Multi-task intelligent seismic acoustic impedance inversion method and apparatus

Abstract

Disclosed in the present application are a multi-task intelligent seismic acoustic impedance inversion method and apparatus. The method comprises: performing well-seismic calibration, and determining samples of a target interval; acquiring a density curve and a velocity curve of logging data, and extracting training sample labels from the density curve and velocity curve of logging data; acquiring, from seismic data, seismic data of the same depth interval as the target interval, and constructing a training data set for a velocity interpretation model; respectively constructing the velocity interpretation model and a density interpretation model, and determining a final density interpretation model P1 and a final velocity interpretation model S1; respectively transferring model parameters of the model P1 and the model S1, so as to construct an acoustic impedance inversion model based on multi-task learning, performing comprehensive training on the acoustic impedance inversion model, and determining a loss function thereof; and completing inversion processing. The present application improves the accuracy of seismic acoustic impedance inversion, achieves higher robustness, and enhances the vertical resolution.

Description

Multi-task intelligent seismic wave impedance inversion method and device

[0001] Cross-reference to Related Applications

[0002] This application claims priority to Chinese Patent Application No. 202411314234.3, filed on September 20, 2024, with the Chinese Patent Office, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the technical field of seismic wave impedance inversion, and in particular to a multi-task intelligent seismic wave impedance inversion method and device. BACKGROUND

[0004] Seismic wave impedance inversion refers to the inversion of the wave impedance of the underground formation by comprehensively utilizing seismic data and logging data with the aid of computer technology. An effective reservoir wave impedance model provides a three-dimensional geological data basis for later numerical simulation of the oil reservoir and plays a key role in oil and gas reserve calculation and other work. The underground formation usually has a multi-scale heterogeneous and continuous structure, but due to technical limitations, these parameters cannot be directly measured.

[0005] At present, geostatistical methods are mainly used for seismic wave impedance inversion. Seismic wave impedance inversion is a comprehensive reservoir characterization technology that combines seismic data, logging data and core data, aiming to maximize the use of all collected data to express the geological information of the target layer under the guidance of geological theory. However, in the process of seismic wave impedance inversion, the multi-solution of wave impedance data is still a problem to be solved in reservoir modeling. Multi-solution leads to the inability to uniquely determine the formation properties, thereby increasing the uncertainty of geological interpretation.

[0006] SUMMARY

[0007] The present application provides a multi-task intelligent seismic wave impedance inversion method and device to at least solve the foregoing technical problems.

[0008] According to a first aspect of an embodiment of the present application, a multi-task intelligent seismic wave impedance inversion method is provided, comprising:

[0009] Calibrating the well and seismic, and determining the target layer sample;

[0010] Obtaining the logging data density curve and the velocity curve, and cutting the training sample label from the logging data density curve and the velocity curve;

[0011] acquire seismic data of the same depth section as the target section from the seismic data, and perform spectral analysis on the acquired seismic data, use the seismic data after spectral analysis and the density data to construct a training data set of the density interpretation model, and use the seismic data after spectral analysis and the velocity data to construct a training data set of the velocity interpretation model;

[0012] Construct the velocity interpretation model and the density interpretation model based on the training data set respectively, and map the seismic data into velocity and density according to the velocity interpretation model and the density interpretation model respectively; wherein the velocity interpretation model and the density interpretation model are both constructed using Gated Recurrent Unit (GRU) neural units, and the number of GRU layers is 3 layers; the dimension of the seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of the output velocity and density is both (B, 1, m), B represents the number of samples of the model at each time of training;

[0013] Use the training data in the training data set of the density interpretation model to train the density interpretation model, as the final density interpretation model P1 after completing the training, and use the training data in the training data set of the velocity interpretation model to train the velocity interpretation model, as the final velocity interpretation model S1 after completing the training;

[0014] Migrate the model parameters of the model P1 and the model S1 respectively, thereby constructing a multi-task learning wave impedance inversion model, wherein the wave impedance inversion model is a parallel connection of the density interpretation model and the velocity interpretation model;

[0015] Comprehensively train the wave impedance inversion model and determine its loss function; use the wave impedance inversion model to complete the inversion processing of the seismic data on density, velocity and wave impedance respectively.

[0016] In an implementation manner, the method further comprises:

[0017] According to the density curve and the velocity curve of the logging data, velocity data and density data are acquired respectively, the product of velocity and density is calculated as wave impedance.

[0018] In an implementation manner, the acquiring seismic data of the same depth section as the target section from the seismic data, and performing spectral analysis on the acquired seismic data comprises:

[0019] Using well coordinates, seismic trace data in a unit area grid with the corresponding position as the center is acquired as model input data, and the following spectral analysis method is used to acquire the frequency domain characteristics of the seismic data;

[0020] Wherein:

[0021] f(t) is a one-dimensional seismic trace signal, F(w) is the frequency spectrum of the one-dimensional seismic trace signal in the frequency domain, j is a complex unit, and w is a base frequency of the one-dimensional seismic trace signal;

[0022] According to the spectral characteristics, a first main frequency is selected, and a band-pass filter is used to obtain a seismic waveform H(f) corresponding to the first main frequency:

[0023] wherein f is the first main frequency, f L is a low-pass frequency value centered on the first main frequency, and f H is a high-pass frequency value centered on the first main frequency.

[0024] In an implementation, the velocity interpretation model and the density interpretation model are respectively constructed based on the training data set, including:

[0025] The number of points of the target layer segment sample is determined to be m, the density and velocity training sample labels are intercepted from the logging data, the label dimension is (1, m), the p*p grid seismic data of the same depth layer segment as the target layer segment is obtained from the seismic data, and the dimension is (p, p, m);

[0026] The dimension of the wave impedance is (1, m), the seismic waveform corresponding to the first main frequency of the seismic data is obtained by using frequency filtering, and the dimension is (p, p, m);

[0027] The seismic data and the seismic waveform corresponding to the main frequency f are merged, the seismic data of different grid positions are merged, and the merged data is used as the input model data, and the dimension is (50, m).

[0028] In an implementation, the operation process of each GRU in the 3-layer GRU of the velocity interpretation model and the density interpretation model includes the following processing steps: controlling how the state of the previous moment affects the current state; controlling how the current input affects the state of the previous moment; calculating the candidate hidden state according to the current input and the reset state of the previous moment; and calculating the final hidden state of the current moment according to the influence of the state of the previous moment on the current state.

[0029] In an implementation, the loss function of the wave impedance inversion model contains three parts, namely the density loss loss_ρ, the velocity loss loss_v and the wave impedance loss loss_e, and the calculation method is as follows:

[0030] Wherein:

[0031] loss_ρ is the inversion loss of the density module;

[0032] loss_v is the inversion loss of the velocity module;

[0033] loss_e is the inversion loss of the wave impedance module;

[0034] N: number of training samples;

[0035] p i : represents the true density value;

[0036] represents the density interpretation model inversion density value;

[0037] v i : represents the true velocity value;

[0038] represents the velocity interpretation model inversion velocity value;

[0039] e i : represents the true wave impedance value;

[0040] represents the seismic wave impedance inversion model inversion value.

[0041] According to the second aspect of the embodiments of the present application, a multi-task intelligent seismic wave impedance inversion device is provided, comprising:

[0042] a first determination unit configured to calibrate well-seismic and determine target layer samples;

[0043] an acquisition unit configured to acquire logging data density curves and velocity curves, and cut training sample labels from the logging data density curves and velocity curves;

[0044] a first construction unit configured to acquire seismic data of the same depth layer as the target layer from the seismic data, and perform spectral analysis on the acquired seismic data, use the spectral analyzed seismic data and density data to construct a training data set of the density interpretation model, and use the spectral analyzed seismic data and velocity data to construct a training data set of the velocity interpretation model;

[0045] a second construction unit configured to construct a velocity interpretation model and a density interpretation model based on the training data set, respectively, and map the seismic data to velocity and density according to the velocity interpretation model and the density interpretation model, respectively; wherein the velocity interpretation model and the density interpretation model are both constructed using a gated recurrent GRU neural unit, and the number of GRU layers thereof is 3 layers; the dimension of the seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of the output velocity and density is both (B, 1, m), B representing the number of samples of the model at each time of training;

[0046] The first training unit is configured to train the density interpretation model by using training data in a training data set of the density interpretation model, as a final density interpretation model P1 after completing training, and train the velocity interpretation model by using training data in a training data set of the velocity interpretation model, as a final velocity interpretation model S1 after completing training;

[0047] The third construction unit is configured to respectively migrate model parameters of the model P1 and the model S1, so as to construct a wave impedance inversion model for multi-task learning, wherein the wave impedance inversion model is a parallel connection of the density interpretation model and the velocity interpretation model.

[0048] The second training unit is configured to comprehensively train the wave impedance inversion model and determine a loss function of the wave impedance inversion model.

[0049] The inversion processing unit is configured to complete inversion processing of the seismic data on the density, the velocity and the wave impedance respectively by using the wave impedance inversion model.

[0050] In an implementation manner, the apparatus further includes:

[0051] The calculation unit is configured to obtain the velocity data and the density data respectively according to the density curve and the velocity curve of the logging data, and calculate a product of the velocity and the density as the wave impedance.

[0052] In an implementation manner, the obtaining unit is further configured to:

[0053] The seismic trace data in a unit area grid with a corresponding position as a center are obtained as model input data by using well coordinates, and the frequency domain features of the seismic data are obtained by using the following spectral analysis method:

[0054] Wherein:

[0055] f(t) is a one-dimensional seismic trace signal, F(w) is a frequency spectrum of the one-dimensional seismic trace signal in the frequency domain, j is a complex unit, and w is a base frequency of the one-dimensional seismic trace signal.

[0056] The first main frequency is selected according to the spectral features, and the following band-pass filter is used to obtain a seismic waveform H(f) corresponding to the first main frequency:

[0057] Wherein, f is the first main frequency, f L is a low-pass frequency value centered on the first main frequency, and f H is a high-pass frequency value centered on the first main frequency.

[0058] In an implementation manner, the second construction unit is further configured to:

[0059] The number of points of the target layer sample is determined as m, the density and velocity training sample labels are intercepted from the logging data, the label dimension is (1, m), the p x p grid seismic data of the same depth layer as the target layer is obtained from the seismic data, and the dimension is (p, p, m);

[0060] The dimension of the wave impedance is (1, m), the seismic waveform corresponding to the first main frequency of the seismic data is obtained by using frequency filtering, and the dimension is (p, p, m);

[0061] The seismic data and the seismic waveform corresponding to the main frequency f are merged, the seismic data of different grid positions are merged, and the merged data is used as the input model data, and the dimension is (50, m).

[0062] The operation process of each GRU in the three-layer GRU of the velocity interpretation model and the density interpretation model includes: controlling how the state of the previous moment affects the current state; controlling how the current input affects the state of the previous moment; calculating the candidate hidden state according to the current input and the reset state of the previous moment; and calculating the final hidden state of the current moment according to the influence of the state of the previous moment on the current state.

[0063] In an implementation manner, the loss function of the wave impedance inversion model includes three parts, namely, the density loss loss_ρ, the velocity loss loss_v and the wave impedance loss loss_e, and the calculation manner is as follows:

[0064] Wherein:

[0065] loss_ρ is the inversion loss of the density module;

[0066] loss_v is the inversion loss of the velocity module;

[0067] loss_e is the inversion loss of the wave impedance module;

[0068] N: the number of training samples;

[0069] ρ i : represents the real density value;

[0070] represents the inversion density value of the density interpretation model;

[0071] v i : represents the real velocity value;

[0072] represents the inversion velocity value of the velocity interpretation model;

[0073] e i : represents the real wave impedance value;

[0074] representing the seismic wave impedance inversion model inversion value.

[0075] The technical scheme provided by the embodiments of the present application can include the following beneficial effects:

[0076] The present application calibrates and determines the seismic data of the target layer sample through well-seismic calibration, constructs a velocity and density interpretation model using a gated recurrent unit GRU neural unit, and constructs a wave impedance inversion model. Therefore, the wave impedance results simulated based on the related model training are more accurate, the geophysical priori knowledge is effectively integrated into the inversion model, the accuracy of the seismic wave impedance inversion is improved, the robustness is higher, and the vertical resolution is improved.

[0077] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0078] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0079] FIG. 1 is a flowchart of a multi-task intelligent seismic wave impedance inversion method according to an embodiment of the present application;

[0080] FIG. 2 is a schematic diagram of seismic one-dimensional waveform data and wave impedance data according to an embodiment of the present application;

[0081] FIG. 3 is a schematic diagram of a seismic wave impedance inversion model framework according to an embodiment of the present application;

[0082] FIG. 4 is a schematic diagram of the composition structure of a multi-task intelligent seismic wave impedance inversion device according to an embodiment of the present application;

[0083] FIG. 5 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0084] The exemplary embodiments will be described in detail herein with reference to 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 implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0085] FIG. 1 is a flowchart of a multi-task intelligent seismic wave impedance inversion method according to an embodiment of the present application, as shown in FIG. 1, the multi-task intelligent seismic wave impedance inversion method according to the embodiments of the present application includes the following steps:

[0086] Step 101, well-to-seismic calibration is performed, and a target layer sample is determined.

[0087] Seismic reservoir prediction is more and more important in oil and gas exploration and development. Well-to-seismic calibration, as a key technology for seismic data structure interpretation and seismic reservoir prediction, is a bridge between geology, logging and seismic. It correlates seismic data with logging data in the depth domain, and combines logging data with seismic data with the advantage of lateral distribution. The quality of well-to-seismic calibration directly affects the effectiveness and accuracy of seismic data structure interpretation and seismic reservoir prediction. The embodiments of the present application need to accurately calibrate the related parameters of well-to-seismic, such as seismic data and logging data, and determine the sample of the target layer to be predicted for seismic reservoir prediction.

[0088] Step 102, obtain the logging data density curve and the velocity curve, and cut the training sample label from the logging data density curve and the velocity curve.

[0089] In the embodiments of the present application, the training sample label can be cut from the density curve and the velocity curve of the logging data, the seismic data of the same depth section as the target layer is obtained from the seismic data in the form of a 5x5 grid, and spectrum analysis is performed. The density interpretation model training data set is established using the seismic data and the density data, and the velocity interpretation model training data set is established using the seismic data and the velocity data. The size of the grid can be set as needed, for example, it can also be a 10x10 grid, a 15x15 grid or a 20x20 grid.

[0090] Step 103, obtain the seismic data of the same depth section as the target layer from the seismic data, and perform spectrum analysis on the obtained seismic data. The training data set of the density interpretation model is constructed using the spectrum-analyzed seismic data and the density data, and the training data set of the velocity interpretation model is constructed using the spectrum-analyzed seismic data and the velocity data.

[0091] In the embodiments of the present application, after well-to-seismic calibration, the number of target layer sample points is determined to be m, the density and velocity training sample labels are cut from the logging data, and the dimensions are both (1, m). The 5x5 grid seismic data of the same depth section as the target layer is obtained from the seismic data, and the dimension is (5, 5, m). The wave impedance data is calculated by multiplying the density and the velocity, and the dimension is (1, m). The seismic waveform corresponding to the first main frequency of the seismic data is obtained by using frequency filtering, and the dimension is (5, 5, m).

[0092] Where, wave impedance e = p.v;

[0093] e is the wave impedance, p is the density, and v is the velocity.

[0094] Using well coordinates, the seismic trace data in the 5*5 grid centered at the corresponding position is obtained as model input data, and the frequency domain characteristics of the seismic data are obtained using a spectral analysis method (such as formula 1).

[0095] Wherein:

[0096] f(t): one-dimensional seismic trace signal.

[0097] F(w): the frequency spectrum of the one-dimensional seismic trace signal in the frequency domain.

[0098] j: complex unit.

[0099] w: base frequency of one-dimensional seismic trace signal.

[0100] Finally, according to the spectral characteristics, the first main frequency is selected, and then the seismic waveform H(f) corresponding to the first main frequency is obtained using a band-pass filter, as follows:

[0101] Wherein:

[0102] f is the first main frequency, f L is the low-pass frequency value centered on the first main frequency, f H is the high-pass frequency value centered on the first main frequency.

[0103] Finally, the seismic data and the seismic waveform corresponding to the main frequency f are merged, the seismic data at different grid positions are merged, and the input model data, i.e., the training data set, is formed, which has a dimension of (50, m).

[0104] Step 104, based on the training data set, respectively constructing a velocity interpretation model and a density interpretation model, and respectively mapping the seismic data into velocity and density according to the velocity interpretation model and the density interpretation model.

[0105] In the embodiments of the present application, the velocity interpretation model and the density interpretation model are both constructed using a gated recurrent GRU neural unit, and the number of GRU layers is 3; the dimension of the seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of the output velocity and density is both (B, 1, m), B representing the number of samples of the model to be trained each time.

[0106] Wherein, the operation process of each GRU in the three layers of GRU of the velocity interpretation model and the density interpretation model includes: controlling how the state of the previous moment affects the current state; controlling how the current input affects the state of the previous moment; calculating the candidate hidden state according to the current input and the reset state of the previous moment; calculating the final hidden state of the current moment according to the influence of the state of the previous moment on the current state.

[0107] Step 105, training the density interpretation model with the training data in the training data set of the density interpretation model as the final density interpretation model P1 after completing the training, and training the velocity interpretation model with the training data in the training data set of the velocity interpretation model as the final velocity interpretation model S1 after completing the training.

[0108] Step 106, migrating the model parameters of the model P1 and the model S1 respectively, thereby constructing a wave impedance inversion model for multi-task learning.

[0109] In the embodiments of the present application, the wave impedance inversion model is in parallel with the density interpretation model and the velocity interpretation model. Specifically, the loss function of the wave impedance inversion model contains three parts, namely the density loss loss_ρ, the velocity loss loss_v and the wave impedance loss loss_e, and the calculation method is as follows:

[0110] Wherein:

[0111] loss_ρ is the inversion loss of the density module;

[0112] loss_v is the inversion loss of the velocity module;

[0113] loss_e is the inversion loss of the wave impedance module;

[0114] N: the number of training samples;

[0115] ρ i : represents the real density value;

[0116] represents the density interpretation model inversion density value;

[0117] v i : represents the real velocity value;

[0118] represents the velocity interpretation model inversion velocity value;

[0119] e i : represents the real wave impedance value;

[0120] represents the wave impedance inversion model inversion value.

[0121] Step 107, comprehensively training the wave impedance inversion model and determining its loss function; using the wave impedance inversion model, completing the inversion processing of the density, velocity and wave impedance of the seismic data respectively.

[0122] The last layer of the wave impedance inversion model is a product operator of vectors. The wave impedance is directly calculated using the output of the velocity interpretation model and the output of the density interpretation model. The input of the wave impedance inversion model is seismic data, and the dimension of the seismic data is (B, 50, m). The output of the wave impedance inversion model includes three parts, namely, velocity, density and wave impedance, and the dimension of the output is (B, 1, m).

[0123] The essence of the technical solution of the embodiment of the application is further illustrated by a specific example.

[0124] Through the above technical means, the technical solution of the embodiment of the application realizes high-precision seismic wave impedance inversion.

[0125] The essence of the technical solution of the embodiment of the application is further illustrated by a specific example.

[0126] The Lu (LU) layer section of M oilfield is used as a research target. FIG. 2 is a schematic diagram of one-dimensional seismic waveform data and wave impedance data according to an embodiment of the application. As shown in FIG. 2, the data required in the implementation process includes one-dimensional seismic data, one-dimensional density data in the logging data, one-dimensional velocity data and one-dimensional wave impedance data.

[0127] The technical solution of the embodiment of the application is implemented in the following specific manner:

[0128] Step (1), after well-seismic calibration, the target layer section LU sample thickness is selected as 40 meters, and the depth direction sample point number is 80. The training sample labels are intercepted from the logging data density curve and the velocity curve. The seismic data of the same depth section is obtained from the seismic data in a 5*5 grid manner, and the frequency spectrum analysis is performed to determine that the seismic main frequency f of the LU layer section is 45HZ. The density interpretation model training data set is established using the seismic data and the density data, the velocity interpretation model training data set is established using the seismic data and the velocity data, and the wave impedance inversion data set is established using the seismic data and the wave impedance data. The dimension of the input data of the wave impedance inversion model is (32, 50, 80), and the output dimension is (32, 1, 80).

[0129] Step (2), the velocity interpretation model and the density interpretation model are constructed to complete the nonlinear mapping from the seismic data to the velocity and the density. Both the velocity interpretation model and the density interpretation model are constructed using GRU neural units, and the number of layers of the GRU neural units is 3 layers. The last layer of the velocity interpretation model and the density interpretation model uses a linear rectification function (Relu). The input dimension of the velocity interpretation model and the density interpretation model is the seismic data (32, 50, 80). The dimensions of the output velocity and the output density are (32, 1, 80).

[0130] Step (3), first, the density interpretation model is trained using seismic data and density data, and the last model P1 after training is saved, and then the velocity interpretation model is trained using seismic data and velocity data, and the last model S1 after training is saved. The hyperparameters used when training the velocity model and the density model independently are respectively: the learning rate is set to 0.001, the epoch is set to 1000, the optimization function is set to the Adam (nonlinear activation function used in neural networks) optimization function, the batch is set to 32, the feature dimension channel of the model is 50 in the first layer, 128 in the second layer, and 256 in the third layer.

[0131] Step (4), the model parameters of the velocity model P1 and the density model S1 are migrated, and a multi-task learning wave impedance inversion model is constructed, which is a parallel connection of the density interpretation model and the velocity interpretation model, and the last layer of the wave impedance inversion model is a vector product operator. The output of the velocity interpretation model and the output of the density interpretation model are used to directly calculate the wave impedance.

[0132] Wave impedance e = p.v

[0133] e is the wave impedance, p is the density, and v is the velocity.

[0134] As shown in FIG. 3, the input of the wave impedance inversion model of the embodiment of the present application is seismic data, and the dimension is (32, 50, 80). The output of the wave impedance inversion model includes three parts, namely velocity, density and wave impedance, and the dimension is (32, 1, 80). During comprehensive training, the loss function includes three parts, namely velocity interpretation loss, density interpretation loss and wave impedance interpretation loss. The hyperparameters used when training the wave impedance inversion model comprehensively are respectively: the learning rate is set to 0.00001, the epoch is set to 5000, the optimization function is set to the Adam optimization function, and the batch is set to 32.

[0135] After the training of the seismic inversion model is completed, the model can be directly used on the non-well seismic data of the target object such as the bottom section. The wave impedance inversion of all seismic traces is completed in a loop, and finally the wave impedance data corresponding to the seismic traces is combined according to the positions of the seismic traces to form a wave impedance volume.

[0136] FIG. 4 is a schematic diagram of the composition structure of the multi-task intelligent seismic wave impedance inversion device according to an embodiment of the present application. As shown in FIG. 4, the multi-task intelligent seismic wave impedance inversion device according to the embodiment of the present application comprises:

[0137] The first determination unit 40 is configured to calibrate well-seismic and determine target layer samples.

[0138] The acquisition unit 41 is configured to acquire a logging data density curve and a velocity curve, and cut a training sample label from the logging data density curve and the velocity curve.

[0139] The first construction unit 42 is configured to acquire seismic data of a same depth section as a target layer section from seismic data, perform spectral analysis on the acquired seismic data, construct a training data set of a density interpretation model using the seismic data after the spectral analysis and the density data, and construct a training data set of a velocity interpretation model using the seismic data after the spectral analysis and the velocity data.

[0140] The second construction unit 43 is configured to construct the velocity interpretation model and the density interpretation model based on the training data sets respectively, and map the seismic data to velocity and density respectively according to the velocity interpretation model and the density interpretation model. The velocity interpretation model and the density interpretation model are both constructed using a gated recurrent unit (GRU) neural unit, and the number of GRU layers is 3. The dimension of the seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of the output velocity and the output density is both (B, 1, m). B represents the number of samples of the model in each training.

[0141] The first training unit 44 is configured to train the density interpretation model using training data in the training data set of the density interpretation model, as a final density interpretation model P1, and train the velocity interpretation model using training data in the training data set of the velocity interpretation model, as a final velocity interpretation model S1.

[0142] The third construction unit 45 is configured to migrate the model parameters of the model P1 and the model S1 respectively, so as to construct a multi-task learning wave impedance inversion model. The wave impedance inversion model is a parallel connection of the density interpretation model and the velocity interpretation model.

[0143] The second training unit 46 is configured to comprehensively train the wave impedance inversion model, and determine a loss function of the wave impedance inversion model.

[0144] The inversion processing unit 47 is configured to complete inversion processing of density, velocity and wave impedance of seismic data respectively using the wave impedance inversion model.

[0145] On the basis of the multi-task intelligent seismic wave impedance inversion device shown in FIG. 4, the multi-task intelligent seismic wave impedance inversion device of the embodiment of the present application further comprises:

[0146] The calculation unit (not shown in FIG. 4) is configured to acquire velocity data and density data respectively according to a logging data density curve and a velocity curve, calculate the product of the velocity and the density as wave impedance.

[0147] In an implementable manner, the acquisition unit 41 is further configured to:

[0148] Using well coordinates, seismic trace data in a unit area grid centered on the corresponding position is obtained as model input data, and the frequency domain characteristics of the seismic data are obtained using the following spectral analysis method;

[0149] Wherein:

[0150] f(t) is a one-dimensional seismic trace signal, F(w) is the frequency spectrum of the one-dimensional seismic trace signal in the frequency domain, j is a complex unit, and w is the base frequency of the one-dimensional seismic trace signal;

[0151] According to the spectral characteristics, a first main frequency is selected, and the following bandpass filter is used to obtain the seismic waveform H(f) corresponding to the first main frequency:

[0152] Wherein, f is the first main frequency, f L is a low-pass frequency value centered on the first main frequency, and f H is a high-pass frequency value centered on the first main frequency.

[0153] In an implementation, the second construction unit 43 is further configured to:

[0154] Determine the number of points of the target interval sample as m, cut the density and velocity training sample labels from the logging data, the label dimension is (1, m), and obtain the p x p grid seismic data of the same depth interval as the target interval from the seismic data, the dimension is (p, p, m);

[0155] The dimension of the wave impedance is (1, m), and the seismic waveform corresponding to the first main frequency of the seismic data is obtained using frequency filtering, and the dimension is (p, p, m);

[0156] Merge the seismic data and the seismic waveform corresponding to the main frequency f, merge the seismic data at different grid positions, and use the merged data as the input model data, the dimension of which is (50, m).

[0157] The operation process of each GRU in the three-layer GRU of the velocity interpretation model and the density interpretation model includes: controlling how the state at the previous moment affects the current state; controlling how the current input affects the state at the previous moment; calculating a candidate hidden state according to the current input and the reset state at the previous moment; and calculating a final hidden state at the current moment according to the influence of the state at the previous moment on the current state.

[0158] In an implementation, the loss function of the wave impedance inversion model includes three parts, namely, the density loss loss_ρ, the velocity loss loss_v and the wave impedance loss loss_e, and the calculation method is as follows:

[0159] wherein:

[0160] loss p is the inversion loss of the density module;

[0161] loss v is the inversion loss of the velocity module;

[0162] loss e is the inversion loss of the wave impedance module;

[0163] N: number of training samples;

[0164] p i : represents the true density value;

[0165] represents the density interpretation model inversion density value;

[0166] v i : represents the true velocity value;

[0167] represents the velocity interpretation model inversion velocity value;

[0168] e i : represents the true wave impedance value;

[0169] represents the seismic wave impedance inversion model inversion value.

[0170] In the example embodiments, the aforementioned units and the like can be implemented by one or more Central Processing Units (CPU), Graphics Processing Units (GPU), Application Specific Integrated Circuits (ASIC), Digital signal processors (DSP), Programmable Logic Devices (PLD), Complex Programmable Logic Devices (CPLD), Field-Programmable Gate Arrays (FPGA), general purpose processors, controllers, microcontrollers (MCU), microprocessors (Microprocessor), or other electronic elements.

[0171] With regard to the apparatus in the above-described embodiments, the specific manner in which the various modules and units perform operations has been described in detail in the embodiments relating to the method, and thus will not be described in detail here.

[0172] FIG. 5 shows a schematic block diagram of an example network element 800 that can be used to implement embodiments of the present disclosure. As shown in FIG. 5, the network element 800 includes a computing unit 801 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. Various programs and data required for the operation of the network element 800 can also be stored in the RAM 803. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0173] A plurality of components in the network element 800 are connected to the I / O interface 805, including an input unit 806 such as a keyboard, a mouse, etc., an output unit 807 such as various types of displays, speakers, etc., a storage unit 808 such as a magnetic disk, an optical disk, etc., and a communication unit 809 such as a network card, a modem, a data processing transceiver, etc. The communication unit 809 allows the network element 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0174] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the multitask intelligent seismic wave impedance inversion method. For example, in some embodiments, the multitask intelligent seismic wave impedance inversion method can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the network element 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the multitask intelligent seismic wave impedance inversion method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the steps of the multitask intelligent seismic wave impedance inversion method by any other appropriate means, such as by means of firmware.

[0175] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0176] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0177] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0178] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0179] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0180] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server is generally established by computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, servers of a distributed system, or servers combined with a blockchain.

[0181] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, without departing from the desired results of the technology disclosed in the present disclosure, which are not limited herein.

[0182] In addition, the terms "first", "second", etc., are used herein only to describe different instances, and do not imply or suggest relative importance or imply the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0183] The above merely provides the specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present disclosure, which should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A multitask intelligent seismic wave impedance inversion method, wherein, The method comprises: well-to-seismic calibration is performed, and a target interval sample is determined; logging data density curves and velocity curves are obtained, and training sample labels are cut from the logging data density curves and the velocity curves; seismic data of the same depth interval as the target interval are obtained from seismic data, and spectral analysis is performed on the obtained seismic data, training data sets of a density interpretation model are constructed using the seismic data after spectral analysis and the density data, and training data sets of a velocity interpretation model are constructed using the seismic data after spectral analysis and the velocity data; the velocity interpretation model and the density interpretation model are respectively constructed based on the training data sets, and seismic data are mapped into velocity and density according to the velocity interpretation model and the density interpretation model; wherein the velocity interpretation model and the density interpretation model are both constructed using a gated recurrent unit (GRU) neural unit, and the number of GRU layers is 3; the dimension of seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of output velocity and density is both (B, 1, m), wherein B represents the number of samples of the model in each training; training data in the training data set of the density interpretation model are used to train the density interpretation model, serving as a final density interpretation model P1 after training is completed, and training data in the training data set of the velocity interpretation model are used to train the velocity interpretation model, serving as a final velocity interpretation model S1 after training is completed; model parameters of the model P1 and the model S1 are migrated respectively, so as to construct a wave impedance inversion model for multi-task learning, wherein the wave impedance inversion model is a parallel connection of the density interpretation model and the velocity interpretation model; the wave impedance inversion model is comprehensively trained, and a loss function thereof is determined; and the wave impedance inversion model is used to complete inversion processing of seismic data on density, velocity and wave impedance.

2. The method of claim 1, wherein, The method further comprises: velocity data and density data are obtained from the logging data density curves and the velocity curves respectively, and the product of the velocity and the density is calculated as wave impedance.

3. The method of claim 1 or 2, wherein, The seismic data of the same depth interval as the target interval are obtained from the seismic data, and the spectral analysis on the obtained seismic data comprises: Using well coordinates, seismic trace data in a unit area grid with the corresponding position as the center is obtained as model input data, and the following spectral analysis method is used to obtain the frequency domain characteristics of the seismic data. wherein: f(t) is a one-dimensional seismic trace signal, F(w) is the frequency spectrum of the one-dimensional seismic trace signal in the frequency domain, j is a complex unit, and w is the base frequency of the one-dimensional seismic trace signal; According to the spectral characteristics, a first main frequency is selected, and a band-pass filter is used to obtain a seismic waveform H(f) corresponding to the first main frequency: where f is the first dominant frequency, f L is a low pass frequency value centered on the first dominant frequency, f H is a high pass frequency value centered on the first dominant frequency.

4. The method of claim 3, wherein, The velocity interpretation model and the density interpretation model are respectively constructed based on the training data sets, comprising: the number of points of the target interval sample is determined as m, the density and velocity training sample labels are cut from the logging data, and the label dimension is (1, m); p×p grid seismic data of the same depth interval as the target interval are obtained from the seismic data, and the dimension is (p, p, m); the dimension of the wave impedance is (1, m), the seismic waveform corresponding to the first main frequency of the seismic data is obtained using frequency filtering, and the dimension is (p, p, m); the seismic data and the seismic waveform corresponding to the main frequency f are merged, the seismic data at different grid positions are merged, and the merged data are used as the data input into the model, and the dimension is (50, m).

5. The method of claim 1, wherein, The operation process of each of the three layers of GRUs of the velocity interpretation model and the density interpretation model includes: controlling how the state at the previous moment affects the current state; controlling how the current input affects the state at the previous moment; calculating a candidate hidden state according to the current input and the reset state at the previous moment; and calculating a final hidden state at the current moment according to the influence of the state at the previous moment on the current state.

6. The method of claim 1, wherein, The loss function of the wave impedance inversion model includes three parts, namely, density loss loss p, velocity loss loss v and wave impedance loss loss e, and is calculated as follows: loss = loss p + loss v + loss e Wherein: loss_p is the inversion loss of the density module; loss_v is the inversion loss of the velocity module; loss_e is the inversion loss of the wave impedance module; N: the number of training samples; ρ i : denotes the true density value; den represents the density value inverted by the density interpretation model; v i : indicates a true speed value; v represents the velocity value inverted by the velocity interpretation model; e i : represents the real wave impedance value; ei represents the wave impedance value inverted by the wave impedance inversion model.

7. A multitask intelligent seismic wave impedance inversion apparatus, wherein, The device comprises: A first determination unit is configured to perform well-log calibration and determine a target layer sample. An acquisition unit is configured to acquire a density curve and a velocity curve of logging data, and cut training sample labels from the density curve and the velocity curve of logging data. A first construction unit is configured to acquire seismic data of the same depth layer as the target layer from seismic data, perform spectral analysis on the acquired seismic data, use the seismic data after spectral analysis and density data to construct a training data set of the density interpretation model, and use the seismic data after spectral analysis and velocity data to construct a training data set of the velocity interpretation model. A second construction unit is configured to construct the velocity interpretation model and the density interpretation model based on the training data sets, respectively, and map the seismic data to velocity and density according to the velocity interpretation model and the density interpretation model, respectively. The velocity interpretation model and the density interpretation model are both constructed using a gated recurrent GRU neural unit, and the number of GRU layers is three. The dimension of the seismic data input into the velocity interpretation model and the density interpretation model is (B, 50, m), and the dimension of the output velocity and density is (B, 1, m). B represents the number of samples of the model at each training time. A first training unit is configured to train the density interpretation model using training data in the training data set of the density interpretation model as a final density interpretation model P1 after training is completed, and train the velocity interpretation model using training data in the training data set of the velocity interpretation model as a final velocity interpretation model S1 after training is completed. A third construction unit is configured to migrate the model parameters of the model P1 and the model S1, respectively, to construct a multi-task learning wave impedance inversion model. The wave impedance inversion model is in parallel with the density interpretation model and the velocity interpretation model. A second training unit is configured to comprehensively train the wave impedance inversion model and determine a loss function thereof. An inversion processing unit is configured to use the wave impedance inversion model to complete inversion processing of density, velocity, and wave impedance of seismic data, respectively.

8. The apparatus of claim 7, wherein, The device further comprises: A calculation unit is configured to acquire velocity data and density data from the density curve and the velocity curve of logging data, respectively, calculate the product of the velocity and the density as wave impedance.

9. The apparatus of claim 7 or 8, wherein, The acquisition unit is further configured to: Using well coordinates, seismic trace data in a unit area grid with the corresponding position as the center is obtained as model input data, and the following spectral analysis method is used to obtain the frequency domain characteristics of the seismic data. Wherein: f(t) is a one-dimensional seismic trace signal, F(w) is the frequency spectrum of the one-dimensional seismic trace signal in the frequency domain, j is a complex unit, and w is a base frequency of the one-dimensional seismic trace signal; According to the spectral characteristics, a first main frequency is selected, and a band-pass filter is used to obtain a seismic waveform H(f) corresponding to the first main frequency: where f is the first dominant frequency, f L is a low-pass frequency value centered on the first dominant frequency, f H is a high-pass frequency value centered on the first dominant frequency.

10. The apparatus of claim 9, wherein, The second construction unit is further configured to: Determine the number of points of the target layer sample as m, intercept the density and velocity training sample label from the logging data, and the label dimension is (1, m); obtain p*p grid seismic data of the same depth layer as the target layer from the seismic data, and the dimension is (p, p, m); The dimension of the wave impedance is (1, m), and the first main frequency of the seismic data is obtained by using frequency filtering to obtain the seismic waveform corresponding to the first main frequency, and the dimension is (p, p, m); The seismic data and the seismic waveform corresponding to the main frequency f are merged, the seismic data at different grid positions are merged, and the merged data is used as the input model data, and the dimension is (50, m).

Images