Geophysical gravity and magnetic anomaly source model construction and rapid forward and reverse modeling method

Abstract

The invention provides a geophysical gravity and magnetic anomaly source model construction and rapid forward and reverse modeling method, and relates to the technical field of intelligent three-dimensional geological modeling and simulation. According to geological information of a research area, source body modeling parameters of a gravity and magnetic anomalous field are set for the research area, and a three-dimensional model matrix M capable of describing a plurality of underground anomalous bodies is obtained; constructing a fast forward modeling network, establishing nonlinear mapping from the three-dimensional model matrix to the simulated abnormal data through the fast forward modeling network, and calculating forward modeling response of the three-dimensional model matrix; and constructing an inversion network based on a concurrent module CFTBlock and combining a CNN and a Transform, and establishing nonlinear mapping from the gravity and magnetic abnormal data to a three-dimensional model matrix. The method has the advantages of fast and accurate forward modeling, high-resolution inversion and the like, and is suitable for better interpretation of actually measured gravity and magnetic data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent three-dimensional geological modeling and simulation, and particularly relates to a geophysical gravity and magnetic anomaly source model construction and fast forward and inverse method. BACKGROUND

[0002] Gravity and magnetic field data inversion can effectively guide geological prospecting and other work. However, the inversion problem is usually ill-posed, and its result is non-unique. In the past research, geophysicists mainly solve it based on the Tikhonov regularization theory, such as smooth inversion based on the maximum smoothing stability functional, focusing inversion based on the minimum gradient support function, etc. However, in this kind of inversion method, it is often necessary to solve large equation sets, which leads to a large amount of matrix operations, and inevitably falls into a local optimal solution in the solving process, thus leading to incomplete recovery of the predicted anomaly body shape and physical properties, so finding a solution close to global optimization and reducing the multi-solution is the main task of the current gravity and magnetic field data inversion. In recent years, deep learning provides a new direction for gravity and magnetic field data inversion due to its nonlinear computing characteristics, and is one of the important development directions of geophysics today. However, it faces the following three problems:

[0003] The first problem is greatly affected by the quality of the data set. The closer the simulated field source body in the data set to the true field source body, the more accurate the prediction result of the network. However, the existing model construction method faces the dual difficulties of insufficient geometric feature representation and low computational efficiency: random block step model and random walk method are difficult to accurately depict complex geological bodies, and although refined subdivision improves accuracy, it significantly increases the amount of calculation. At the same time, Berlin noise is widely used in the field of visual elements, reproduction of natural textures, simulation of the atmosphere due to its ability to generate two-dimensional continuous differentiable functions. Compared with other natural noise algorithms, Berlin noise algorithm can generate continuous and smooth noise through interpolation and smoothing function processing, avoiding the generation of sharp edges, and performs well in simulating natural phenomena, which provides a new way of thinking for model construction with geological reality.

[0004] The second problem is how to quickly and batch generate data sets close to the real anomaly field. The problem faced by the traditional forward method is that it occupies a large space and has low computational efficiency, and the root cause is the storage and solution of large-scale matrix equations. In view of this problem, scholars have proposed methods such as wavelet transform, matrix compression, and fast Fourier transform to reduce the memory occupation problem of large-scale matrices. However, as the data size expands and the grid is finely subdivided, the amount of calculation and memory consumption increases dramatically, and the above methods are difficult to meet the requirements of fast forward. In recent years, scholars have tried to use deep learning for approximate fitting of forward, which has significantly accelerated the forward process, but the accuracy needs to be further improved. Therefore, research on fast forward method based on deep learning can effectively solve the problems faced by traditional forward method.

[0005] The third problem is that traditional deep learning methods are mainly based on convolutional neural networks (CNN), mainly including traditional CNN and Unet methods. Among them, Unet3+, ResUnet and other Unet variants can reconstruct density anomalies with clear boundaries and have strong robustness to noise, and are also gradually applied. However, in the case where high-resolution spatial information is crucial, traditional networks may inadvertently homogenize key features, leading to inaccurate inversion results due to loss of details. Combining CNN and Transformer architecture can solve the limitation and take advantage of both models, which can well obtain global features due to the self-attention mechanism and multi-layer perceptron (MLP) structure. Some scholars have begun to try to combine local and global features and have achieved good results. However, in the field of deep learning gravity and magnetic field data inversion, scholars still mainly focus on local features, and lack of simultaneous application of local and global features. Therefore, the simultaneous use of local and global features of gravity and magnetic field data is an important direction of current deep learning inversion.

[0006] The above research shows that developing a suitable gravity and magnetic anomaly source model construction and fast forward and inverse method can provide technical support for future gravity and magnetic exploration work. SUMMARY

[0007] The technical problem to be solved by the present application is to solve the deficiencies of the prior art, and to provide a geophysical gravity and magnetic anomaly source model construction and fast forward and inverse method, which realizes the construction and forward and inverse of the geophysical gravity and magnetic anomaly source model.

[0008] To solve the above technical problems, the technical solution adopted by the present application is: a geophysical gravity and magnetic anomaly source model construction and fast forward and inverse method, comprising: According to the geological information of the study area, the source body modeling parameters of the gravity and magnetic anomaly field in the study area are set; Select a rectangular region in the study area, generate a Berlin noise noise, perform hard thresholding, and divide the rectangular region into multiple two-dimensional independent blocks through connection relationship; A three-dimensional model matrix is generated for each independent block, and the final three-dimensional model matrix M describing multiple underground abnormal bodies is obtained by merging; For the three-dimensional model matrix describing multiple underground abnormal bodies, the kernel function matrix is calculated using the forward calculation formula combined with the terrain and observation surface, and then the simulated gravity and magnetic anomaly data are obtained, and the training set and test set of the simulated gravity and magnetic anomaly forward response are obtained combined with the parallel computing method; A fast forward network is constructed, and a nonlinear mapping from the three-dimensional model matrix to the simulated anomaly data is established through the fast forward network to calculate the forward response of the three-dimensional model matrix; The fast forward modeling network with optimal weight parameters was trained using the training and test sets of gravity and magnetic anomaly forward modeling responses. Generate a training set and a test set for gravity and magnetic anomalies to train the inversion network; An inversion network based on the concurrent module CFTBlock, combining CNN and Transformer, is constructed to establish a nonlinear mapping from gravity and magnetic anomaly data to a three-dimensional model matrix. Using the gravity and magnetic anomaly training set used to train the inversion network, a random noise addition strategy is employed to train the network and obtain an inversion network with optimal weight parameters.

[0009] Furthermore, the specific method for setting the source body modeling parameters for the gravity and magnetic anomaly field in the study area is as follows: Based on the geological information of the study area, the lithological distribution and physical property values ​​of different regions are statistically analyzed to obtain the spatial location and range of physical property values ​​of the exploration target. Thus, when simulating the gravity and magnetic anomaly field, the physical property intervals and distribution ranges are set as modeling parameters for the gravity and magnetic anomaly. At the same time, based on the measurement results, the observation plane and topographic information are set.

[0010] Furthermore, the specific method for selecting a rectangular region in the study area, generating Berlin noise, performing hard thresholding, and dividing the rectangular region into multiple independent two-dimensional blocks through connectivity is as follows: In the study area, the size of the rectangular region where the Berlin noise is to be generated is set, and the Berlin noise is generated; hard thresholding refers to setting the element values ​​in the two-dimensional noise matrix below a preset minimum physical property threshold. α The elements are set to 0; then, based on whether each non-zero element has a non-zero element in its four adjacent positions (above, below, left, and right), the entire rectangular area is divided into... N A set of independent two-dimensional blocks that are not connected to each other 1 ...A N .

[0011] Furthermore, the specific method for generating a three-dimensional model matrix for each independent block and merging them to obtain the final three-dimensional model matrix M that can describe multiple underground anomalies is as follows: The continuous underground geometric space is discretized into a finite number of small units, and a three-dimensional matrix is ​​used to describe the distribution and physical properties of gravity and magnetic anomalies in the underground space, thus obtaining a three-dimensional model matrix. For each independent two-dimensional block A l Create a 3D model matrix mt with an initial state of all zeros. l , l = 1, 2,..., N Its dimensions are nx × ny × nz , nx, ny , nz These represent the number of grid cells in the study area along the X, Y, and Z directions, respectively; mt is defined as... l,n Indicates block A l The corresponding three-dimensional model matrix mt l In the depth layer n Two-dimensional slices at the location, n = 1, 2,..., nz Randomly generate a depth layer number. n ,1≤ n ≤ nz ;Transfer two-dimensional block A l Copy the data to its corresponding 3D model matrix mt l Two-dimensional slice mt l,n In this context, it serves as the initial core layer of the block's anomaly; based on this core layer... n Starting from, and simultaneously moving towards n Decrease direction and n Diffusion occurs across all depth layers in the increasing direction; this diffusion process ultimately results in each independent two-dimensional block A. l Generate a three-dimensional model matrix mt l ; Generate a 3D model matrix mt from all independent 2D blocks l By merging, we obtain the final three-dimensional model matrix M that can describe multiple underground anomalies;

[0012] Furthermore, when the initial core layer of the block anomaly diffuses upward or downward, the attenuation value β is first subtracted from each element of the two-dimensional matrix of the core layer, and then the current attenuation coefficient γ is used to reduce the overall result. Finally, the horizontal direction is adjusted according to the shift step size to generate a stepped or tilted model.

[0013] Furthermore, the method employs parallel computing technology to generate in batches three-dimensional model matrices describing gravity and magnetic anomalies with different spatial distributions and physical properties, specifically: Each thread independently completes the entire construction process of a single 3D model matrix, including the generation of Berlin noise, independent block division, 3D model matrix diffusion calculation, and finally outputting the final 3D model matrix M, ensuring that the 3D model matrices generated by different threads are independent and diverse in spatial distribution and physical properties.

[0014] Furthermore, the method constructs a boundary reconstruction module and combines it with a CNN to construct a fast forward modeling network; The fast forward network takes ResUNet as the backbone architecture, is composed of an encoder for feature extraction and data compression and a decoder for forward calculation and data reconstruction, and after feature reconstruction at each layer of the decoder, adopts a boundary reconstruction module based on a double residual mechanism for feature enhancement reconstruction to further improve the feature reconstruction accuracy. The boundary recognition module first extracts primary boundary features through a residual block; then introduces residual connection to fuse the original features and the reconstructed features, retaining key feature information; and finally implements secondary feature reconstruction through a composite output layer containing a 3x3 convolution and a 1x1 convolution.

[0015] Further, the specific method for generating the gravity and magnetic anomaly training set and test set for training the inversion network is: According to the existing geological and geophysical data and prior information of the target exploration area, a 1x10 4 ~1x10 5 three-dimensional model matrix conforming to the geological characteristics of the region is created; then, the fast forward network with optimal weight parameters is used to calculate the gravity and magnetic anomaly data corresponding to the three-dimensional model matrix, thereby forming the training data set of the inversion network.

[0016] Further, the inversion network is composed of an input layer, a core processing layer, and an output layer; the input layer is used to receive the gravity and magnetic anomaly data, the core processing layer is composed of an encoder-decoder containing four encoding-decoding blocks, the inversion network extracts high-dimensional features of the gravity and magnetic anomaly data through three layers of downsampling, each encoding-decoding block integrates multiple concurrent modules CFTBlock, and jump connection is introduced to realize multi-scale feature fusion, thereby effectively retaining the detail information of the anomaly data; finally, the features output by the decoder are mapped and converted into a three-dimensional model matrix through a 1x1 convolution of the output layer, which can reflect the spatial form and physical characteristics of the underground anomaly body.

[0017] Further, the concurrent module CFTBlock includes a CNN branch and a Transformer branch, the CNN branch adopts a two-stage design to gradually extract and enhance local features: the first stage is composed of two convolution blocks and a residual connection, local features are captured through convolution operation, and the residual connection is used to alleviate the gradient vanishing problem; the second stage is based on convolution and residual connection, and the global features of the Transformer branch are fused to realize the combination of local features and global context; The Transformer branch is composed of a multi-head self-attention module and an MLP module; the multi-head self-attention mechanism is used to capture global features; the MLP module performs nonlinear transformation on the output of the self-attention module to further enhance the feature expression ability.

[0018] The beneficial effects generated by the above technical scheme are that the geophysical gravity anomaly source model construction and fast forward and inverse method allows the underground anomaly body model to have variable properties, is more in line with the actual situation, shows that the central property value is high, gradually decreases to both sides, presents the characteristics of large in the middle and small at both ends in the three-dimensional model form, and allows small anomaly bodies to exist, which is consistent with the actual geological model; the fast forward method based on the boundary reconstruction module can quickly perform batch forward of the underground anomaly body model with low error; the high-resolution inversion method combined with the random noise strategy can obtain better inversion results, improve the accuracy of the recovered underground property results, and quickly obtain the results. It has been verified that the method proposed in the application has the advantages of source body modeling being more in line with the actual situation, being able to realize fast and accurate forward, and being able to perform high-resolution inversion, and is suitable for better interpretation of measured gravity and magnetic data. The application has important significance for improving the processing and interpretation accuracy of geophysical gravity and magnetic field data. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of a geophysical gravity anomaly source model construction and fast forward and inverse method provided for an embodiment of the application is shown in the figure; Figure 2 A simulated gravity anomaly model schematic diagram provided for an embodiment of the application is shown in the figure, wherein (a) is a three-dimensional model, and (b) is a profile graph; Figure 3 A gravity anomaly three-dimensional model fast forward network architecture diagram based on a CNN and a boundary reconstruction module provided for an embodiment of the application is shown in the figure; Figure 4 A box plot of a test data set E provided for an embodiment of the application is shown in the figure; d Figure 5 An inversion network architecture diagram of a joint CNN and Transformer based on a concurrent module CFTBlock provided for an embodiment of the application is shown in the figure; Figure 6 A module architecture diagram of the concurrent module CFTBlock provided for an embodiment of the application is shown in the figure, (a) is dimensionality increasing and dimensionality decreasing for spatial alignment of feature maps, and (b) is implementation details of a CNN branch, a Transformer branch and a FCU; Figure 7 A residual density model, noisy data and a profile graph provided for an embodiment of the application are shown in the figure, (a) is a three-dimensional model (results of residual density absolute value >0.3 are drawn), (b) is noisy data, and (c) is a residual density model z=425m profile graph; Figure 8 Inversion results and a profile graph provided for an embodiment of the application are shown in the figure, (a) is a three-dimensional result graph (results of residual density absolute value >0.3 are drawn), and (b) is an inversion result z=425m profile graph.​ Figure 9 The following are gravity anomaly data and inversion results of Venton Salt Dome provided in this embodiment of the invention: (a) is the gravity anomaly data of Venton Salt Dome; (b) is the three-dimensional spatial diagram of the prediction model (plotting the part where the absolute value of the remaining density is >0.3); (c) is the profile of the prediction model at z=525m; (d) is the profile of the prediction model at y=3334.25km; and (e) is the profile of the prediction model at x=442.550km. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] In this embodiment, a method for constructing a geophysical gravity and magnetic anomaly source model and performing rapid forward and inverse modeling is described, such as... Figure 1 As shown, it includes the following steps: Step 1: Parameter setting: Based on geological information such as drilling, seismic profiles, and geological profiles, the source body modeling parameters for the gravity and magnetic anomaly field in the study area are set. Before setting parameters, based on previous work, including geological information such as drilling, seismic profiles, and geological profiles, we statistically analyze the lithological distribution and physical property values ​​in different regions to obtain the approximate spatial location and physical property values ​​of the exploration target. This allows us to set the physical property range and distribution range, among other gravity and magnetic anomaly modeling parameters, when simulating gravity and magnetic anomalies. At the same time, we set the observation plane and topographic information based on the measurement data. In this embodiment, parameters are set for the measured data to be used later, with the minimum physical property value set to 0.55 and the maximum burial depth range to 200m.

[0022] Step 2: Select a rectangular region in the study area, generate Berlin noise, perform hard thresholding, and divide the rectangular region into multiple two-dimensional independent blocks through connectivity; In the study area, the size of the rectangular region where the Berlin noise is to be generated is set, and the Berlin noise is generated; hard thresholding refers to setting the element values ​​in the two-dimensional noise matrix below a preset minimum physical property threshold. α The elements are set to 0; then, based on whether each non-zero element has a non-zero element in its four adjacent positions (above, below, left, and right), the entire rectangular area is divided into... N A set of independent two-dimensional blocks that are not connected to each other 1 ...A N ; In this embodiment, the study area is divided into a 64×64 grid, and a breadth-first search method is used to divide the independent blocks. Step 3: Assign different independent blocks to different depth layers in the three-dimensional space of the study area, and attenuate and diffuse in the depth direction, to generate a three-dimensional model matrix for each independent block, and combine to obtain the final three-dimensional model matrix M which can describe multiple abnormal bodies in the subsurface; For convenience of calculation, the continuous geometric space in the subsurface is discretized into a finite number of small units, i.e. using finite element meshing to divide the three-dimensional space of the study area into nx×ny×nz grids (nx, ny, nz represent the number of mesh units in the X, Y, Z directions of the study area, and the grid is consistent with the size of the study area in the plane), and use a three-dimensional matrix to describe the distribution and physical property values of the gravity and magnetic anomaly bodies in the subsurface, to obtain a three-dimensional model matrix (i.e. a three-dimensional model matrix describing the spatial distribution and physical property characteristics of the gravity and magnetic anomaly bodies), wherein the minimum physical property threshold α is set based on the geological information obtained in step 1, representing the minimum non-zero value (residual density or magnetic intensity) allowed to exist in the three-dimensional model matrix; For each independent two-dimensional block A l Create a three-dimensional model matrix mt l , l = 1, 2,..., N , whose dimensions are nx × ny × nz ( nx , ny , nz represent the number of mesh units in the X, Y, Z directions of the study area) ; define mt l,n to represent the two-dimensional slice of the three-dimensional model matrix mt l corresponding to the block A l at the depth layer n ( n = 1, 2,..., nz ); randomly generate a depth layer number n , 1≤ n ≤ nz ; copy the data of the two-dimensional block A l to the two-dimensional slice mt l of its corresponding three-dimensional model matrix mt l,n as the initial core layer of the block anomaly body; take this core layer n as the starting point and diffuse to all depth layers in the n decrease direction and n increase direction; in the diffusion process, the value of each element in the two-dimensional matrix gradually decreases with the increase of the distance from the core layer; this diffusion process generates a three-dimensional model matrix mt l for each independent two-dimensional block A l; Generate a 3D model matrix mt from all independent 2D blocks l By merging, we obtain the final three-dimensional model matrix M that can describe multiple underground anomalies; When the initial core layer of the block anomaly diffuses upward or downward, the attenuation value β is first subtracted from each element of the two-dimensional matrix of the core layer. Then, the current attenuation coefficient γ is used to reduce the overall result. Finally, the horizontal direction is adjusted according to the shift step size in order to generate a stepped or tilted model. The shift step size represents the number of grids that the data moves in one direction when it diffuses upward or downward. 3D model matrix mt l The two-dimensional slice at depth n+k is shown in the following formula: (1); (2); (3); Among them, mt l,n+k Represents the three-dimensional model matrix mt l In the depth layer n+k Two-dimensional slices at the location, ( p , q () represents an element in the two-dimensional slice. k =±1, ±2…,± MS , MS This indicates the preset maximum number of unidirectional diffusion steps. FL The filter function is represented by , the base attenuation coefficient is represented by , and the current attenuation coefficient is represented by . γ Based on diffusion steps step Attenuation, s x , s y The horizontal shift step size is for different directions, and its positive and negative signs represent the direction. If during the diffusion process, the number of diffusion steps step Exceed MS The iteration stops; each independent two-dimensional block undergoes diffusion to obtain its three-dimensional model matrix mt. l The final three-dimensional model matrix M describing multiple underground anomalies is: (4); For different 3D model matrix generation processes, all variables are random values ​​except for the spatial partitioning method and the threshold α. The three-dimensional model matrix generated in this embodiment supports the spatially continuous variation of physical property parameters (such as residual density and magnetic susceptibility), and presents a typical distribution rule of "high in the center and low at the edges", that is, the absolute value of the physical property is highest in the central region of the three-dimensional model matrix, gradually decays and decreases along the central axis to both sides, and the three-dimensional form gradually decreases to both sides from the largest central scale, while allowing to contain independent anomaly bodies of smaller scales, which can effectively simulate the typical properties of actual geological structures Figure 2 ).

[0023] Step 4: Repeat steps 2-3 and combine parallel computing technology to quickly and batch generate three-dimensional model matrices describing different spatial distributions and physical properties of gravity and magnetic anomaly bodies; Repeat steps 2-3, and use OpenMP multi-thread parallel technology to efficiently generate three-dimensional model matrix M. The specific implementation is that each thread independently completes the complete construction process of a single model, including Berlin noise generation, independent block division, three-dimensional model matrix diffusion calculation, and finally outputting the final three-dimensional model matrix M, ensuring that the three-dimensional model matrices generated by different threads are independent and diverse in spatial distribution and physical properties. By combining parallel computing technology, the three-dimensional model matrix can be quickly generated, thereby achieving batch and rapid generation of the required number of three-dimensional model matrices in a short time.

[0024] Step 5: For the three-dimensional model matrix describing multiple underground anomaly bodies obtained in step 4, combine the terrain and observation surface, use the forward calculation formula to calculate the kernel function matrix, and then obtain the simulated gravity and magnetic anomaly data, and combine the parallel computing method to obtain the training set and test set of the simulated gravity and magnetic anomaly forward response for subsequent rapid forward method research;

[0025] The formula for obtaining the simulated gravity and magnetic anomaly data d using the kernel function matrix is: (5); Where G is the kernel function matrix calculated by the forward formula, and m is the vectorization result of the three-dimensional model matrix M obtained in step 4.

[0026] The calculation of formula 5 is accelerated in parallel using the OpenMP method, which specifically means that multiple matrix multiplications are calculated simultaneously using the multi-threading technology of OpenMP, and the results are saved.

[0027] In this embodiment, the forward calculation is performed in the Cartesian coordinate system, and the distribution ranges of the X, Y, and Z axes are 0 to 3200 m, 0 to 3200 m, and 0 to 1000 m, respectively. The distance between the measurement point and the measurement line is 50 m. The gravity anomaly data V zand total field magnetic anomaly ΔT values. A total of 50,000 sets of gravity anomaly data and magnetic anomaly data were generated, and were divided into a training set and a test set in a ratio of 4:1.

[0028] Step 6: A boundary reconstruction module is constructed, and a fast forward network is constructed in combination with a CNN, a nonlinear mapping from a three-dimensional model matrix to simulated anomaly data is established through the fast forward network, and the forward response of the three-dimensional model matrix is calculated; the training set and the test set created in step 5 are used for training, and a fast forward network with optimal weight parameters is obtained, and a validation set is established for training effect verification; The fast forward network takes ResUNet as the backbone architecture, is composed of an encoder for feature extraction and data compression and a decoder for forward calculation and data reconstruction, and after feature reconstruction at each layer of the decoder, a boundary reconstruction module based on a double residual mechanism is used for feature enhancement reconstruction, further improving the feature reconstruction accuracy; The boundary recognition module first performs primary feature extraction through a residual block (ResBlock) to obtain reconstruction features; then residual connection is introduced to fuse the primary features and the reconstruction features, retaining key feature information; finally, a composite output layer containing a 3x3 convolution and a 1x1 convolution is used to implement secondary feature reconstruction; this double residual optimization structure enhances the feature expression ability of the network and improves the calculation accuracy of the forward task.

[0029] The fast forward network takes a three-dimensional model matrix M describing multiple subsurface anomaly bodies as network input, and the input data first passes through an encoder; the encoder gradually reduces the data size through multiple downsampling operations, and combines a residual module ResBlock that can effectively train deep networks to perform deep feature extraction, obtaining high-dimensional feature representations containing key information; the high-dimensional features are then input to the decoder, which gradually recovers the data size through upsampling operations; in each layer of the decoding process, not only is upsampling performed, but also feature enhancement is performed after feature reconstruction of the current layer by a boundary reconstruction module, and the corresponding forward result output of the layer is calculated; in order to fuse information of different scales, the output of each layer of the decoder is fused with more detailed low-dimensional feature information from the corresponding layer of the encoder through nearest neighbor interpolation; this design allows the network to effectively utilize more original and detailed information of the input data when calculating the current scale result.

[0030] The three-dimensional model matrix describing multiple subsurface anomaly bodies and its corresponding forward response data generated in step 5 are used to construct a gravity-magnetic anomaly training set, and the fast forward network is trained; during the training process, the test set is used to monitor the training effect in real time, preventing the model from over-relying on the training data (overfitting) and evaluating its generalization ability; finally, the weight parameters that perform best (with the highest accuracy) on the test set are selected for fast calculation of the forward response of the three-dimensional model matrix; In this embodiment, the fast forward network architecture is as shown in Figure 3 The encoder part takes a three-dimensional model matrix with resolution nz×64×64 as input (nz is the number of partitions in the depth direction), first performs channel dimension transformation through 1×1 convolution, and then inputs into a multi-level ResBlock for feature extraction. The ResBlock resolutions are 64×64, 32×32, 16×16, 8×8, and 4×4, which follow the principle of doubling the channel number and halving the resolution step by step to realize multi-scale compression and extraction of features. The decoder is symmetrical to the encoder, and uses upsampling operation to gradually restore the feature map resolution until it is consistent with the observation plane size. Finally, the fused features are mapped to the simulated anomaly data (i.e. the forward response of the three-dimensional model matrix) through 1×1 convolution, and the output resolution remains consistent with the input, ensuring the spatial consistency of the forward result.

[0031] The parameters used are: learning rate 3×10 -4 , learning rate decay strategy using Cosine Annealing, batch size 4, iteration number 100, optimizer using Adam, and loss function using SmoothL1. 10,000 pairs of geological models with highly random physical properties are generated as a validation set, of which 5,000 pairs of gravity anomaly data and 5,000 pairs of magnetic anomaly data, and the error of the model prediction is counted, as shown in Figure 4 The evaluation index is the relative error E d , and the smaller E d , the more accurate the predicted forward result. The formula is as follows: (6); Where d true represents the true data, and d pre represents the predicted data.

[0032] As can be seen from Figure 4 , the E d of gravity anomaly and magnetic anomaly is mostly below 5×10 -6 , and the median is 1.25×10 -6 and 1.28×10 -6 , respectively. The distribution range of E d of the magnetic anomaly is larger than that of the gravity anomaly, which is due to the larger amplitude of the magnetic anomaly than the gravity anomaly, resulting in larger error of the predicted data. However, the maximum value of E d of the gravity anomaly is below 2×10 -5 , and the maximum value of E d of the magnetic anomaly is below 6×10 -5 .

[0033] Step 7: Based on the three-dimensional model matrix generation method of steps 2-5 and the fast forward network model with the optimal weight parameter obtained in step 6, generate the gravity and magnetic anomaly training set and test set used for subsequent training of the inversion network; According to the existing geological and geophysical data (such as historical exploration data, geological maps, etc.) and prior information (such as known rock physical property range, etc.) of the target exploration area, 1x10 4 ~1x10 5 groups of three-dimensional model matrices conforming to the geological characteristics of the region are created; then, the fast forward network with the optimal weight parameter obtained in step 6 is used to efficiently calculate the gravity and magnetic anomaly data corresponding to these three-dimensional model matrices, forming the training data set of the inversion network; this method of generating gravity and magnetic anomaly data based on prior knowledge of the target area aims to make the optimal three-dimensional model matrix obtained by training the inversion network more accurately invert the spatial distribution and physical characteristics of the actual underground gravity and magnetic anomaly body in the region.

[0034] In this embodiment, 100,000 data sets are generated for training the inversion network. According to the test, it takes 5 minutes and 35 seconds to generate 100,000 three-dimensional model matrix data of different gravity anomaly bodies, and 4 minutes and 21.4 seconds to generate the corresponding simulated gravity and magnetic anomaly data, wherein the simulated gravity and magnetic anomaly data is calculated using the forward network, which uses GPU parallel mode with parallel parameter settings of 8 threads for parallel file reading and a batch size of 256.

[0035] Step 8: Construct an inversion network based on the joint CNN and Transformer of the concurrent module CFTBlock to establish a nonlinear mapping from gravity and magnetic anomaly data to three-dimensional model matrix; use the gravity and magnetic anomaly training set created in step 7 to train the inversion network with a random noise strategy and obtain an inversion network with optimal weight parameters; randomly select a model from the validation set and combine it with the measured data to verify the effect of the inversion network.

[0036] The inversion network is composed of an input layer, a core processing layer, and an output layer; the input layer is used to receive gravity and magnetic anomaly data, the core processing layer is composed of an encoder-decoder containing 4 encoding-decoding blocks; the inversion network extracts high-dimensional features of gravity and magnetic anomaly data through three layers of downsampling, each encoding-decoding block integrates 3-4 concurrent modules CFTBlock, and introduces a skip connection to realize multi-scale feature fusion, effectively preserving the details of the anomaly data; finally, the 1x1 convolution of the output layer maps the features output by the decoder to a three-dimensional model matrix, which can reflect the spatial form and physical characteristics of the underground anomaly body.

[0037] The concurrent module CFTBlock includes a CNN branch and a Transformer branch. The CNN branch adopts a two-stage design, aiming to gradually extract and enhance local features: the first stage is composed of two convolutional blocks with residual connection, capturing local features through convolution operation and alleviating the gradient vanishing problem through the use of residual connection; the second stage is to fuse the global features of the Transformer branch on the basis of convolution and residual connection, realizing the efficient combination of local features and global context. With the increase of network depth, the resolution of feature maps gradually decreases, while the number of channels increases accordingly. The second stage aims to reduce the parameter amount while retaining more rich semantic information.

[0038] Correspondingly, the Transformer branch is composed of a multi-head self-attention module and an MLP module. The multi-head self-attention mechanism is used to capture global features and improve the model's attention to key features; the MLP module performs nonlinear transformation on the output of the self-attention module to further enhance the feature expression ability.

[0039] Compared with traditional CNN, the CNN branch realizes down-sampling through MaxPool, which can expand the receptive field and reduce the parameter amount, but inevitably leads to the loss of some detailed information. The Transformer branch can dynamically allocate feature weights through the self-attention mechanism, effectively alleviating the information loss problem of the pooling operation, thereby generating inversion results with clear boundaries and high resolution. In each training process, random noise is added to the input gravity and magnetic anomaly data, rather than directly adding noise during the process of making the training set. This can make the same data maintain differences through the method of adding noise in real time in each training, thereby improving the generality and stability of the method.

[0040] In this embodiment, the inversion network architecture diagram and the concurrent module CFTBlock are shown in Figure 5 and Figure 6 . Among them, CFTBlock is the core component of ICT-Inv Net, which combines the local features of CNN and the global features of Transformer, significantly improving the reconstruction accuracy of predicting the geometric shape of the anomaly body and the physical property parameters. The CNN branch extracts local spatial features, containing spatial position information, the Transformer branch obtains long-distance global encoding features through the self-attention mechanism, enhancing the global feature expression ability, and the FCU module performs feature fusion and exchange between the CNN branch and the Transformer branch, improving the robustness of feature representation.

[0041] In this embodiment, a gravity anomaly data different from the training set and the test set is established, as shown in Figure 7Meanwhile, 0%~5% random noise is added to the input data randomly each time of training. In the theoretical test stage, in order to test the anti-noise ability of the algorithm, 5% Gaussian noise is added to the theoretical data. The constructed high-resolution inversion method is used for testing and the results are as shown in Figure 8 .

[0042] In order to verify the practicability of the method of the present application, the measured geophysical gravity anomaly data of the Ventnor salt dome area in the United States is also used for testing, and a survey area with an area of about 3200m x 3200m is selected for measured data testing, and the underground is divided into 64 x 64 x 20 small cuboids, each with a size of 50 x 50 x 50m 3 . The data is gridded and interpolated using Kriging interpolation, and a high-pass filter with a spatial wavelength of 5500m is used to remove the regional background field. According to the prior information, the generated minimum density absolute value is set to 0.55g / cm 3 . The inversion results are as shown in Figure 9 . Using the method proposed in the present application, high-resolution inversion of gravity anomaly data in the Ventnor salt dome area is realized, and the obtained inversion results have high resolution, which is beneficial to the processing and interpretation of gravity and magnetic field data.

[0043] It can be seen that the geophysical gravity and magnetic anomaly source model construction and fast forward and inverse method proposed in the present application can realize fast generation of large amount of data sets while having low error, and can provide technical support for subsequent inversion work. The inversion method combining local features and global features can obtain high-resolution results, which is beneficial to improving the application of gravity and magnetic exploration in practical work.

[0044] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part or all of the technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope defined by the present application.

Claims

1. A method for constructing and quickly inverting a geophysical gravity anomaly source model, characterized in that, The application relates to a method for training a fast forward network and an inversion network based on a concurrent module CFTBlock. The method comprises the following steps: According to the geological information of a research area, source body modeling parameters of a gravity and magnetic anomaly field in the research area are set; A rectangular area is selected in the research area, Berlin noise is generated, hard thresholding is performed, and the rectangular area is divided into multiple two-dimensional independent blocks through a connection relationship; A three-dimensional model matrix is generated for each independent block, and finally a three-dimensional model matrix M capable of describing multiple abnormal bodies underground is obtained by merging; For the three-dimensional model matrix describing multiple abnormal bodies underground, a kernel function matrix is calculated by using a forward calculation formula in combination with a terrain and an observation surface, so that simulated gravity and magnetic anomaly data are obtained, and a training set and a test set of simulated gravity and magnetic anomaly forward response are obtained in combination with a parallel computing method; A fast forward network is constructed, a nonlinear mapping from the three-dimensional model matrix to the simulated anomaly data is established through the fast forward network, and the forward response of the three-dimensional model matrix is calculated; The fast forward network is trained by using the training set and the test set of the gravity and magnetic anomaly forward response to obtain a fast forward network with optimal weight parameters; A gravity and magnetic anomaly training set and a test set for training an inversion network are generated; 2. The method according to claim 1, characterized in that, An inversion network based on a concurrent module CFTBlock is constructed, a nonlinear mapping from the gravity and magnetic anomaly data to the three-dimensional model matrix is established, and the inversion network is trained by using the gravity and magnetic anomaly training set in combination with a random noise adding strategy to obtain an inversion network with optimal weight parameters. The specific method for setting the source body modeling parameters of the gravity and magnetic anomaly field in the research area is as follows:

3. The method according to claim 1, characterized in that, According to the geological information of the research area, the lithology distribution and physical property values in different regions are counted, the range of the spatial position and the physical property values of the exploration target is obtained, and the physical property interval and the distribution range, which are the gravity and magnetic anomaly body modeling parameters, are set when the gravity and magnetic anomaly field simulation is performed; meanwhile, the observation plane and the terrain information are set according to the measurement condition. The specific method for selecting a rectangular area in the research area, generating Berlin noise, performing hard thresholding, and dividing the rectangular area into multiple independent two-dimensional blocks is as follows: Subsequently, according to whether each non-0 element has non-zero elements in its upper, lower, left and right four adjacent positions, the entire rectangular region is divided into N independent two-dimensional blocks A which are not connected to each other 1 … A N .

4. The method according to claim 3, characterized in that, The size of the rectangular area to be generated with Berlin noise is set in the research area, and Berlin noise is generated; hard thresholding refers to setting the elements with a value lower than a preset minimum physical property threshold alpha in the two-dimensional matrix noise to 0. The specific method for generating a three-dimensional model matrix for each independent block and obtaining the final three-dimensional model matrix M capable of describing multiple abnormal bodies underground is as follows: For each independent two-dimensional block A l Create a three-dimensional model matrix mt with an initial state of all zeros l , l = 1, 2,..., N , whose dimensions are A continuous underground geometric space is discretized into a limited number of small units, and a three-dimensional matrix is used to describe the distribution and physical property values of the gravity and magnetic anomaly bodies in the underground space, so that the three-dimensional model matrix is obtained. × nx × ny , nz , nx , ny respectively represent the number of grid units in the X, Y, Z directions of the study area; define mt l,n A l corresponding three-dimensional model matrix mt l two-dimensional slice of the block A l at the depth layer n , n = 1, 2,..., nz ; randomly generate a depth layer number n , 1≤ n ≤ nz ; copy the data of the two-dimensional block A l to the two-dimensional slice mt l,n of its corresponding three-dimensional model matrix mt l as the initial core layer of the anomaly body of the block; take this core layer n as the starting point and simultaneously diffuse to all depth layers in the n decrease direction and n increase direction; This diffusion process eventually generates a three-dimensional model matrix mt for each independent two-dimensional block A l Generate a three-dimensional model matrix mt l ; all independent two-dimensional block generated three-dimensional model matrix mt l Merging, get the final three-dimensional model matrix M which can describe multiple underground anomaly bodies.

5. The method according to claim 4, characterized in that, When the initial core layer of the block anomaly body spreads upward or downward, each element in the two-dimensional matrix of the core layer is first reduced by the attenuation value β, and then the current attenuation coefficient nz The results are reduced as a whole, and finally adjusted in the horizontal direction according to the displacement step to generate a stepped or inclined model.

6. The method according to claim 5, wherein, gamma The method adopts a parallel computing technology to batch generate three-dimensional model matrices describing gravity and magnetic anomaly bodies with different spatial distributions and physical property characteristics, and the specific method is as follows: Each thread independently completes the complete construction process of a single three-dimensional model matrix, including Berlin noise generation, independent block division, three-dimensional model matrix diffusion calculation and finally outputting the three-dimensional model matrix M, so that the three-dimensional model matrices generated by different threads are independent and diverse in spatial distribution and physical property characteristics.

7. The method according to claim 6, characterized in that, The method constructs a boundary reconstruction module and a fast forward network in combination with a CNN; The fast forward network takes ResUNet as a backbone architecture, is composed of an encoder for feature extraction and data compression and a decoder for forward calculation and data reconstruction, and after feature reconstruction at each layer of the decoder, adopts a boundary reconstruction module based on a double residual mechanism for feature enhancement reconstruction, further improving the feature reconstruction accuracy; The boundary recognition module first extracts primary boundary features through a residual block; then introduces residual connection to fuse the original features and the reconstructed features, retaining key feature information; Finally, a secondary feature reconstruction is implemented through a composite output layer containing a 3*3 convolution and a 1*1 convolution.

8. The method according to claim 7, characterized in that, The specific method for generating the gravity-magnetic anomaly training set and test set for training the inversion network is: According to the existing geological and geophysical data and prior information of the target exploration area, 1x10 4 ~1x10 5 The three-dimensional model matrix is consistent with the geological characteristics of the area; then, the rapid forward network with the optimal weight parameter is used to calculate the gravity and magnetic anomaly data corresponding to the three-dimensional model matrix, thereby forming the training data set of the inversion network.

9. The method according to claim 8, characterized in that, The inversion network is composed of an input layer, a core processing layer and an output layer; wherein the input layer is used to receive gravity-magnetic anomaly data, the core processing layer is composed of an encoder-decoder containing four encoding-decoding blocks; the inversion network extracts high-dimensional features of gravity-magnetic anomaly data through three layers of downsampling, each encoding-decoding block integrates multiple concurrent modules CFTBlock, and jump connection is introduced to realize multi-scale feature fusion, effectively retaining the detail information of anomaly data; finally, through the 1*1 convolution of the output layer, the feature mapping output by the decoder is converted into a three-dimensional model matrix, which can reflect the spatial form and physical characteristics of the underground anomaly body.

10. The method according to claim 9, characterized in that, The concurrent module CFTBlock includes a CNN branch and a Transformer branch, the CNN branch adopts a two-stage design, gradually extracting and enhancing local features: the first stage is composed of two convolution blocks and residual connection, capturing local features through convolution operation and alleviating the gradient vanishing problem through residual connection; the second stage is to fuse the global features of the Transformer branch on the basis of convolution and residual connection, realizing the combination of local features and global context; The Transformer branch is composed of a multi-head self-attention module and an MLP module; the multi-head self-attention mechanism is used to capture global features; The MLP module performs nonlinear transformation on the output of the self-attention module, further enhancing the feature expression ability.

Images