A primary wave imaging method integrating multiple wave imaging information

Abstract

The present invention discloses a primary wave imaging method that integrates multiple wave imaging information. The method first applies a migration operator to the separated primary and multiple waves for migration imaging, obtaining high-precision primary and multiple wave imaging profiles. A sliding window is then used to partition the primary and multiple wave imaging profiles, completing the production of training and test data sets. A U-net network is then trained to obtain a network model that can characterize the mapping relationship between multiple wave imaging and primary wave imaging. Finally, the test set data is input into the trained network model for processing, and the processed windowed data is weightedly superimposed to obtain a complete primary wave imaging profile that integrates multiple wave imaging information. Based on deep learning, the present invention establishes a nonlinear relationship between multiple wave imaging and primary wave imaging, fully exploiting the additional imaging information in the multiple wave imaging profile, and providing richer stratigraphic structural information for primary wave imaging in weakly illuminated areas.

Description

Technical Field

[0001] The present invention belongs to the field of geological exploration technology and relates to a stratum structure imaging technology, in particular to a primary wave imaging method that fuses multiple wave imaging information. Background Art

[0002] In actual oil and gas seismic exploration, the distribution of seismic sources and receivers is limited and irregular due to factors such as acquisition costs and field obstacles. This results in incomplete data acquisition, insufficient illumination and angular illumination of the underground area, and thus compromises imaging quality. Furthermore, as exploration targets become increasingly complex, primary waves often fail to effectively illuminate complex underground structures. Therefore, traditional primary wave imaging methods struggle to obtain effective imaging information in weakly illuminated areas of complex structures.

[0003] In traditional seismic data processing, multiples are considered noise and must be removed before imaging. As exploration becomes increasingly challenging, fully exploiting the stratigraphic information carried by various wave fields in seismic data to meet the needs of imaging complex media has become a key development trend in precision seismic exploration. Essentially, multiples can be viewed as seismic records generated by returning the primary wave as a virtual source. Compared to the primary wave, multiples have a smaller reflection angle, a longer propagation path, and a wider illumination range, thus carrying richer stratigraphic information.

[0004] Numerous studies have demonstrated that imaging multiples can effectively expand the scope of underground illumination, overcome the lack of imaging information caused by incomplete acquisition and observation systems, and more effectively image weakly illuminated areas of complex structures, thereby obtaining high-quality imaging results containing richer stratigraphic information. The rational integration of primary and multiple imaging results can effectively improve the quality of stratigraphic imaging. However, current mainstream processing and interpretation methods use the primary reflection wave as the effective wave, while current research on multiple imaging methods focuses on individual multiple imaging. There is still a lack of reasonable and effective measures to ensure that multiple imaging results can serve the primary imaging and provide a beneficial supplement to primary imaging. Summary of the Invention

[0005] In view of the above problems existing in the prior art, the present invention provides a primary wave imaging method that fuses multiple wave imaging information. The method can achieve the purpose of multiple wave imaging to supplement the primary wave imaging in weakly illuminated areas.

[0006] In order to achieve the above object, the present invention provides a primary wave imaging method for fusing multiple wave imaging information, comprising the following steps:

[0007] 1) Using the surface multiple suppression (SRME) method to obtain separated primaries and multiples from the acquired seismic shot gathers;

[0008] 2) Apply the migration operator to migrate and image the primary wave and the multiple wave respectively, and obtain the primary wave and the multiple wave imaging profiles;

[0009] 3) A sliding window is used to block the data of the primary and multiple wave imaging sections, and data augmentation is performed by flipping, rotating, adding different degrees of noise, etc. to increase the data volume and complete the preparation of training and test data sets;

[0010] 4) Inputting the training data set into the U-net network for training to obtain a network model that can characterize the mapping relationship between multiple wave imaging and primary wave imaging;

[0011] 5) Inputting the test set data into the trained network model to obtain primary wave imaging window data fused with multiple wave imaging information;

[0012] 6) Perform window weighted superposition on the processed primary wave window data to obtain the final complete primary wave imaging profile that integrates the multiple wave imaging information.

[0013] Preferably, the migration operator in step 2) can be a one-way wave migration operator, a two-way wave migration operator, a least squares migration operator, etc. The migration imaging process can be regarded as the result of the migration operator acting on the seismic data. The primary wave data migration imaging process is expressed as:

[0014] R p =O[s(x s )]P (1)

[0015] Where, P represents the primary wave data, O[s(x s )] represents the primary wave imaging migration operator, R p It is the result of the primary wave data migration;

[0016] The multiple wave data migration imaging process is expressed as:

[0017] R m =O[d(x r )]M (2)

[0018] Where M represents multiple wave data, O[d(x r )] represents the multiple wave imaging migration operator, R m It is the result of multiple wave data migration. In order to ensure the accuracy of multiple wave migration imaging, the principle of multiple wave order classification method is used to obtain the multiple wave migration imaging section to eliminate the imaging artifacts introduced by the cross-correlation of multiple waves of different orders.

[0019] Preferably, the specific method of the multiple wavelet order imaging is:

[0020] First, the focusing transform and SRME method are applied to achieve multiple wave reduction, complete the order multiple wave separation, and obtain multiple waves of different orders. Then, the order multiple waves are imaged to obtain high-precision multiple wave migration imaging profiles. The order multiple wave migration imaging process is expressed as follows:

[0021] R m =O[d(x r )](M1+M2+...) (3)

[0022] Where M1 and M2 represent the first-order and second-order multiple waves, respectively.

[0023] Preferably, in the data set preparation process in step 3), windowed data of an area with full coverage of the imaging section and simple structure are selected to prepare a training data set, wherein the multiple wave imaging windowed data are used as training input data and the primary wave imaging windowed data are used as labels; and the multiple wave imaging windowed data of other areas are used as test data sets.

[0024] Preferably, the process of inputting the training data set into the U-net network for training in step 4) is expressed as:

[0025]

[0026] Where R mi is the i-th multiple wave imaging window data, represents the primary imaging window data obtained after network training on the i-th multiple imaging window data, Net represents the network structure, and θ represents the training parameters. During network training, the training data is first input for forward propagation, and the neural network calculates the output data layer by layer. Then, the difference between the network output data and the real data is calculated, and backpropagation is used to continuously update the network parameters to minimize the loss function. Finally, the optimal network model is obtained by repeatedly training with a large amount of labeled data.

[0027] Preferably, the loss function is a mean square error loss function, which is defined as:

[0028]

[0029] Where R pi is the i-th primary wave imaging window data, and N is the number of input seismic data samples.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention provides a primary wave imaging method that integrates multiple wave imaging information. By leveraging the advantage of multiple waves having a wider illumination range, the method establishes a nonlinear mapping relationship between primary wave imaging and multiple wave imaging based on deep learning. This provides a new approach for multiple wave imaging to serve primary wave imaging. The method utilizes a trained network model to extract richer imaging information from the multiple wave imaging results to assist primary wave imaging, thereby achieving the purpose of improving the quality of primary wave imaging. This method has important application value in supplementing the imaging of weak primary wave illumination areas, such as imaging gaps and low coverage areas caused by limited observation systems. The present invention can be widely used in the field of exploration seismic data processing in which multiple waves are developed. By integrating multiple wave imaging information, the method can be used to supplement primary wave imaging, thereby improving the accuracy of stratigraphic structure migration imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of a primary wave imaging method for fusing multiple wave imaging information according to an embodiment of the present invention;

[0033] Figure 2 A schematic diagram of the U-net network structure used in an embodiment of the present invention;

[0034] Figure 3 The velocity model used in the embodiment of the present invention;

[0035] Figure 4 The original primary wave migration imaging section of the seismic record of the embodiment of the present invention;

[0036] Figure 5 The original multiple wave migration imaging section of the seismic record of the embodiment of the present invention;

[0037] Figure 6 This is a primary wave migration imaging section based on deep learning fusion of multiple wave imaging information in an embodiment of the present invention;

[0038] Figure 7 It is the difference between the primary wave migration imaging section obtained by fusion of multiple wave imaging information and the original primary wave migration imaging section in the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention is described in detail below by way of exemplary embodiments, but it should be understood that elements, structures, and features of one embodiment may be beneficially combined in other embodiments without further description.

[0040] The present invention provides a primary imaging method that fuses multiple imaging information. This method utilizes deep learning to establish a network model capable of characterizing the mapping relationship between multiple and primary imaging results. Based on the trained network model, the method extracts richer structural information from multiple imaging, thereby enabling primary migration imaging that fuses multiple imaging information and improves primary imaging quality in weakly illuminated areas. The following describes this primary imaging method in detail, in conjunction with the accompanying drawings.

[0041] See also Figure 1 The embodiment of the present invention provides a primary wave imaging method for fusing multiple wave imaging information, and the specific steps are as follows:

[0042] S1. Surface multiple suppression (SRME) is used to obtain separated primaries and multiples from the acquired seismic shot gathers.

[0043] S2. Apply the migration operator to migrate and image the primary wave and the multiple wave respectively, and obtain the primary wave and the multiple wave imaging profiles;

[0044] Specifically, in the primary and multiple wave migration imaging process, the migration operator can be a one-way wave migration operator, a two-way wave migration operator, or a least squares migration operator. The migration imaging process can be regarded as the result of the migration operator acting on the seismic data. The primary wave data migration imaging process is expressed as:

[0045] R p =O[s(x s )]P (1)

[0046] Where, P represents the primary wave data, O[s(x s )] represents the primary wave imaging migration operator, R p It is the result of the primary wave data migration;

[0047] The multiple wave data migration imaging process is expressed as:

[0048] R m =O[d(x r )]M (2)

[0049] Where M represents multiple wave data, O[d(x r )] represents the multiple wave imaging migration operator, R m It is the result of multiple wave data migration. In order to ensure the accuracy of multiple wave migration imaging, the principle of multiple wave order classification method is used to obtain the multiple wave migration imaging section to eliminate the imaging artifacts introduced by the cross-correlation of multiple waves of different orders.

[0050] Specifically, the method for multiple-order imaging is as follows: first, focus transformation and SRME are applied to reduce the multiple order, complete the order-separation of multiples, obtain multiples of different orders, and then image the order-separated multiples to obtain high-precision multiple migration imaging profiles. The order-separated multiple migration imaging process is expressed as follows:

[0051] R m =O[d(x r )](M1+M2+...) (3)

[0052] Where M1 and M2 represent the first-order and second-order multiple waves, respectively.

[0053] S3. Use a sliding window to block the data of the primary wave and multiple wave imaging profiles, and perform data augmentation by flipping, rotating, adding different degrees of noise, etc. to increase the data volume and complete the preparation of training data sets and test data sets;

[0054] Specifically, during the dataset preparation process, windowed data from areas with full coverage of the imaging profile and simple structure are selected to prepare the training dataset, where the multiple-wave imaging windowed data are used as the training input data and the single-wave imaging windowed data are used as the label; and the multiple-wave imaging windowed data from other areas are used as the test dataset.

[0055] S4. Inputting the training data set into the U-net network for training to obtain a network model that can characterize the relationship between multiple wave imaging and primary wave imaging;

[0056] Specifically, the process of inputting the training data set into the U-net network for training is expressed as:

[0057]

[0058] Where R mi is the i-th multiple wave imaging window data, represents the primary wave imaging window data obtained after network training of the i-th multiple wave imaging window data, Net represents the network structure, and θ represents the training parameters. The schematic diagram of the U-net network structure used can be found in Figure 2The network contains 18 hidden layers, using a 3×3 convolution kernel size. The number of convolution kernels in each layer is 64, 64, 128, 128, 256, 256, 512, 512, 1024, 1024, 512, 512, 256, 256, 128, 128, 64, and 64, respectively. A Relu activation function is added after each layer, with the last layer having 1 convolution kernel and no activation function. During network training, training data is first input for forward propagation, and the neural network calculates the output data layer by layer. The difference between the network output and the real data is then calculated, and backpropagation is used to continuously update the network parameters to minimize the loss function.

[0059] Specifically, the loss function is selected as the mean square error loss function, which is defined as:

[0060]

[0061] Where R pi is the i-th primary wave imaging window data, and N is the number of input seismic data samples. Finally, the optimal network model is obtained by repeatedly training with a large amount of labeled data.

[0062] S5. Input the test set data into the trained network model to obtain primary wave imaging window data fused with multiple wave imaging information;

[0063] S6. Perform window weighted superposition on the processed primary wave window data to obtain a complete primary wave imaging profile that is finally integrated with the multiple wave imaging information.

[0064] The above-mentioned primary wave imaging method that integrates multiple wave imaging information of the present invention uses deep learning tools for the first time to establish a nonlinear mapping relationship between multiple wave imaging and primary wave imaging, thereby providing a method for extracting additional information from multiple wave imaging and using it to improve the quality of seismic data imaging. Especially in weak lighting conditions such as acquisition loss and complex structures, the primary wave imaging profile that integrates multiple wave imaging information can provide richer and more detailed stratigraphic structural information.

[0065] In order to more clearly illustrate the above method of the present invention, a specific embodiment is used to illustrate it below.

[0066] Example

[0067] The present invention is applied to Figure 3 In order to verify the effectiveness of the present invention, an incomplete observation system was used when setting up the multi-layer medium model data. Obstacles were set between 1.1 km and 2.42 km in the horizontal direction, making it impossible to deploy shot points and detectors. As a result, a gap was created in the data collection between the 36th and 65th shots, as shown in Figure 2. Figure 3Indicated by the middle triangle.

[0068] The above method of the present invention first uses the SRME method to obtain separated primary waves and multiple waves from the acquired seismic shot gathers; then, in order to ensure the accuracy of imaging, the multiple waves are separated by order based on SRME and focusing transformation, and the primary waves and multiple waves are migrated and imaged using the least squares reverse time migration operator to obtain the original primary wave imaging section (such as Figure 4 ) and the original multiple wave imaging section (as shown in Figure 5 As shown in the figure, both the multiple and primary imaging results well reflect the subsurface tectonic morphology and exhibit high imaging accuracy. However, due to an incomplete observation system, an imaging gap appears in the primary imaging profile approximately 2 km from the obstacle area. Because the multiples themselves carry richer stratigraphic information, this imaging gap is effectively compensated for in the multiple imaging results. However, there are significant differences between the multiple and primary imaging results in terms of amplitude, phase, and frequency. Due to the lack of a nonlinear mapping relationship between the multiple and primary imaging, the multiple imaging results cannot be directly used to compensate for the primary imaging results.

[0069] In order to solve the above problems, the present invention introduces deep learning theory to establish a nonlinear mapping relationship between multiple wave imaging and primary wave imaging. First, a network training data set is prepared. In order to increase the number of samples, a sliding window is used to block the single wave and multiple wave imaging profile data of the selected fully covered and simply constructed imaging area, and data augmentation is performed by a combination of flipping, rotating, adding different degrees of noise, etc. to increase the data volume and complete the data set preparation. The single wave window data is used as the label, the multiple wave window data is used as the network training input, and the multiple wave imaging window data of other areas are used as the prediction set. Then the prepared training data set is input into the U-net network (such as Figure 2 ) to obtain a network model that can characterize the mapping relationship between multiple wave imaging and primary wave imaging. Next, the test set data is input into the trained network model to obtain the primary wave imaging window data that integrates the multiple wave imaging information in the low coverage and complex structure area. Finally, the primary wave window data of the fully covered and simple structure imaging area and the processed primary wave window data are window-weighted superpositioned to obtain the final complete primary wave imaging profile that integrates the multiple wave imaging information (as shown in Figure 6 As shown in the figure, after fusing the multiple wave imaging information, the shallow imaging gap in the primary wave imaging section is effectively compensated. In order to highlight the advantages of the present invention, the primary wave migration imaging section ( Figure 6 ) and the original first-wave migration imaging section ( Figure 4 ) to make a difference, the difference profile is as follows Figure 7As shown in the figure, compared with the original primary-wave migration imaging section, the primary-wave migration imaging result obtained by the present invention, which integrates the multiple-wave imaging information, not only obtains richer imaging gap information, but also obtains richer imaging information in the low-coverage areas near the left and right boundaries. The feasibility and effectiveness of the present invention have been well verified.

[0070] The above embodiments are used to explain the present invention rather than to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A primary wave imaging method integrating multiple wave imaging information, characterized in that: The following steps are involved: 1) Using the free surface multiple suppression (SRME) method on the acquired seismic shot gathers to obtain separated primaries and multiples; 2) Apply the migration operator to migrate and image the primary wave and the multiple wave respectively, and obtain the primary wave and the multiple wave imaging profiles; 3) A sliding window is used to block the data of the primary and multiple wave imaging sections, and data augmentation is performed by flipping, rotating, adding different degrees of noise, etc. to increase the data volume and complete the preparation of training and test data sets; 4) Inputting the training data set into the U-net network for training to obtain a network model that can characterize the mapping relationship between multiple wave imaging and primary wave imaging; 5) Inputting the test set data into the trained network model to obtain primary wave imaging window data fused with multiple wave imaging information; 6) Performing window-weighted superposition on the processed primary wave window data to obtain the final complete primary wave imaging profile fused with multiple wave imaging information; During the data set preparation process in step 3), windowed data of an area with full coverage of the imaging section and simple structure are selected to prepare a training data set, wherein the multiple wave imaging windowed data are used as training input data and the primary wave imaging windowed data are used as labels; and the multiple wave imaging windowed data of other areas are used as test data sets; The process of inputting the training data set into the U-net network for training in step 4) is expressed as follows: Where R mi is the i-th multiple wave imaging window data, represents the primary wave imaging window data obtained after network training on the i-th multiple wave imaging window data, Net represents the network structure, and θ represents the training parameters. During the network training process, the training data is first input for forward propagation, and the output data is obtained by layer-by-layer calculation through the neural network. Then, the difference between the network output data and the real data is obtained, and the network parameters are continuously updated through backpropagation to minimize the loss function. Finally, the optimal network model is obtained by repeatedly training with a large amount of labeled data.

2. The primary wave imaging method for fusing multiple wave imaging information according to claim 1, characterized in that: The migration operator in step 2) is one of the one-way wave migration operator, the two-way wave migration operator and the least squares migration operator. The migration imaging process is regarded as the result of the migration operator acting on the seismic data. The primary wave data migration imaging process is expressed as: R p =O[s(x s )]P Where, P represents the primary wave data, O[s(x s )] represents the primary wave imaging migration operator, R p It is the result of the primary wave data migration; The multiple wave data migration imaging process is expressed as: R m =O[d(x r )]M Where M represents multiple wave data, O[d(x r )] represents the multiple wave imaging migration operator, R m It is the result of multiple wave data migration. In order to ensure the accuracy of multiple wave migration imaging, the principle of multiple wave order classification method is used to obtain the multiple wave migration imaging section to eliminate the imaging artifacts introduced by the cross-correlation of multiple waves of different orders.

3. The primary wave imaging method for fusing multiple wave imaging information according to claim 2, characterized in that: The specific method of the multiple wavelet order imaging is as follows: First, the focusing transform and SRME method are applied to achieve multiple wave reduction, complete the order multiple wave separation, and obtain multiple waves of different orders. Then, the order multiple waves are imaged to obtain high-precision multiple wave migration imaging profiles. The order multiple wave migration imaging process is expressed as follows: R m =O[d(x r )](M1+M2+...) Where M1 and M2 represent the first-order and second-order multiple waves, respectively.

4. The primary wave imaging method for fusing multiple wave imaging information according to claim 1, characterized in that: The loss function is the mean square error loss function, which is defined as: Where R pi is the i-th primary wave imaging window data, and N is the number of input seismic data samples.

Images