A variable porosity digital core modeling method based on conditional variational autoencoder

Abstract

The application provides a variable porosity digital core modeling method based on a conditional variational autoencoder, and belongs to the technical field of digital cores, and comprises the following steps: collecting reservoir core samples of the same type to construct a three-dimensional digital core; cutting the three-dimensional digital core into a cubic three-dimensional digital core small block with 128*128*128 voxels, and calculating the porosity of each small block as the conditional information of the small block; inputting each three-dimensional digital core and the porosity into a conditional variational autoencoder model as training data and conditional information respectively to train the conditional variational autoencoder model to obtain a trained conditional variational autoencoder model; and inputting an arbitrary porosity and a random vector sampled from a latent space into the trained conditional variational autoencoder model to obtain a new three-dimensional digital core. The method can construct a digital core with an arbitrary porosity.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital core and specifically relates to a variable porosity digital core modeling method based on a conditional variational autoencoder. BACKGROUND

[0002] In recent years, digital core technology has become an effective means for geological exploration and geophysical research, and X-ray CT scanning has been widely used in digital core modeling, and deep learning technology has played a key role in digital core modeling and analysis. By using deep learning methods, digital cores can be efficiently constructed based on existing CT scan data, thereby improving the efficiency of digital core modeling.

[0003] However, this method is difficult to accurately control the parameters of the digital core model, resulting in low accuracy of the digital core modeling and failing to meet the requirements of specific digital core modeling. SUMMARY

[0004] In order to overcome the deficiencies of the prior art, the application provides a variable porosity digital core modeling method based on a conditional variational autoencoder.

[0005] In order to achieve the above purpose, the application provides the following technical scheme:

[0006] A variable porosity digital core modeling method based on a conditional variational autoencoder, comprising:

[0007] Collecting reservoir core samples to construct a three-dimensional digital core;

[0008] Cutting the three-dimensional digital core into 128x128x128 voxel cubic three-dimensional digital core pieces, and calculating the porosity of each piece as the conditional information of each cubic three-dimensional digital core piece;

[0009] Inputting the porosity of each three-dimensional digital core and each cubic three-dimensional digital core piece as training data and conditional information into the encoder of the conditional variational autoencoder model, the encoder mapping the three-dimensional digital core and the porosity to a distribution in a latent space, the decoder of the conditional variational autoencoder model generating a new three-dimensional digital core from the sampling of the latent space and the porosity, and the loss function of the conditional variational autoencoder model reaching a minimum value when the new three-dimensional digital core is the same as the three-dimensional digital core input into the encoder of the conditional variational autoencoder model, thereby obtaining a trained conditional variational autoencoder model;

[0010] Sampling a random vector subject to a standard normal distribution from the latent space, inputting the porosity value of the digital core model to be generated and the random vector into the decoder of the trained conditional variational autoencoder model, and obtaining the three-dimensional digital core model to be generated.

[0011] Furthermore, the collection of reservoir core samples and the construction of a three-dimensional digital core include:

[0012] Core samples of the same type of reservoir from a certain region were collected and subjected to CT scans and FIB-SEM scans. After filtering, noise reduction, and image segmentation, a three-dimensional digital core was constructed.

[0013] Furthermore, the encoder and decoder include: convolutional layers, batch normalization layers, and activation function layers.

[0014] Furthermore, the loss function of the conditional variational autoencoder includes reconstruction loss, KL divergence loss, and conditional loss;

[0015] Reconstruction loss is used to measure the distance between the generated digital core and the original digital core; KL divergence loss is used to measure the distance between the distribution of latent variables and the standard normal distribution; conditional loss is used to add porosity to the generative model.

[0016] Furthermore, the reconstruction loss is:

[0017]

[0018] Where n is the number of samples, and x and y are the actual value and the predicted value, respectively.

[0019] Furthermore, the KL divergence loss is:

[0020]

[0021] Where μ represents the mean of the learned latent distribution, σ 2 This represents the variance of the learned latent distribution.

[0022] The variable porosity digital core modeling method based on conditional variational autoencoder provided by this invention has the following beneficial effects:

[0023] This invention proposes a modeling method for constructing digital cores with different porosities using a conditional variational autoencoder, thereby controlling the porosity range in digital core modeling and improving the efficiency of digital core construction. Utilizing deep learning to construct digital cores of the same type of sample with different porosities is of significant value for pore structure analysis and numerical simulation of rock physical properties, providing effective technical support for geological and geophysical research. Attached Figure Description

[0024] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram of a variable porosity digital core modeling process based on a conditional variational autoencoder;

[0026] Figure 2 Three-dimensional digital core samples for training purposes;

[0027] Figure 3 The result of constructing a three-dimensional digital core with arbitrary porosity. Detailed Implementation

[0028] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0029] Example:

[0030] This invention provides a method for modeling variable porosity digital cores based on a conditional variational autoencoder, specifically as follows: Figure 1 As shown, the process includes: collecting reservoir core samples of the same type to construct a three-dimensional digital core; cutting the three-dimensional digital core into 128×128×128 voxel cube three-dimensional digital core blocks, calculating the porosity of each block as the conditional information of that block; inputting each three-dimensional digital core and porosity as training data and conditional information into a conditional variational autoencoder model for training, to obtain a trained conditional variational autoencoder model; and inputting arbitrary porosity and random vectors sampled from the potential space into the trained conditional variational autoencoder model to obtain a new three-dimensional digital core.

[0031] The following are the specific implementation details of this invention:

[0032] 1. Sample Collection: Core samples of the same type of reservoir from a certain area were collected for CT scanning, FIB-SEM scanning, and other experiments. After processing methods such as filtering and noise reduction, and image segmentation, three-dimensional digital cores were constructed. All constructed three-dimensional digital cores were cut into 128×128×128 voxel cubes, and the porosity of each cube was calculated as the condition information for that cube.

[0033] 2. Model Architecture:

[0034] The conditional variational autoencoder (CVAE) consists of two parts: an encoder and a decoder.

[0035] The encoder maps the input data (three-dimensional digital core) and the condition information (porosity) to a distribution in a latent space, and the decoder generates data (new three-dimensional digital core) from the sampling of the latent space and the condition information (porosity).

[0036] The encoder and decoder are composed of convolutional layers, batch normalization and activation functions.

[0037] 3. Loss function: The loss function of CVAE includes reconstruction loss, KL divergence loss and condition loss. The reconstruction loss is used to measure the distance between the generated digital core and the original digital core, i.e. the quality of the reconstructed digital core; the KL divergence loss is used to measure the distance between the distribution of the latent variable and the standard normal distribution, and the condition loss is used to add the condition information (porosity) to the generative model.

[0038] where the reconstruction loss is calculated using the mean square error (MSE):

[0039]

[0040] where n is the number of samples, x and y are the true value and the predicted value respectively.

[0041] KL divergence loss:

[0042]

[0043] The condition loss also uses the MSE loss.

[0044] 4. Train the model: input each three-dimensional digital core and porosity as training data and condition information respectively into the conditional variational autoencoder model for training. For each training sample, calculate the loss and update the model parameters. This process is completed using the backpropagation algorithm.

[0045] 5. Model validation: after training, use the test data set to evaluate the performance of the model, calculate the reconstruction loss, condition loss and the quality of the generated samples to evaluate the performance of the model.

[0046] 6. Use the model to generate new three-dimensional digital core samples: the trained CVAE can be used to generate new samples with any porosity. Input any porosity and a random vector sampled from the latent space to generate a new three-dimensional digital core.

[0047] The above embodiments are only the preferred embodiments of the present application, and the protection scope of the present application is not limited thereto, and any simple change or equivalent replacement of the technical solutions which can be obviously obtained by those skilled in the art within the technical range disclosed by the present application shall belong to the protection scope of the present application.

Claims

1. A method for modeling variable porosity digital cores based on a conditional variational autoencoder, characterized in that, include: Collect reservoir core samples and construct a three-dimensional digital core. The three-dimensional digital core was cut into cubic three-dimensional digital core blocks of 128×128×128 voxels, and the porosity of each block was calculated as the condition information of each cubic three-dimensional digital core block. The porosity of each 3D digital core and each cubic 3D digital core block is used as training data and conditional information, respectively, and input into the encoder of the conditional variational autoencoder model. The encoder maps the 3D digital core and porosity to a distribution in a latent space. The decoder of the conditional variational autoencoder model generates a new 3D digital core from the sampling and porosity in the latent space. When the new 3D digital core is the same as the 3D digital core input to the encoder of the conditional variational autoencoder model, the loss function of the conditional variational autoencoder model reaches its minimum value. Obtain a well-trained conditional variational autoencoder model; A random vector following a standard normal distribution is sampled from the latent space. The porosity value of the digital core model to be generated and the random vector are input into the decoder of the trained conditional variational autoencoder model to obtain the three-dimensional digital core model to be generated. The loss function of the conditional variational autoencoder model includes reconstruction loss, KL divergence loss, and conditional loss; Reconstruction loss is used to measure the distance between the generated digital core and the original digital core; KL divergence loss is used to measure the distance between the distribution of latent variables and the standard normal distribution; conditional loss is used to add porosity to the generative model.

2. The method for modeling variable porosity digital cores based on a conditional variational autoencoder according to claim 1, characterized in that, The collection of reservoir core samples and the construction of a three-dimensional digital core include: Core samples of the same type of reservoir from a certain region were collected and subjected to CT scans and FIB-SEM scans. After filtering, noise reduction, and image segmentation, a three-dimensional digital core was constructed.

3. The method for modeling variable porosity digital cores based on a conditional variational autoencoder according to claim 1, characterized in that, The encoder and decoder include: a convolutional layer, a batch normalization layer, and an activation function layer connected in sequence.

4. The method for modeling variable porosity digital cores based on a conditional variational autoencoder according to claim 1, characterized in that, The reconstruction loss is: Where n is the number of samples, and x and y are the actual value and the predicted value, respectively.

5. The method for modeling variable porosity digital cores based on a conditional variational autoencoder according to claim 1, characterized in that, The KL divergence loss is: Where μ represents the mean of the learned latent distribution, σ 2 This represents the variance of the learned latent distribution.

Images