Fracture attribute seismic inversion method and system based on model-data dual drive

Abstract

The invention provides a crack attribute seismic inversion method and system based on model-data dual drive, and relates to the technical field of seismic reservoir identification, and the method comprises the steps: obtaining pre-stack seismic data; inputting the pre-stack seismic data into the model-data dual-drive fracture attribute inversion network, and performing fracture weakness parameter inversion to obtain a prediction result of a fracture weakness parameter; the model-data dual-drive fracture attribute inversion network takes pre-stack seismic data as the input of the inversion network, and introduces an initial model as an inversion constraint; seismic features are obtained by using the pre-stack seismic data; initial model features are obtained through the initial model, the seismic features and the initial model features are fused, and fracture parameter predicted values are obtained; and inputting the predicted value into a forward model for forward simulation to obtain a synthetic azimuth seismic record difference value, introducing the synthetic azimuth seismic record difference value into a consistency loss function, and performing parameter updating on an inversion network to obtain a final model-data dual-drive fracture attribute inversion network.

Description

Technical Field

[0001] The present disclosure relates to the technical field of seismic reservoir identification, and particularly to a method and system for seismic inversion of fracture attributes based on model-data dual drive. Background Art

[0002] The statements in this part merely provide background technical information related to the present disclosure, and do not necessarily constitute prior art.

[0003] Fractures are key structural elements of oil and gas reservoirs. Accurately predicting fracture attributes is of great significance for oil and gas exploitation.

[0004] Under complex geological backgrounds, traditional methods predict fracture attributes by constructing analytical functions, but it is difficult to characterize the complex non-linear relationship between seismic data and fracture attributes, and the prediction accuracy is limited. The development of deep learning has brought new opportunities to the field of geophysical inversion. Deep neural networks represented by Convolutional Neural Network (CNN) are widely used in fracture parameter prediction.

[0005] In fracture parameter prediction and fracture attribute inversion, the convolutional neural network has currently demonstrated powerful feature extraction capabilities. However, this method highly depends on labeled data. The finiteness and insufficiency of labeled data seriously affect the prediction accuracy. In addition, all existing methods lack geophysical principle constraints and cannot ensure that the inversion process follows geophysical laws during prediction, resulting in low inversion accuracy of fracture attributes. Summary of the Invention

[0006] To solve the above problems, the present disclosure proposes a method and system for seismic inversion of fracture attributes based on model-data dual drive, and proposes a model-data dual drive network that integrates semi-supervised learning and geophysical constraints. By designing a semi-supervised learning framework that includes an unlabeled data consistency loss function, the problem of scarce labeled data is alleviated. Secondly, an initial model is introduced into the inversion network to provide geological prior information for inversion. A fracture weakness forward model is constructed based on HTI media and embedded into the semi-supervised learning framework to achieve deep fusion and collaborative drive of the model and data.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] A method for seismic inversion of fracture attributes based on model-data dual drive includes:

[0009] Obtain pre-stack seismic data;

[0010] Input the pre-stack seismic data into the model-data dual drive fracture attribute inversion network to perform fracture weakness parameter inversion and obtain the prediction result of the fracture weakness parameter;

[0011] Among them, the training process of the model-data dual-driven fracture attribute inversion network includes: using pre-stack seismic data as the input of the inversion network, introducing an initial model as the inversion constraint, and constructing a loss function for a semi-supervised learning mechanism; First, extract features from the pre-stack seismic data to obtain seismic features; extract features from the geological structure information in the initial model to obtain initial model features, fuse the seismic features and the initial model features, and output the predicted value of the fracture weakness parameter; input the predicted value of the fracture weakness parameter into the forward modeling of the fracture weakness based on the HTI medium to obtain the synthetic azimuth seismic record difference, introduce the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, update the parameters of the inversion network, and obtain the final model-data dual-driven fracture attribute inversion network.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions:

[0013] A seismic inversion system for fracture attributes based on model-data dual-driving, comprising:

[0014] A data acquisition module for acquiring pre-stack seismic data;

[0015] A prediction module for inputting pre-stack seismic data into the model-data dual-driven fracture attribute inversion network to perform inversion of fracture weakness parameters and obtain the prediction result of the fracture weakness parameters;

[0016] Among them, the training process of the model-data dual-driven fracture attribute inversion network includes: using pre-stack seismic data as the input of the inversion network, introducing an initial model as the inversion constraint, and constructing a loss function for a semi-supervised learning mechanism; First, extract features from the pre-stack seismic data to obtain seismic features; extract features from the geological structure information in the initial model to obtain initial model features, fuse the seismic features and the initial model features, and output the predicted value of the fracture weakness parameter; input the predicted value of the fracture weakness parameter into the forward modeling of the fracture weakness based on the HTI medium to obtain the synthetic azimuth seismic record difference, introduce the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, update the parameters of the inversion network, and obtain the final model-data dual-driven fracture attribute inversion network.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions:

[0018] A computer program product includes a computer program, and when the computer program is executed by a processor, it implements the seismic inversion method for fracture attributes based on model-data dual-driving described above.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium for storing computer instructions, which when executed by a processor, implement the model-data dual-driven fracture property seismic inversion method described above.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device includes: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory to enable the electronic device to implement the model-data dual-driven fracture property seismic inversion method described above.

[0023] Compared with the prior art, the beneficial effects of the present disclosure are:

[0024] The model-data dual-driven fracture property seismic inversion method of the present disclosure introduces a semi-supervised learning architecture based on data augmentation technology, fully mines the information of unlabeled data, expands the diversity of the training set, and further enhances the robustness and stability of network learning. At the same time, to ensure that the inversion process follows geophysical laws, the initial model constraint is incorporated into the inversion network, enabling the inversion work to start from an initial point that conforms to physical principles; then based on the horizontal transverse isotropic (HTI) medium theory, a forward model is constructed that can clearly describe the relationship between fracture properties and azimuth seismic data. This model is realized through the convolution operation of the azimuth PP-wave reflection coefficient equation and the seismic wavelet, generating the seismic response characteristics of the fracture weakness parameter, laying a theoretical foundation for subsequent inversion work, and significantly improving the inversion accuracy and calculation efficiency.

[0025] The model-data dual-driven fracture property seismic inversion method of the present disclosure, based on the principle of the physics-informed neural network (PINN), organically integrates the constructed forward model and the inversion network into a semi-supervised framework, creating a set of high-precision model-data dual-driven fracture property inversion network systems, realizing the deep integration and collaborative driving of the model and data. The introduction of the initial model constraint significantly improves the lateral resolution, enabling a more refined characterization of the lateral variation of the fracture reservoir, and providing a reliable technical support for fracture weakness inversion under complex geological backgrounds. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The specification drawings forming a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure.

[0027] Figure 1 It is a schematic structural diagram of the model-data dual-driven fracture property inversion network of the embodiment of the present disclosure;

[0028] Figure 2 Schematic diagram of the working process of the model-data dual-driven fracture property inversion network according to an embodiment of the present disclosure;

[0029] Figure 3 Data-driven forward network structure according to an embodiment of the present disclosure;

[0030] Figure 4 True fracture weakness of the three-dimensional reverse thrust model according to an embodiment of the present disclosure; where, Figure 4 (a) in is Δ N , Figure 4 (b) in is Δ T ;

[0031] Figure 5 Normalized seismic data difference of the three-dimensional reverse thrust model according to an embodiment of the present disclosure;

[0032] Wherein, Figure 5 (a) in has an incident angle of 20°, Figure 5 (b) in has an incident angle of 10°.

[0033] Figure 6 Predicted fracture weakness of the reverse thrust model by different networks according to an embodiment of the present disclosure;

[0034] Wherein, Figure 6 (a) in is the Δ N predicted by Data-Driven-net, Figure 6 (b) in is the Δ T predicted by Data-Driven-net; Figure 6 (c) in is the Δ N predicted by Data-Driven-IM-net, Figure 6 (d) in is the Δ T predicted by Data-Driven-IM-net; Figure 6 (d) in is the Δ N predicted by Model-Data-Driven-net, Figure 6 (f) in is the Δ T predicted by Model-Data-Driven-net.

[0035] Figure 7 Absolute error of the predicted fracture weakness by different networks according to an embodiment of the present disclosure;

[0036] Wherein, Figure 7 (a) and (b) in are respectively the Δ N , Δ T predicted by Data-Driven-net; Figure 7(c) and (d) in [it] are respectively the Δ predicted by Data-Driven-IM-net N , Δ T ; Figure 7 (e) and (f) in [it] are respectively the Δ predicted by Model-Data-Driven-net N , Δ T .

[0037] Figure 8 is the comparison between the network prediction value and the true value of the test set of the present disclosure embodiment;

[0038] Figure 9 is the predicted crack weakness of the actual data by different networks of the present disclosure embodiment;

[0039] Among them, Figure 9 (a) and (b) in [it] are respectively the Δ predicted by Data-Driven-IM-net N , Δ T ; Figure 9 (c) and (d) in [it] are respectively the Δ predicted by Model-Data-Driven-net N , Δ T .

[0040] Figure 10 (a) in [it] is the predicted azimuth seismic record difference of Model-Data-Driven-net of the present disclosure embodiment;

[0041] Figure 10 (a) in [it] is the predicted azimuth seismic absolute error of Model-Data-Driven-net of the present disclosure embodiment;

[0042] Figure 11 is the comparison between the network prediction value and the well logging data of the test well of the present disclosure embodiment. Detailed implementation manners

[0043] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further descriptions of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0046] Embodiment 1

[0047] In an embodiment of the present disclosure, a model-data dual-driven seismic inversion method for fracture properties is provided. A model-data dual-driven fracture property inversion network that fuses semi-supervised learning and geophysical constraints is proposed. First, by designing a semi-supervised framework that includes an unlabeled data consistency loss function, the problem of scarce labeled data is alleviated. Second, an initial model is introduced into the inversion network to provide geological prior information for the inversion. Then, a fracture weakness forward model based on HTI media is constructed and embedded into the semi-supervised learning framework to achieve deep fusion and collaborative driving of the model and data. The implementation steps include:

[0048] Step 1: Obtain pre-stack seismic data;

[0049] Step 2: Input the pre-stack seismic data into the model-data dual-driven fracture property inversion network to perform fracture weakness parameter inversion and obtain the prediction result of the fracture weakness parameter;

[0050] Among them, the training process of the model-data dual-driven fracture property inversion network in the model-data dual-driven seismic inversion method of the present disclosure is as follows: Using the pre-stack seismic data as the input of the inversion network and introducing an initial model as the inversion constraint, a loss function of the semi-supervised learning mechanism is constructed. First, extract the seismic features from the pre-stack seismic data; extract the geological structure information features from the initial model to obtain the initial model features, fuse the seismic features and the initial model features, and output the predicted value of the fracture weakness parameter; input the predicted value of the fracture weakness parameter into the fracture weakness forward model based on HTI media for forward simulation to obtain the synthetic azimuth seismic record difference, introduce the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, update the parameters of the inversion network, and obtain the final model-data dual-driven fracture property inversion network.

[0051] As an embodiment, the construction steps of the model-data dual-driven fracture property inversion network in the model-data dual-driven seismic inversion method of the present disclosure are as follows:

[0052] Step 1: Construct a consistency loss function for unlabeled data, and use a weighted combination strategy to organically integrate it with the supervised learning loss function, thus successfully constructing a loss function suitable for the semi-supervised learning mechanism, and then building a semi-supervised learning framework.

[0053] Step 2: According to the principles of convolutional neural networks and recurrent neural networks, construct an inversion network for predicting fracture weakness from azimuthal seismic data, and introduce an initial model as an inversion constraint through the loss function, so as to provide geological information and low-frequency information for inversion.

[0054] Step 3: Based on the PP-wave reflection coefficient equation of HTI media, derive the AVAZ approximation function under the premise of the assumption that the upper interface is an isotropic medium, and use the convolution theory to construct a forward model of fracture weakness based on HTI media for fracture weakness.

[0055] Step 4: Based on the principle of physics-informed neural networks, integrate the constructed AVAZ (Amplitude Variation with Azimuth) forward model into the semi-supervised framework, construct a model-data dual-driven fracture property inversion network for predicting fracture weakness using azimuthal seismic records, and form a workflow for specific implementation and application.

[0056] As an example, the specific process of the construction and training of the model-data dual-driven fracture property inversion network in the model-data dual-driven fracture property seismic inversion method is as follows:

[0057] S1: Construct a consistency loss function for unlabeled data, and use a weighted combination strategy to organically integrate it with the supervised learning loss function, thus successfully constructing a loss function suitable for the semi-supervised learning mechanism;

[0058] Specifically, in supervised learning, the mean square error (MSE) is often used as the loss function L sup (·), and its expression is:

[0059]

[0060] Among them, Θ represents the parameters of the inversion network, and N l is the number of labeled samples, x i (i = 1, 2,..., N l ) is the labeled sample, y i is the known sample label, and F Θ (x i ) is the network prediction value of the sample.

[0061] For unlabeled samples x j (j = 1, 2,..., N u),the network will give the corresponding predicted value F Θ (x j ). To mine the information of unlabeled samples, the present disclosure assumes that the predicted value F Θ (x j ) is transformed by a certain function f to obtain the transformed value f(F Θ (x j ))), and defines the consistency loss function:

[0062]

[0063] Based on the above formula, the present disclosure constructs the loss function of the semi-supervised learning mechanism by combining the supervised learning loss function and the consistency loss function in a weighted combination manner:

[0064] L semi (Θ, f) = αL sup (Θ) + βL cons (Θ, f) (3)

[0065] where L semi (Θ) is the total loss function of the semi-supervised learning network, α, β ∈ [0, 1] are weight coefficients used to balance the weights of labeled data and unlabeled data, and α + β = 1.

[0066] S2: According to the principles of convolutional neural network and recurrent neural network, construct an inversion network for predicting fracture weakness from azimuthal seismic data, and introduce an initial model as an inversion constraint through the loss function, so as to provide geological information and low-frequency information for inversion;

[0067] Specifically, based on the operation characteristics of convolutional neural network and recurrent neural network (Recurrent Neural Network, RNN), the present disclosure constructs an inversion network composed of two parts: a seismic feature extraction module and an initial model constraint, as shown in Figure 1 .

[0068] Furthermore, the inversion network is composed of a seismic feature extraction module and an initial model constraint. The pre-stack seismic data and the initial model are input into the inversion network, and the initial model serves as an inversion constraint; the seismic feature extraction module includes a global feature extraction module and a local feature extraction module; the global feature extraction module is composed of a series of gated recurrent units GRUs, and the local feature extraction module is composed of one-dimensional convolutional Conv1d blocks with different dilation factors in parallel, which are used to extract the global features and local features of the pre-stack seismic data respectively, and fuse the global features and local features to obtain seismic features.

[0069] Specifically, the inversion network takes pre-stack seismic data as input. First, the pre-stack seismic data is simultaneously input into two feature extraction modules, namely the global feature extraction module and the local feature extraction module. The global feature extraction module is composed of a series of gated recurrent units (GRUs). The gating mechanism in the GRU can effectively process the information flow within the sequence, avoiding the common problems of gradient vanishing or explosion in traditional RNNs, and thus can efficiently and accurately extract global features from complex time-series data such as seismic data. The local feature extraction module is composed of one-dimensional convolutional (Conv1d) blocks with different dilation factors in parallel to capture local features at different scales. Each Conv1d block consists of a one-dimensional convolutional layer, a batch normalization layer (BN), and a hyperbolic tangent activation function (Tanh) in sequence. The extracted local features are integrated through a fully connected layer and then output after passing through three convolutional blocks. Finally, the output result of feature extraction is seismic features, which are the fusion of the local features and global features of the pre-stack seismic data. However, during this process, due to the convolutional operation, the dimension of the input data will change. To ensure that the sampling rates of seismic data and well logging data match, the present disclosure uses a deconvolution block to upsample the high-dimensional features, and its structure is the same as that of the convolutional block. The end of the inversion network is a regression module, which consists of a GRU and a fully connected layer, and is used to normalize the output data and achieve the conversion from feature representation to fracture parameter estimation values.

[0070] Furthermore, an initial model is constructed based on well logging data, seismic attributes, and horizon information, that is, the initial distribution of fracture parameters. Among them, the initial model contains geological structure information and low-frequency information, which are used as important constraint terms in fracture parameter inversion. The initial model is processed by the input network of the initial model constraint part. In the inversion network, the input network of the initial model constraint part consists of three convolutional blocks. The first two convolutional blocks are used for feature extraction, and the last convolutional block is used for regression prediction of the parameter domain, and the initial model features are output.

[0071] Furthermore, the seismic features and the initial model features are fused to output the predicted value of the fracture weakness parameter. The loss function of this inversion network is:

[0072]

[0073] where, Θ I1 and Θ I2 respectively represent the input network parameters of seismic data and the initial model, m label and s label are the labeled samples and their corresponding labels respectively, and m0 is the initial model. λ is the weight coefficient. When the initial model does not match the true fracture parameters due to complex underground structures or other reasons, its size can be adjusted to avoid the network overlearning the error information in the initial model.

[0074] S3: Based on the PP-wave reflection coefficient equation in HTI media, under the premise of the assumption that the upper interface is an isotropic medium, the AVAZ approximation function is derived, and the forward modeling model of fracture weakness based on HTI media is constructed using the convolution theory;

[0075] Specifically, in the forward modeling model of fracture weakness in HTI media, a PP-wave reflection coefficient equation based on HTI media is constructed, including: for a formation with vertical fractures developed in an isotropic background formation, which is characterized as a transversely isotropic medium with a horizontal axis of symmetry, i.e., HTI media. Based on the Born approximation and the stationary phase method, the PP-wave reflection coefficient equation in HTI media is expressed as the sum of two components, namely, the azimuth-independent reflection coefficient in the isotropic background medium and the azimuth-dependent reflection coefficient caused by fractures. The specific implementation is as follows:

[0076] According to the definition of fractured media by HSU et al., ignoring the specific shape of the fractures and assuming that the fractures are infinitely thin planes, the concept of fracture weakness is proposed, and its specific expression is:

[0077]

[0078] where M b and μ b represent the longitudinal wave and shear wave moduli of the isotropic background respectively, Z N and Z T represent the normal compliance and tangential compliance of the fractures respectively, Δ N is the normal fracture weakness, which is used to indicate the fluid situation in the fractures, and Δ T is the tangential fracture weakness, which is used to indicate the fracture density.

[0079] For a formation with vertical fractures developed in an isotropic background formation, it can be characterized as a transversely isotropic medium with a horizontal axis of symmetry, i.e., HTI media. Based on the Born approximation and the stationary phase method (2006), the PP-wave reflection coefficient equation in HTI media can be expressed as the sum of two components, namely, the azimuth-independent reflection coefficient in the isotropic background medium and the azimuth-dependent reflection coefficient

[0080]

[0081] sum:

[0082]

[0083] where θ represents the longitudinal wave incident angle, φ represents the azimuth angle, ρ represents the density, ΔM b and Δμ band Δρ represent the differences in the longitudinal wave, shear wave moduli, and density between the reflection interfaces, and represents the average value of the reflection interface. and represent the difference in the fracture weakness of the two layers above and below the interface. g = μ b / M b represents the ratio of the longitudinal wave modulus to the shear wave modulus.

[0084] Furthermore, in order to focus on the seismic response characteristics of fracture weakness and thus eliminate the influence of PP wave reflection in an isotropic background medium, under the assumption that the upper layer is a homogeneous isotropic medium, taking two azimuth angles φ1 and φ2 as examples, the difference in the reflection coefficient is calculated:

[0085]

[0086] In the actual seismic data acquisition process, seismic waves will undergo multiple reflections and transmissions with the underground medium. Based on the convolution model, this disclosure can approximately simulate this forward process, that is, the seismic data s(θ, φ) is obtained by convolving the reflection coefficient with the seismic wavelet w, and considering the influence of random noise e:

[0087]

[0088] Since the seismic data response is more sensitive to elastic parameters, in order to ensure the stability and accuracy of fracture weakness inversion, this disclosure calculates the difference of seismic data in different azimuths, and according to Equation (9), the following formula can be derived:

[0089]

[0090] where, Δs(θ, φ1, φ2) represents the difference in seismic amplitudes corresponding to different azimuths. m * =[Δ N ,Δ T T represents the fracture weakness to be inverted, and F(·) represents the forward simulation process of fracture weakness.

[0091] ​S4: Based on the principle of the physics-informed neural network, the constructed AVAZ forward model is integrated into the semi-supervised framework to construct and complete the model-data dual-driven fracture property inversion network for predicting fracture weakness using azimuthal seismic records, including: introducing the HTI medium fracture weakness forward model into the inversion network, and the residual between the synthetic azimuthal seismic record difference generated by it and the actual azimuthal seismic record difference is introduced into the consistency loss function as an evaluation of the consistency degree of the prediction result, and can also participate in the parameter update process of the inversion network to complete the training of the model-data dual-driven fracture property inversion network.

[0092] Specifically, in order to organically integrate the geophysical constraints related to fracture parameters into the inversion process, the present disclosure draws on the core principle of the physics-informed neural network (PINN) proposed by Raissi, relies on the semi-supervised inversion framework, and on the basis of the network with initial model constraints, defines the transformation as the AVAZ forward model to form a workflow of a dual-driven inversion method with both physical models and data-driven models, as Figure 2 shown.

[0093] During the network training process of the present disclosure, due to the successful introduction of the forward model, the residual between the synthetic azimuthal seismic record difference generated by it and the actual azimuthal seismic record difference is incorporated into the consistency loss function and is an important part of it. It can not only reflect the consistency degree of the network prediction result, but also participate in the parameter update process of the inversion network to improve the accuracy and reliability of the inversion:

[0094]

[0095] S5: Use the trained model-data dual-driven fracture property inversion network for fracture property seismic inversion to obtain the fracture weakness prediction result, including: obtaining pre-stack seismic data; inputting the pre-stack seismic data into the model-data dual-driven fracture property inversion network for fracture weakness parameter inversion to obtain the prediction result of the fracture weakness parameter.

[0096] Simulation experiment

[0097] To prove the significant superiority of the proposed method, the present disclosure uses the network for comparative testing: (1) Data-Driven-net, where the forward network is as Figure 3 shown, and there is no initial model constraint in the inversion network; (2) Data-Driven-IM-net with initial model constraint; (3) Model-Data-Driven-net.

[0098] The present disclosure adopts a 3D inverse mask model of 801×801 gathers, and intercepts 100×100 gathers at equal intervals for subsequent processing to reduce computational complexity. Figure 4 The real fracture weakness is shown. The model has 187 sampling points per channel and a sampling interval of 2ms. The synthetic seismic data is generated based on the convolution of the 20Hz main frequency Ricker wavelet and the PP wave reflection coefficient equation. The incident angles are 10° and 20°, and the azimuths are 45° and 60°. In order to simulate the real geological exploration environment, random noise with a signal-to-noise ratio (SNR) of 5 is injected into the synthetic data, such as Figure 5 As shown. Then, 0.2% of the data in the inverse mask model was randomly selected as the training set. The network training used the Adam optimization algorithm, with an initial learning rate of 0.001, weight coefficients α and β set to 1 and 0.1 respectively, and the number of iterative training rounds set to 500 rounds.

[0099] The prediction results and residual analysis based on different network architectures are shown in Figure 6 and Figure 7 The results show that compared with the network without initial model constraints, the predicted parameters of the other two networks are more consistent with the true values, and the lateral resolution is significantly improved. To further verify the network performance, this paper selects a random single channel of the test set for comparison, such as Figure 8 As shown in the figure, the comparison results show that in the data mutation area, the prediction results of the pure data-driven network deviate significantly and cannot accurately capture the characteristics of rapid data changes, while the model-data dual-driven network can accurately describe this change.

[0100] In order to quantitatively evaluate the network inversion performance, this paper introduces the mean square error (MSE) and Pearson correlation coefficient (PCC) dual indicators for analysis, as shown in Table 1.

[0101] Table 1. Dual index analysis of mean square error (MSE) and Pearson correlation coefficient (PCC)

[0102]

[0103] The data in Table 1 show that Model-Data-Driven-net has the smallest MSE value compared with other network prediction results, indicating that the deviation between the predicted value and the true value is smaller; at the same time, it has the largest PCC, which indicates that the linear correlation between the predicted result and the true value is stronger, which strongly proves that the present invention has higher accuracy and reliability in crack weakness prediction.

[0104] In addition, the present disclosure also compared the training times of different networks. The training times of Data-Driven-net, Data-Driven-IM-net, and Model-Data-Driven-net were 45, 44, and 28 minutes respectively. Therefore, it was verified that the present disclosure can significantly reduce the network training time cost and effectively improve the fracture weakness prediction efficiency.

[0105] To further verify the applicability of the proposed method, the present disclosure applied the model-data dual-driven network to an oil and gas field in the Sichuan Basin. The target reservoir in the study area is mainly composed of sandstone, and fractures are widely developed and have good continuity. Seismic data with azimuth angles of 20° and 125° were selected as inputs and processed through noise reduction and amplitude preservation. The present disclosure selected two wells as the training set and used horizon information combined with a filtering method to interpolate the logging data to generate an initial model. Additionally, one well was selected for testing.

[0106] Figure 9 and Figure 10 The fracture weakness prediction results of the model-data dual-driven, the difference between the forward simulation-generated synthetic azimuth seismic data, and their errors are shown respectively. The results show that the peak of the fracture weakness predicted by the network appears in the time interval of 2.35 to 2.4 seconds, indicating a relatively high degree of fracture development in this area. The fracture weakness predicted by the model-data dual-driven network is highly consistent with the actual logging curve in the overall trend. In addition, the difference between the synthetic seismic data predicted by the forward simulation matches the actual data.

[0107] To further prove the effectiveness of introducing the forward model, the present disclosure compared the inversion results of the data-driven network without the forward model and the model-data dual-driven network for the test well with the actual dataset, as Figure 11 shown. It can be observed that in the target reservoir area after 2.4 seconds, the data-driven-initial model network failed to effectively invert the peak of the fracture weakness. In contrast, the model-data dual-driven network accurately captured this key information, further highlighting its advantages and reliability in the fracture weakness inversion of complex reservoirs.

[0108] To verify the effectiveness of the method, this disclosure conducts a comparative analysis on synthetic data using three different network structures: Data-Driven-net, Data-Driven-IM-net, and Model-Data-Driven-net. The results show that even when the labeled data is limited, the semi-supervised framework can still obtain ideal prediction results. The introduction of the initial model significantly improves the lateral resolution, enabling a more refined characterization of the lateral variations in the fractured reservoir. In the comparison of the prediction performance metrics of MSE, PCC, and training duration, Model-Data-Driven-net exhibits lower MSE, higher PCC, and optimal training efficiency, highlighting the importance of the forward modeling process in endowing the network with strong geophysical constraints, which can effectively improve the prediction accuracy and inversion efficiency. This disclosure further applies the proposed method to actual data. Model-Data-Driven-net not only has the highest prediction accuracy, but also the synthetic seismic difference is closest to the actual data, significantly improving the reliability of the inversion results. Therefore, it is confirmed that this disclosure plays a role in improving both the inversion accuracy and efficiency, providing a reliable technical support for the fracture weakness inversion under complex geological backgrounds.

[0109] Example 2

[0110] In an embodiment of this disclosure, a seismic inversion system for fracture properties based on model-data dual driving is provided, including:

[0111] A data acquisition module for acquiring pre-stack seismic data;

[0112] A prediction module for inputting the pre-stack seismic data into the model-data dual-driven fracture property inversion network to perform inversion of fracture weakness parameters and obtain the prediction result of the fracture weakness parameters;

[0113] Among them, the training process of the model-data dual-driven fracture property inversion network includes: using the pre-stack seismic data as the input of the inversion network, and introducing an initial model as the inversion constraint to construct a loss function for the semi-supervised learning mechanism; first, extracting features from the pre-stack seismic data to obtain seismic features; extracting features from the geological structure information in the initial model to obtain initial model features, fusing the seismic features and the initial model features, and outputting to obtain the predicted value of the fracture weakness parameters; inputting the predicted value of the fracture weakness parameters into the forward simulation based on the HTI medium fracture weakness forward model to obtain the synthetic azimuth seismic record difference, introducing the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, and updating the parameters of the inversion network to obtain the final model-data dual-driven fracture property inversion network.

[0114] Example 3

[0115] In one embodiment of the present disclosure, a computer program product is provided, including a computer program which, when executed by a processor, implements the model-data dual-driven fracture property seismic inversion method described above.

[0116] Embodiment 4

[0117] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium is used to store computer instructions which, when executed by a processor, implement the model-data dual-driven fracture property seismic inversion method described above.

[0118] Embodiment 5

[0119] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes and implements the model-data dual-driven fracture property seismic inversion method described above.

[0120] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present disclosure.

[0121] It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0122] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate computer-implemented processing, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0123] Although the specific embodiments of the present disclosure have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that various modifications or variations that can be made without creative efforts on the basis of the technical solutions of the present disclosure are still within the scope of protection of the present disclosure.

Claims

1. A fracture attribute seismic inversion method based on model-data dual drive, characterized in that: include: Acquire pre-stack seismic data; The pre-stack seismic data is input into the model-data dual-driven fracture attribute inversion network to perform fracture weakness parameter inversion and obtain the fracture weakness parameter prediction results; Among them, the training process of the model-data dual-driven fracture attribute inversion network includes: taking pre-stack seismic data as the input of the inversion network, introducing the initial model as the inversion constraint, and constructing the loss function of the semi-supervised learning mechanism; first, extracting the features of the pre-stack seismic data to obtain the seismic features; extracting the features of the geological structure information in the initial model to obtain the initial model features, fusing the seismic features with the initial model features, and outputting the predicted values ​​of the fracture weakness parameters; inputting the predicted values ​​of the fracture weakness parameters into the HTI medium fracture weakness forward model for forward simulation to obtain the synthetic azimuth seismic record difference, introducing the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, updating the parameters of the inversion network, and obtaining the final model-data dual-driven fracture attribute inversion network.

2. The fracture attribute seismic inversion method based on model-data dual drive according to claim 1, characterized in that: Constructing an inversion network, the inversion network consists of two parts: a seismic feature extraction module and an initial model constraint. Prestack seismic data and an initial model are input into the inversion network, and the initial model is used as an inversion constraint. The seismic feature extraction module includes a global feature extraction module and a local feature extraction module. The global feature extraction module is composed of a series of gated recurrent units (GRUs), and the local feature extraction module is composed of parallel one-dimensional convolution Conv1d blocks with different dilation factors, which are used to extract the global and local features of pre-stack seismic data respectively, and the global and local features are fused to obtain the seismic features.

3. The fracture attribute seismic inversion method based on model-data dual drive according to claim 1, characterized in that: The initial model contains geological structure information and low-frequency information, which are used as constraints in the inversion of fracture parameters. The initial model is processed by the initial model constraint part. In the inversion network, the initial model constraint part consists of three convolution blocks. The first two convolution blocks are used for feature extraction, and the last convolution block is used for regression prediction of the parameter domain. The initial model features are output, and the seismic features and the initial model features are fused to output the predicted value of the fracture weakness parameter.

4. The fracture attribute seismic inversion method based on model-data dual drive according to claim 1, characterized in that: In the HTI medium fracture weakness forward model, a PP wave reflection coefficient equation based on HTI medium is constructed, including: a stratum with vertical fractures developed in an isotropic background stratum is characterized as a transversely isotropic medium with a horizontal symmetry axis, namely, an HTI medium. Based on the Born approximation and the steady phase method, the PP wave reflection coefficient equation in the HTI medium is expressed as the sum of two components, namely, the sum of the orientation-independent reflection coefficient under the isotropic background medium and the orientation-dependent reflection coefficient caused by the fracture.

5. The fracture attribute seismic inversion method based on model-data dual drive according to claim 4, characterized in that: Focusing on the seismic response characteristics of fracture weakness, the influence of PP wave reflection under isotropic background medium is eliminated. Under the assumption that the upper layer is a uniform isotropic medium, the difference of reflection coefficient is calculated using two azimuth angles. The forward modeling process is approximately simulated based on the convolution model, that is, the seismic data is obtained by convolution operation of reflection coefficient and seismic wavelet, and the influence of random noise is considered. The forward modeling process of deriving fracture weakness is realized by making a difference on seismic data in different azimuths.

6. The fracture attribute seismic inversion method based on model-data dual drive according to claim 1, characterized in that: The HTI medium fracture weakness forward model is introduced into the inversion network. The residual between the synthetic azimuth seismic record difference and the actual azimuth seismic record difference generated by the model is introduced into the consistency loss function as a measure to evaluate the consistency of the prediction results. At the same time, it can also participate in the parameter update process of the inversion network to complete the training of the model-data dual-driven fracture attribute inversion network.

7. A fracture attribute seismic inversion system based on model-data dual drive, characterized in that: include: A data acquisition module, used for acquiring pre-stack seismic data; A prediction module is used to input pre-stack seismic data into a model-data dual-driven fracture attribute inversion network to perform fracture weakness parameter inversion and obtain fracture weakness parameter prediction results; Among them, the training process of the model-data dual-driven fracture attribute inversion network includes: taking pre-stack seismic data as the input of the inversion network, introducing the initial model as the inversion constraint, and constructing the loss function of the semi-supervised learning mechanism; first, extracting the features of the pre-stack seismic data to obtain the seismic features; extracting the features of the geological structure information in the initial model to obtain the initial model features, fusing the seismic features with the initial model features, and outputting the predicted values ​​of the fracture weakness parameters; inputting the predicted values ​​of the fracture weakness parameters into the HTI medium fracture weakness forward model for forward simulation to obtain the synthetic azimuth seismic record difference, introducing the synthetic azimuth seismic record difference into the consistency loss function of the semi-supervised learning mechanism, updating the parameters of the inversion network, and obtaining the final model-data dual-driven fracture attribute inversion network.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for seismic inversion of fracture attributes based on model-data dual drive described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the fracture attribute seismic inversion method based on model-data dual drive as described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the model-data dual-driven fracture attribute seismic inversion method as described in any one of claims 1 to 6.

Images