Fault modeling method and system based on 3D convolutional neural network

Abstract

The invention relates to a fault modeling method and system based on a 3D convolutional neural network, and the method comprises the steps: obtaining seismic data and logging data, and carrying out the preprocessing of the seismic data and logging data; carrying out data fusion on the seismic data and the logging data; constructing a 3D convolutional neural network model; performing cross validation on the 3D convolutional neural network model by adopting a cross entropy loss function, and further optimizing parameters of the 3D convolutional neural network model; performing fault prediction on the input data through the 3D convolutional neural network model, outputting a three-dimensional fault model, and predicting fault coordinates and geometric morphology; gaussian filtering is adopted to carry out smooth optimization on the three-dimensional fault model, the boundary definition is improved, and the fault model is displayed through a three-dimensional visualization tool. According to the method, the seismic data and the logging data are fused, the fault features are automatically extracted through the 3D convolutional neural network model, fault modeling automation is achieved, the modeling precision and the calculation efficiency are improved, manual intervention is reduced, and visualization and interactivity are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional geological modeling, and particularly to a fault modeling method and system based on a 3D convolutional neural network. Background Art

[0002] Three-dimensional fault modeling is a key technical means for understanding underground geological structures, mainly used in fields such as geological exploration and oil and gas development. Existing three-dimensional fault modeling methods include the holistic method, the unified modeling method of strata and faults, and the modeling method based on machine learning.

[0003] In the prior art, the holistic method for three-dimensional fault modeling usually assumes that faults do not exist. First, a three-dimensional stratum model without faults is constructed, and then the fault data is introduced for repair. Although the holistic method is applicable to fault models with small fault offsets and simple structures, when applied in complex geological regions, the adaptability and accuracy of the model are poor, and it cannot effectively handle large-scale complex fault structures. The unified modeling method of strata and faults relies on expert experience. Especially under complex geological conditions, the subjective judgment of experts has a great influence on the model results, making it difficult to achieve full automation, with low work efficiency and prone to errors. Although existing machine learning methods can improve the degree of automation to a certain extent, they usually require a large amount of labeled data for training and lack sufficient generalization ability, making it difficult to adapt to the complex fault characteristics of different geological regions.

[0004] Therefore, it is necessary to provide a new fault modeling method based on a 3D convolutional neural network. Summary of the Invention

[0005] Based on the above problems existing in the prior art, the purpose of the embodiments of the present invention is to provide a fault modeling method and system based on a 3D convolutional neural network. By fusing seismic and logging data, the 3D convolutional neural network model is used to automatically extract fault features, realizing the automation of fault modeling, improving the modeling accuracy, calculation efficiency, reducing manual intervention, and enhancing visualization and interactivity.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is: A fault modeling method based on a 3D convolutional neural network, comprising:

[0007] S1, obtaining seismic data and logging data, and respectively preprocessing the seismic data and logging data;

[0008] S2, performing data fusion on the seismic data and the logging data, and the data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion;

[0009] S3. Construct a 3D convolutional neural network model, which includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is used to extract tomographic spatial features, the pooling layer is used to reduce the dimension, and the fully connected layer is used to integrate and output.

[0010] S4. Use the cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and then optimize the parameters of the 3D convolutional neural network model.

[0011] S5. Use the 3D convolutional neural network model to perform tomographic prediction on the input data, and output a three-dimensional tomographic model, predicted tomographic coordinates, and geometric shapes.

[0012] S6. Use Gaussian filtering to smooth and optimize the three-dimensional tomographic model to improve the boundary clarity, and display the tomographic model through a three-dimensional visualization tool.

[0013] Furthermore, the seismic data is SEG-Y format seismic data. The SEG-Y format seismic data includes three-dimensional spatial coordinates and the amplitude information of seismic waves. The three-dimensional spatial coordinates include the X-axis, Y-axis, and T time axis. The SEG-Y format seismic data is stored as a three-dimensional matrix, which is represented as (X×Y×T); the logging data includes detailed physical property parameters of underground horizons. The detailed physical property parameters of underground horizons include porosity, permeability, and lithology. The logging data is in the form of three-dimensional physical property parameters along the vertical depth axis (Z).

[0014] Furthermore, the spatial alignment of the seismic data includes converting the time axis T of the seismic data into the depth axis Z to align it with the logging data in depth. The conversion formula is:

[0015]

[0016] where Z is the depth coordinate of the seismic data, V avg is the average velocity model, and T [[ID=2,4]] time is the seismic time data.

[0017] Furthermore, the feature splicing includes splicing the three-dimensional amplitude matrix of the seismic data and the three-dimensional physical property parameters of the logging data in the channel dimension. Among them, the three-dimensional physical property parameter volume of the logging data is generated from discrete well point data through Kriging interpolation to form a four-dimensional input tensor; the four-dimensional input tensor is input∈X×Y×Z×(C SEISMIC +S logging ), where C SEISMIC is the number of channels of the seismic data, S logging is the physical property parameter of the logging data. The four-dimensional input tensor integrates the spatial distribution information of the seismic data and the physical property details of the logging data.

[0018] Further, the normalization includes performing Z-Score standardization on seismic data and logging data respectively to ensure that the scales of all data sources are consistent. The normalization formula is:

[0019]

[0020] where x norm is the normalized data, x is the original data, μ is the mean of the original data, and σ is the standard deviation of the original data;

[0021] The weighted fusion includes performing weighted fusion on the normalized seismic data and logging data using an adaptive weight; the adaptive weight takes α = 0.6, and the weighted fusion calculation formula is:

[0022] Faused = α·Seismic+(1-α)·Logging

[0023] where Faused is the fused data, α is the adaptive weight, Seismic is the seismic data, and Logging is the logging data.

[0024] Further, the specific steps for constructing the 3D convolutional neural network model include:

[0025] Step S31, the convolutional layer extracts the fault spatial features, including sliding convolution along the x, y, and z directions through a three-dimensional convolutional kernel to extract local features such as faults and lithology changes, introducing non-linearity through the ReLU activation function to form a higher-level feature map, containing information such as fault strike and dip;

[0026] Step S32, uses max pooling to reduce the dimension of the convolved feature map, reduces the spatial dimension, reduces the computational complexity, and retains significant geological features such as the position of the fault plane;

[0027] Step S33, flattens the features after multiple layers of convolution and pooling, integrates them through a fully connected layer, and the output layer predicts the position, strike, and dip of the fault plane to generate a three-dimensional fault model.

[0028] Further, the cross-entropy loss function is used to measure the difference between the model prediction value and the true label. Let the fault probability prediction value output by the 3D convolutional neural network model be y ′ , and the true label be y, then the cross-entropy loss function L is defined as:

[0029]

[0030] where L is the cross-entropy loss value, N is the number of samples, y i is the true label representing the i-th sample, and y i ′ is the predicted probability value of the i-th sample.

[0031] A tomographic modeling system based on a 3D convolutional neural network, which is applied to the above-mentioned tomographic modeling method based on a 3D convolutional neural network. The system includes:

[0032] A data acquisition module, which is used to acquire seismic data and logging data, and preprocess the seismic data and logging data respectively;

[0033] A data fusion module, which is used to fuse the seismic data and the logging data. The data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion;

[0034] A modeling module, which is used to build a 3D convolutional neural network model. The 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer. The tomographic spatial features are extracted through the convolutional layer, the dimension is reduced through the pooling layer, and integration and output are performed through the fully connected layer;

[0035] A model training and optimization module, which is used to perform cross-validation on the 3D convolutional neural network model by using a cross-entropy loss function, and then optimize the parameters of the 3D convolutional neural network model;

[0036] A prediction output module, which is used to perform tomographic prediction on the input data through the 3D convolutional neural network model, and output a three-dimensional tomographic model, predicted tomographic coordinates, and geometric shapes;

[0037] A result optimization and display module, which is used to perform smoothing optimization on the three-dimensional tomographic model by using Gaussian filtering to improve the boundary clarity, and display the tomographic model through a three-dimensional visualization tool.

[0038] An embodiment of the present invention also provides a network-side server, including:

[0039] At least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned tomographic modeling method based on a 3D convolutional neural network.

[0040] An embodiment of the present invention also provides a computer-readable storage medium, storing a computer program, and when the computer program is executed by a processor, the above-mentioned tomographic modeling method based on a 3D convolutional neural network is implemented.

[0041] The beneficial effects of the present invention are as follows: A tomographic modeling method based on a 3D convolutional neural network of the present invention includes: acquiring seismic data and logging data, and respectively preprocessing the seismic data and the logging data; performing data fusion on the seismic data and the logging data, and the data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion; constructing a 3D convolutional neural network model, the 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer, extracting tomographic spatial features through the convolutional layer, reducing the dimension through the pooling layer, and integrating and outputting through the fully connected layer; using a cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and further optimizing the parameters of the 3D convolutional neural network model; performing tomographic prediction on the input data through the 3D convolutional neural network model, and outputting a three-dimensional tomographic model, predicted tomographic coordinates, and geometric shapes; using Gaussian filtering to perform smoothing optimization on the three-dimensional tomographic model to improve the boundary clarity, and displaying the tomographic model through a three-dimensional visualization tool. The tomographic modeling method based on a 3D convolutional neural network of the present invention effectively fuses seismic data and logging data. Seismic data provides extensive spatial information, while logging data supplements detailed physical properties, improving the accuracy and reliability of the model; adopting a modeling method of a 3D convolutional neural network can efficiently process large-scale seismic data, reducing the problems of high computational cost and long modeling process in traditional modeling methods. Especially in the processing of multi-dimensional data, the convolutional operation of the 3D convolutional neural network model can process a large amount of data in parallel, greatly improving the computational efficiency of modeling. And adopting modern three-dimensional visualization technology, through Web-side rendering technology, real-time tomographic modeling and dynamic display are realized; users can view the spatial distribution, geometric features, etc. of the tomographic model through interactive operations, greatly improving the operability and visualization effect of the model, and helping geological analysis and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described below with reference to the drawings and embodiments.

[0043] In the figures:

[0044] Figure 1 is a flowchart of the tomographic modeling method based on a 3D convolutional neural network provided by the first embodiment of the present invention;

[0045] Figure 2 is a schematic diagram of the modules of the tomographic modeling system based on a 3D convolutional neural network provided by the second embodiment of the present invention;

[0046] Figure 3 is a schematic diagram of the structure of the network-side server provided by the third embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] First Embodiment:

[0049] The first embodiment of the present invention provides a tomographic modeling method based on a 3D convolutional neural network, including: obtaining seismic data and logging data, and respectively preprocessing the seismic data and the logging data; performing data fusion on the seismic data and the logging data, where the data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion; constructing a 3D convolutional neural network model, the 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer, extracting tomographic spatial features through the convolutional layer, reducing the dimension through the pooling layer, and integrating and outputting through the fully connected layer; using a cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and then optimizing the parameters of the 3D convolutional neural network model; performing tomographic prediction on the input data through the 3D convolutional neural network model, and outputting a three-dimensional tomographic model, predicted tomographic coordinates, and geometric morphology; using Gaussian filtering to perform smoothing optimization on the three-dimensional tomographic model to improve the boundary clarity, and displaying the tomographic model through a three-dimensional visualization tool. The tomographic modeling method based on the 3D convolutional neural network of the present invention effectively fuses seismic data and logging data through time-depth conversion. Seismic data provides extensive spatial information, while logging data supplements detailed physical properties, improving the accuracy and reliability of the model; fusing seismic data and logging data can accurately extract important features such as fault planes and stratigraphic changes when dealing with complex geological bodies. Using a 3D convolutional neural network model to learn the deep spatial relationships in the data can provide higher-precision tomographic modeling results; the process of fault identification and modeling is completely based on the training and prediction of the network model, greatly improving the automation level, increasing work efficiency, and reducing the influence of human factors on the modeling results; using the modeling method of a 3D convolutional neural network can efficiently process large-scale seismic data, reducing the problems of high computational cost and long modeling process in traditional modeling methods. Especially in the processing of multi-dimensional data, the convolutional operation of the 3D convolutional neural network model can parallelly process a large amount of data, greatly improving the computational efficiency of modeling; and using modern three-dimensional visualization technology, through Web-side rendering technology, real-time tomographic modeling and dynamic display are realized; users can view the spatial distribution, geometric features, etc. of the tomographic model through interactive operations, greatly improving the operability and visualization effect of the model, and helping geological analysis and decision support.

[0050] The implementation details of the tomographic modeling method based on 3D convolutional neural network in this embodiment will be specifically described below. The following content is only the implementation details provided for easy understanding and is not necessary for implementing this solution. The specific process of this embodiment is as Figure 1 shown.

[0051] Step S1: Obtain seismic data and logging data, and preprocess the seismic data and logging data respectively.

[0052] Specifically, the seismic data is SEG-Y format seismic data. The SEG-Y format seismic data includes three-dimensional spatial coordinates and amplitude information of seismic waves. The three-dimensional spatial coordinates include the X-axis, Y-axis, and T time axis. The SEG-Y format seismic data is stored as a three-dimensional matrix, which is represented as (X×Y×T), where each position corresponds to the amplitude value of seismic reflection and contains information about the reflecting surface, and is further used to infer the underground geological structure; by using professional tools to read the SEG-Y format seismic data, extract the time series, amplitude information, and reflecting surface information, and construct a three-dimensional volume data.

[0053] Furthermore, label the seismic data, and manually label the formation, fault, and reflecting surface information in the seismic data by geological experts. The labeled seismic data is stored as an additional data set and used in conjunction with the SEG-Y data in the form of a label file for training and supervising the 3D convolutional neural network model for fault plane prediction.

[0054] The logging data includes detailed physical property parameters of underground horizons, such as porosity, permeability, lithology, etc. The logging data is in the form of three-dimensional physical property parameters along the vertical depth axis (Z). The logging data can provide more detailed physical property information, and there is a strong spatial correlation between the logging data and the seismic data.

[0055] Step S2: Perform data fusion on the seismic data and the logging data. The data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion.

[0056] Specifically, the spatial alignment of the seismic data includes converting the time axis T of the seismic data into the depth axis Z to align with the logging data in depth. The conversion formula is:

[0057]

[0058] where Z is the depth coordinate of the seismic data, V avg is the average velocity model, and T time is the seismic time data.

[0059] Feature splicing includes splicing the three-dimensional amplitude matrix of seismic data and the three-dimensional physical property parameters of logging data in the channel dimension. Among them, the three-dimensional physical property parameter volume of logging data is generated from discrete well point data through Kriging interpolation to form a four-dimensional input tensor. The four-dimensional input tensor is input ∈ X × Y × Z × (C SEISMIC +S logging ). Among them, C SEISMIC is the number of seismic data channels, and S logging is the physical property parameter of logging data. The four-dimensional input tensor integrates the spatial distribution information of seismic data and the physical property details of logging data.

[0060] Normalization includes performing Z-Score standardization on seismic data and logging data respectively to ensure that the scales of all data sources are consistent. The normalization formula is:

[0061]

[0062] Among them, x norm is the normalized data, x is the original data, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0063] Weighted fusion includes performing weighted fusion on the normalized seismic data and logging data using an adaptive weight. The adaptive weight takes α = 0.6. The weighted fusion calculation formula is:

[0064] Faused = α·Seismic + (1 - α)·Logging

[0065] Among them, Faused is the fused data, α is the adaptive weight, Seismic is the seismic data, and Logging is the logging data.

[0066] Step S3, construct a 3D convolutional neural network model. The 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer extracts the fault spatial features, the pooling layer reduces the dimension, and the fully connected layer integrates and outputs.

[0067] Specifically, the specific steps for constructing the 3D convolutional neural network model include:

[0068] Step S31, the convolutional layer extracts the fault spatial features by sliding the three-dimensional convolutional kernel along the x, y, and z directions for convolution, extracting local features such as faults and lithology changes, introducing non-linearity through the ReLU activation function, and forming a higher-level feature map containing information such as fault strike and dip.

[0069] Specifically, use the three-dimensional convolutional kernel to slide along the three spatial dimensions of X, Y, and Z to perform convolution operations on the four-dimensional input tensor.

[0070] The first convolutional layer extracts local features, such as basic geological features like fault plane edges, lithology change interfaces, reflection layer boundaries, etc. The second convolutional layer gradually extracts high-level features through superposition, such as fault strikes, for example, dip directions, dips, multi-fault intersection relationships, formation offset patterns, etc.

[0071] After convolution, the ReLU activation function is used for non-linear transformation. The formula of the ReLU activation function is:

[0072] ReLU(x) = max(0, x)

[0073] By introducing non-linear mapping through the ReLU activation function, the 3D convolutional neural network model can learn complex geological structure patterns.

[0074] As an example, the convolutional kernel size is 3×3×3 (three-dimensional kernel) to balance the local feature capture ability and computational efficiency; the stride is 1 to ensure the continuity of feature extraction and spatial resolution.

[0075] Step S32, use max pooling to reduce the dimension of the convolved feature map, reduce the spatial dimension, lower the computational complexity, and retain significant geological features such as the position of the fault plane.

[0076] Specifically, a 3D max pooling layer is connected after each convolutional layer to reduce the dimension of the feature map; the pooling kernel size is 2×2×2, and the stride is set to 2 to slide, and the maximum value in the local area is selected as the output. The pooling layer reduces the spatial dimension of the feature map, such as the X×Y×Z size is reduced, and the computational complexity is lowered; in addition, it can retain significant geological features. For example, the position of the fault plane, strong reflection interfaces, suppressing noise and secondary information. After pooling, the dimension of the feature map decreases layer by layer, but the key fault features (such as position, strike) are strengthened and retained.

[0077] Therefore, through the pooling operation, the 3D convolutional neural network model can effectively reduce the dimension of the feature map, while retaining the most critical information and improving computational efficiency.

[0078] Step S33, flatten the features after multiple convolutions and poolings, integrate them through the fully connected layer, and the output layer predicts the position, strike, and dip direction of the fault plane to generate a three-dimensional fault model.

[0079] Specifically, after completing multiple convolution and pooling operations, the high-level features of the network need to be further integrated. The fully connected layer flattens the features extracted by convolution and pooling and performs further processing to output the final fault prediction result. The role of the fully connected layer is to integrate the features extracted by different convolutional layers to obtain a more comprehensive representation.

[0080] Flatten the three-dimensional feature map output by the last pooling layer into a one-dimensional vector and input it into the fully connected layer. The neurons in the fully connected layer are connected to all neurons in the previous layer, enabling the 3D convolutional neural network model to fuse features from different convolutional layers. By fusing the features extracted from different convolutional layers, such as the spatial distribution of faults and the correlation of physical property parameters, a global feature representation is formed.

[0081] The last layer of the fully connected layer serves as the output layer. Through linear transformation, the output layer predicts the fault probability value or geometric parameters of each voxel. For example, it predicts the position (three-dimensional coordinates), strike, and dip of the fault. After the values output by the neurons in the output layer are processed by the activation function, the final fault prediction results are obtained, such as the probability of the existence of a fault at each position and the geometric parameters of the fault, thereby generating a complete three-dimensional fault model.

[0082] Step S4: Use the cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and then optimize the parameters of the 3D convolutional neural network model.

[0083] Specifically, the preprocessed and fused seismic data and logging data are divided into a training set, a validation set, and a test set. The training set is used to train the 3D convolutional neural network model, the validation set is used to evaluate the performance of the 3D convolutional neural network model and adjust the hyperparameters during the training process, and the test set is used to finally evaluate the generalization ability of the trained 3D convolutional neural network model.

[0084] The cross-entropy loss function is used to measure the difference between the model prediction value and the true label. For the fault modeling problem, let the predicted fault probability value output by the 3D convolutional neural network model be y ′ , and the true label be y, then the cross-entropy loss function L is defined as:

[0085]

[0086] where L is the cross-entropy loss value, N is the number of samples, y i represents the true label of the i-th sample, and y i ′ is the predicted probability value of the i-th sample.

[0087] Adopt k-fold cross-validation. Divide the training set into k subsets again. Each time, select k - 1 of these subsets as the training data, and the remaining 1 subset as the validation data. Repeat this k times to ensure that each subset has the opportunity to be used as the validation set. During each training process, calculate the cross-entropy loss value of the 3D convolutional neural network model on the validation set and record the optimal parameters of the 3D convolutional neural network model.

[0088] An optimizer is used to optimize and update the parameters of the 3D convolutional neural network model. As an example, the Adam optimizer is adopted to dynamically adjust the learning rate of each parameter according to the first-moment estimate and second-moment estimate of the gradient of each parameter. In each training iteration, the gradient of the loss function with respect to the parameters of the 3D convolutional neural network model is calculated, and then the parameters are adjusted according to the update rule of the optimizer, so that the value of the loss function continuously decreases, thereby optimizing the 3D convolutional neural network model.

[0089] As an example, the performance comparison between the existing network model and the optimized 3D convolutional neural network model in this embodiment is shown in Table 1 below.

[0090] Table 1 Performance Comparison of SVM, KNN and 3D Convolutional Neural Network Models

[0091]

[0092] In step S5, the input data is tomographically predicted through the 3D convolutional neural network model, and a three-dimensional tomographic model, predicted tomographic coordinates and geometric shapes are output.

[0093] Specifically, the four-dimensional input tensor after preprocessing and fusion processing is input into the 3D convolutional neural network model after parameter optimization. Among them, the four-dimensional input tensor is the fusion result of seismic data and logging data.

[0094] The fused seismic data and logging data are successively propagated forward through the convolutional layer, pooling layer and fully connected layer of the 3D convolutional neural network model. In the convolutional layer, the three-dimensional convolutional kernel extracts the local features related to the faults in the seismic data and logging data; the pooling layer reduces the dimension of the feature map and retains the key features; the fully connected layer integrates the extracted features, and finally the output layer outputs the tomographic prediction results at each voxel position according to the integrated features.

[0095] The output result is a three-dimensional tomographic model, and each voxel in the three-dimensional tomographic model corresponds to a predicted tomographic probability value. By setting a probability threshold, the voxel positions with probability values greater than the threshold are determined as the predicted tomographic positions, thereby obtaining the predicted tomographic coordinates. According to these coordinate points and the information such as the strike and dip of the faults learned in the model, the geometric shapes of the faults are determined, such as the inclination angle and extension direction of the fault plane.

[0096] In step S6, Gaussian filtering is used to smooth and optimize the three-dimensional tomographic model to improve the boundary clarity, and the tomographic model is displayed through a three-dimensional visualization tool.

[0097] Specifically, the use of Gaussian filtering to smooth and optimize the three-dimensional tomographic model to improve the boundary clarity includes: performing smoothing processing on the obtained three-dimensional tomographic model using three-dimensional Gaussian filtering.

[0098] The core of Gaussian filtering is a three-dimensional Gaussian kernel, and the probability density function of the three-dimensional Gaussian is:

[0099]

[0100] where (x, y, z) are the three-dimensional space coordinates and σ is the standard deviation.

[0101] The standard deviation σ is used to control the smoothness of the Gaussian kernel. By setting appropriate kernel sizes and standard deviations, convolving the Gaussian kernel with the three-dimensional tomographic model, and performing weighted averaging on each voxel value in the model, noise can be suppressed, the model surface can be smoothed, and thus the clarity of the tomographic boundary can be improved.

[0102] As an example, the Gaussian kernel is 3×3×3 and the standard deviation σ = 1.0.

[0103] Use a three-dimensional visualization tool combined with Web-based three-dimensional rendering technology to display the processed three-dimensional tomographic model. First, convert the data of the smoothed three-dimensional tomographic model to meet the input requirements of the visualization tool; then set parameters such as viewing angle, lighting, and color mapping in the visualization interface to intuitively display information such as the spatial distribution, geometric shape, and trend of the tomographic model. Users can view the three-dimensional tomographic model from different angles through interactive operations, which is convenient for analyzing and studying the geological structure.

[0104] As an example, the three-dimensional visualization tool can use PyVista and VTK.

[0105] A tomographic modeling method based on a 3D convolutional neural network according to the present invention includes: acquiring seismic data and logging data, and preprocessing the seismic data and the logging data respectively; performing data fusion on the seismic data and the logging data, and the data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion; constructing a 3D convolutional neural network model, the 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer, extracting tomographic spatial features through the convolutional layer, reducing the dimension through the pooling layer, and integrating and outputting through the fully connected layer; using a cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and further optimizing the parameters of the 3D convolutional neural network model; performing tomographic prediction on the input data through the 3D convolutional neural network model, and outputting a three-dimensional tomographic model, predicted tomographic coordinates, and geometric morphology; using Gaussian filtering to perform smoothing optimization on the three-dimensional tomographic model to improve the boundary clarity, and displaying the tomographic model through a three-dimensional visualization tool. The tomographic modeling method based on a 3D convolutional neural network according to the present invention effectively fuses seismic data and logging data through time-depth conversion. Seismic data provides extensive spatial information, while logging data supplements detailed physical properties, improving the accuracy and reliability of the model; fusing seismic data and logging data can accurately extract important features such as fault planes and stratigraphic changes when dealing with complex geological bodies. Using a 3D convolutional neural network model to learn the deep spatial relationships in the data can provide a higher-precision tomographic modeling result; the process of fault identification and modeling is completely based on the training and prediction of the network model, greatly improving the automation degree, improving work efficiency, and reducing the influence of human factors on the modeling result; using the modeling method of a 3D convolutional neural network can efficiently process large-scale seismic data, reducing the problems of high computational cost and long modeling process in traditional modeling methods. Especially in the processing of multi-dimensional data, the convolutional operation of the 3D convolutional neural network model can process a large amount of data in parallel, greatly improving the computational efficiency of modeling; and using modern three-dimensional visualization technology, through Web-side rendering technology, real-time tomographic modeling and dynamic display are realized; users can view the spatial distribution, geometric features, etc. of the tomographic model through interactive operations, greatly improving the operability and visualization effect of the model, and helping geological analysis and decision support.

[0106] Second Embodiment:

[0107] As Figure 2 shown, the second embodiment of the present invention provides a tomographic modeling system based on a 3D convolutional neural network. The system includes: a data acquisition module 201, a data fusion module 202, a modeling module 203, a model training and optimization module 204, a prediction and output module 205, and a result optimization and display module 206.

[0108] Specifically, the data acquisition module 201 is used to acquire seismic data and logging data, and preprocess the seismic data and logging data respectively; the data fusion module 202 is used to perform data fusion on the seismic data and the logging data, and the data fusion includes spatial alignment, feature splicing, normalization and weighted fusion; the modeling module 203 is used to construct a 3D convolutional neural network model, and the 3D convolutional neural network model includes a convolutional layer, a pooling layer and a fully connected layer, and the fault spatial features are extracted through the convolutional layer, the dimension is reduced by the pooling layer, and the integration and output are performed by the fully connected layer; the model training and optimization module 204 is used to perform cross-validation on the 3D convolutional neural network model by using the cross-entropy loss function, and then optimize the parameters of the 3D convolutional neural network model; the prediction output module 205 is used to perform fault prediction on the input data through the 3D convolutional neural network model, and output a three-dimensional fault model, predicted fault coordinates and geometric shapes; the result optimization and display module 206 is used to perform smoothing optimization on the three-dimensional fault model by using Gaussian filtering to improve the boundary clarity, and display the fault model through a three-dimensional visualization tool.

[0109] It is not difficult to find that this embodiment is a system embodiment corresponding to the first embodiment, and this embodiment can be implemented in cooperation with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and in order to reduce repetition, they will not be elaborated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied in the first embodiment.

[0110] It is worth mentioning that each module involved in this embodiment is a logical module. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of the present invention, units that are not closely related to solving the technical problems proposed by the present invention are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

[0111] The third embodiment of the present invention relates to a network-side server, as Figure 3 shown, including at least one processor 302; and a memory 301 communicatively connected to the at least one processor 302; wherein, the memory 301 stores instructions executable by the at least one processor 302, and the instructions are executed by the at least one processor 302 to enable the at least one processor 302 to execute the above data processing method.

[0112] Among them, the memory 301 and the processor 302 are connected in a bus manner. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors 302 and the memory 301 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, etc. These are well known in the art, so they will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium. The data processed by the processor 302 is transmitted on the wireless medium through the antenna. Further, the antenna also receives data and transmits the data to the processor 302.

[0113] The processor 302 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. The memory 301 can be used to store the data used by the processor 302 when performing operations.

[0114] The fourth embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the tomographic modeling method based on a 3D convolutional neural network in the first embodiment.

[0115] That is, those skilled in the art can understand that all or part of the steps in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions for causing a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0116] The above are only embodiments of the present invention. Common general knowledge such as specific structures and characteristics known in the art is not described in detail herein. Those of ordinary skill in the art know all the general technical knowledge in the technical field to which the invention pertains before the filing date or the priority date, can obtain all the prior art in this field, and have the ability to apply conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, complete and implement this solution in combination with their own abilities. Some typical well-known structures or well-known methods should not become an obstacle for those of ordinary skill in the art to implement this application. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several modifications and improvements can also be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope claimed in this application should be based on the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

[0117] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A tomographic modeling method based on a 3D convolutional neural network, characterized in that, Including: S1. Obtain seismic data and logging data, and preprocess the seismic data and logging data respectively; S2. Perform data fusion on the seismic data and logging data. The data fusion includes spatial alignment, feature splicing, normalization, and weighted fusion; S3. Construct a 3D convolutional neural network model. The 3D convolutional neural network model includes a convolutional layer, a pooling layer, and a fully connected layer. Extract the fault spatial features through the convolutional layer, reduce the dimension through the pooling layer, and integrate and output through the fully connected layer; S4. Use the cross-entropy loss function to perform cross-validation on the 3D convolutional neural network model, and then optimize the parameters of the 3D convolutional neural network model; S5. Perform fault prediction on the input data through the 3D convolutional neural network model, and output a three-dimensional fault model, predicted fault coordinates, and geometric morphology; S6. Use Gaussian filtering to smooth and optimize the three-dimensional fault model to improve the boundary clarity, and display the fault model through a three-dimensional visualization tool.

2. The tomographic modeling method based on a 3D convolutional neural network according to claim 1, wherein The seismic data is seismic data in SEG-Y format. The SEG-Y format seismic data includes three-dimensional spatial coordinates and the amplitude information of seismic waves. The three-dimensional spatial coordinates include the X-axis, Y-axis, and T time axis. The SEG-Y format seismic data is stored as a three-dimensional matrix, which is expressed as (X×Y×T); the logging data includes the detailed physical property parameters of underground horizons. The detailed physical property parameters of underground horizons include porosity, permeability, and lithology. The form of the logging data is three-dimensional physical property parameters along the vertical depth axis (Z).

3. The tomographic modeling method based on a 3D convolutional neural network according to claim 1, characterized in that The spatial alignment of the seismic data includes converting the time axis T of the seismic data into the depth axis Z to align with the logging data in depth. The conversion formula is: where Z is the depth coordinate of seismic data, V avg is the average velocity model, and T time is the seismic time data.

4. The tomographic modeling method based on 3D convolutional neural network according to claim 1, characterized in that, The feature concatenation includes concatenating the three-dimensional amplitude matrix of seismic data and the three-dimensional physical property parameters of logging data in the channel dimension, where the three-dimensional physical property parameter volume of logging data is generated from discrete well point data by Kriging interpolation to form a four-dimensional input tensor; the four-dimensional input tensor is input∈X×Y×Z×(C SEISMIC +S logging ), where C SEISMIC is the number of seismic data channels, S logging is the physical property parameter of logging data, and the four-dimensional input tensor integrates the spatial distribution information of seismic data and the physical property details of logging data.

5. The tomographic modeling method based on a 3D convolutional neural network according to claim 1, wherein, The normalization includes performing Z-Score standardization on the seismic data and logging data respectively to ensure that the scales of all data sources are consistent. The normalization formula is: where x norm is the normalized data, x is the original data, μ is the mean of the original data, and σ is the standard deviation of the original data; The weighted fusion includes using an adaptive weight to perform weighted fusion on the normalized seismic data and logging data; the adaptive weight takes α = 0.6, and the weighted fusion calculation formula is: Faused = α·Seismic+(1-α)·Logging where Faused is the fused data, α is the adaptive weight, Seismic is the seismic data, and Logging is the logging data.

6. The tomographic modeling method based on a 3D convolutional neural network according to claim 1, characterized in that The specific steps for constructing the 3D convolutional neural network model include: Step S31. The convolutional layer extracts the fault spatial features by sliding the three-dimensional convolutional kernel in the x, y, and z directions to extract local features such as faults and lithology changes, and introduces non-linearity through the ReLU activation function to form a higher-level feature map, which contains information such as fault strike and dip; Step S32. Use max pooling to reduce the dimension of the convolutional feature map, reduce the spatial dimension, reduce the computational complexity, and retain significant geological features such as the position of the fault plane; Step S33. Flatten the features after multiple layers of convolution and pooling, integrate them through the fully connected layer, and the output layer predicts the position, strike, and dip of the fault plane to generate a three-dimensional fault model.

7. The tomographic modeling method based on 3D convolutional neural network according to claim 1, wherein, The cross-entropy loss function is used to measure the difference between the model's predicted values and the true labels. Let the predicted tomogram probability value output by the 3D convolutional neural network model be y ′ , and the true label be y. Then the cross-entropy loss function L is defined as: Among them, L is the cross-entropy loss value, N is the number of samples, y i represents the true label of the i-th sample, y i ′ is the predicted probability value of the i-th sample.

8. A tomographic modeling system based on a 3D convolutional neural network, characterized in that, Applied to the fault modeling method based on a 3D convolutional neural network according to claims 1-7, the system includes: A data acquisition module, configured to acquire seismic data and logging data, and preprocess the seismic data and the logging data respectively; A data fusion module, configured to fuse the seismic data and the logging data, and the data fusion includes spatial alignment, feature splicing, normalization and weighted fusion; A modeling module, configured to construct a 3D convolutional neural network model, the 3D convolutional neural network model includes a convolutional layer, a pooling layer and a fully-connected layer, extracting fault spatial features through the convolutional layer, reducing the dimension through the pooling layer, and integrating and outputting through the fully-connected layer; A model training and optimization module, configured to perform cross-validation on the 3D convolutional neural network model by using a cross-entropy loss function, and further optimize the parameters of the 3D convolutional neural network model; A prediction output module, configured to perform fault prediction on the input data through the 3D convolutional neural network model, and output a three-dimensional fault model, predicted fault coordinates and geometric shapes; A result optimization and display module, configured to perform smoothing optimization on the three-dimensional fault model by using Gaussian filtering to improve the boundary clarity, and display the fault model through a three-dimensional visualization tool.

9. A network-side server, characterized in that, Comprising: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fault modeling method based on a 3D convolutional neural network according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the fault modeling method based on a 3D convolutional neural network according to any one of claims 1 to 7.

Images