A VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs

By constructing the TransU-net model, using angle superposition and logging data augmentation methods, the problems of medium and low frequency anisotropy parameters and training sample dependence in VTI medium seismic AVO inversion in shale reservoirs are solved, and the elastic parameters and Inversion accuracy of anisotropic parameters.

CN118625382BActive Publication Date: 2024-11-12SANYA MARINE OIL & GAS RESEARCH INSTITUTE NORTHEAST PETROLEUM UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411110634.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-12
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

The existing VTI medium seismic AVO inversion method is difficult to accurately obtain low-frequency anisotropic parameters and training samples in shale reservoirs, which affects the inversion accuracy of elastic parameters and anisotropic parameters.

Method used

Using TransU-net model built on Transformer network and convolutional neural network, the TransU-net model is used to superimpose the prestack seismic data of the well bypass channel and augment the well log data, and train samples are generated to train the model to achieve elastic parameters and anisotropy. Inversion of parameters.

Benefits of technology

The inversion accuracy of elastic parameters and anisotropic parameters of shale reservoirs is improved, the dependence on low-frequency anisotropic parameters and training samples is reduced, and the robustness and accuracy of inversion is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118625382B_ABST
    Figure CN118625382B_ABST
Patent Text Reader

Abstract

The present invention discloses a VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs, which belongs to the field of geophysical technology and includes the following steps: constructing a TransU‑net model based on a Transformer network and a convolutional neural network; stacking pre-stack seismic data of well bypasses to obtain angle stacked seismic data to form sample input; augmenting well logging data to form sample output; forming training samples based on sample input and sample output, and training the TransU‑net model based on the training samples; and obtaining inverted elastic parameters and anisotropy parameters based on the trained TransU‑net model. The present invention can alleviate the dependence of inversion on low-frequency anisotropy parameters and training samples with anisotropy parameters as sample output, and improve the inversion accuracy of shale reservoir elastic parameters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of geophysical technology, and in particular relates to a VTI medium data driven seismic AVO inversion method for obtaining elastic parameters of shale oil reservoirs. Background Art

[0002] Elastic parameters are key parameters in the exploration and development of shale reservoirs, and play an important role in sweet spot identification and fracturing design. Seismic AVO (amplitude variation with offset) inversion is a common method for obtaining elastic parameters based on seismic data and logging data. Shale reservoirs mainly develop periodic thin layers with strong heterogeneity and strong anisotropy. Compared with isotropic media, the AVO inversion method based on VTI (vertical transverse isotropy) media can improve the inversion accuracy of elastic parameters of shale reservoirs by simultaneously inverting elastic parameters and anisotropic parameters.

[0003] At present, the commonly used VTI medium seismic AVO inversion methods mainly include model-driven inversion methods based on convolution forward theory and data-driven inversion methods based on neural network technology. The accuracy of the inversion parameters of the former depends on low-frequency elastic parameters and low-frequency anisotropy parameters; the latter requires a large number of representative training samples, in which the sample output includes real elastic parameters and anisotropy parameters. The low-frequency parameters and elastic parameters and anisotropy parameter information in the training samples are usually obtained from well logging data. However, for most actual shale reservoir work areas, well logging data usually do not include accurate anisotropy parameters, which makes it difficult to generate low-frequency parameters and training samples that meet the requirements of VTI medium AVO inversion, affecting the inversion accuracy of elastic parameters and anisotropy parameters.

[0004] Therefore, a VTI medium seismic AVO inversion method is needed that has less dependence on low-frequency anisotropy parameters and training samples with anisotropy parameters as sample output. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs to solve the problems existing in the above prior art.

[0006] To achieve the above object, the present invention provides a VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs, comprising the following steps:

[0007] Based on Transformer network and convolutional neural network, build TransU-net model;

[0008] Stack the pre-stack seismic data of the well bypass to obtain angle stack seismic data to form sample input;

[0009] Performing augmentation processing on the well logging data, and forming sample output based on the augmented well logging data;

[0010] Based on the sample input and sample output, a training sample is formed, and the TransU-net model is trained based on the training sample;

[0011] Based on the trained TransU-net model, the inversion elastic parameters and anisotropic parameters of the test area are obtained.

[0012] Optionally, the TransU-net model includes a CNN part for extracting local information, a Transformer part for extracting global information, and an upsampling part for restoring data features.

[0013] Optionally, the CNN part includes a first feature map, a second feature map and a third feature map, and each feature map is connected by a convolution operation; the third feature map forms a local hidden feature after dimension adjustment, and is input into the Transformer part after a maximum pooling operation.

[0014] Optionally, the Transformer part includes a first Transformer layer, a second Transformer layer, a third Transformer layer, a fourth Transformer layer, a fifth Transformer layer, a sixth Transformer layer, a seventh Transformer layer and an eighth Transformer layer, and each Transformer layer is connected by a multi-layer perceptron; a global hidden feature is formed based on the eighth Transformer layer and input into the upsampling part after dimension adjustment.

[0015] Optionally, the upsampling part includes a first feature map, a second feature map, a third feature map, a fourth feature map, a fifth feature map, a sixth feature map, a seventh feature map and an eighth feature map. After the upsampling operation, each feature map is fused with the feature map corresponding to the CNN part through a jump connection, and then a convolution operation is performed to generate a feature map of the next layer.

[0016] Optionally, the process of obtaining angle-stacked seismic data includes: performing angle stacking on pre-stack seismic data to obtain angle-stacked seismic data.

[0017] Optionally, the process of augmenting the well logging data includes: replacing the low-frequency part of the well logging data, distorting the high-frequency part of the well logging data, and increasing the amount of the well logging data by adding random numbers to the well logging data.

[0018] Optionally, the angle-stacked seismic data include 9°, 16°, 23°, 30° and 37° angle-stacked seismic data obtained by stacking pre-stack seismic data in angle ranges of 3-15°, 10-22°, 17-29°, 24-36° and 31-43°, respectively.

[0019] Optionally, the process of training the TransU-net model includes: inputting training samples into the TransU-net model to calculate the objective function, and iteratively updating the TransU-net model parameters with a back-propagation algorithm until the objective function converges to obtain a trained TransU-net model.

[0020] Optionally, the objective function is expressed as:

[0021]

[0022] in, represents the L2 norm, It indicates the influence ratio of the objective function on the training samples. represents the predicted elasticity parameter, represents the sample elasticity parameter, represents angle stacked seismic data, Represents sample earthquake data.

[0023] Compared with the prior art, the present invention has the following advantages and technical effects:

[0024] The present invention proposes a VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale oil reservoirs. The method can use the TransU-net model constrained by the convolution forward theory to establish a mapping relationship between angle stacking seismic data and elastic parameters and anisotropy parameters; the TransU-net model can be trained using training samples with velocity and density as sample outputs; the trained network model can be used to implement inversion of elastic parameters and anisotropy parameters, thereby alleviating the dependence of inversion on low-frequency anisotropy parameters and training samples with anisotropy parameters as sample outputs, and improving the inversion accuracy of shale oil reservoir elastic parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0026] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0027] Figure 2 A schematic diagram of a TransU-net model constructed according to an embodiment of the present invention;

[0028] Figure 3 A training flow chart of the TransU-net model constructed in an embodiment of the present invention;

[0029] Figure 4 Schematic diagram of the actual parameters of some Marmousi models of the embodiments of the present invention: (a) is a schematic diagram of the longitudinal wave velocity; (b) is a schematic diagram of the shear wave velocity; (c) is a schematic diagram of the density; (d) is an anisotropy parameter Schematic diagram; (e) is the anisotropy parameter Schematic diagram;

[0030] Figure 5 Schematic diagrams of stacked seismic data at different angles according to an embodiment of the present invention: (a) is a schematic diagram of stacked seismic data at 9°; (b) is a schematic diagram of stacked seismic data at 16°; (c) is a schematic diagram of stacked seismic data at 23°; (d) is a schematic diagram of stacked seismic data at 30°; (e) is a schematic diagram of stacked seismic data at 37°;

[0031] Figure 6 A schematic diagram of a group of training samples according to an embodiment of the present invention;

[0032] Figure 7 Schematic diagram of inversion parameters of the method proposed in this embodiment of the present invention: (a) is a schematic diagram of longitudinal wave velocity; (b) is a schematic diagram of shear wave velocity; (c) is a schematic diagram of density; (d) is anisotropy parameter Schematic diagram; (e) is the anisotropy parameter Schematic diagram;

[0033] Figure 8 Schematic diagram of inversion parameters of the data-driven method according to an embodiment of the present invention: (a) is a schematic diagram of longitudinal wave velocity; (b) is a schematic diagram of shear wave velocity; (c) is a schematic diagram of density; (d) is anisotropy parameter Schematic diagram; (e) is the anisotropy parameter Schematic diagram;

[0034] Fig. 9 Schematic diagram of inversion parameters of the model-driven method of an embodiment of the present invention: (a) is a schematic diagram of longitudinal wave velocity; (b) is a schematic diagram of shear wave velocity; (c) is a schematic diagram of density; (d) is anisotropy parameter Schematic diagram; (e) is the anisotropy parameter Schematic diagram;

[0035] Fig.10 The comparison diagrams at the 2001th CDP of the embodiment of the present invention are as follows: (a) is a comparison diagram of the true parameter, the low-frequency parameter and the inversion parameter; (b) is a comparison diagram of the absolute error between the true parameter and the inversion parameter. DETAILED DESCRIPTION

[0036] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0037] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] Embodiment 1

[0039] A VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs, the process is as follows: Figure 1 As shown, the following steps are included:

[0040] Constructing the TransU-net model for VTI media data-driven seismic AVO inversion:

[0041] VTI medium data-driven AVO inversion predicts five parameters from one data, which requires the neural network model to have strong feature extraction capabilities. TransU-net combines the advantages of the Transformer network and the convolutional neural network (CNN), which can capture the long-term correlation features within the data while retaining the local detail features, thereby improving the comprehensiveness and accuracy of feature extraction. In addition, the self-attention mechanism of the Transformer network allows the TransU-net model to consider the interactions within the data during the feature extraction process, which helps to better understand the information contained in the data. Furthermore, TransU-net adopts the encoding-decoding structure of U-net, which can realize feature extraction at different scales, and effectively fuses these features using jump connections, thereby enhancing the richness and robustness of feature extraction.

[0042] The TransU-net model constructed in this embodiment for VTI medium data-driven seismic AVO inversion is as follows: Figure 2 As shown, it includes a CNN part for extracting local information, a Transformer part for extracting global information, and an upsampling part for restoring data features.

[0043] The CNN part contains three feature maps, namely the first feature map, the second feature map and the third feature map. Each feature map is connected by a convolution operation with a convolution kernel of 3×3. The last layer of feature maps is dimensionally adjusted to form local hidden features, which are input to the Transformer part after the maximum pooling operation. The Transformer part includes 8 Transformer layers, namely the first Transformer layer, the second Transformer layer, the third Transformer layer, the fourth Transformer layer, the fifth Transformer layer, the sixth Transformer layer. The sformer layer, the seventh Transformer layer and the eighth Transformer layer are connected by a multi-layer perceptron. The last Transformer layer forms a global hidden feature, which is input to the upsampling part after dimension adjustment. The upsampling part includes 8 feature maps, namely the first feature map, the second feature map, the third feature map, the fourth feature map, the fifth feature map, the sixth feature map, the seventh feature map and the eighth feature map. After the upsampling operation, each feature map is fused with the feature map corresponding to the CNN part through a jump connection, and then the feature map of the next layer is generated after a convolution operation with a convolution kernel of 3×3. In the entire network model, the ReLU activation function is used.

[0044] Generate and augment training samples for training the above TransU-net model:

[0045] The training samples for training the TransU-net model proposed in this embodiment are composed of sample input and sample output. The sample input is 9°, 16°, 23°, 30° and 37° angle stacked seismic data based on the pre-stack angle seismic data of the well bypass track according to the angle ranges of 3-15°, 10-22°, 17-29°, 24-36° and 31-43°, referred to as sample seismic data; the sample output is the P-wave velocity, S-wave velocity and density obtained from the well logging data, referred to as sample elastic parameters.

[0046] Since there are few well logging data in the actual work area and they are unevenly distributed, resulting in few training samples, this embodiment augments the training samples. First, the number of well logging data is increased by replacing the low-frequency part of the well logging data, distorting the high-frequency part of the well logging data, adding random numbers to the well logging data, etc.; then, the longitudinal wave reflection coefficient is calculated based on the precise AVO forward equation of the VTI medium, and it is convolved with different seismic wavelets extracted from the seismic data of the target work area to synthesize pre-stack seismic data; then, the synthesized pre-stack seismic data is superimposed into angle stack seismic data; finally, the added well logging data is combined with the corresponding synthetic angle stack seismic data to form a training sample.

[0047] Train the TransU-net model constructed above:

[0048] like Figure 3 As shown, firstly, the sample earthquake data Input the constructed TransU-net model and output the predicted parameters , including the predicted elastic parameters and predict anisotropic parameters; then, based on the predicted parameters, the VTI medium precise AVO forward equation is used to calculate the P-wave reflection coefficient, and the P-wave reflection coefficient is combined with the seismic wavelet convolution extracted from the target area seismic data to synthesize the angle stacking seismic data. , forming a closed loop. In each closed loop, this embodiment calculates the normalized predicted elasticity parameter , sample elasticity parameters The error between the synthetic pre-stack seismic data and the sample seismic data is combined to form the objective function, which is expressed as:

[0049] (1)

[0050] in, represents the L2 norm, which is used to calculate the errors between elastic parameters and seismic data; It represents the weight coefficient of the first term on the right side of the equal sign in formula (1), that is, the proportion of the objective function affected by the training samples, which is determined by testing.

[0051] In this embodiment, the sample seismic data are sequentially input into the TransU-net model, the objective function is calculated based on formula (1) in each closed loop, and the TransU-net model parameters are iteratively updated using the back propagation algorithm until the objective function converges, thus completing the training of the TransU-net model.

[0052] Inversion of shale reservoir elastic parameters:

[0053] The pre-stack seismic data of the shale oil reservoir area to be inverted are stacked at 9°, 16°, 23°, 30° and 37° angles according to the angle ranges of 3-15°, 10-22°, 17-29°, 24-36° and 31-43°, and then output to the trained TransU-net model to obtain the inversion elastic parameters and anisotropy parameters.

[0054] A VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs proposed in this embodiment is applied to part of the Marmousi 2 model for trial calculation, and a model-driven inversion method and a data-driven inversion method are used as comparison algorithms.

[0055] The selected Marmousi 2 model has similar parameter characteristics to the shale oil reservoir, and its true P-wave velocity, S-wave velocity and density are as follows: Figure 4 As shown in (a), (b) and (c), there are 4001 CDPs, each of which includes 1001 sampling points with a sampling interval of 2ms. The true anisotropic parameters are randomly generated under the constraints of the layer information. and ,like Figure 4 Based on the above elastic parameters and anisotropy parameters, the VTI medium precise AVO forward equation is used to calculate the longitudinal wave reflection coefficient with an incident angle of 3-43°, and it is convolved with the Ricker wavelet with a main frequency of 25 Hz to synthesize the pre-stack seismic data. Then, the 3-15°, 10-22°, 17-29°, 24-36° and 31-43° pre-stack seismic data are stacked into 9°, 16°, 23°, 30° and 37° angle stacked seismic data, respectively, as shown in Fig. Figure 5 shown.

[0056] First, from Figure 4 The data of 20 CDPs are randomly selected from the elastic parameters shown in the figure, and the training samples are made by stacking the seismic data with the corresponding angles and performing sample augmentation. A total of 100 sets of training samples are generated. Figure 6 A set of training samples is shown. Next, the model data is inverted using the method proposed in this embodiment. After multiple tests, the formula (1) Set to 0.63, the objective function converges after 140 iterations, and the training of the TransU-net model is completed. Figure 5 The angle stacked seismic data in is input into the trained TransU-net model, and the inverted elastic parameters and anisotropic parameters are output, such as Figure 7 As shown in the figure, the inversion parameters of the method proposed in this embodiment have good similarity with the real parameters and have a high resolution, indicating that the method proposed in this embodiment has good feasibility.

[0057] For comparison, the elastic parameters and anisotropic parameters obtained by the data-driven method and the model-driven method are shown in Figure 8 and 9 As shown in the figure, the parameters inverted by the two methods are also similar to the true parameters in form. For a clearer comparison, the 2100th CDP, that is, the true parameters and inverted parameters that were not involved in making the training samples, are extracted and displayed in Fig.10 Middle (a), Fig.10(b) is the absolute value of the error between the true parameter and the inversion parameter. As can be seen from the figure, compared with the data-driven method and the model-driven method, the curve of the inversion parameter of the method proposed in this embodiment is closer to the curve of the true parameter, and the absolute value of the error between them is smaller. In addition, the MSE between the inversion parameters of different methods and the true parameters is calculated and shown in Table 1. As can be seen from the table, among the three methods, the MSE corresponding to the inversion parameters of the method proposed in this embodiment is the smallest, indicating that it has the highest accuracy. In summary, the method proposed in this embodiment can invert the elastic parameters and anisotropic parameters of shale reservoirs with higher accuracy without relying on low-frequency anisotropic parameters and training samples with anisotropic parameters as sample output.

[0058] Table 1

[0059]

[0060] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs, characterized in that: The following steps are involved: Based on Transformer network and convolutional neural network, build TransU-net model; The pre-stack seismic data of the well bypass are stacked to obtain angle stacked seismic data to form sample input; Performing augmentation processing on the well logging data, and forming sample output based on the augmented well logging data; Based on the sample input and sample output, a training sample is formed, and the TransU-net model is trained based on the training sample; Based on the trained TransU-net model, the inversion elastic parameters and anisotropic parameters of the area to be measured are obtained; The TransU-net model includes a CNN part for extracting local information, a Transformer part for extracting global information, and an upsampling part for restoring data features; The CNN part includes a first feature map, a second feature map and a third feature map, and each feature map is connected by a convolution operation; the third feature map is dimensionally adjusted to form a local hidden feature, and is input into the Transformer part after a maximum pooling operation; The Transformer part includes a first Transformer layer, a second Transformer layer, a third Transformer layer, a fourth Transformer layer, a fifth Transformer layer, a sixth Transformer layer, a seventh Transformer layer and an eighth Transformer layer, and each Transformer layer is connected by a multi-layer perceptron; a global hidden feature is formed based on the eighth Transformer layer, and is input into the upsampling part after dimension adjustment; The upsampling part includes a first feature map, a second feature map, a third feature map, a fourth feature map, a fifth feature map, a sixth feature map, a seventh feature map and an eighth feature map. After the upsampling operation, each feature map is subjected to feature fusion with the feature map corresponding to the CNN part through a jump connection, and then a feature map of the next layer is generated after a convolution operation; The process of training the TransU-net model includes: inputting the training samples into the TransU-net model to calculate the objective function, and iteratively updating the TransU-net model parameters by the back propagation algorithm until the objective function converges to obtain the trained TransU-net model; The expression of the objective function is: in, represents the L2 norm, It indicates the influence ratio of the objective function on the training samples. represents the predicted elasticity parameter, represents the sample elasticity parameter, represents angle stacked seismic data, Represents sample earthquake data.

2. The VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs according to claim 1, characterized in that: The process of augmenting the logging data includes: replacing the low-frequency part of the logging data, distorting the high-frequency part of the logging data, and increasing the amount of logging data by adding random numbers to the logging data.

3. The VTI medium data-driven seismic AVO inversion method for obtaining elastic parameters of shale reservoirs according to claim 1, characterized in that: The angle stacked seismic data include 9°, 16°, 23°, 30° and 37° angle stacked seismic data obtained by stacking pre-stack seismic data in angle ranges of 3-15°, 10-22°, 17-29°, 24-36° and 31-43° respectively.