Three-dimensional geological structure data modeling method and device and medium

Abstract

The invention discloses a three-dimensional geologic structure data modeling method and device and a medium. The method comprises the steps of obtaining multi-scale three-dimensional geologic structure data of the same area; the multiple scales are a plurality of different resolution acquisition conditions; preprocessing the three-dimensional geological structure data to obtain target geological structure data; performing multi-scale feature fusion coding on the target geological structure data through a pre-constructed multi-scale feature coding model to obtain a feature coding vector; and performing feature decoding on the feature coding vector through a pre-constructed multi-scale feature decoding model to generate a three-dimensional geologic model. Therefore, on the basis of the multi-scale feature coding model and the multi-scale feature decoding model, modeling can be directly carried out on the three-dimensional geological structure data, and the modeling precision is improved. Besides, the three-dimensional geological data of different scales in the same area are acquired for cross-resolution feature coding, so that information between different resolutions is effectively captured, the feature extraction depth is enhanced, and the modeling precision is further improved.

Description

Technical Field

[0001] The present application relates to the field of earth science technology, and in particular to a three-dimensional geological structure data modeling method, device and medium. Background Art

[0002] 3D geological structure modeling refers to the digital three-dimensional representation of underground geological bodies, which is used to gain a deeper understanding of the geological environment. With the continuous development of the field of earth science, 3D geological structure modeling has become a key tool in various fields such as resource exploration, environmental protection, disaster prevention, and urban planning.

[0003] Traditional three-dimensional geological structure modeling is mainly based on geologists' understanding of geological phenomena and the analysis of geological data. This method requires a lot of time and manpower costs, is inefficient, and the modeling accuracy is easily limited by the geologist's personal experience and knowledge level.

[0004] With the rapid development of deep learning technology, rapid modeling can be achieved through deep learning technology. Specifically, during the modeling process, three-dimensional geological structure data of the same scale is first converted into two-dimensional structure data, and then deep learning technology is used to perform geological structure modeling on the two-dimensional structure data. This method can only model data of the same scale, and the data uniformity leads to reduced modeling accuracy. In addition, in the process of converting three-dimensional geological structures into two-dimensional geological structures, three-dimensional structure, spatial information, and other information may be lost, further reducing modeling accuracy, thereby affecting the work efficiency, work accuracy, and application scope of downstream tasks such as earthquake identification, fault identification, and resource exploration.

[0005] Therefore, how to improve the accuracy of three-dimensional geological structure modeling, thereby improving the work efficiency, accuracy and application scope of downstream tasks, is an urgent problem to be solved by technical personnel in this field. Summary of the Invention

[0006] In view of this, one aspect of the present application provides a three-dimensional geological structure data modeling method, the method comprising:

[0007] Acquire multi-scale 3D geological structure data of the same area; the multi-scale refers to acquisition conditions with multiple different resolutions;

[0008] Preprocessing the three-dimensional geological structure data to obtain target geological structure data;

[0009] Performing multi-scale feature fusion coding on the target geological structure data through a pre-built multi-scale feature coding model to obtain a feature coding vector;

[0010] The feature coding vector is feature decoded by a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

[0011] Optionally, preprocessing the three-dimensional geological structure data to obtain target geological structure data includes:

[0012] Converting the three-dimensional geological structure data into a specified format to obtain structure data;

[0013] According to the principle of minimization of cropping, blank data at the exploration edge of the structural data in the specified format are cropped to obtain cropped structural data;

[0014] The average value and standard deviation of the trimmed structural data are calculated; and the trimmed structural data are standardized according to the average value and the standard deviation to obtain the target geological structural data.

[0015] Optionally, the multi-scale feature fusion encoding of the target geological structure data is performed using a pre-built multi-scale feature encoding model to obtain a feature encoding vector, including:

[0016] Segmenting the target geological structure data of different scales respectively to obtain a plurality of data segmentation sets corresponding to the different scales;

[0017] Performing cross-resolution encoding on the data blocks in the data block set to obtain fused encoding;

[0018] Performing random masking on the fusion code to obtain a masked data code;

[0019] The masked data is encoded using a preset number of Transformer encoders to obtain the feature encoding vector.

[0020] Optionally, performing cross-resolution encoding on the data blocks in the data block set to obtain fused encoding includes:

[0021] By using a convolutional neural network, encoding the data blocks in each of the data block sets with preset dimensions is performed to obtain block codes;

[0022] Establishing a three-dimensional coordinate system based on the target geological structure data of the largest scale among the multiple scales;

[0023] Based on the three-dimensional coordinate system, generating first three-dimensional position coordinates of the largest scale data slice and second three-dimensional position coordinates of other scale data slices;

[0024] Encoding the data blocks in the data block set according to the preset dimension, the first three-dimensional position coordinates, and the second three-dimensional position coordinates to obtain a position code;

[0025] The fusion code is determined according to the block code and the position code.

[0026] Optionally, determining the fusion code according to the block code and the position code includes:

[0027] Summing the block code and the position code of the same scale to obtain a target code;

[0028] The target codes are spliced together to obtain the fused code.

[0029] Optionally, encoding the masked data using a preset number of Transformer encoders to obtain the feature encoding vector includes:

[0030] Normalizing the masked data code to obtain a normalized code;

[0031] Generating a query matrix, a key matrix, and a value matrix corresponding to the normalized code through linear transformation;

[0032] Calculating a multi-head attention score based on the query matrix, the key matrix, and the value matrix;

[0033] The multi-head attention score is calculated through a feedforward neural network to obtain the feature encoding vector.

[0034] Optionally, the performing feature decoding on the feature coding vector using a pre-built multi-scale feature decoding model includes:

[0035] Performing feature mapping and feature restoration processing on the feature coding vector respectively to determine a decoding dimension and an input feature dimension;

[0036] Splitting the feature coding vector into a plurality of feature coding sub-vectors corresponding to the multiple scales under the conditions of the decoding dimension and the input feature dimension;

[0037] Calculating cross-attention scores of the plurality of feature encoding sub-vectors based on a cross-attention mechanism;

[0038] Processing the cross attention scores through a preset number of Transformer decoders to generate the three-dimensional geological model;

[0039] Perform data prediction for a specified task on the three-dimensional geological model to obtain a prediction result.

[0040] Another aspect of the present application provides a three-dimensional geological structure data modeling device, the device comprising:

[0041] An acquisition module is used to acquire multi-scale three-dimensional geological structure data of the same area; the multi-scale refers to multiple different resolutions;

[0042] A preprocessing module, used for preprocessing the three-dimensional geological structure data to obtain target geological structure data;

[0043] A feature coding model is used to perform multi-scale feature fusion coding on the target geological structure data using a pre-built multi-scale feature coding model to obtain a feature coding vector;

[0044] The feature decoding module is used to perform feature decoding on the feature coding vector through a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

[0045] Another aspect of the present application provides a three-dimensional geological structure data modeling device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the steps of the three-dimensional geological structure data modeling method are implemented.

[0046] Another aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the three-dimensional geological structure data modeling method when the program is executed by a processor.

[0047] The present application provides a method, device, and medium for modeling 3D geological structure data. The beneficial effects of this method include: directly modeling 3D geological structure data based on a multi-scale feature encoding model and a multi-scale feature decoding model, avoiding the loss of spatial information and other information caused by converting 3D geological structure data into 2D data, thereby improving modeling accuracy. Furthermore, cross-resolution feature encoding is performed on 3D geological data of different scales acquired within the same region, effectively capturing and correlating information across different resolutions. This enhances the depth and breadth of 3D geological data feature extraction, further improving modeling accuracy and thus ensuring the efficiency, accuracy, and application scope of downstream tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of a flow chart of a three-dimensional geological structure data modeling method provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of the structure of a three-dimensional geological structure data modeling system provided in an embodiment of the present application;

[0050] Figure 3 This is an example flow chart of a three-dimensional geological structure data modeling method provided in another embodiment of the present application;

[0051] Figure 4 A schematic flow chart of a three-dimensional geological structure data modeling method provided in another embodiment of the present application;

[0052] Figure 5 A schematic diagram of the structure of a three-dimensional geological structure data modeling device provided in an embodiment of the present application;

[0053] Figure 6 This is a structural schematic diagram of a three-dimensional geological structure data modeling device provided in another embodiment of the present application.

[0054] The reference numerals are as follows: 60 is a memory, 61 is a processor, 62 is a display screen, 63 is an input and output interface, 64 is a communication interface, 65 is a power supply, 66 is a communication bus, 601 is a computer program, 602 is an operating system, and 603 is data. DETAILED DESCRIPTION

[0055] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0056] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0057] Figure 1 A flow chart of a three-dimensional geological structure data modeling method provided in an embodiment of the present application is shown as follows: Figure 1 As shown, the method includes:

[0058] S10: Acquire multi-scale 3D geological structure data of the same area; multi-scale refers to acquisition conditions with multiple different resolutions;

[0059] It's understandable that 3D geological structure data of varying resolutions can provide geological information at varying scales. High-resolution data can capture more detailed geological features, such as small-scale faults and dykes, while low-resolution data can provide more macroscopic geological structure information.

[0060] Furthermore, data of varying resolutions can provide a multifaceted understanding of complex geological phenomena. For example, when analyzing complex faults, high-resolution data can clearly demonstrate the fault's strike, dip, and distribution of rock layers within the fault zone, while low-resolution data can help understand the fault's impact on a larger scale.

[0061] Therefore, in an optional embodiment, in order to improve the modeling accuracy and reliability of the three-dimensional geological model, three-dimensional geological structure data of multiple different resolution conditions are obtained for the same area, that is, multi-scale three-dimensional geological structure data are obtained.

[0062] It should be noted that the three-dimensional geological structure data can be acquired through geological exploration technology or geophysical exploration. This application does not limit the method of acquiring the three-dimensional geological structure data. In addition, it should be noted that this application does not limit the number of multi-scales acquired.

[0063] S11: preprocessing the three-dimensional geological structure data to obtain target geological structure data;

[0064] After obtaining the initially collected 3D geological structure data, in order to improve the data quality of 3D geological modeling and ensure that the characteristic information in the geological structure data can be quickly captured during modeling, in a specific embodiment, the collected 3D geological structure data is preprocessed to obtain target geological structure data.

[0065] It should be noted that the initially collected three-dimensional geological structure data includes multiple scales. After preprocessing, the number of geological data scales does not change, that is, the generated target geological structure data still includes multiple resolutions.

[0066] In addition, it should be noted that preprocessing may include but is not limited to noise removal, filling in missing values, standardization, and format conversion.

[0067] Figure 2 This is a schematic diagram of a three-dimensional geological structure data modeling system provided in an embodiment of the present application. In an optional embodiment, the three-dimensional geological structure data modeling method provided in this application can be applied to Figure 2 The modeled system shown.

[0068] like Figure 2 As shown, the 3D geological structure data modeling system provided in the embodiments of the present application includes a data preprocessing unit, a multi-scale encoding unit, and a multi-scale decoding unit. In a specific embodiment, the collected multi-scale 3D geological structure data is used as input to the system, that is, as input to the data preprocessing unit, and preprocessed to obtain target geological structure data.

[0069] S12: Perform multi-scale feature fusion encoding on the target geological structure data through a pre-built multi-scale feature encoding model to obtain a feature encoding vector;

[0070] Further, such as Figure 2 As shown in Figure 1, the data preprocessing unit transmits the output target geological structure data to the multi-scale encoding unit for encoding. Specifically, the pre-built multi-scale feature encoding model is used to perform cross-resolution multi-scale feature fusion encoding on the input target geological structure, generating a feature encoding vector. It should be noted that the parameters of the multi-scale encoding unit are only updated during pre-training and are frozen during the inference phase.

[0071] In a specific embodiment, the multi-scale feature encoding model can capture the characteristics of target geological structure data at different scales (i.e., different resolutions), capturing multi-level information in the data from coarse to fine granularity, thereby performing fusion encoding of multi-scale features to obtain a multi-scale feature encoding vector. This helps enhance the feature extraction capabilities of complex geological structures.

[0072] S13: The feature encoding vector is feature decoded by a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

[0073] like Figure 2 As shown in Figure 1, the multi-scale encoding unit uses the output feature encoding vector as input to the multi-scale decoding unit to decode the feature decoding vector. Specifically, the decoding process gradually restores the feature encoding vector to the original data or generates a new 3D geological model. The resulting 3D geological model is the output of the decoding process and can intuitively demonstrate the spatial distribution and characteristics of geological structures.

[0074] In a specific embodiment, the multi-scale feature decoding model is capable of processing feature information of different scales, that is, it can process feature information of different resolutions, ensuring that geological structure information from local to global is retained during the decoding process, so that the generated three-dimensional geological model not only has high-resolution details, but also reflects the overall geological structure.

[0075] Therefore, the 3D geological structure data modeling method provided in the embodiments of this application, based on a multi-scale feature encoding model and a multi-scale feature decoding model, directly models 3D geological structure data, avoiding the loss of spatial information and other information caused by converting 3D geological structure data into 2D data, thereby improving modeling accuracy. Furthermore, cross-resolution feature encoding is performed on 3D geological data of different scales acquired within the same region, effectively capturing and correlating information across different resolutions. This enhances the depth and breadth of 3D geological data feature extraction, further improving modeling accuracy and ensuring the efficiency, accuracy, and application scope of downstream tasks.

[0076] In an optional embodiment, preprocessing the three-dimensional geological structure data to obtain target geological structure data includes:

[0077] Convert the format of three-dimensional geological structure data to obtain structure data in a specified format;

[0078] According to the principle of minimizing clipping, the blank data at the exploration edge of the structural data in the specified format is clipped to obtain the clipped structural data;

[0079] The mean value and standard deviation of the cropped structural data are calculated; and the cropped structural data are standardized according to the mean value and standard deviation to obtain the target geological structure data.

[0080] like Figure 2 As shown, the data preprocessing unit includes a format conversion layer, a data clipping layer, and a standardization layer. In a specific embodiment, the original collected 3D geological structure data includes but is not limited to seismic data format (.segy), lidar data format (.las), and point cloud data format (.xyz). In order to facilitate subsequent data encoding and ensure data consistency, it can be processed by Figure 2 The format conversion layer shown converts the initially collected three-dimensional geological structure data into a specified format. In an optional embodiment, the specified format may be .npy format (a format used to store array data in the NumPy library).

[0081] It is understandable that in a specific geological exploration embodiment, the acquired three-dimensional geological structure data may contain blank data that does not contain useful geological information. In order to avoid wasting computing resources, these blank data can be cropped to reduce data redundancy and make subsequent processing more efficient.

[0082] Specifically, the process of clipping blank data minimizes the retention of information at key scales, ensuring that the clipped data still contains geological features from local to global scales. This facilitates multiscale analysis and supports more comprehensive geological modeling. Clipped data can be more efficient in multiscale analysis, enabling rapid generation of geological models at different scales, supporting geological research from macro to micro scales.

[0083] Therefore, by minimizing the number of blank data points at the edges, we can improve the quality and processing efficiency of 3D geological structural data. This process not only reduces data redundancy and improves model accuracy and reliability, but also enhances the model's generalization capabilities, improves data visualization, supports multi-scale analysis, and optimizes resource allocation.

[0084] Furthermore, the trimmed structural data is expressed as X0, and the average value X of the trimmed structural data is calculated. mean and standard deviation X std, and then the cropped structure data is standardized according to formula (1):

[0085]

[0086] Among them, X is the standardized target geological structure data, and X0 is the cropped structure data.

[0087] In a specific embodiment, standardization is performed to ensure that each multi-scale feature has the same weight in the model, so that each feature can participate in feature fusion more fairly, improve the effect of feature fusion, and enhance the expressiveness of the model.

[0088] Therefore, the data preprocessing unit uses the three-dimensional geological structure data obtained under different exploration conditions, converts it into a unified format and sizes it, and performs standardized operations to form a standardized data corpus, which is saved in a unified file format to ensure that the model can quickly obtain training corpus as needed.

[0089] Figure 3 This is a flow chart of a three-dimensional geological structure data modeling method provided in another embodiment of the present application. In an optional embodiment, as Figure 3 As shown in the figure, the target geological structure data is fused with multi-scale features through the pre-built multi-scale feature coding model to obtain a feature coding vector, including:

[0090] S30: Segmenting target geological structure data of different scales to obtain multiple data segmentation sets corresponding to different scales;

[0091] Based on the above embodiment, target geological structure data X of different scales is segmented separately. In an optional embodiment, the segmentation can be performed according to a preset fixed size. It is understood that in the specific embodiment, the target geological structure data X includes multiple different resolutions. However, for ease of understanding in this application, the following example will be described using the same region including two different resolutions, that is, including target geological structure data X at two scales.

[0092] Specifically, the target geological structure data X is divided into target geological structure data X1 of the first scale (H1, W1, D1) and target geological structure data X2 of the second scale (H2, W2, D2), that is, like Figure 2As shown, the multi-scale encoding unit includes a data segmentation layer. In a specific embodiment, the data segmentation layer segments the target geological structure data X1 and the target geological structure data X2 according to a fixed size s×s×s to obtain a first data segment P1 and a second data segment P2, where P1 = (H1 / / s) × (W1 / / s) × (D1 / / s), and P2 = (H2 / / s) × (W2 / / s) × (D2 / / s), where / / represents an integer division operation. It is understandable that the first data segment P1 and the second data segment P2 each include multiple segments, each corresponding to a different scale.

[0093] It should be noted that this application does not impose any restrictions on the fixed size of the segmentation, and it can be set according to actual business needs. It is understandable that if the actual business needs require more accurate 3D geological modeling and the computing power of the modeling system is high enough, a smaller granularity size can be selected for segmentation as much as possible. If the computing power of the modeling system is insufficient, or the modeling accuracy requirements are relatively low, a larger granularity size can be selected for segmentation.

[0094] S31: performing cross-resolution encoding on the data blocks in the data block set to obtain fused encoding;

[0095] Furthermore, cross-resolution encoding is performed on the data blocks after segmentation, that is, feature information of data blocks at different scales is fused to obtain fused code. For example, for the first data block P1 and the second data block P2 in the above example, cross-resolution feature information encoding is performed to obtain fused code X pe .

[0096] S32: Perform random masking on the fusion code to obtain a masked data code;

[0097] In an optional embodiment, in order to improve the performance of subsequent 3D geological modeling in downstream tasks, such as geological structure classification, resource prediction, etc., the model can more accurately identify and predict geological features. pe After that, the fusion code X pe Random masking is performed. By hiding some data, the model is forced to learn more general feature representations instead of relying on local details. This reduces the model's overfitting to specific data and improves its generalization ability in different geological environments.

[0098] Specifically, for the fusion code X pe Perform random masking according to the preset mask ratio r and record the masked data encoding At the same time, record the mask position like Figure 2As shown, the multi-scale coding unit includes a random mask layer, which mainly determines whether each block is masked by generating random numbers. The mask ratio r can be dynamically adjusted according to actual business needs.

[0099] S33: Encode the masked data using a preset number of Transformer encoders to obtain a feature encoding vector.

[0100] Get the masked data encoding X mask After that, the data is encoded by a preset number of Transformer encoders X mask Feature encoding is further performed to obtain a feature encoding vector. It should be noted that, in a specific embodiment, the preset number of Transformer encoders can be set according to actual business needs, and the preset number is greater than 2. In addition, it should be noted that, in a specific embodiment, the preset number of Transformer encoders selected can be the same encoder.

[0101] In an optional embodiment, when actual business needs require high accuracy in 3D geological modeling and the computing power available for modeling can meet these high-precision requirements, a larger number of Transformer encoders can be selected for feature encoding. Of course, in another optional embodiment, if the 3D geological modeling accuracy requirements are lower, or the computing power available for modeling is insufficient to support high-power modeling, a smaller number of Transformer encoders can be selected for feature encoding.

[0102] In specific embodiments, it is understood that the core of the Transformer encoder is the self-attention mechanism, which can simultaneously consider the relationship between all positions in the input sequence, regardless of their distance in the sequence. Therefore, in 3D geological modeling, geological features at different locations (for example, rock layer distribution, fault location, etc.) may have long-range correlations. The Transformer encoder can effectively capture these long-range dependencies, thereby more accurately understanding the overall characteristics of the geological structure.

[0103] Furthermore, in a specific embodiment, a preset number of Transformer encoders have no loop structure and can perform encoding calculations in parallel. This allows data from different locations to be processed simultaneously when processing multi-scale 3D geological structure data, improving training and inference efficiency and reducing computing time and resource consumption.

[0104] Based on the above embodiments, Figure 3 As shown, cross-resolution encoding is performed on the data blocks in the data block set to obtain fusion encoding, including:

[0105] S310: Encoding the data blocks in each data block set with a preset dimension using a convolutional neural network to obtain block codes;

[0106] In an optional embodiment, a three-dimensional convolutional neural network can be used to encode each data block, and the number of encoding channels can be set in advance, that is, the encoding dimension d is set in advance. In a specific embodiment, the encoding dimension d can be set according to actual business needs.

[0107] It is understood that in a specific embodiment, data blocks divided at the same scale (i.e., the same resolution) constitute a data block set, and the number of data block sets is the same as the number of scale types. When encoding the data blocks, it is necessary to encode the data blocks in different data block sets separately. For ease of understanding, the following description will be based on two different scales as an example.

[0108] Specifically, for the target geological structure data X1 of the first scale (H1, W1, D1) in the above embodiment, the first data block P1 after segmentation is encoded to obtain the block code X e1 ,in, like Figure 2 As shown, the multi-scale feature coding unit includes a block coding layer, which is mainly used to perform block coding on data blocks under different scale conditions.

[0109] The second data block P2 corresponding to the target geological structure data X2 of the second scale (H2, W2, D2) is encoded to obtain the block code X e2 ,in,

[0110] S311: Establishing a three-dimensional coordinate system based on the target geological structure data of the largest scale among the multiple scales;

[0111] Furthermore, among the acquired multi-scale target geological structure data, a three-dimensional coordinate system is established with the largest-scale target geological structure data as a reference.

[0112] It is understandable that geological structure data of different scales can reflect the characteristics of geological bodies in different spatial ranges, and comprehensively consider the acquisition range of large-scale three-dimensional geological structure data and the acquisition accuracy of small-scale geological structure data. In order to capture more features and improve modeling accuracy, a three-dimensional coordinate system is established based on the largest-scale geological structure data.

[0113] S312: Generate first three-dimensional position coordinates of the largest scale data slice and second three-dimensional position coordinates of other scale data slices based on the three-dimensional coordinate system;

[0114] Furthermore, in the three-dimensional coordinate system of the maximum-scale data, first three-dimensional position coordinates of the data block corresponding to the maximum scale and second three-dimensional position coordinates of the data blocks corresponding to other scales in the multi-scale data except the maximum scale are generated.

[0115] It is understood that when the number of multi-scales is greater than two, the second three-dimensional position coordinates include coordinates corresponding to different scales. For example, when the multi-scale is two, in addition to the maximum scale, two different scales are included. The data blocks corresponding to these two different scales have different relative three-dimensional position coordinates in the three-dimensional coordinate system. In fact, it can be understood that after establishing the three-dimensional coordinate system of the geological data at the maximum scale, the relative three-dimensional position coordinates of the geological data at other scales are found in this coordinate system.

[0116] S313: Encoding the data blocks in the data block set according to the preset dimension, the first three-dimensional position coordinates, and the second three-dimensional position coordinates to obtain a position code;

[0117] After obtaining the first three-dimensional position coordinates and the second three-dimensional position coordinates, position encoding is performed on the data blocks. In a specific embodiment, the three-dimensional position coordinates can be expressed as p(h i ,w i ,d i ), where h i is the coordinate on the horizontal axis in the three-dimensional space coordinate system, w i is the coordinate on the vertical axis, d i is the coordinate on the elevation axis.

[0118] Furthermore, the data blocks in all data blocks are encoded according to the preset dimension d and the three-dimensional position coordinates p. The three-dimensional position coordinates p include the first three-dimensional position coordinates and the second three-dimensional position coordinates. Specifically, the encoding is performed according to the position encoding formulas (2) and (3):

[0119]

[0120] Among them, p is the three-dimensional position coordinate (including the first three-dimensional position coordinate and the second three-dimensional position coordinate), i is the index of the encoding dimension, representing the i-th dimension in the encoding vector, and each dimension i ranges from 0 to d-1, and d is the preset dimension.

[0121] For ease of understanding, two scales are used as an example. Thus, under the two scales, the first three-dimensional position coordinates corresponding to the target geological structure data X1 with a larger scale and the second three-dimensional position coordinates corresponding to the target geological structure data X2 with a smaller scale can be obtained. Further, PE is calculated according to formula (2) and formula (3) respectively. p,2i and PEp,2i After that, they are combined to obtain a position code vector of length d. That is, the position code corresponding to each dimensional coordinate is calculated according to formula (2) and formula (3), and a position code is assigned to each 3D geological structure data block according to its 3D position coordinates.

[0122] Thus, position codes at different scales can be obtained. For target geological structure data X1 with a larger scale, position code PE can be obtained. e1 ,in, For the smaller-scale target geological structure data X2, the position code PE can be obtained. e2 ,in,

[0123] like Figure 2 As shown, in an optional embodiment, position coding can be performed by the position coding layer in the multi-scale coding unit to obtain the position coding PE e1 and position encoding PE e2 .

[0124] S314: Determine the fusion code according to the block code and the position code.

[0125] Furthermore, in an optional embodiment, determining a fusion code according to the block code and the position code includes:

[0126] The target code is obtained by summing the block code and position code of the same scale;

[0127] Perform designated splicing on each target code to obtain a fusion code.

[0128] In a specific embodiment, different scales of target geological structure data can obtain different block codes and position codes. When determining the fusion code, the block codes and position codes of the same scale are first summed. For example, in the above example, the block codes X of the same scale are summed. e1 and position encoding PE e1 Sum and get the target code X pe1 , that is, X pe1 =PE e1 +X e1 At the same time, the blocks of the same scale are encoded as X e2 and position encoding PE e2 Sum and get the target code X pe2 , that is, X pe2 =PE e2 +X e2 .

[0129] In a specific embodiment, Figure 2As shown, the block coding layer will obtain the block coding X e1 and block code X e2 Transmitted to the position coding layer so that the position coding layer can obtain the position coding PE e1 and position encoding PE e2 Then, the target code is calculated.

[0130] Further, Figure 2 The position encoding layer shown encodes the target X pe1 and target code X pe2 Perform vertical splicing to obtain the fusion code X pe ,in Vertical splicing refers to splicing two matrices vertically. For example, splicing two two-by-two target encoding matrices together can produce a four-by-two fused encoding matrix. It should be noted that when the target code is a vector, horizontal splicing is performed.

[0131] In an optional embodiment, a preset number of Transformer encoders are used to encode the masked data to obtain a feature encoding vector, including:

[0132] Normalize the masked data encoding to obtain normalized encoding;

[0133] Generate the query matrix, key matrix and value matrix corresponding to the normalized encoding through linear transformation;

[0134] Calculate the multi-head attention score based on the query matrix, key matrix and value matrix;

[0135] The multi-head attention scores are calculated through a feedforward neural network to obtain the feature encoding vector.

[0136] like Figure 2 As shown in the figure, the multi-scale encoding unit includes a Transformer encoding layer, which is composed of a preset number of Transformer encoders connected in front and behind. Each Transformer encoder is composed of five parts: a normalization layer, a multi-head self-attention layer, a normalization layer, a feedforward neural network layer, and a normalization layer.

[0137] In a specific embodiment, when the Transformer encoder encodes the masked data code, the normalization layer normalizes the masked data code to obtain a normalized code. The specific normalization formula is formula (4):

[0138]

[0139] Among them, X normis the normalized code after normalization, α and β are learnable parameters, X mask For masked data encoding, μ and δ represent X mask The mean and standard deviation of .

[0140] In the multi-head self-attention layer, the query matrix Q, key matrix K and value matrix V corresponding to the normalized encoding are generated by linear transformation, and the multi-head attention score X is calculated based on the query matrix Q, key matrix K and value matrix V. attention Specifically, MultiHead(Q,K,V)=Concat(head1,...,head h )W O , Furthermore, the multi-head attention score Attention(Q,K,V) is calculated according to formula (5):

[0141]

[0142] Furthermore, in the feedforward neural network layer, the multi-head attention scores are calculated to obtain the feature encoding vector. Specifically, the calculation is performed according to formula (6):

[0143] X encoder =max(0,X attention w1+b1)w2+b2

[0144] Among them, X encoder is the feature encoding vector, X attention is the multi-head attention score, w1, w2, b1 and b2 are learnable parameters.

[0145] Therefore, in order to utilize the strong correlation between three-dimensional geological structure data of different resolutions, the multi-scale self-attention mechanism promotes cross-resolution feature encoding between multiple different inputs through the multi-resolution self-attention mechanism, ensuring that the model can effectively capture and associate information between different resolutions, enhancing the depth and breadth of feature extraction, and promoting the portability and adaptability of the model in various three-dimensional geological structure data analysis applications.

[0146] Figure 4 A flow chart of a three-dimensional geological structure data modeling method provided in another embodiment of the present application is shown as follows: Figure 4 As shown, in an optional embodiment, feature decoding is performed on the feature coding vector using a pre-built multi-scale feature decoding model, including:

[0147] S40: performing feature mapping and feature restoration processing on the feature coding vector respectively to determine the decoding dimension and the input feature dimension;

[0148] In an optional embodiment, the multi-scale decoding module mainly consists of two 3D Transformer decoders with the same structure and based on the cross-attention mechanism, wherein the parameters involved are only updated during pre-training and frozen during the inference phase, and the decoder output can be selected by setting the output parameter Output, such as Figure 2 As shown in the figure, the multi-scale decoding module includes a feature mapping layer, a feature restoration layer, a feature splitting layer, a cross attention layer, a Transformer decoding layer, and a data prediction layer.

[0149] In a specific embodiment, the feature encoding vector X is encoded by the feature mapping layer encoder Perform feature mapping and determine the decoding dimension according to formula (6):

[0150] d decoder =X encoder w3+b3 (6)

[0151] Among them, d decoder is the decoding dimension, w3 and b3 are learnable parameters. In a specific embodiment, the feature encoding vector X can be encoder Convert from encoding dimension d to decoding dimension d decoder .

[0152] At the same time, the feature encoding vector X is encoder Perform feature restoration processing to determine the input feature dimension. Specifically, the feature restoration layer converts the feature encoding vector X into encoder The dimension is ((H1 / / s+H2 / / s)×(W1 / / s+W2 / / s)×(D1 / / s+D1 / / s)×(1-r))×d decoder Restore to ((H1 / / s+H2 / / s)×(W1 / / s+W2 / / s)×(D1 / / s+D1 / / s))×d decoder .

[0153] S41: splitting the feature encoding vector into multiple feature encoding sub-vectors corresponding to multiple scales under the conditions of the decoding dimension and the input feature dimension;

[0154] Furthermore, under the conditions of decoding dimension and input feature dimension, the feature splitting layer converts the feature encoding vector X encoder Split into multiple feature coding sub-vectors corresponding to multiple scales. For example, in the above example, the first data block P1 and the second data block P2 of the three-dimensional geological structure data of two different scales can be split into feature coding sub-vectors X encoder Split into feature encoding sub-vector X encoder1 and the feature encoding subvector X encoder2 ,in,

[0155] S42: Based on the cross-attention mechanism, calculate the cross-attention scores of multiple feature encoding sub-vectors;

[0156] Furthermore, the cross attention layer will be based on the feature encoder vector X encoder1 and the feature encoding subvector X encoder2 , generate query matrices Q1 and Q2 respectively through linear transformation. At the same time, according to the feature encoding vector X encoder , through linear transformation, key matrices K1 and K2, and value matrices V1 and V2 are generated respectively. Similarly, the cross attention scores are calculated based on formula (5) to obtain the cross attention scores Cross1 and Cross2.

[0157] S43: Processing the cross attention scores through a preset number of Transformer decoders to generate a 3D geological model;

[0158] In an optional embodiment, the Transformer decoding layer is composed of a preset number of Transformer decoders connected front to back. The preset number can be set according to actual business needs. The Transformer decoder consists of five parts: a normalization layer, a multi-head self-attention layer, a normalization layer, a feedforward neural network layer, and a normalization layer.

[0159] In a specific embodiment, the cross attention scores Cross1 and Cross2 are calculated by a preset number of Transformer decoders to generate a three-dimensional geological model X decoder1 and 3D geological model X decoder2 .

[0160] S44: Perform data prediction for a specified task on the three-dimensional geological model to obtain a prediction result.

[0161] In an optional embodiment, as Figure 2 As shown, the multi-scale decoding unit also includes a data prediction layer. In a specific embodiment, step S43 generates a three-dimensional geological model X decoder1 and 3D geological model X decoder2 Transmitted to the data prediction layer, the data prediction layer performs data prediction according to formula (7) and formula (8):

[0162] X predict1 =X decoder1 w4+b4 (7)

[0163] X predict2 =X decoder1 w5+b5 (8)

[0164] Among them, Xpredict1 For the three-dimensional geological model X decoder1 For the prediction result of a specified task, X predict2 For the three-dimensional geological model X decoder2 The prediction results for a specified task, w4, b4, w5 and b5 are learnable parameters.

[0165] It should be noted that when the 3D geological structure data input to the data preprocessing unit includes multi-scale data, the number of corresponding output 3D geological models and the corresponding number of prediction results are the same as the number of scales of the 3D geological structure data. When calculating the prediction results, the calculation formula refers to Formula (7) or Formula (8).

[0166] In addition, it should be noted that the designated tasks can be set in advance, and the designated tasks may include but are not limited to fault determination and earthquake phase determination.

[0167] Thus, after pre-training on a large-scale publicly available 3D geological structure dataset, the multi-scale feature decoding model has the ability to extract general and regular feature representations. The effectiveness of these feature representations is verified in different application datasets to evaluate the performance of the pre-trained model on specific downstream tasks.

[0168] In summary, this application provides a deep learning framework based on a 3D spatial Transformer by combining a 3D masked autoencoder spatial Transformer, a multi-scale self-attention mechanism, and multi-scale feature decoding. Leveraging large-scale unlabeled 3D geological structure datasets, robust feature representations are extracted to improve the accuracy and performance of identifying geological structures, stratigraphic distributions, or other exploration tasks, thus advancing the field of 3D geological structure data processing, analysis, and interpretation.

[0169] In the above embodiments, the three-dimensional geological structure data modeling method is described in detail. The present application also provides a corresponding embodiment of a three-dimensional geological structure data modeling device.

[0170] Figure 5 A schematic diagram of a three-dimensional geological structure data modeling device provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the device includes:

[0171] The acquisition module 50 is used to acquire multi-scale three-dimensional geological structure data of the same area; multi-scale means multiple different resolutions;

[0172] A preprocessing module 51 is used to preprocess the three-dimensional geological structure data to obtain target geological structure data;

[0173] The feature coding model 52 is used to perform multi-scale feature fusion coding on the target geological structure data through a pre-built multi-scale feature coding model to obtain a feature coding vector;

[0174] The feature decoding module 53 is used to perform feature decoding on the feature coding vector using a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

[0175] In addition, the three-dimensional geological structure data modeling device provided in the embodiment of the present application also includes:

[0176] A format conversion module is used to convert the format of three-dimensional geological structure data to obtain structure data in a specified format;

[0177] A clipping module is used to clip blank data at the edge of the structural data exploration in a specified format according to the principle of minimization of clipping, so as to obtain clipped structural data;

[0178] The standardization module is used to calculate the mean value and standard deviation of the cropped structural data; and standardize the cropped structural data according to the mean value and standard deviation to obtain the target geological structure data.

[0179] The data segmentation module is used to segment the target geological structure data of different scales to obtain multiple data segmentation sets corresponding to different scales;

[0180] A fusion coding module, configured to perform cross-resolution coding on the data blocks in the data block set to obtain fusion coding;

[0181] The mask module is used to randomly mask the fusion code to obtain the masked data code;

[0182] The Transformer encoding module is used to encode the masked data using a preset number of Transformer encoders to obtain a feature encoding vector.

[0183] A block encoding module is used to encode the data blocks in each data block set with preset dimensions through a convolutional neural network to obtain block codes;

[0184] A coordinate system establishment module is used to establish a three-dimensional coordinate system based on the target geological structure data of the largest scale among the multiple scales;

[0185] A coordinate generation module, configured to generate, based on a three-dimensional coordinate system, first three-dimensional position coordinates of a data block of the largest scale and second three-dimensional position coordinates of data blocks of other scales;

[0186] A position encoding module, configured to encode the data blocks in the data block set according to the preset dimension, the first three-dimensional position coordinates, and the second three-dimensional position coordinates to obtain a position code;

[0187] The fusion coding submodule is used to determine the fusion coding according to the block coding and the position coding.

[0188] The summation module is used to sum the block codes and position codes of the same scale to obtain the target code;

[0189] The splicing module is used to perform specified splicing on each target code to obtain a fused code.

[0190] The normalization module is also used to normalize the masked data encoding to obtain normalized encoding;

[0191] A linear transformation module, used to generate a query matrix, a key matrix, and a value matrix corresponding to the normalized encoding through linear transformation;

[0192] The multi-head attention module is used to calculate the multi-head attention score based on the query matrix, key matrix and value matrix;

[0193] The feedforward neural network module is used to calculate the multi-head attention scores through the feedforward neural network to obtain the feature encoding vector.

[0194] A restoration module is used to perform feature mapping and feature restoration on the feature encoding vector to determine the decoding dimension and the input feature dimension;

[0195] A splitting module is used to split the feature encoding vector into multiple feature encoding sub-vectors corresponding to multiple scales under the conditions of decoding dimension and input feature dimension;

[0196] The crisscross attention module is used to calculate the crisscross attention scores of multiple feature encoding sub-vectors based on the crisscross attention mechanism;

[0197] A generation module is used to process the cross-attention scores through a preset number of Transformer decoders to generate a 3D geological model;

[0198] The prediction module is used to perform data prediction of a specified task on the three-dimensional geological model and obtain prediction results.

[0199] Figure 6 A schematic diagram of a three-dimensional geological structure data modeling device provided in another embodiment of the present application is shown in FIG. Figure 6 As shown, the three-dimensional geological structure data modeling device includes: a memory 60 for storing computer programs;

[0200] The processor 61 is configured to implement the steps of the three-dimensional geological structure data modeling method mentioned in the above embodiment when executing the computer program.

[0201] The three-dimensional geological structure data modeling device provided in this embodiment may include but is not limited to a tablet computer, a laptop computer, or a desktop computer.

[0202] Among them, the processor 61 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 61 can be implemented in at least one hardware form of a digital signal processor (DSP), a field programmable gate array (FPGA), and a programmable logic array (PLA). The processor 61 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 61 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 61 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0203] The memory 60 may include one or more computer-readable storage media, which may be non-transitory. The memory 60 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory storage devices. In this embodiment, the memory 60 is used to store at least the following computer program 601, wherein, after being loaded and executed by the processor 61, the computer program can implement the relevant steps of the three-dimensional geological structure data modeling method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 60 may also include an operating system 602 and data 603, and the storage method may be temporary or permanent. The operating system 602 may include Windows, Unix, Linux, etc. The data 603 may include, but is not limited to, relevant data involved in the three-dimensional geological structure data modeling method.

[0204] In some embodiments, the three-dimensional geological structure data modeling device may further include a display screen 62 , an input / output interface 63 , a communication interface 64 , a power supply 65 , and a communication bus 66 .

[0205] Those skilled in the art will understand that Figure 6 The structure shown in the figure does not constitute a limitation on the three-dimensional geological structure data modeling device, and may include more or fewer components than those shown in the figure.

[0206] The three-dimensional geological structure data modeling device provided in the embodiment of the present application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the three-dimensional geological structure data modeling method in the above embodiment.

[0207] It should be noted that although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

Claims

1. A three-dimensional geological structure data modeling method, characterized in that: The method comprises: Acquire multi-scale 3D geological structure data of the same area; the multi-scale refers to acquisition conditions with multiple different resolutions; Preprocessing the three-dimensional geological structure data to obtain target geological structure data; Performing multi-scale feature fusion coding on the target geological structure data through a pre-built multi-scale feature coding model to obtain a feature coding vector; The feature coding vector is feature decoded by a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

2. The three-dimensional geological structure data modeling method according to claim 1, characterized in that: The preprocessing of the three-dimensional geological structure data to obtain target geological structure data includes: Converting the three-dimensional geological structure data into a specified format to obtain structure data; According to the principle of minimization of cropping, blank data at the exploration edge of the structural data in the specified format are cropped to obtain cropped structural data; The average value and standard deviation of the trimmed structural data are calculated; and the trimmed structural data are standardized according to the average value and the standard deviation to obtain the target geological structural data.

3. The three-dimensional geological structure data modeling method according to claim 1, wherein: The target geological structure data is coded using a pre-built multi-scale feature fusion model. Get the feature encoding vector, including: Segmenting the target geological structure data of different scales respectively to obtain a plurality of data segmentation sets corresponding to the different scales; Performing cross-resolution encoding on the data blocks in the data block set to obtain fused encoding; Performing random masking on the fusion code to obtain a masked data code; The masked data is encoded using a preset number of Transformer encoders to obtain the feature encoding vector.

4. The three-dimensional geological structure data modeling method according to claim 3, wherein: The performing cross-resolution encoding on the data blocks in the data block set to obtain fusion encoding includes: By using a convolutional neural network, encoding the data blocks in each of the data block sets with preset dimensions is performed to obtain block codes; Establishing a three-dimensional coordinate system based on the target geological structure data of the largest scale among the multiple scales; Based on the three-dimensional coordinate system, generating first three-dimensional position coordinates of the largest scale data slice and second three-dimensional position coordinates of other scale data slices; Encoding the data blocks in the data block set according to the preset dimension, the first three-dimensional position coordinates, and the second three-dimensional position coordinates to obtain a position code; The fusion code is determined according to the block code and the position code.

5. The three-dimensional geological structure data modeling method according to claim 4, characterized in that: The determining the fusion code according to the block code and the position code includes: Summing the block code and the position code of the same scale to obtain a target code; The target codes are spliced together to obtain the fused code.

6. The three-dimensional geological structure data modeling method according to claim 3, wherein: The step of encoding the masked data using a preset number of Transformer encoders to obtain the feature encoding vector includes: Normalizing the masked data code to obtain a normalized code; Generating a query matrix, a key matrix, and a value matrix corresponding to the normalized code through linear transformation; Calculating a multi-head attention score based on the query matrix, the key matrix, and the value matrix; The multi-head attention score is calculated through a feedforward neural network to obtain the feature encoding vector.

7. The three-dimensional geological structure data modeling method according to claim 1, wherein: The feature decoding of the feature coding vector using a pre-built multi-scale feature decoding model includes: Performing feature mapping and feature restoration processing on the feature coding vector respectively to determine a decoding dimension and an input feature dimension; Splitting the feature coding vector into a plurality of feature coding sub-vectors corresponding to the multiple scales under the conditions of the decoding dimension and the input feature dimension; Calculating cross-attention scores of the plurality of feature encoding sub-vectors based on a cross-attention mechanism; Processing the cross attention scores through a preset number of Transformer decoders to generate the three-dimensional geological model; Perform data prediction for a specified task on the three-dimensional geological model to obtain a prediction result.

8. A three-dimensional geological structure data modeling device, characterized in that: The device comprises: An acquisition module is used to acquire multi-scale three-dimensional geological structure data of the same area; the multi-scale refers to multiple different resolutions; A preprocessing module, used for preprocessing the three-dimensional geological structure data to obtain target geological structure data; A feature coding model is used to perform multi-scale feature fusion coding on the target geological structure data using a pre-built multi-scale feature coding model to obtain a feature coding vector; The feature decoding module is used to perform feature decoding on the feature coding vector through a pre-built multi-scale feature decoding model to generate a three-dimensional geological model.

9. A three-dimensional geological structure data modeling device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps of the three-dimensional geological structure data modeling method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the three-dimensional geological structure data modeling method according to any one of claims 1 to 7 are implemented.

Images