A modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network

Abstract

The application discloses a modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network, and on the basis of comprehensively processing drilling data and geophysical profile data in a modeling area, the drilling data and the geophysical profile data are sampled to a unified grid size by using a data upsampling method, and known grids are assigned to the modeling area; then, the adjusted U-Net3D network is used to perform feature extraction by downsampling and feature completion by upsampling on the modeling area containing part of the known data, and finally, a three-dimensional geological model is constructed from the multi-source data known in part. The application can effectively realize the organization of multi-source geological data, realize the construction of a three-dimensional geological model from multi-source data known in part, and has the advantages of simple realization, remarkable effect and application requirement.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional geological modeling technology, and in particular relates to a modeling method based on a pre-trained U-Net3D network to process multi-source geological data and construct a three-dimensional geological model. Background Technology

[0002] Understanding the strike of underground strata, the distribution of ore bodies, and the extent of structural extension helps geologists and other construction personnel effectively design surface construction plans and determine underground construction directions. The creation of accurate three-dimensional geological models allows geologists to more intuitively understand the distribution of underground strata, ore bodies, and structures.

[0003] Existing methods for constructing 3D geological models mainly include explicit modeling and implicit modeling. Explicit modeling includes automatic modeling methods based on generalized triangular prism elements and free-form curve surface modeling techniques. Existing implicit modeling methods mainly include radial basis function (RBF), Hermite RBF (HRBF), moving least squares (MLS), anisotropic RBF (ARBF), and generalized RBF (GRBF). Explicit modeling directly constructs a 3D geological model based on borehole or profile data through manual interaction; implicit modeling extracts boundary information from borehole or profile data using implicit functions, calculates the structural surfaces of strata and rock masses, and constructs a 3D geological model. However, existing modeling methods encounter various problems when used to construct 3D geological models. First, explicit modeling methods require a lot of manual operation and experience, which makes them very time-consuming and labor-intensive when processing large amounts of data. Second, implicit 3D modeling methods face a series of problems when dealing with multi-source heterogeneous deep exploration data in underground engineering areas, such as difficulty in coordinating the interpretation of structural boundaries, time-consuming and labor-intensive geometric modeling, difficulty in updating models, and lack of uncertainty analysis and accuracy assessment. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned in the background art and to propose a method for processing multi-source data and constructing three-dimensional geological models based on a pre-trained U-Net3D network. This method enables the organization and comprehensive analysis of borehole data and geophysical profile data, and ultimately achieves rapid analysis and automatic modeling of multi-source data.

[0005] To achieve the objectives of this invention, a modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network is disclosed, comprising the following steps:

[0006] Step 1: Extract borehole data and geophysical profile data; perform data upsampling on multi-source data by calculating the minimum cell size;

[0007] Step 2: Assign the known multi-source data to the corresponding regions of the modeling area raster;

[0008] Step 3: Construct the U-Net3D deep learning network;

[0009] Step 4: Use the U-Net3D pre-trained deep learning network to invert the modeling area containing known data, and realize the construction of a three-dimensional geological model from multiple sources.

[0010] Furthermore, in step 1, due to the differences in storage format and representation range between borehole data and geophysical data, it is necessary to upsample the multi-source data. The specific steps are as follows:

[0011] Step 1-1: Determine the data types of borehole data and geophysical data;

[0012] Steps 1-2: Determine the representation range of different data, and determine the minimum raster size based on the resolution of different data;

[0013] Steps 1-3: Determine a uniform depth and perform data upsampling.

[0014] Furthermore, in step 2, the known multi-source data is assigned to the corresponding region of the modeling region raster; based on the calculated unit raster size, the modeling region is divided into N raster of the same size, and the known multi-source data is assigned to the corresponding raster; the unknown region raster only contains coordinate information (X, Y, Z), and does not contain attribute information or lithological information.

[0015] Furthermore, in step 3, the U-Net3D network includes 4 downsampling operations, 4 upsampling operations, and one 3D horizontal convolution.

[0016] Furthermore, in step 4, the U-Net3D pre-trained deep learning network is used to invert the modeling area containing known data, thereby realizing the construction of a three-dimensional geological model from multiple sources.

[0017] Furthermore, in step 1-1, the header information of the borehole data includes lithology name, X, Y, depth from, and depth to; the header information of geophysical data varies depending on the storage format; geophysical data formats include SEG-Y, LAS, NetCDF, GRD, HDF5, XYZ, and Matlab.mat.

[0018] Further, in steps 1-2, the coverage area of ​​the data is determined. For geophysical data, its coverage area is determined and denoted as [d]. profile min ,d profile max For borehole data, determine its coverage area, denoted as [d]. drill min ,d drill maxSelect a target raster size that will be used to evenly divide the geophysical and borehole data. The target raster size is determined based on the smallest cell size in the geophysical and borehole datasets.

[0019] Furthermore, in steps 1-3, the depth range [d] is unified. min ,d max It needs to cover the entire range of geophysical data and borehole data, that is...

[0020] d min =min(d profile min ,d drill min )

[0021] d max =maxx(d profile max d drill max )

[0022] The entire depth range [d min ,d max ][d min ,d max Divide the grid into sections based on the target grid size Δd, and calculate the grid size S. grids

[0023]

[0024] Data upsampling involves redistributing all data according to the calculated raster size, upsampling a single data point of known length into multiple raster cells of the same size.

[0025] Furthermore, step 3 specifically includes the following steps:

[0026] Step 3-1: The length, width, and height of the region to be modeled are X_Length, Y_Length, and Z_Length. During training, the modeling region containing some known grids is input into the U-Net3D network and undergoes four downsampling operations. The specific downsampling parameters are set as follows:

[0027] The first downsampling involves a 4x4 3D convolutional layer with a stride of 2, padding of 1, and leakyReLU activation. After the first downsampling, the dimensions of the region to be modeled become half of the initial modeling region, and the number of channels becomes 32.

[0028] The second downsampling involves a 3x3 convolutional layer with a stride of 1, padding of 1, and leakyReLU activation. After the second downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels is increased to 64.

[0029] The third downsampling involves a 3x3 convolutional layer with a stride of 1, padding of 1, and leakyReLU activation. After the third downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels is increased to 128.

[0030] The fourth downsampling step involves a 3x3 convolutional layer with a stride of 1, padding of 1, and LeakyReLu activation. After the fourth downsampling step, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels increases to 256.

[0031] Step 3-2: Further feature extraction operations are performed on the modeling region after four downsampling operations. The horizontal feature extraction operation includes a 4*4 three-dimensional convolutional layer with a stride of 2, padding of 1, and a leakyReLU activation function. After horizontal convolution, the length, width, and height of the region to be modeled are reduced to 1 / 4 of the initial modeling region, and the number of channels is increased to 256.

[0032] Step 3-3: After performing four downsampling and horizontal convolution operations, the extracted 3D model feature values ​​need to be upsampled. Upsampling is also performed four times, and the specific upsampling parameters are set as follows:

[0033] The first upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the first upsampling, the dimensions of the region to be modeled are increased to half of the initial modeling region, and the number of channels decreases to 128. In addition, the data from the first upsampling and the third downsampling are skipped to ensure model accuracy.

[0034] The second upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the second upsampling, the modeling region retains half the dimensions of the initial modeling region, while the number of channels decreases to 64. The data from the second upsampling and the second downsampling are then connected in a skip connection to ensure model accuracy.

[0035] The third upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the third upsampling, the modeling region retains half the dimensions of the initial modeling region, and the number of channels decreases to 32. The third upsampling is followed by a skip connection with the data after the first downsampling to ensure model accuracy.

[0036] The fourth upsampling step involves a 4x4 3D convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the fourth upsampling step, the dimensions of the region to be modeled are consistent with those of the initial modeling region, and the prediction results for each cell can be output.

[0037] Further, in step 4, the pre-trained model of the 3D geological model is inverted based on U-Net3D, and then the area to be modeled is inverted by loading the pre-trained model, including the following steps:

[0038] Step 4-1: Manually generate a large number of three-dimensional geological models with random occurrences, random tectonic events, and random lithology for training purposes;

[0039] Step 4-2: Divide the generated 3D geological models of different events into training set, test set and validation set according to the classification method of 8:1:1, and input the training set into the U-Net3D deep learning network mentioned in step 3 for training; use precision, recall and F1 score to evaluate the modeling effect on the test set and validation set.

[0040] Step 4-3: Save the pre-trained model whose precision, recall, and F1 score on both the test and validation sets are greater than 90%.

[0041] Step 4-4: Load the pre-trained model to perform inversion modeling on the region to be modeled, which contains known multi-source data.

[0042] Compared with existing technologies, the significant advancements of this invention lie in its ability to effectively organize multi-source geological data and construct a three-dimensional geological model. This is manifested in two aspects: first, it utilizes data upsampling methods to unify borehole data and geophysical profile data into the same raster size and assign them to the modeling area; second, it employs an adjusted U-Net3D network to perform downsampling feature extraction and upsampling feature completion on the modeling area containing partially known data, ultimately achieving the construction of a three-dimensional geological model from partially known multi-source data. This invention is simple to implement, highly effective, and meets application requirements.

[0043] To more clearly illustrate the functional characteristics and structural parameters of the present invention, further explanation is provided below in conjunction with the accompanying drawings and specific embodiments. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0045] Figure 1 This is a schematic diagram of the U-Net3D deep learning network structure used for multi-source data inversion;

[0046] Figure 2 This is a flowchart of constructing a three-dimensional geological model based on U-Net3D multi-source data. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] This invention, based on the comprehensive processing of borehole data and geophysical profile data in the modeling area, utilizes a data upsampling method to sample the borehole data and geophysical profile data to a uniform raster size, and assigns the known raster values ​​to the area to be modeled. Then, using an adjusted U-Net3D network, it performs downsampling feature extraction and upsampling feature completion on the modeling area containing partially known data, ultimately achieving the construction of a 3D geological model from partially known multi-source data. This invention is simple to implement, has significant effects, and meets application requirements.

[0049] like Figure 2 As shown, a modeling method for constructing a 3D geological model based on pre-trained U-Net3D network using multi-source data can effectively extract and process borehole data and geophysical profile data, realizing the construction of a 3D geological model based on multi-source data. The method specifically includes the following steps:

[0050] Step 1: Extract borehole data and geophysical profile data; perform data upsampling on multi-source data by calculating the minimum cell size.

[0051] Because borehole data and geophysical data differ in their storage formats and representation ranges, it is necessary to upsample the multi-source data. The specific steps are as follows:

[0052] Step 1-1: Determine the data types of borehole data and geophysical data.

[0053] The header information of the borehole data includes: lithology name, X, Y, depth from, and depth to.

[0054] Geophysical data header information varies depending on the storage format. Common geophysical data formats include: SEG-Y, LAS, NetCDF, GRD, HDF5, XYZ, and Matlab.mat, etc.

[0055] Step 1-2: Determine the representation range of different data, and determine the minimum raster size based on the resolution of different data.

[0056] 1. Determine the coverage area of ​​the data.

[0057] For geophysical data, its coverage area is determined and denoted as [d]. profile min ,d profile max ].

[0058] For borehole data, determine its coverage area, denoted as [d]. drill min ,d drill max ].

[0059] 2. Select a target raster size Δd, which will be used to uniformly divide the geophysical data and borehole data. This can be determined based on the smallest cell size in both datasets.

[0060] Steps 1-3: Determine a uniform depth and perform data upsampling.

[0061] 1. Determine a uniform depth range

[0062] Unified depth range [d min ,d max It needs to cover the entire range of geophysical data and borehole data, namely:

[0063] d min =min(d profile min d drill min )

[0064] d max =max(d profile max ,d drill max )

[0065] 2. The entire depth range [d] min ,d max ][d min ,d max Divide the grid into sections based on the target grid size Δd, and calculate the grid size S. grids :

[0066]

[0067] 3. Data upsampling:

[0068] All data are redistributed according to the calculated raster size, upsampling a single data point of known length into multiple raster cells of the same size.

[0069] Step 2: Assign the known multi-source data to the corresponding area of ​​the modeling area raster.

[0070] Based on the calculated unit raster size, the modeling area is divided into N raster cells of uniform size, and known multi-source data are assigned to the corresponding raster cells. The raster cells for unknown areas only contain coordinate information (X, Y, Z) and do not contain attribute or lithological information.

[0071] Step 3: Construct the U-Net3D deep learning network;

[0072] like Figure 1 As shown, the U-Net3D network designed in this invention includes 4 downsampling operations, 4 upsampling operations, and one 3D horizontal convolution.

[0073] Step 3-1: The dimensions of the region to be modeled are X_Length, Y_Length, and Z_Length. During training, the modeling region, which includes some known grid cells, is input into the U-Net3D network and undergoes four downsampling operations. The specific downsampling parameters are set as follows:

[0074] First downsampling: The parameters for the first downsampling include a 4x4 3D convolutional layer with a stride of 2, padding of 1, and leakyReLU activation function. After the first downsampling, the dimensions of the region to be modeled become half of the initial modeling region, and the number of channels becomes 32.

[0075] Second downsampling: The parameters for the second downsampling include a 3x3 three-dimensional convolutional layer with a stride of 1, padding of 1, and leakyReLU activation function. After the second downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels increases to 64.

[0076] Third downsampling: The parameters for the third downsampling include a 3x3 three-dimensional convolutional layer with a stride of 1, padding of 1, and leakyReLU activation function. After the third downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels increases to 128.

[0077] Fourth downsampling: The parameters for the fourth downsampling include a 3x3 convolutional layer with a stride of 1, padding of 1, and leakyReLU activation. After the fourth downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels increases to 256.

[0078] Step 3-2: Further feature extraction is performed on the modeling region after four downsampling operations. The horizontal feature extraction operation includes a 4*4 3D convolutional layer with a stride of 2, padding of 1, and leakyReLU activation function. After horizontal convolution, the length, width, and height of the region to be modeled are reduced to 1 / 4 of the initial modeling region, and the number of channels is increased to 256.

[0079] Step 3-3: After performing four downsampling and horizontal convolution operations, the feature values ​​of the extracted 3D model need to be upsampled. Upsampling is also performed four times, and the specific upsampling parameters are set as follows:

[0080] First upsampling: The parameters for the first upsampling include a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the first upsampling, the dimensions of the region to be modeled are increased to half of the initial modeling region, and the number of channels decreases to 128. Furthermore, skip connections are made between the data from the first upsampling and the data from the third downsampling to ensure model accuracy.

[0081] Second upsampling: The parameters for the second upsampling include a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the second upsampling, the dimensions of the region to be modeled remain half of the initial modeling region, while the number of channels decreases to 64. Skip connections are made between the data from the second upsampling and the data from the second downsampling to ensure model accuracy.

[0082] Third upsampling: The parameters for the third upsampling include a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the third upsampling, the dimensions of the region to be modeled remain half of the initial modeling region, and the number of channels decreases to 32. The data from the third upsampling is then connected to the data from the first downsampling to ensure model accuracy.

[0083] Fourth upsampling: The parameters for the fourth upsampling include a 4x4 3D convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the fourth upsampling, the dimensions of the region to be modeled are consistent with the initial modeling region, and the prediction result for each cell can be output.

[0084] Step 4: Use the U-Net3D pre-trained deep learning network to invert the modeling area containing known data, and realize the construction of a three-dimensional geological model from multiple sources.

[0085] Untrained deep learning networks cannot predict or classify events in unknown or blank areas. Therefore, this invention designs a pre-trained model based on U-Net3D to invert a 3D geological model, and then inverts the region to be modeled by loading the pre-trained model.

[0086] Step 4-1: Manually generate a large number of three-dimensional geological models with random occurrences, random tectonic events, and random lithology for training.

[0087] Step 4-2: Divide the generated 3D geological models of different events into training set, test set and validation set according to the classification method of 8:1:1, and input the training set into the U-Net3D deep learning network mentioned in Step 3 for training; use precision, recall and F1 score to evaluate the modeling effect on the test set and validation set.

[0088] Step 4-3: Save the pre-trained model whose precision, recall, and F1 score on both the test and validation sets are greater than 90%.

[0089] Step 4-4: Load the pre-trained model to perform inversion modeling on the region to be modeled, which contains known multi-source data.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network, characterized in that, Includes the following steps: Step 1: Extract borehole data and geophysical profile data; perform data upsampling on multi-source data by calculating the minimum cell size; Step 2: Assign the known multi-source data to the corresponding regions of the modeling area raster; Step 3: Construct the U-Net3D deep learning network; Step 4: Use the U-Net3D pre-trained deep learning network to invert the modeling area containing known data to realize the construction of a three-dimensional geological model from multiple source data; In step 1, because the storage formats and representation ranges of borehole data and geophysical data differ, it is necessary to upsample the multi-source data. The specific steps are as follows: Step 1-1: Determine the data types of borehole data and geophysical data; Steps 1-2: Determine the representation range of different data, and determine the minimum raster size based on the resolution of different data; Steps 1-3: Determine a uniform depth and perform data upsampling; In step 2, the known multi-source data is assigned to the corresponding region of the modeling area raster; based on the calculated unit raster size, the modeling area is divided into N raster of the same size, and the known multi-source data is assigned to the corresponding raster; the unknown area raster only contains coordinate information (X, Y, Z), and does not contain attribute information or lithology information; In step 3, the U-Net3D network includes 4 downsampling operations, 4 upsampling operations, and one 3D horizontal convolution.

2. The modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network according to claim 1, characterized in that, In step 1-1, the header information of the borehole data includes lithology name, X, Y, depth from, and depth to; the header information of geophysical data varies depending on the storage format; geophysical data formats include SEG-Y, LAS, NetCDF, GRD, HDF5, XYZ, and Matlab .mat.

3. The modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network according to claim 1, characterized in that, In steps 1-2, the coverage area of ​​the data is determined. For geophysical data, the coverage area is determined and denoted as... For borehole data, determine its coverage area, denoted as . Select a target raster size that will be used to evenly divide the geophysical and borehole data. The target raster size is determined based on the smallest cell size in the geophysical and borehole datasets.

4. The modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network according to claim 1, characterized in that, In steps 1-3, a unified depth range is established. It needs to cover the entire range of geophysical data and borehole data, that is The entire depth range target grid size Divide the grid and calculate the grid size. Data upsampling involves redistributing all data according to the calculated raster size, upsampling a single data point of known length into multiple raster cells of the same size.

5. The modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network according to claim 1, characterized in that, Step 3 specifically includes the following steps: Step 3-1: The length, width, and height of the region to be modeled are X_Length, Y_Length, and Z_Length. During training, the modeling region containing some known grids is input into the U-Net3D network and undergoes four downsampling operations. The specific downsampling parameters are set as follows: The first downsampling involves a 4x4 3D convolutional layer with a stride of 2, padding of 1, and leakyReLU activation. After the first downsampling, the dimensions of the region to be modeled become half of the initial modeling region, and the number of channels becomes 32. The second downsampling involves a 3x3 convolutional layer with a stride of 1, padding of 1, and leakyReLU activation. After the second downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels is increased to 64. The third downsampling involves a 3x3 convolutional layer with a stride of 1, padding of 1, and leakyReLU activation. After the third downsampling, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels is increased to 128. The fourth downsampling step involves a 3x3 convolutional layer with a stride of 1, padding of 1, and LeakyReLu activation. After the fourth downsampling step, the dimensions of the region to be modeled remain half of the initial modeling region, but the number of channels increases to 256. Step 3-2: Further feature extraction operations are performed on the modeling region after four downsampling operations. The horizontal feature extraction operation includes a 4*4 three-dimensional convolutional layer with a stride of 2, padding of 1, and a leakyReLU activation function. After horizontal convolution, the length, width, and height of the region to be modeled are reduced to 1 / 4 of the initial modeling region, and the number of channels is increased to 256. Step 3-3: After performing four downsampling and horizontal convolution operations, the extracted 3D model feature values ​​need to be upsampled. Upsampling is also performed four times, and the specific upsampling parameters are set as follows: The first upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the first upsampling, the dimensions of the region to be modeled are increased to half of the initial modeling region, and the number of channels decreases to 128. In addition, the data from the first upsampling and the third downsampling are skipped to ensure model accuracy. The second upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the second upsampling, the modeling region retains half the dimensions of the initial modeling region, while the number of channels decreases to 64. The data from the second upsampling and the second downsampling are then connected in a skip connection to ensure model accuracy. The third upsampling involves a 3x3 convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the third upsampling, the modeling region retains half the dimensions of the initial modeling region, and the number of channels decreases to 32. The third upsampling is followed by a skip connection with the data after the first downsampling to ensure model accuracy. The fourth upsampling step involves a 4x4 3D convolutional layer with a stride of 2, padding of 1, and ReLU activation. After the fourth upsampling step, the dimensions of the region to be modeled are consistent with those of the initial modeling region, and the prediction results for each cell are output.

6. The modeling method for processing multi-source geological data and constructing a three-dimensional geological model based on a pre-trained U-Net3D network according to claim 1, characterized in that, In step 4, the pre-trained model of the 3D geological model is inverted based on U-Net3D, and then the area to be modeled is inverted by loading the pre-trained model, including the following steps: Step 4-1: Manually generate a large number of three-dimensional geological models with random occurrences, random tectonic events, and random lithology for training purposes; Step 4-2: Divide the generated 3D geological models of different events into training set, test set and validation set according to the classification method of 8:1:1, and input the training set into the U-Net3D deep learning network mentioned in step 3 for training; use precision, recall and F1 score to evaluate the modeling effect on the test set and validation set. Step 4-3: Save the pre-trained model whose precision, recall, and F1 score on both the test and validation sets are greater than 90%. Step 4-4: Load the pre-trained model to perform inversion modeling on the region to be modeled, which contains known multi-source data.

Images