compared with the reflection coefficient obtained after the first round of training. The inversion network is trained with the loss term between .

[0033] In some embodiments, the forward modeling network is a network based on a convolution physical model, and the forward modeling network parameters W F One-dimensional convolution kernel for modeling seismic wavelets.

[0034] In some embodiments, the inversion network adopts a one-dimensional U-Net network structure.

[0035] In some embodiments, the iteration termination condition is that the number of iterations reaches a preset number of iterations.

[0036] According to an embodiment of the present disclosure, a high-resolution inversion device for seismic wave impedance based on physical model constraints is proposed, comprising:

[0037] Forward network training unit, used to train the forward network using logging wave impedance data Z and corresponding seismic data S

[0038] The first initial training unit of the inversion network is used to use the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L, seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints;

[0039] The second initial training unit of the inversion network is used to make the forward network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Conduct the first round of training based on cycle consistency constraints;

[0040] Initial prediction unit, used to utilize the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the reflection coefficient

[0041] The inversion network iterative training unit is used to iteratively train the inversion network. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on well logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

[0042] According to one embodiment of the present disclosure, an electronic device is provided, the device comprising a memory and a processor, the memory being used to store computer instructions executable on the processor, the processor being used to implement any of the above methods when executing the computer instructions.

[0043] According to an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, any of the above methods is implemented.

[0044] The high-resolution inversion scheme of seismic wave impedance based on physical model constraints proposed in the present disclosure effectively overcomes the shortcomings of the prior art by combining the forward physical model with the inversion network model to form a closed-loop network structure, and introducing unlabeled data to participate in the training process. The scheme uses the forward physical model to constrain the inversion results, so that the inversion results have good physical interpretability; at the same time, spatial constraints are realized by introducing seismic trace position text data, further improving the controllability of the inversion results. The present disclosure adopts an iterative optimization strategy, and uses the wave impedance data and reflection coefficient obtained from the previous round of inversion as new constraints to train the inversion network. With the increase in the number of iterations, the accuracy of the inversion results continues to improve, especially in the recovery of high-frequency details. In addition, the present disclosure makes full use of unlabeled data to participate in network training, which significantly improves the generalization performance of the model under limited logging data conditions, reduces the multi-solution problem of high-resolution inversion of seismic wave impedance, and makes the inversion results more reliable.

[0045] Other features and advantages of the present disclosure are described in detail below. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.

[0047] It should be noted that Figure 2 to Figure 10 Some characters may not be clear enough due to resolution, but these characters do not affect the understanding of the solution by those skilled in the art.

[0048] Figure 1 A flowchart of a high-resolution inversion method for seismic wave impedance based on physical model constraints according to an embodiment of the present disclosure is shown.

[0049] Figure 2 A schematic diagram of a closed-loop network training process according to an exemplary embodiment of the present disclosure is shown.

[0050] Figure 3 The original seismic data and the predicted reflection coefficient data profiles at different iteration numbers are shown.

[0051] Figure 4 Spectral plots of the original seismic data and predicted reflection coefficients are shown.

[0052] Figure 5 The real wave impedance data and the wave impedance data profiles inverted at different iteration numbers are displayed.

[0053] Figure 6 The comparison between the real high-frequency wave impedance data and the inverted high-frequency wave impedance data is shown.

[0054] Figure 7 A cross-section of the true wave impedance data and the inversion error data is shown.

[0055] Figure 8 The wave impedance of the 1st to 6th inversion of single-channel data is shown.

[0056] Fig. 9 The reflection coefficients of the 1st to 6th inversions of the single-channel data are shown.

[0057] Fig.10 The relationship between the number of iterations and the inversion mutual correlation coefficient and inversion error is shown.

[0058] Fig.11 It is a schematic diagram of the structure of an electronic device shown in at least one embodiment of the present disclosure. DETAILED DESCRIPTION

[0059] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same 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 disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0060] The disclosed embodiments may be applied to a computer system / server that may operate with numerous other general or special computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with a computer system / server include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above systems, and the like.

[0061] Computer systems / servers may be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. In general, program modules may include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers may be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules may be located on local or remote computing system storage media including storage devices.

[0062] Figure 1A flowchart of a method for high-resolution inversion of seismic wave impedance based on physical model constraints according to an embodiment of the present disclosure is shown. As shown in the figure, the method includes steps 1 to 5.

[0063] Step 1: Use the well logging impedance data Z and the corresponding seismic data S to train the forward modeling network

[0064] In some embodiments, the forward modeling network is a network based on a convolution physical model, through a one-dimensional convolution kernel W F To simulate the seismic wavelet. The physical model of this forward modeling network can be expressed as:

[0065]

[0066] Where r(t) is the reflection coefficient, w(t) is the seismic wavelet, and * represents the convolution operation. The reflection coefficient r(t) can be calculated from the wave impedance using the following formula:

[0067]

[0068] In some embodiments, the forward modeling network is based on the following forward physical constraint loss term L f (W F ) for training:

[0069]

[0070] Among them, W F is the forward modeling network parameter. In some examples, W F Represented as a one-dimensional convolution kernel, used to simulate seismic wavelets.

[0071] The forward modeling physical constraint loss term can be optimized by using the stochastic gradient descent method to obtain the optimal forward modeling network parameter W F The training process of the forward modeling network is essentially the estimation process of the seismic wavelet. The trained forward modeling network will be used as the physical constraint in the subsequent inversion network training process.

[0072] Step 2: Using the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L , seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints.

[0073] In some embodiments, the inversion network A one-dimensional U-Net network structure is adopted.

[0074] In the first round of training, the inversion network is trained with the well logging impedance data Z corresponding to the well bypass seismic data S and the initial low-frequency impedance data Z L, seismic trace position text data T as input, and output predicted wave impedance data.

[0075] In some embodiments, the first logging constraint loss term L may be used as follows: b1 (W B ) for the inversion network To train:

[0076]

[0077] Among them, W B is the inversion network parameter; Z is the logging wave impedance data; S is the seismic data; Z L is the initial low-frequency wave impedance data; T is the seismic trace position text data; By adopting this loss term, the wave impedance data output by the inversion network can be made as close as possible to the known logging wave impedance data, thereby achieving preliminary training of the inversion network.

[0078] The seismic trace position text data T introduced in this step can realize spatial constraints and improve the controllability of the inversion results. L The introduction of provides low-frequency background information for the inversion network, which helps to improve the reliability of the inversion results.

[0079] Step 3: Make the forward modeling network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Perform the first round of training based on cycle consistency constraints.

[0080] This step will invert the network The output of the inversion impedance data is used as the forward network The input of the forward network and inversion network The series connection forms a closed loop structure. Take the unlabeled earthquake data S * , unlabeled initial low-frequency wave impedance data and unlabeled seismic trace position text data T * As input, it outputs predicted wave impedance data, which is input into the forward modeling network In the forward modeling network Synthetic seismic data is generated based on the wave impedance data. Through this closed-loop structure, the constraints of the forward physical model can be introduced into the inversion network. during the training process.

[0081] On this basis, in some implementations, the following first cycle consistency constraint loss term L can be used: bf1 (W B , W F ) for the inversion network To train:

[0082]

[0083] Among them, S * is unlabeled earthquake data; is the unlabeled initial low-frequency wave impedance data; T * is the unlabeled seismic trace position text data; W F is the forward modeling network parameter obtained in step 1; W B is the inversion network parameter. The loss term L bf1 (W B , W F ) requires synthetic seismic data generated by a closed-loop network Should be compared with the original unlabeled earthquake data S * As close as possible, so that in the unlabeled data S * Inversion network further optimization.

[0084] Step 4: Use the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the reflection coefficient

[0085] After completing the inversion network through steps 2 and 3 After the initial training of * By inversion, we can obtain:

[0086]

[0087] Predicted After that, the corresponding reflection coefficient can be further calculated according to the above formula 2

[0088] The predicted wave impedance data and the corresponding reflection coefficient will be used as input conditions for the subsequent iterative training process.

[0089] Step 5: Iteratively train the inversion network. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on well logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

[0090] The value of i starts from 1, which means that iterative training begins from the second round.

[0091] Compared with the first round of training, the wave impedance data predicted after the previous round of training is As an inversion network Input and add reflection coefficient as inversion network Because at the beginning of the first round of training, the unlabeled earthquake data S * The corresponding wave impedance data did not obtain the corresponding reflection coefficient. After completing the first round of training, the inversion network trained initially can be used For seismic data S and S * Perform inversion (as in step 4) to predict the wave impedance data and calculate the corresponding reflection coefficient. Since the reflection coefficient directly reflects the wave impedance variation characteristics of the underground medium, introducing the reflection coefficient as the input of the inversion network can provide additional constraint information for the inversion process, which helps to improve the accuracy of the inversion results.

[0092] In some embodiments, in the i+1th round of iterative training, firstly based on the second logging constraint loss term L b2 (W B ) for the inversion network To train:

[0093]

[0094] Among them, W F is the forward modeling network parameter, W B is the inversion network parameter. (i) is the inversion network based on the i-th round of training Predicted wave impedance data, r (i) Based on Z (i)The reflection coefficient is calculated using Formula 2. Compared with the first round of training, the reflection coefficient is added to the inversion network input here, so that the logging constraint term can use more prior information.

[0095] Next, training is performed using unlabeled data under the constraints of the forward network. In some embodiments, the second cycle consistency constraint term L bf2 (W B , W F ) for the inversion network Perform training based on cycle consistency constraints:

[0096]

[0097] Among them, W F is the forward modeling network parameter, W B is the inversion network parameter. Similarly, is the inversion network based on the i-th round of training Predicted wave impedance data, Based on The reflection coefficient is calculated using Equation 2.

[0098] In some embodiments, during the cycle consistency constraint training, the forward modeling network may also output a reflection coefficient. By comparing the reflection coefficient output by the forward modeling network with the initial reflection coefficient obtained in step 4 after the first round of training, A reflection coefficient loss term can be constructed to provide additional constraint information for inversion network training and further improve the reliability of the inversion results.

[0099] In each round of training, after the training based on the cycle consistency constraint is completed, the inversion network currently trained can be used For seismic data S and S * Perform inversion to obtain wave impedance data and the corresponding reflection coefficient in The reflection coefficient can be further calculated using Formula 2

[0100] Then, it can be determined whether the iteration termination condition is met. In some embodiments, the iteration termination condition is that the number of iterations reaches a preset number of iterations. If not, and Update the corresponding input of the inversion network and enter the next round of iteration; if satisfied, terminate the iteration and output the final inversion result

[0101] The high-resolution inversion scheme of seismic wave impedance based on physical model constraints proposed in this embodiment effectively overcomes the shortcomings of the prior art by combining the forward physical model with the inversion network model to form a closed-loop network structure, and introducing unlabeled data to participate in the training process. This embodiment uses the forward physical model to constrain the inversion results, so that the inversion results have good physical interpretability; at the same time, spatial constraints are realized by introducing seismic trace position text data, further improving the controllability of the inversion results. This embodiment adopts an iterative optimization strategy, and uses the wave impedance data and reflection coefficient obtained from the previous round of inversion as new constraints to train the inversion network. With the increase in the number of iterations, the accuracy of the inversion results continues to improve, especially in the recovery of high-frequency details. In addition, this embodiment makes full use of unlabeled data to participate in network training, significantly improves the generalization performance of the model under limited logging data conditions, reduces the multi-solution problem of high-resolution inversion of seismic wave impedance, and makes the inversion results more reliable.

[0102] Figure 2 FIG. 2 shows a schematic diagram of a closed-loop network training process according to an exemplary embodiment of the present disclosure. Figure 2 As shown in Figure 2, the network training process includes two main stages. The first stage ( Figure 2 The upper part shows the initial training process, including the training of the forward network and the first round of training of the inverse network. A one-dimensional convolution kernel is used to simulate seismic wavelets, and the well logging wave impedance data and the corresponding seismic data are input into the network for training; the inversion network The training consists of two parts: one is to use the logging wave impedance data for training based on logging constraints, and the other is to use unlabeled data for training based on cycle consistency constraints under the constraints of the forward network.

[0103] Phase II Figure 2 The lower part shows the iterative training process. Compared with the first stage, the input of the inversion network is increased by reflection coefficient data. In each round of iteration, the inversion network The training is also based on logging constraints and cycle consistency constraints, and after the training is completed, a new wave impedance prediction result is output. The prediction result is used to calculate the new reflection coefficient and as the input for the next round of iterative training until the predetermined number of iterations is reached.

[0104] Through this closed-loop network structure design, the forward physical model can continuously provide constraints for the training of the inversion network, while making full use of unlabeled data to effectively improve the accuracy and reliability of the inversion results.

[0105] In order to verify the effectiveness and superiority of this scheme, a specific application example is given below to demonstrate the inversion effect obtained according to this scheme.

[0106] This example uses the wave impedance data of the Marmousi II model for the experiment. The reflection coefficient is calculated using the differential operator, and the Ricker wavelet with a main frequency of 20Hz is convolved with the reflection coefficient to obtain the synthetic seismic data. The data contains 476 seismic traces, with 1651 time sampling points and a sampling interval of 1ms. In the experiment, the 250th wave impedance data is selected as the well logging label data, and the corresponding seismic trace data, low-frequency wave impedance data and seismic trace position data are used as labeled data, and the remaining data are used as unlabeled data to participate in network training.

[0107] Figure 3 The original seismic data and the predicted reflection coefficient data profiles at different iteration numbers are shown.

[0108] in Figure 3 (a) shows the original seismic data. Figure 3 (b) to (f) show the reflection coefficients predicted from the 1st to the 5th time, respectively. It can be seen from the figure that with the increase of the number of iterations, the predicted reflection coefficient becomes clearer and the characteristics of the formation interface become more prominent.

[0109] Figure 4 Spectral plots of the original seismic data and predicted reflection coefficients are shown. Figure 4 (a) is the spectrum of the original seismic data. Figure 4 (b) to (f) are the spectra of the reflection coefficients predicted from the 1st to the 5th time. Spectral analysis shows that through iterative training, the predicted reflection coefficients contain richer frequency components.

[0110] Figure 5 The real wave impedance data and the wave impedance data profiles inverted at different iteration numbers are displayed. Figure 5 (a) shows the real wave impedance data, Figure 5 (b) to (f) show the wave impedance data of the 1st to 5th inversion respectively. Figure 6 The comparison between the real high-frequency wave impedance data and the inverted high-frequency wave impedance data is further shown. From the area marked by the arrow, it can be observed that with the increase in the number of iterations, the inversion result's ability to depict high-frequency details continues to improve. The average cross-correlation coefficient of the high-frequency inversion part gradually increases from the initial 0.8569 to 0.9040.

[0111] Figure 7 A cross-section of the true wave impedance data and the inversion error data is shown. Figure 7 (a) is the real wave impedance data, Figure 7(b) to (f) are the error data of the 1st to 5th inversion. The mean absolute error of the inversion gradually decreases from 316.70 to 277.34. It can be seen that using the inversion results of the previous training as the input condition for the next training can improve the subsequent inversion accuracy and better reflect the high-frequency detail components.

[0112] Figure 8 and Fig. 9 The inversion results of single-channel data with a CDP of 350 in the test set are shown. Figure 8 From left to right, the wave impedances of the 1st to 6th inversions are shown. Fig. 9 From left to right, the reflection coefficients predicted from the 1st to the 6th time are shown, where the solid black lines represent the true values ​​and the dashed lines represent the inversion results. From the single-channel results, it can be seen that with the increase in the number of iterations, the degree of fit between the inversion results and the true values ​​continues to improve.

[0113] Fig.10 The relationship between the number of iterations and the inversion mutual correlation coefficient and inversion error is shown. Fig.10 The middle horizontal axis represents the number of iterations, the left vertical axis represents the mean absolute error (MAE), and the corresponding bar graph shows the trend of the error gradually decreasing with the increase of the number of iterations; the right vertical axis represents the cross-correlation coefficient, and the corresponding broken line shows the trend of the cross-correlation coefficient gradually increasing with the increase of the number of iterations. Fig.10 It is intuitively demonstrated that the high-resolution inversion method of seismic wave impedance based on physical model constraints proposed in the present disclosure can simultaneously achieve error reduction and improvement of cross-correlation coefficients through iterative training.

[0114] The above experimental results show that the high-resolution inversion method of seismic wave impedance based on physical model constraints proposed in the present invention can effectively improve the inversion accuracy, especially showing significant advantages in recovering high-frequency details.

[0115] The present disclosure also proposes a high-resolution inversion device for seismic wave impedance, comprising a forward network training unit, a first initial training unit for an inversion network, a second initial training unit for an inversion network, an initial prediction unit and an iterative training unit for an inversion network.

[0116] The forward network training unit is used to train the forward network using the logging wave impedance data Z and the corresponding seismic data S

[0117] The first initial training unit of the inversion network is used to use the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L , seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints.

[0118] The second initial training unit of the inversion network is used to make the forward network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Perform the first round of training based on cycle consistency constraints.

[0119] The initial prediction unit is used to utilize the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the reflection coefficient

[0120] The inversion network iterative training unit is used to iteratively train the inversion network. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on well logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

[0121] Fig.11 An electronic device provided for at least one embodiment of the present disclosure comprises a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the high-resolution inversion method of seismic wave impedance based on physical model constraints as described in any embodiment or implementation of the present disclosure when executing the computer instructions.

[0122] At least one embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the high-resolution inversion method for seismic wave impedance based on physical model constraints as described in any embodiment or implementation of the present disclosure.

[0123] It should be understood by those skilled in the art that one or more embodiments of the present specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of the present specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of the present specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the data processing device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0125] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0126] The embodiments of the subject matter and functional operations described in this specification may be implemented in the following: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. The embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules in computer program instructions encoded on a tangible non-temporary program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical, optical or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0127] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits), and the apparatus can also be implemented as special purpose logic circuits.

[0128] Computers suitable for executing computer programs include, for example, general and / or special microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, the computer will also include one or more large-capacity storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to this large-capacity storage device to receive data from it or to transmit data to it, or both. However, the computer does not necessarily have such a device. In addition, the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0129] Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks. The processor and memory may be supplemented by, or incorporated in, special purpose logic circuitry.

[0130] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of the specific embodiments of specific inventions. Certain features described in multiple embodiments in this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claim protection, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of a sub-combination.

[0131] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or requiring that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0132] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0133] The above description is merely a preferred embodiment of one or more embodiments of the present specification and is not intended to limit one or more embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present specification shall be included in the scope of protection of one or more embodiments of the present specification.

Claims

1. A high-resolution inversion method for seismic wave impedance based on physical model constraints, characterized in that: include: The forward modeling network is trained using the well logging impedance data Z and the corresponding seismic data S. Using the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L , seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints; Forward Modeling Network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Conduct the first round of training based on cycle consistency constraints; Using the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the corresponding reflection coefficient The inversion network is iteratively trained. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

2. The method according to claim 1, characterized in that The training forward network include: Based on the forward physical constraint loss term L f (W F ) Train the forward network Among them, W F are the forward modeling network parameters.

3. The method according to claim 1, characterized in that The inversion network The first round of training based on well logging constraints includes: Based on the first well logging constraint loss term L b1 (W B ) for the inversion network Perform the first round of training based on well logging constraints: Among them, W B are the inversion network parameters.

4. The method according to claim 1, characterized in that The inversion network The first round of training based on cycle consistency constraints includes: Based on the first cycle consistency constraint L bf1 (W B ,W F ) for the inversion network Perform the first round of training based on cycle consistency constraints: Among them, W F is the forward modeling network parameter, W B are the inversion network parameters.

5. The method according to claim 1, characterized in that In iterative training, the inversion network Training based on well logging constraints includes: Based on the second well logging constraint loss term L b2 (W B ) for the inversion network Perform training based on well logging constraints: Among them, W B are the inversion network parameters.

6. The method according to claim 1, characterized in that In iterative training, the inversion network Training with cycle consistency constraints involves: Based on the second cycle consistency constraint L bf2 (W B ,W F ) for the inversion network Perform training based on cycle consistency constraints: Among them, W F is the forward modeling network parameter, W B are the inversion network parameters.

7. The method according to claim 6, characterized in that In iterative training, the inversion network During the training based on the cycle consistency constraint, the forward modeling network also outputs the reflection coefficient, and the reflection coefficient output by the forward modeling network is compared with the reflection coefficient obtained after the first round of training. The inversion network is trained with the loss term between .

8. The method according to claim 1, characterized in that The forward modeling network is a network based on a convolution physical model. The forward modeling network parameters W F One-dimensional convolution kernel for modeling seismic wavelets.

9. The method according to claim 1, characterized in that: The inversion network adopts a one-dimensional U-Net network structure.

10. The method according to claim 1, characterized in that The iteration termination condition is that the number of iterations reaches a preset number of iterations.

11. A high-resolution inversion device for seismic wave impedance based on physical model constraints, characterized in that: include: Forward network training unit, used to train the forward network using logging wave impedance data Z and corresponding seismic data S The first initial training unit of the inversion network is used to use the logging wave impedance data Z and the corresponding seismic data S, the initial low-frequency wave impedance data Z L , seismic trace position text data T for inversion network Conduct the first round of training based on well logging constraints; The second initial training unit of the inversion network is used to make the forward network and inversion network Constitute a closed-loop structure in the forward modeling network Using unlabeled seismic data S under constraints * , unlabeled initial low-frequency wave impedance data Unlabeled seismic trace location text data T * Inversion Network Conduct the first round of training based on cycle consistency constraints; Initial prediction unit, used to utilize the currently trained inversion network For seismic data S and S * Perform inversion to obtain the initial predicted wave impedance data And according to Calculate the reflection coefficient The inversion network iterative training unit is used to iteratively train the inversion network. The i+1th round of training includes using the wave impedance data predicted after the previous round of training. and the corresponding reflection coefficient As an inversion network Input, in turn, to the inversion network Perform training based on logging constraints and training based on cycle consistency constraints, and use the currently trained inversion network after this round of training is completed For seismic data S and S * Perform inversion to obtain wave impedance data And calculate the corresponding reflection coefficient Then determine whether the iteration termination condition is met. If not, enter the next round of iteration. If it is met, terminate the iteration and output the final inversion result.

12. An electronic device, characterized in that: The device comprises a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the method according to any one of claims 1 to 10 when executing the computer instructions.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images