Model generation method, well-to-seismic joint velocity modeling method, device and equipment

Abstract

The application provides a model generation method, a well-seismic joint velocity modeling method, a device and equipment, a texture migration network model is built, the texture migration network includes a generator, a discriminator and an autoencoder; the generator and the discriminator are used for constituting a conditional generative adversarial network, and the autoencoder is used for reconstructing logging velocity; sample data is acquired, the conditional generative adversarial network constituted based on the generator and the discriminator is used for training the texture migration network model; the sample data includes initial velocity and a depth profile; the logging velocity is input into the autoencoder, and the texture migration network model is subjected to parameter optimization; the texture migration network model is used for outputting a velocity modeling image according to the input initial velocity and depth profile, and the velocity and the reflection surface form are matched in the velocity modeling image. The application can be applied to unseen data by only training the network by using artificial synthetic data, and the generalization capability of the network is improved; and high-precision velocity modeling is realized by only using a small amount of logging labels and a small amount of data.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of oil and gas exploration and development, and particularly to a model generation method, a well-seismic combined velocity modeling method, an apparatus, an electronic device, and a readable storage medium. Background Technology

[0002] Depth migration is extremely sensitive to velocity, therefore the accuracy of velocity modeling directly determines subsequent geological interpretation and reservoir parameter modeling. As exploration and development deepen, complex structural blocks, such as igneous intrusions and complex fractured melt bodies, present new challenges to velocity modeling. Current modeling methods include: inversion = tomography + migration. Tomography updates the background velocity, and migration repositions the reflecting surface, ultimately achieving a match between velocity and the reflecting surface. This conclusion can also be drawn from the simple time-depth relationship T (travel time) = D (reflecting surface depth) / V (velocity); successful full-waveform inversion requires simultaneous repositioning of the reflecting surface and updating of the background velocity.

[0003] Current velocity modeling strategies that utilize the velocity-reflection surface matching relationship, while requiring less computation compared to reflection waveform inversion, do not achieve good modeling results in terms of automatic selection of stratigraphic information and seamless integration of stratigraphic information into the inversion process. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide a model generation method, a well-seismic combined velocity modeling method, apparatus and equipment that overcome or at least partially solve the above problems.

[0005] In a first aspect, embodiments of this application disclose a model generation method, the method comprising:

[0006] A texture transfer network model is constructed, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct the logging velocity;

[0007] Acquire sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile;

[0008] The logging velocity is input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained texture transfer network model. The trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image.

[0009] Secondly, embodiments of this application disclose a well-seismic combined velocity modeling method, the method comprising:

[0010] Acquire the data to be predicted, which includes initial velocity and depth profiles;

[0011] The data to be predicted is input into the trained texture transfer network model, which outputs a velocity modeling image in which velocity and reflective surface morphology are matched; wherein the texture transfer network model is trained by the model generation method described in the first aspect.

[0012] Thirdly, embodiments of this application disclose a model generation apparatus, the apparatus comprising:

[0013] A construction module is used to build a texture transfer network model, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct logging velocity;

[0014] The first training module is used to acquire sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional generation adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile;

[0015] The second training module is used to input the logging velocity into the autoencoder and optimize the parameters of the texture transfer network model to obtain the trained texture transfer network model. The trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image.

[0016] Fourthly, embodiments of this application disclose a well-seismic combined velocity modeling device, the device comprising:

[0017] The acquisition module is used to acquire the data to be predicted, which includes initial velocity and depth profiles;

[0018] The generation module, as disclosed in this application embodiment, is used to input the data to be predicted into a trained texture transfer network model and output a velocity modeling image, wherein the velocity in the velocity modeling image matches the shape of the reflective surface; wherein the texture transfer network model is trained by the model generation method described in the first aspect.

[0019] Fifthly, embodiments of this application disclose an electronic device, including a processor and a memory, wherein the memory stores a program or instructions executable on the processor, and the program or instructions, when executed by the processor, implement the steps of the method as described in either the first or second aspect.

[0020] Fifthly, embodiments of this application disclose a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the method described in either the first or second aspect.

[0021] In this embodiment, a texture transfer network model is constructed, comprising a generator, a discriminator, and an autoencoder. The generator and discriminator form a conditional generative adversarial network (GAN), and the autoencoder reconstructs logging velocities. Sample data is acquired, input into the generator, and trained on the texture transfer network model based on the GAN formed by the generator and discriminator. The sample data includes initial velocity and depth profiles. The logging velocities are input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained model. The trained model outputs a velocity modeling image based on the input initial velocity and depth profiles, where velocity and reflector surface morphology are matched. This application proposes a well-seismic joint velocity modeling method based on a texture transfer network, starting from the velocity-reflector surface matching relationship. The network can be trained using only artificially synthesized data and applied to unseen data, improving its generalization ability. Transfer learning utilizes only a small number of logging labels, eliminating the need for additional labels collected for supervised learning in the target area, achieving high-precision velocity modeling with limited data. Attached Figure Description

[0022] Figure 1 This is a flowchart of the steps of a model generation method provided in an embodiment of the present invention;

[0023] Figure 2 This is a network structure diagram of a texture transfer network provided in an embodiment of the present invention;

[0024] Figure 3 This is a flowchart illustrating the steps of a well-seismic combined velocity modeling method provided in an embodiment of the present invention;

[0025] Figure 4 This is a flowchart of a texture transfer network generation method provided in an embodiment of the present invention;

[0026] Figure 5 This is an experimental comparison diagram provided in an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of velocity distribution provided in an embodiment of the present invention;

[0028] Figure 7 This is an application effect diagram provided by an embodiment of the present invention;

[0029] Figure 8 This is a block diagram of a model generation device provided in an embodiment of this application;

[0030] Figure 9 This application provides a well-seismic combined velocity modeling device;

[0031] Figure 10 A block diagram of an electronic device provided in this application embodiment;

[0032] Figure 11 A block diagram of another electronic device provided in the embodiments of this application. Detailed Implementation

[0033] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0034] refer to Figure 1 The diagram illustrates a flowchart of a model generation method provided in an embodiment of this application, the method comprising:

[0035] Step 101: Construct a texture transfer network model, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct logging velocity.

[0036] In this application, seismic wave velocity plays an indispensable role in the acquisition, processing, interpretation, and evaluation of seismic data, and is also crucial in various stages of oil and gas exploration and development. The accuracy of seismic wave velocity determination directly affects the final results of seismic exploration. During seismic interpretation, only by obtaining an accurate velocity field can the depth, dip angle, and dip direction of the target layer be precisely determined, the reliability of well locations be confirmed, and accurate geological maps be provided for oil and gas exploration and development, thereby determining the properties of reservoirs and fluids. Therefore, this application proposes a model generation method to build a texture migration network model, achieving automatic fusion of stratigraphic information and velocity. Here, stratigraphic information can refer to the geological layers in the geological depth profile, and velocity refers to the velocity corresponding to each geological layer.

[0037] Furthermore, the texture transfer network can include a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct logging velocities. The conditional generative adversarial network composed of the generator and discriminator is used for adversarial training of the model, and the autoencoder is used to reconstruct logging velocities, fine-tuning some parameters in the texture transfer network so that the texture transfer network can be applied to velocity modeling of unseen data.

[0038] Optionally, the generator includes an encoder and a decoder; the decoder and the encoder are connected by a jumper; the decoder has one output channel.

[0039] In this embodiment of the application, reference is made to Figure 2 , Figure 2 The network structure diagram of the texture transfer network in this application is shown. The network consists of three parts: a generator, a discriminator, and an autoencoder. The parameter settings of one texture transfer network can be as follows: Ck represents a Conv-BatchNorm-ReLu (convolution-acceleration-activation) layer with k channels. Tk represents a TransConv-AdaIN-LeakyReLu (transfer convolution-transfer-activation) layer with k channels. All convolution kernels are set to 4×4, and the convolution stride is set to 2. The Conv layer with a stride of 2 implements downsampling, while the TransConv layer with a stride of 2 implements upsampling. The slope of all LeakyReLu layers is 0.2. For the generator, a UNet architecture with jumper connections is used. The encoder is C64-C128-C256-C512-C512-C512-C512-C512. The i-th layer of the decoder and the (8-i)-th layer of the encoder are connected by jumpers, resulting in a decoder structure of T512-T1024-T1024-T1024-T1024-T512-T256-T128. This means connecting the first layer of the decoder to the seventh layer of the encoder, the second layer of the decoder to the sixth layer of the encoder, and so on, completing the jumper connections between the encoder and decoder. The last layer of the decoder has no activation function and outputs 1 channel. For the discriminator, a PatchGAN (Adversarial Network) architecture is used, namely C64-C128-C256-C512. The activation function of the last layer of the discriminator is Sigmoid. For the autoencoder, let Lk denote a Linear-ReLU layer with k channels. The autoencoder architecture is L16-L32-L64-L128-L64-L32-L16. Implicit encoding uses linear layers to make its feature channel count the same as the generator.

[0040] Step 102: Obtain sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile.

[0041] In this embodiment, the sample data can be an artificially synthesized dataset used to train the texture transfer network model. The sample data may include initial velocity, depth profile, true velocity, and corresponding well logging labels. The initial velocity may be seismic velocity, and the depth profile may be a geological depth profile. By training the model with sample data, the model can generate a corresponding true velocity modeling image given the initial velocity, depth profile, and well logging labels, resulting in a reasonable velocity distribution and enabling the application of previously unseen data to the texture transfer network model.

[0042] Furthermore, a conditional generative adversarial network (GAN) consisting of a generator and a discriminator is used to train the texture transfer network model. The generator takes initial velocity and depth profiles as input, receives random noise, and uses this noise to generate samples similar to the training data. The generator's goal is to generate realistic samples so that the discriminator cannot accurately distinguish between generated and real samples. The generator can be viewed as a generative model that learns the distribution characteristics of the training data to generate new samples similar to it. The discriminator takes samples (which can be real or generated by the generator) as input and predicts their realism. The discriminator's goal is to classify the samples, determining whether they are real or generated. The discriminator can be viewed as a discriminative model that learns how to distinguish between real and generated samples and provides feedback signals to the generator regarding generated samples. The generator and discriminator compete and cooperate with each other through adversarial training. The generator's goal is to deceive the discriminator, making the generated samples increasingly similar to real samples, to the point that the discriminator cannot accurately distinguish them. The discriminator aims to classify samples as accurately as possible, making the differences between real and generated samples more apparent. Through an iterative adversarial training process, both the generator and discriminator continuously adjust their parameters to reach an equilibrium. Ultimately, the generator can produce realistic samples, while the discriminator cannot accurately distinguish between real and generated samples. This adversarial training mechanism allows the texture transfer network model to learn the distribution of real data and generate diverse and creative samples. The game-like process between the generator and discriminator drives the learning and improvement of the texture transfer network model, making the generated output data increasingly realistic.

[0043] Optionally, step 102 specifically includes:

[0044] Sub-step 1021: Input the initial velocity and depth profile into the encoder of the generator to obtain the feature information corresponding to the initial velocity and depth profile;

[0045] Sub-step 1022: Input the feature information into the decoder of the generator and output new sample data similar to the sample data;

[0046] Sub-step 1023: Input the sample data and new sample data into the discriminator, and train the texture transfer network model under the drive of the loss function.

[0047] In this embodiment, for sub-steps 1021 to 1023, the generator may include an encoder and a decoder. The encoder processes the input, and the decoder generates the output. Specifically, the initial velocity and depth profiles are input into the encoder of the generator. The encoder can extract feature information from the input sample data. Then, the decoder upsamples the feature information and passes it to a convolutional layer for processing to generate new sample data similar to the sample data. The new sample data can be used for adversarial training of the texture transfer model.

[0048] Furthermore, the training of the texture transfer network can be based on a loss function to obtain the loss of the adversarial training process and steps such as velocity reconstruction and logging velocity reconstruction, thereby improving the generalization of the texture transfer network.

[0049] Optionally, sub-step 1023 specifically includes:

[0050] Sub-step A1: Determine a first loss function based on the output of the generator and the output of the discriminator. The first loss function is used to determine the adversarial loss of the texture transfer network.

[0051] In this embodiment of the application, the first loss function is used to determine the adversarial loss of the texture transfer network, and the first loss function can be defined as:

[0052]

[0053] in, This represents the adversarial loss, where x represents the initial velocity and depth profile of the input, and y represents the true velocity label. Let G represent the loss, G represent the generator, and D represent the discriminator.

[0054] Sub-step A2: Determine the second loss function based on the input logging rate and the logging rate reconstructed by the autoencoder. The second loss function is used to determine the logging rate reconstruction loss.

[0055] In this embodiment, the second loss function is used to determine the logging rate reconstruction loss based on the input logging rate and the logging rate reconstructed by the autoencoder. The second loss function can be defined as:

[0056]

[0057] in, The logging velocity reconstruction loss is represented by AE, where AE represents the logging velocity autoencoder and w represents the input logging velocity.

[0058] Sub-step A3: Determine the third loss function based on the actual speed and the speed output by the generator. The third loss function is used to determine the speed reconstruction loss.

[0059] In this embodiment, a third loss function is determined based on the actual speed and the speed output by the generator. The speed reconstruction loss is defined as follows:

[0060]

[0061] Sub-step A4: Determine the target loss function based on the first loss function, the second loss function, and the third loss function, and train the texture transfer network model using the target loss function.

[0062] In this embodiment, the target loss function consists of three parts: adversarial loss, velocity field reconstruction loss, and well logging velocity reconstruction loss. Utilizing the concept of conditional adversarial generation, the generator G aims to achieve velocity-reflection surface matching as much as possible, while the discriminator D aims to accurately distinguish between the network-generated result and the true answer. For example, the true answer can be output as 1, while the network-generated result can be output as 0. Therefore, the two engage in a dynamic game process, and ideally, the generator G's generated result can be indistinguishable from the real answer.

[0063] Optionally, sub-step A4 specifically includes:

[0064] The first loss function, the second loss function, and the third loss function are added together to form the target loss function, which is expressed by the following first formula:

[0065]

[0066] in, Describes the target loss function. Denotes the first loss function. This represents the second loss function. Let G represent the third loss function, G represent the generator, AE represent the autoencoder, and D represent the discriminator. This represents the preset weight parameters, for example, It can be set to 100. In this application, the artificially synthesized dataset, including initial velocity, depth profile, true velocity, and corresponding well logging labels, is provided to the texture transfer network model, enabling the texture transfer network model to learn the network weights.

[0067] Step 103: Input the logging velocity into the autoencoder to optimize the parameters of the texture transfer network model and obtain the trained texture transfer network model; the trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, and the velocity and reflection surface morphology are matched in the velocity modeling image.

[0068] In this embodiment, for unseen data, the texture transfer network model can reliably achieve velocity-reflection surface matching, but a velocity distribution difference still exists between the initial velocity and the true velocity. Considering that the velocity labels of actual data are unavailable, this application only uses well logging velocity to fine-tune the texture transfer network model. Starting from the velocity-reflection surface matching relationship, this application proposes a well-seismic joint velocity modeling method based on texture transfer networks. The network can be trained using only artificially synthesized data and applied to unseen data, improving the network's generalization ability. Transfer learning utilizes only a small number of well logging labels, eliminating the need to collect additional labels required for supervised learning in the target area, thus achieving high-precision velocity modeling with limited data.

[0069] Optionally, step 103 specifically includes:

[0070] Sub-step 1031: Freeze the encoder parameters of the generator, and adjust the parameters of the decoder and autoencoder of the generator based on the logging speed.

[0071] In this embodiment, by freezing the encoder parameters of the generator, only the decoder of the grower and the logging velocity autoencoder are fine-tuned, so that the texture transfer network can be applied to unseen data.

[0072] Optionally, sub-step 1031 specifically includes:

[0073] Sub-step B1: Determine the fourth loss function based on the input logging rate and the reconstruction rate at the logging location;

[0074] Sub-step B2: Determine the transfer learning loss function based on the fourth loss function and the second loss function. The transfer learning loss function is used to adjust the parameters of the texture transfer network. The transfer learning loss function is expressed by the following second formula:

[0075]

[0076] in, The loss function represents transfer learning. This represents the second loss function. This represents the fourth loss function. This represents the decoder parameters of the generator, and AE represents the autoencoder.

[0077] In this embodiment, for sub-steps B1 and B2, the velocity reconstruction loss under the logging mask requires that the reconstruction velocity at the logging location be the same as the logging velocity. The fourth loss function formula is:

[0078]

[0079] in This represents the decoder parameters of the generator. Combined with logging velocity reconstruction loss. The loss function for transfer learning is...

[0080]

[0081] The initial velocity, depth profile, and logging labels of the unseen data are provided to the network. The encoder part of the network generator is frozen, and only the decoder of the generator and the logging autoencoder are fine-tuned. The Adam (Adaptive Moment Estimation) optimizer is used to perform transfer learning on the network. The main function of the Adam optimizer is to update the neural network parameters based on gradient information, thereby minimizing the loss function and enabling the network to be applied to the unseen data.

[0082] In summary, in this embodiment, a texture transfer network model is constructed, comprising a generator, a discriminator, and an autoencoder. The generator and discriminator form a conditional generative adversarial network (GAN), and the autoencoder reconstructs logging velocities. Sample data is acquired and input into the generator. Based on the GAN formed by the generator and discriminator, the texture transfer network model is trained. The sample data includes initial velocity and depth profiles. The logging velocities are input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained model. The trained model is used to output a velocity modeling image matching the velocity and reflection surface morphology based on the input initial velocity and depth profile. This application proposes a well-seismic joint velocity modeling method based on a texture transfer network, starting from the velocity-reflection surface matching relationship. The network can be trained using only artificially synthesized data and applied to unseen data, improving the network's generalization ability. Transfer learning utilizes only a small number of logging labels, eliminating the need for additional labels for supervised learning in the target area, achieving high-precision velocity modeling with limited data.

[0083] refer to Figure 3 It illustrates a flowchart of a well-seismic combined velocity modeling method provided in an embodiment of this application, the method comprising:

[0084] Step 201: Obtain the data to be predicted, which includes the initial velocity and depth profile;

[0085] Step 202: Input the data to be predicted into the trained texture transfer network model and output a velocity modeling image, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image; wherein the texture transfer network model is trained by the model generation method described above.

[0086] In this embodiment, after obtaining the trained texture transfer network model, it can be used to model the velocity of unseen data. The unseen data can be the data to be detected. The data to be predicted is input into the trained texture transfer network model, and the output is a velocity modeling image that matches the velocity with the reflection surface morphology. This application proposes a well-seismic joint velocity modeling method based on the velocity-reflection surface matching relationship. The network can be trained using only artificially synthesized data and applied to unseen data, improving the network's generalization ability. Transfer learning utilizes only a small number of well logging labels, eliminating the need for additional labels collected for supervised learning in the target area, thus achieving high-precision velocity modeling with limited data.

[0087] refer to Figure 4 , Figure 4 A flowchart of a texture transfer network generation method provided by the present invention is shown, including:

[0088] 1) Constructing a texture transfer network: The network consists of three parts: a generator, a discriminator, and an autoencoder. The generator and discriminator form a conditional generative adversarial network to achieve velocity-reflector morphology matching; the autoencoder is used to reconstruct logging velocities and provide their implicit encoding; the implicit encoding is connected to the generator through an adaptive instance normalization layer to correct the distribution difference between logging velocities and initial velocities.

[0089] 2) Network training using artificially synthesized training data: The loss function consists of three parts: adversarial loss, velocity field reconstruction loss, and well logging velocity reconstruction loss. Artificially synthesized data is provided to the network, and the Adam optimizer is used to learn the network weights.

[0090] 3) Unseen Data Transfer Learning: Texture transfer networks exhibit strong generalization ability in velocity-reflection surface morphology matching, but struggle to generalize to differences in velocity distribution. For unseen data with velocity distributions significantly different from the training data, the network is fine-tuned using only logging velocities. The encoder parameters of the generator are frozen, and network weights are fine-tuned only for its decoder and the logging velocity autoencoder. The objective function consists of two parts: logging velocity reconstruction loss and velocity field reconstruction loss under the logging mask. The network weights are fine-tuned using the Adam optimizer, with only logging data provided to the network.

[0091] The effectiveness and advantages of this application will be verified by applying the method proposed in this invention to Marmousi model data and comparing it with deep learning methods based on conditional generative adversarial networks and traditional interpolation methods for well logging data guided by seismic data, so as to demonstrate the application effect of this invention.

[0092] This invention modifies the Marmousi model. The Marmousi model is downsampled by a factor of 3, resulting in a model size of 251*767, with a spatial sampling interval of 12m*12m. Five grid points are added vertically to the water layer, resulting in a final model size of 256*767. Figure 5 As shown in f. The Marmousi model uses a 10Hz Ricker wavelet, 767 fixed detectors, and 21 shots uniformly distributed to simulate observation records. One channel is randomly selected as the initial 1D model, as shown in f. Figure 3 As shown in diagram a. Three wells were selected for fine-tuning the network, as follows: Figure 5 d and 5e are shown by white dashed lines. Figure 5 Figures a and 5b show the initial velocity and depth profiles. Comparisons with deep learning methods, the method of this invention, and traditional methods are shown in [see...]. Figure 5 c-5e. It is worth noting that the velocity distributions of the Marmousi model and the artificially synthesized training set differ significantly, such as... Figure 6 As shown, the velocity distribution range of the training set is between 2500-5500 m / s, while the velocity distribution range of the test data is between 1500-5500 m / s. (Comparison) Figure 5 c and 5d, with the assistance of well logging data, although the overall morphology is not significantly different, the speed of the method of this invention is closer to the actual result, and the effect is better in both shallow and deep layers. (Comparison) Figure 5 d and 5e, the method of this invention achieves well interpolation that better matches the characteristics of the depth profile structure under the condition of only 3 wells, reducing the dependence of traditional well interpolation methods on the number of wells logged. Table 1 shows... Figure 5 The quantitative comparison results of different methods quantitatively demonstrate the advantages of this invention. Further, using the initial input velocity ( Figure 5 a) Comparison of output speed of deep learning methods ( Figure 5 c) Output speed of the present invention ( Figure 5 d) and the output speed compared to traditional methods ( Figure 5 e) As the initial model for full waveform inversion, the final inversion result is as follows: Figure 7 As shown. Figure 7 The figure shows the application effect of using the output speed of this invention as the initial speed for full waveform inversion. Figures 7a-7d show the inversion results using the input initial speed, the output speed of the comparison deep learning method, the output speed of this invention, and the output speed of the comparison traditional method as the initial models for full waveform inversion, respectively. It can be seen that the output speed of this invention avoids the cycle jump problem in full waveform inversion, achieves convergence, and has the best effect.

[0093] Table 1

[0094]

[0095] In summary, in this embodiment, a texture transfer network model is constructed, comprising a generator, a discriminator, and an autoencoder. The generator and discriminator form a conditional generative adversarial network (GAN), and the autoencoder reconstructs logging velocities. Sample data is acquired and input into the generator. Based on the GAN formed by the generator and discriminator, the texture transfer network model is trained. The sample data includes initial velocity and depth profiles. The logging velocities are input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained model. The trained model outputs a velocity modeling image based on the input initial velocity and depth profiles, where the velocity and reflector surface morphology are matched. This application proposes a well-seismic joint velocity modeling method based on a texture transfer network, starting from the velocity-reflector surface matching relationship. The network can be trained using only artificially synthesized data and applied to unseen data, improving the network's generalization ability. Transfer learning utilizes only a small number of logging labels, eliminating the need for additional labels for supervised learning in the target area, achieving high-precision velocity modeling with limited data.

[0096] refer to Figure 8 This application illustrates a model generation apparatus provided in an embodiment of the present application, the apparatus comprising:

[0097] Module 301 is used to build a texture transfer network model, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct logging velocity.

[0098] The first training module 302 is used to acquire sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile;

[0099] The second training module 303 is used to input the logging velocity into the autoencoder and optimize the parameters of the texture transfer network model to obtain the trained texture transfer network model. The trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image.

[0100] Optionally, the generator includes an encoder and a decoder; the decoder and the encoder are connected by a jumper; the decoder has one output channel.

[0101] Optionally, the first training module includes:

[0102] The first submodule is used to input the initial velocity and depth profile into the encoder of the generator to obtain the feature information corresponding to the initial velocity and depth profile;

[0103] The second submodule is used to input the feature information into the decoder of the generator and output new sample data that is similar to the sample data.

[0104] The third submodule is used to input the sample data and new sample data into the discriminator and train the texture transfer network model under the drive of the loss function.

[0105] Optionally, the third submodule includes:

[0106] The first loss function determination submodule is used to determine a first loss function based on the output of the generator and the output of the discriminator. The first loss function is used to determine the adversarial loss of the texture transfer network.

[0107] The second loss function determination submodule is used to determine the second loss function based on the input logging rate and the logging rate reconstructed by the autoencoder. The second loss function is used to determine the logging rate reconstruction loss.

[0108] The third loss function determination submodule is used to determine the third loss function based on the actual speed and the speed output by the generator. The third loss function is used to determine the speed reconstruction loss.

[0109] The target loss function determination submodule is used to determine the target loss function based on the first loss function, the second loss function, and the third loss function, and to train the texture transfer network model using the target loss function.

[0110] Optionally, the target loss function determination submodule includes:

[0111] The combination submodule is used to add the first loss function, the second loss function, and the third loss function to obtain the target loss function, which is expressed by the following first formula:

[0112]

[0113] in, Represents the target loss function. Denotes the first loss function. This represents the second loss function. Let G represent the third loss function, G represent the generator, AE represent the autoencoder, and D represent the discriminator. This represents the preset weight parameters.

[0114] Optionally, the second training module includes:

[0115] The adjustment submodule is used to freeze the encoder parameters of the generator and adjust the parameters of the decoder and autoencoder of the generator based on the logging rate.

[0116] Optionally, the adjustment submodule includes:

[0117] The fourth loss function determination submodule is used to determine the fourth loss function based on the input logging rate and the reconstruction rate at the logging location;

[0118] The transfer learning loss function determination submodule is used to determine the transfer learning loss function based on the fourth loss function and the second loss function. The transfer learning loss function is used to adjust the parameters of the texture transfer network, and the transfer learning loss function is expressed by the following second formula:

[0119]

[0120] in, The loss function represents transfer learning. This represents the second loss function. This represents the fourth loss function. This represents the decoder parameters of the generator, and AE represents the autoencoder.

[0121] refer to Figure 9 This application illustrates a well-seismic combined velocity modeling device according to an embodiment of the present application, the device comprising:

[0122] The acquisition module is used to acquire the data to be predicted, which includes initial velocity and depth profiles;

[0123] The generation module is used to input the data to be predicted into the trained texture transfer network model and output a velocity modeling image that matches the velocity and the shape of the reflective surface; wherein the texture transfer network model is trained by the above-mentioned model generation method.

[0124] In summary, in this embodiment, a texture transfer network model is constructed, comprising a generator, a discriminator, and an autoencoder. The generator and discriminator form a conditional generative adversarial network (GAN), and the autoencoder reconstructs logging velocities. Sample data is acquired and input into the generator. Based on the GAN formed by the generator and discriminator, the texture transfer network model is trained. The sample data includes initial velocity and depth profiles. The logging velocities are input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained model. The trained model outputs a velocity modeling image based on the input initial velocity and depth profiles, where the velocity and reflector surface morphology are matched. This application proposes a well-seismic joint velocity modeling method based on a texture transfer network, starting from the velocity-reflector surface matching relationship. The network can be trained using only artificially synthesized data and applied to unseen data, improving the network's generalization ability. Transfer learning utilizes only a small number of logging labels, eliminating the need for additional labels collected for supervised learning in the target area, thus achieving high-precision velocity modeling with limited data.

[0125] Figure 10 A block diagram of an electronic device 600 is shown according to an exemplary embodiment. For example, the electronic device 600 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0126] Reference Figure 10 The sub-device 600 may include one or more of the following components: processing component 602, memory 604, power supply component 606, multimedia component 608, audio component 610, input / output (I / O) interface 612, sensor component 614, and communication component 616.

[0127] Processing component 602 typically controls the overall operation of electronic device 600, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 602 may include one or more processors 620 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 602 may include one or more modules to facilitate interaction between processing component 602 and other components. For example, processing component 602 may include a multimedia module to facilitate interaction between multimedia component 608 and processing component 602.

[0128] Memory 604 is used to store various types of data to support the operation of electronic device 600. Examples of this data include instructions for any application or method operating on electronic device 600, contact data, phonebook data, messages, pictures, multimedia, etc. Memory 604 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0129] Power supply component 606 provides power to various components of electronic device 600. Power supply component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 600.

[0130] Multimedia component 608 includes a screen that provides an output interface between the electronic device 600 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may not only sense the boundaries of touch or swipe actions but also detect the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 608 includes a front-facing camera and / or a rear-facing camera. When the electronic device 600 is in an operating mode, such as a shooting mode or a multimedia mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0131] Audio component 610 is used to output and / or input audio signals. For example, audio component 610 includes a microphone (MIC) used to receive external audio signals when electronic device 600 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 604 or transmitted via communication component 616. In some embodiments, audio component 610 also includes a speaker for outputting audio signals.

[0132] I / O interface 612 provides an interface between processing component 602 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, start buttons, and lock buttons.

[0133] Sensor assembly 614 includes one or more sensors for providing state assessments of various aspects of electronic device 600. For example, sensor assembly 614 can detect the on / off state of electronic device 600, the relative positioning of components such as the display and keypad of electronic device 600, changes in position of electronic device 600 or a component of electronic device 600, the presence or absence of user contact with electronic device 600, orientation or acceleration / deceleration of electronic device 600, and temperature changes of electronic device 600. Sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 614 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 614 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0134] Communication component 616 facilitates wired or wireless communication between electronic device 600 and other devices. Electronic device 600 can access wireless networks based on communication standards, such as WiFi, carrier networks (such as 2G, 3G, 4G, or 5G), or combinations thereof. In one exemplary embodiment, communication component 616 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 616 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0135] In an exemplary embodiment, the electronic device 600 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to implement the model generation method and the well-seismic combined velocity modeling method provided in the embodiments of this application.

[0136] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 604 including instructions, which can be executed by a processor 620 of an electronic device 600 to perform the above-described method. For example, the non-transitory storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0137] Figure 11A block diagram of an electronic device 700 is shown according to an exemplary embodiment. For example, the electronic device 700 may be provided as a server. (Refer to...) Figure 11 The electronic device 700 includes a processing component 722, which further includes one or more processors, and memory resources represented by memory 732 for storing instructions, such as application programs, that can be executed by the processing component 722. The application programs stored in memory 732 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 722 is configured to execute instructions to perform the methods provided in the embodiments of this application.

[0138] Electronic device 700 may also include a power supply component 726 configured to perform power management of electronic device 700, a wired or wireless network interface 750 configured to connect electronic device 700 to a network, and an input / output (I / O) interface 758. Electronic device 700 may operate on an operating system stored in memory 732, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0139] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0140] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A model generation method, characterized in that, The method includes: A texture transfer network model is constructed, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct the logging velocity; Acquire sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile; The logging velocity is input into the autoencoder to optimize the parameters of the texture transfer network model, resulting in a trained texture transfer network model. The trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image. The step of inputting the logging velocity into the autoencoder to optimize the parameters of the texture transfer network model includes: The encoder parameters of the freeze generator are adjusted based on the logging rate, and the decoder and autoencoder parameters of the generator are adjusted accordingly. The parameter adjustment of the decoder and autoencoder of the generator based on the logging velocity includes: The fourth loss function is determined based on the input logging rate and the reconstruction rate at the logging location; The transfer learning loss function is determined based on the fourth loss function and the second loss function. This transfer learning loss function is used to adjust the parameters of the texture transfer network, and is expressed by the following second formula: in, The loss function represents transfer learning. This represents the second loss function. This represents the fourth loss function. The decoder parameters of the generator are represented by AE, which represents the autoencoder; the second loss function is determined based on the input logging rate and the logging rate reconstructed by the autoencoder.

2. The method according to claim 1, characterized in that, The generator includes an encoder and a decoder; the decoder and the encoder are connected by a jumper; the decoder has one output channel.

3. The method according to claim 1 or 2, characterized in that, The step of inputting the sample data into the generator and training the texture transfer network model based on the conditional generative adversarial network composed of the generator and the discriminator includes: The initial velocity and depth profiles are input into the encoder of the generator to obtain the feature information corresponding to the initial velocity and depth profiles; The feature information is input into the decoder of the generator, which outputs new sample data that is similar to the sample data. The sample data and new sample data are input into the discriminator, and the texture transfer network model is trained under the drive of the loss function.

4. The method according to claim 3, characterized in that, The sample data and new sample data are input into the discriminator, and the texture transfer network model is trained under the drive of the loss function, including: A first loss function is determined based on the outputs of the generator and the discriminator. This first loss function is used to determine the adversarial loss of the texture transfer network. A second loss function is determined based on the input logging rate and the logging rate reconstructed by the autoencoder. The second loss function is used to determine the logging rate reconstruction loss. A third loss function is determined based on the actual speed and the speed output by the generator, and the third loss function is used to determine the speed reconstruction loss; The target loss function is determined based on the first loss function, the second loss function, and the third loss function, and the texture transfer network model is trained using the target loss function.

5. The method according to claim 4, characterized in that, The step of determining the target loss function based on the first loss function, the second loss function, and the third loss function includes: The first loss function, the second loss function, and the third loss function are added together to form the target loss function, which is expressed by the following first formula: in, Describes the target loss function. Denotes the first loss function. This represents the second loss function. Let G represent the third loss function, G represent the generator, AE represent the autoencoder, and D represent the discriminator. This represents the preset weight parameters.

6. A method for combined well-seismic velocity modeling, characterized in that, The method includes: Acquire the data to be predicted, which includes initial velocity and depth profiles; The data to be predicted is input into the trained texture transfer network model, and a velocity modeling image is output, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image; wherein the texture transfer network model is trained by the model generation method according to any one of claims 1-5.

7. A model generation apparatus, characterized in that, The device includes: A construction module is used to build a texture transfer network model, which includes a generator, a discriminator, and an autoencoder; the generator and discriminator are used to form a conditional generative adversarial network, and the autoencoder is used to reconstruct logging velocity; The first training module is used to acquire sample data, input the sample data into the generator, and train the texture transfer network model based on the conditional generation adversarial network composed of the generator and the discriminator; the sample data includes: initial velocity and depth profile; The second training module is used to input the logging velocity into the autoencoder and optimize the parameters of the texture transfer network model to obtain the trained texture transfer network model. The trained texture transfer network model is used to output a velocity modeling image based on the input initial velocity and depth profile, wherein the velocity and the reflection surface morphology are matched in the velocity modeling image. The second training module includes: The adjustment submodule is used to freeze the encoder parameters of the generator and adjust the parameters of the decoder and autoencoder of the generator based on the logging rate. The adjustment submodule includes: The fourth loss function determination submodule is used to determine the fourth loss function based on the input logging rate and the reconstruction rate at the logging location; The transfer learning loss function determination submodule is used to determine the transfer learning loss function based on the fourth loss function and the second loss function. The transfer learning loss function is used to adjust the parameters of the texture transfer network, and the transfer learning loss function is expressed by the following second formula: in, The loss function represents transfer learning. This represents the second loss function. This represents the fourth loss function. The decoder parameters of the generator are represented by AE, which represents the autoencoder; the second loss function is determined based on the input logging rate and the logging rate reconstructed by the autoencoder.

8. A well-seismic combined velocity modeling device, characterized in that, The device includes: The acquisition module is used to acquire the data to be predicted, which includes initial velocity and depth profiles; The generation module is used to input the data to be predicted into the trained texture transfer network model and output a velocity modeling image, wherein the velocity in the velocity modeling image matches the shape of the reflective surface; wherein the texture transfer network model is trained by the model generation method according to any one of claims 1-5.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a program or instructions that can run on the processor, and the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 6.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 6.

Images