A source body positioning method based on intelligent gravity-magnetic field separation and Euler deconvolution

Abstract

The present application belongs to the technical field of gravity and magnetic exploration data processing, and relates to a source body positioning method based on intelligent gravity-magnetic field separation and Euler deconvolution, comprising: setting gravity-magnetic anomaly field modeling parameters according to the geological and physical data of a study area; generating an initial seed point and expanding growth in a three-dimensional underground grid to generate an independent anomaly field model; Q Mapping the model position to a real coordinate and calculating a forward anomaly field data set; extracting the global absolute maximum value of the data set and sorting in ascending order, and taking the first p superposition to generate a composite field, taking the first p as a target field and the composite field to construct a field separation data set; training the field separation network with the field separation data set until convergence; inputting the measured data to obtain the separation field and the residual field data, and selecting the data for iterative separation according to the requirements to obtain the field separation result; using the field separation result to obtain a preliminary positioning result by a joint Euler deconvolution method, and determining the final source body positioning after screening. The method has the beneficial effect of realizing multi-scale geological target field separation and determining the source body positioning.

Description

Technical Field

[0001] This invention relates to the field of gravity and magnetic exploration data processing technology, and in particular to a source body localization method based on intelligent gravity and magnetic field separation and Euler deconvolution. Background Technology

[0002] The purpose of gravity and magnetic data source location is to obtain information such as the distribution location and structural morphology of target geological bodies based on gravity or magnetic data obtained in the field. Eulerian deconvolution is an automated calculation method that can determine the location and depth of geological bodies using gravity and magnetic grid data. It can be used to calculate parameters such as the burial depth and location of geological bodies and is widely used in various resource exploration projects. However, due to the complex types of mineral alteration and varied geological structures in the deep crust, the obtained gravity and magnetic anomaly parameters usually exhibit the superposition effect of anomalies from different target geological bodies. This makes it difficult for the Eulerian deconvolution method to accurately determine the boundary range and morphology of underground ore bodies, reducing the accuracy of gravity and magnetic anomaly field interpretation and analysis.

[0003] Traditionally, field separation in gravity and magnetic exploration data primarily involves distinguishing between residual gravity and magnetic anomalies and regional gravity and magnetic anomalies. However, with the advancement of related research, issues such as improving the accuracy of field separation urgently need to be addressed. Therefore, a multi-scale series method has been proposed, which, in addition to regional gravity and magnetic anomalies, also separates anomaly information generated by different underground geological bodies. Multi-scale weak anomalies are relative; for example, between two geological bodies with similar scales but different depths, the anomaly generated by the deeper geological body can be considered a weak anomaly; conversely, between two geological bodies with the same depth but different scales, the anomaly generated by the smaller-scale geological body is considered a weak anomaly.

[0004] Traditional separation methods, such as analytical continuation, wavelet multi-scale analysis, and polynomial fitting, primarily focus on regional and local fields, with less emphasis on residual anomalies. Traditional methods for field separation and weak anomaly extraction, such as wavelet multi-scale analysis, blind source separation, and empirical mode decomposition, rely on threshold settings in experiments, which are susceptible to subjective human influence. Therefore, further research is needed on field separation of residual fields and the potential value of residual anomalies. Summary of the Invention

[0005] Technical problems to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a source volume localization method based on intelligent gravity and magnetic field separation and Euler deconvolution. It solves the technical problem that traditional separation methods are mostly aimed at regional and local field studies and rarely involve residual anomalies, and that methods that can perform field separation and weak anomaly extraction rely on threshold settings that are affected by subjective factors.

[0007] Technical solution

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] This invention provides a source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution, comprising the following steps:

[0010] Step 1: Based on the geological and physical property data of the study area, set the modeling parameters for the gravity and magnetic anomaly field;

[0011] Step 2: Randomly generate multiple initial seed points in the underground three-dimensional grid space of the study area, and expand them in random directions according to preset rules until Q independent anomaly field random three-dimensional models are generated.

[0012] Step 3: Map the random 3D model positions to real physical coordinates, calculate for each model, and obtain the forward modeling anomaly field dataset;

[0013] Step 4: Extract the maximum global absolute value of the forward anomaly dataset, and sort all data in ascending order accordingly to obtain an ordered model sequence. ;

[0014] Step 5: Extracting the ordered model sequence before p The superposition of data generates a composite field. , with the first The data is for separating the target field. This constitutes a field separation dataset. ;

[0015] Step 6: Use field separation dataset The training environment separates the network until the network reaches the preset convergence criterion;

[0016] Step 7: Input the collected measured gravity and magnetic exploration data into the trained field separation network, output the separated field data, and calculate the remaining field data as the difference between the measured gravity and magnetic exploration data and the separated field data;

[0017] Step 8: Select the separated field data or residual field data as new input according to the geological interpretation requirements, input the trained field separation network for iterative separation, and obtain the field separation results;

[0018] Step 9: Based on the field separation results, use the joint Euler deconvolution method to solve for the preliminary source localization results;

[0019] Step 10: Filter the preliminary source location results and finally determine the source location results.

[0020] As a further improvement of the present invention, step 1 specifically involves: based on the geological mapping data and physical property database of the study area, statistically analyzing the density and magnetic susceptibility ranges of different lithological zones, setting the physical property parameter range, initial seed size, and maximum growth volume constraints for anomaly field modeling; and defining the observation plane position and grid resolution.

[0021] The anomalous field modeling parameters include gravity gradient data, gravity anomaly data, magnetic anomaly data, and magnetic gradient data.

[0022] As a further improvement of the present invention, the step 2 of generating a stochastic three-dimensional model of Q independent anomaly fields includes the following steps:

[0023] Step 2.1: Build a growth path stack for each initial seed and record the current growth coordinates;

[0024] Step 2.2: Determine whether the current growth coordinates of the initial seed meet the growth conditions;

[0025] Specifically, if the neighboring grids in the initial seed's underground 3D grid space are not occupied and are within the preset growth space, they are marked as growable areas; if growth is blocked, the process reverts to the previous growth node and selects a new growth direction.

[0026] Step 2.3: Expand the seed volume in a random direction within the legal growth space until the preset number of growths or the upper limit of the volume is reached;

[0027] Step 2.4: Repeat the above steps until Q independent stochastic three-dimensional models of anomalies are generated.

[0028] As a further improvement of the present invention, in step 3, the model obtained by growing the random seed in a three-dimensional mesh is transformed to a real physical space location, and calculated according to the following formula:

[0029] ;

[0030] In the formula, and They represent the first, second, and third elements in the random three-dimensional model, respectively. The actual physical location and grid location corresponding to each grid point This represents the spacing between the meshes in the real space. This represents the minimum value of the grid point coordinates in a certain direction. Each represents a different coordinate axis. This represents the coordinates of a grid point in a certain direction.

[0031] As a further improvement of the present invention, in step 3, gravity or magnetic forward modeling is performed on each random three-dimensional model to obtain the corresponding forward modeling anomaly field data.

[0032] As a further improvement of the present invention, the specific method for constructing the field separation dataset in step 5 is: taking the first... One normal modeling anomaly field data Superposition generates composite superposition field As network input, with the first A single, separate target field As the target output Calculate according to the following formula:

[0033] ;

[0034] Composite superposition field and separation target field Constructing a field separation dataset The training set and validation set are divided proportionally.

[0035] As a further improvement of the present invention, step 6 utilizes a field separation dataset. The training ground separation network is specifically as follows:

[0036] Construct a field separation network, establish a nonlinear mapping from the composite superimposed field to the target separation field through the field separation network, and optimize the weight parameters of the field separation network using the partitioned training set and validation set until the field separation network reaches the preset convergence criterion;

[0037] The field separation network consists of a field separation estimation network and a field separation sub-network. The field separation estimation network consists of multiple convolutional layers and activation functions, used to extract multi-scale features of the composite superimposed field. The field separation sub-network consists of an encoder for feature extraction and data compression and a decoder for data reconstruction.

[0038] The calculation formula for the convolutional layer of the field separation estimation network is as follows:

[0039] ;

[0040] ;

[0041] ;

[0042] In the formula, , Indicates the current convolutional layer Input data, This represents the output of the current convolutional layer after performing a convolution operation on the input data; This represents the weight values ​​of the current convolutional layer. Indicates the number of channels for the input data; , These represent the width and depth of the current convolutional layer, respectively. and This represents the stride and padding values ​​used in the convolution operation on the input data. This represents the bias coefficient of the current convolutional layer; Represents a convolutional layer Size, It can be represented as:

[0043] ;

[0044] In the formula, This represents the activation function applied during the current model training.

[0045] As a further improvement of the present invention, step 7 specifically involves: inputting the collected measured gravity and magnetic exploration data into the trained field separation network, outputting the separated field data, and calculating the difference between the measured gravity and magnetic exploration data and the separated field data as the remaining field data according to the following formula:

[0046] ;

[0047] In the formula, For the remaining field data, The collected measured gravity and magnetic exploration data, This is the data for the separation field.

[0048] As a further improvement of the present invention, in step 9, based on the field separation results, a joint Euler deconvolution method is used for solving, specifically as follows:

[0049] A joint Euler deconvolution method for gravity and magnetic anomaly data is proposed by calculating the vertical derivative. The formula is based on the Euler homogeneous equation.

[0050] ;

[0051] in, and These are the coordinates of the observation point and the center of the geological body, respectively. To construct the index, and These are background field and gravity / magnetic anomaly data, respectively.

[0052] Transformed into a vector product form:

[0053] ;

[0054] To reduce interference from the choice of the construction exponent N and the background field B, second-order derivatives are taken in the x, y, and z directions to derive the joint Euler deconvolution formula for gravity gradient data or magnetic gradient data:

[0055] ;

[0056] Transformed into a vector product form:

[0057] ;

[0058] The equations are solved using the sliding window method, with each sliding window providing the corresponding solution. This is the preliminary result of the source localization. The number of solutions.

[0059] The beneficial effects of this invention are:

[0060] It can achieve field separation of geological targets of different scales, depths and sizes, and finally determine the source body location result; the method was applied to theoretical models and measured data to verify the feasibility of the algorithm.

[0061] It has the advantages of rapid and accurate field separation, can effectively identify the boundary and shape information of underground anomalies, and is suitable for the interpretation of measured gravity and magnetic data. It is of great significance for improving the processing and interpretation accuracy of geophysical gravity and magnetic field data. Attached Figure Description

[0062] Figure 1 A flowchart of a source localization method based on intelligent gravity and magnetic field separation and Euler deconvolution is provided for an embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the spatial location of the model generated by the random generation method in this embodiment;

[0064] Figure 3 This is a diagram showing the spatial location and physical properties of the theoretical model in this embodiment;

[0065] Figure 4 This is the superimposed observation data generated by Model I-1 and Model I-2 in this embodiment;

[0066] Figure 5 This refers to the observation data generated by I-1 in this embodiment;

[0067] Figure 6 This refers to the observation data generated by I-1 in this embodiment;

[0068] Figure 7 This refers to the abnormal data of I-1 obtained by field separation in this embodiment;

[0069] Figure 8 This refers to the abnormal data of I-2 obtained by field separation in this embodiment;

[0070] Figure 9 This is the processed magnetic anomaly data in this embodiment;

[0071] Figure 10This is the actual magnetic anomaly data field separation result in this embodiment;

[0072] Figure 11 The source volume result is obtained by performing the Euler deconvolution method on the field separation result in this embodiment. Detailed Implementation

[0073] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0075] In this embodiment, a source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution is described, such as... Figure 1 As shown, it includes the following steps:

[0076] Step 1: Based on the geological and physical property data of the study area, set the modeling parameters for the gravity and magnetic anomaly field;

[0077] Specifically, based on geological mapping data and physical property database of the study area, the density and magnetic susceptibility ranges of different lithological zones are statistically analyzed, and the physical property parameter ranges, initial seed size and maximum growth volume constraints for gravity and magnetic anomaly field modeling are set; the observation plane position and grid resolution are defined.

[0078] The gravity and magnetic anomaly field parameters include gravity gradient data, gravity anomaly data, magnetic anomaly data, and magnetic gradient data.

[0079] In this embodiment, parameters are set for the region where the subsequently used measured data is located, and the remaining density generation range is set to [-1.0, 1.0] g / cm³. 3 The residual magnetization intensity generation range is set to [1, 100] A / m, the initial seed size is set to 50m×50m×50m, and the number of seeds to be grown is set to [2, 5].

[0080] Step 2: Randomly generate multiple initial seed points in the underground three-dimensional grid space of the study area, and expand them in random directions according to preset rules until Q independent anomaly field random three-dimensional models are generated.

[0081] Step 2.1: Build a growth path stack for each initial seed and record the current growth coordinates;

[0082] Step 2.2: Determine whether the current growth coordinates of the initial seed meet the growth conditions;

[0083] Specifically, if the neighboring grids in the initial seed's underground 3D grid space are not occupied and are within the preset growth space, they are marked as growable areas; if growth is blocked, the process reverts to the previous growth node and selects a new growth direction.

[0084] Step 2.3: Expand the seed volume in a random direction within the legal growth space until the preset number of growths or the upper limit of the volume is reached;

[0085] Step 2.4: Repeat the above steps until Q (number of seeds × number of times a single seed grows) independent stochastic three-dimensional models of anomalies are generated.

[0086] In this embodiment, for ease of calculation, the continuous underground geometric space is discretized into a finite number of small units, i.e., finite element meshing is used to divide the three-dimensional underground space of the study area into a 64×64×20 mesh; the generated model schematic diagram is shown below. Figure 2 As shown. The seed number generation range is [2, 5], and the single seed growth count range is [16, 81].

[0087] Step 3: Map the random 3D model positions to real physical coordinates, calculate for each model, and obtain the forward modeling anomaly field dataset;

[0088] Specifically, the model obtained by growing a random seed in a 3D mesh undergoes coordinate transformation to its real physical location, calculated according to the following formula:

[0089]

[0090] In the formula, and They represent the first, second, and third elements in the random three-dimensional model, respectively. The actual physical location and grid location corresponding to each grid point This represents the spacing between the meshes in the real space. This represents the minimum value of the grid point coordinates in a certain direction. They represent different directions. This represents the coordinates of a grid point in a certain direction.

[0091] The random 3D model is generated using virtual coordinates, which need to be converted to real physical coordinates before subsequent forward modeling calculations can be performed. Each random 3D model is discretized into a grid, and a gravity or magnetic forward modeling algorithm based on discrete integral equations is used to calculate the forward modeling anomaly field data generated by the discretized model on the observation plane.

[0092] Step 4: Extract the maximum global absolute value of the forward anomaly dataset, and sort all data in ascending order accordingly to obtain an ordered model sequence. ;

[0093] Specifically, the maximum value of the global absolute value of the anomaly data is extracted, and all forward-modeled anomaly data are sorted in ascending order according to this maximum value to generate an ordered model sequence. .

[0094] In this embodiment, the forward modeling used is calculated in a Cartesian coordinate system, with the coordinates ranging from 0m to 3200m, 0m to 3200m, and 0m to 1000m along the X, Y, and Z axes, respectively. The distance between the measuring point and the measuring line is 50m. Gravity and magnetic anomalies and gradient data are calculated for the randomly grown anomaly model.

[0095] Step 5: Extracting the ordered model sequence before The superposition of data generates a composite field. , with the first The data is for separating the target field. This constitutes a field separation dataset. ;

[0096] Specifically, take the first... One normal modeling anomaly field data Superposition generates composite superposition field As network input, with the first A single, separate target field As the target output Calculate according to the following formula:

[0097]

[0098] Composite superposition field and separation target field Constructing a field separation dataset Divided according to a preset ratio Generate training and validation sets.

[0099] In this embodiment, each anomalous field data generates 2×10 4 The dataset was divided into a training set and a validation set in a 4:1 ratio.

[0100] Step 6: Use the field separation dataset The training environment separates the network until the network reaches the preset convergence criterion;

[0101] Specifically, first, a field separation network is constructed, and a nonlinear mapping from the composite superimposed field to the target separation field is established through the field separation network. The weight parameters of the field separation network are optimized using the divided training set and validation set until the field separation network reaches the preset convergence criterion.

[0102] The field separation network consists of a field separation estimation network and a field separation sub-network. The field separation estimation network consists of multiple convolutional layers and activation functions, used to extract multi-scale features of the composite superimposed field. The field separation sub-network consists of an encoder for feature extraction and data compression and a decoder for data reconstruction.

[0103] The calculation formula for the convolutional layer of the field separation estimation network is as follows:

[0104]

[0105]

[0106]

[0107] In the formula, , Indicates the current convolutional layer Input data, This represents the output of the current convolutional layer after performing a convolution operation on the input data. This represents the weight values ​​of the current convolutional layer. This indicates the number of channels for the input data. , These represent the width and depth of the current convolutional layer, respectively. and This represents the stride and padding values ​​used in the convolution operation on the input data. This represents the bias coefficient of the current convolutional layer. Represents a convolutional layer Size, It can be represented as:

[0108]

[0109] In the formula, This represents the activation function applied to the currently trained network.

[0110] In this embodiment, the field separation estimation network consists of 5 convolutional layers and 5 ReLU nonlinear activation function layers to separate the field dataset. The estimation and preliminary feature extraction of the data are performed. The estimation results and preliminary features are then input into a field separation sub-network. In this network, the encoder gradually reduces the data size through multiple downsampling operations. The high-dimensional features are then input into the decoder, where the data size is gradually restored through upsampling operations. Finally, the result is output through a 1×1 convolutional output layer. The parameters used are: learning rate 1×10⁻⁶. -3 The learning rate decay strategy uses an equal-interval adjustment of the learning rate, reducing it by half every 200 iterations. The batch size is 128, the number of iterations is 200, the mean squared error (MSE) is used as the loss function, and the adaptive moment estimation (Adam) is used as the optimizer. When the validation set loss is continuous... Training terminates when the number of iterations has not decreased or the target number of iterations has been reached, and the optimal weights are saved, thus completing the training of the field separation network.

[0111] Step 7: Input the collected measured gravity and magnetic exploration data into the trained field separation network, output the separated field data, and calculate the remaining field data based on the difference between the measured gravity and magnetic exploration data and the separated field data;

[0112] Specifically, the collected measured gravity and magnetic exploration data are input into a trained field separation network, which outputs separated field data. The remaining field data is calculated as the difference between the measured data and the separated field data according to the following formula:

[0113]

[0114] In the formula, For the remaining field data, The collected measured gravity and magnetic exploration data, This is the data for the separation field.

[0115] Step 8: Select the separated field data or residual field data as new input according to the geological interpretation requirements, input the trained field separation network for iterative separation, and obtain the field separation results;

[0116] Specifically, based on the needs of geological interpretation, or As new input data, it is fed into the trained field separation network, and the process is repeated to achieve iterative separation of multi-scale anomalies, thereby obtaining field separation results. Then, for each separation result, the next step of source localization is carried out.

[0117] The purpose of field separation mentioned here is to decompose the superimposed anomalies generated by multiple magnetic bodies into multiple independent magnetic anomaly fields. By calculating each individual anomaly field separately, the positioning accuracy can be improved.

[0118] In this embodiment, a gravity anomaly model different from the training and validation sets is established. Models I-1 and I-2 are obtained after 22 and 17 seed growth cycles, respectively, with densities of -0.24 g / cm³.3 With 0.17 g / cm 3 The burial depths are 750m and 550m respectively. Their spatial locations are as follows: Figure 3 As shown. Calculate its gravity anomaly data V. z And the field separation network was used for testing, and the results are as follows: Figure 4-8 As shown in Table 1, the mean square error of the field separation results and the observed anomalies generated by theoretical models I-1 and I-2 is as follows. From the image distribution, the field separation results of models I-1 and I-2 maintain a good correspondence with the observed data in terms of spatial location, amplitude, and anomaly distribution. Subsequently, for the gravity gradient data of model I... , , , , , With gravity anomaly data Quantitative analysis using the mean squared error shows that the separation results are relatively close to the actual data. This indicates that the training results achieve good separation even with a dispersed model.

[0119] Table 1 shows the mean square error between the separation results of Model I and the actual data.

[0120]

[0121] To verify the practicality of the method of this invention, magnetic anomaly data from Macheng area, Hebei Province, China, were used for testing. A test area of ​​approximately 3200m × 3200m was selected for the experimental data. The underground area was divided into 64 × 64 × 20 small cuboids, each with dimensions of 50 × 50 × 50m. 3 After undergoing diurnal variation correction, IGRF correction, and polarization treatment, the processed magnetic anomaly data can be found in [link to data]. Figure 9 .

[0122] Field separation results are as follows Figure 10 As shown, the method proposed in this invention is used to extract the main anomalies from magnetic anomaly data in Macheng area, Hebei Province.

[0123] Step 9: Based on the field separation results, use the joint Euler deconvolution method to solve for the preliminary source localization results;

[0124] Specifically, a joint Euler deconvolution method for gravity and magnetic anomaly data is proposed by calculating the vertical derivative. The formula is based on the Euler homogeneous equation:

[0125] (8)

[0126] in, and These are the coordinates of the observation point and the center of the geological body, respectively. To construct the index, and These are background field and gravity / magnetic anomaly data, respectively.

[0127] Transformed into a vector product form:

[0128] (9)

[0129] To reduce interference from the choice of the construction exponent N and the background field B, second-order derivatives are taken in the x, y, and z directions to derive the joint Euler deconvolution formula for gravity gradient data or magnetic gradient data:

[0130] (10)

[0131] Transformed into a vector product form:

[0132] (11)

[0133] The equations are solved using the sliding window method, with each sliding window providing the corresponding solution. This is the preliminary result of the source localization. The number of solutions.

[0134] Step 10: Filter the preliminary source location results and finally determine the source location results;

[0135] Specifically, the solution obtained by solving the Euler deconvolution equation is used to perform validity screening to obtain the final result, which is the final determined source body localization result. The specific validity screening methods include, but are not limited to, horizontal gradient filtering, main body anomaly distance criterion, and clustering and divergence criterion.

[0136] In this embodiment, the horizontal gradient filtering coefficient is set to 1, the convergence index is set to 6, and the radius of action of the convergence criterion is set to 1.5 times the survey line spacing.

[0137] The source volume was located using the Euler deconvolution method based on the field separation results. (See attached image.) Figure 11 It reflects the outline and boundary of underground anomalies.

[0138] Through the above calculation process, the source body localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution proposed in this invention can perform rapid and accurate field separation, highlight the main anomaly field data, and the proposed Eulerian deconvolution joint method can reflect the spatial location of underground anomalies, which is beneficial to improving the application of gravity and magnetic exploration in practical work.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A source volume localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution, characterized in that, Includes the following steps: Step 1: Based on the geological and physical property data of the study area, set the modeling parameters for the gravity and magnetic anomaly field; Step 2: Randomly generate multiple initial seed points in the underground three-dimensional grid space of the study area, and expand them in random directions according to preset rules until Q independent anomaly field random three-dimensional models are generated. Step 3: Map the random 3D model positions to real physical coordinates, calculate for each model, and obtain the forward modeling anomaly field dataset; Step 4: Extract the maximum global absolute value of the forward anomaly dataset, and sort all data in ascending order accordingly to obtain an ordered model sequence. ; Step 5: Extract the first p data points of the ordered model sequence and superimpose them to generate a composite field. , with the first The data is for separating the target field. This constitutes a field separation dataset. ; Step 6: Use field separation dataset The training environment separates the network until the network reaches the preset convergence criterion; Step 7: Input the collected measured gravity and magnetic exploration data into the trained field separation network, output the separated field data, and calculate the remaining field data as the difference between the measured gravity and magnetic exploration data and the separated field data; Step 8: Select the separated field data or residual field data as new input according to the geological interpretation requirements, input the trained field separation network for iterative separation, and obtain the field separation results; Step 9: Based on the field separation results, use the joint Euler deconvolution method to solve for the preliminary source localization results; Step 10: Filter the preliminary source location results and finally determine the source location results.

2. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 1, characterized in that, Specifically, step 1 involves: based on geological mapping data and physical property database of the study area, statistically analyzing the density and magnetic susceptibility ranges of different lithological zones, and setting the physical property parameter ranges, initial seed size, and maximum growth volume constraints for anomaly field modeling. Define the observation plane location and grid resolution; The anomalous field modeling parameters include gravity gradient data, gravity anomaly data, magnetic anomaly data, and magnetic gradient data.

3. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 2, characterized in that, The step 2 of generating a stochastic three-dimensional model of Q independent anomaly fields includes the following steps: Step 2.1: Build a growth path stack for each initial seed and record the current growth coordinates; Step 2.2: Determine whether the current growth coordinates of the initial seed meet the growth conditions; Specifically, if the neighboring grids in the initial seed's underground 3D grid space are not occupied and are within the preset growth space, they are marked as growable areas; if growth is blocked, the process reverts to the previous growth node and selects a new growth direction. Step 2.3: Expand the seed volume in a random direction within the legal growth space until the preset number of growths or the upper limit of the volume is reached; Step 2.4: Repeat the above steps until Q independent stochastic three-dimensional models of anomalies are generated.

4. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 3, characterized in that, In step 3, the model obtained by growing the random seed in a 3D mesh is transformed to its real physical space location using the following formula: ； In the formula, and They represent the first, second, and third elements in the random three-dimensional model, respectively. The actual physical location and grid location corresponding to each grid point This represents the spacing between the meshes in the real space. This represents the minimum value of the grid point coordinates in a certain direction. Each represents a different coordinate axis. This represents the coordinates of a grid point in a certain direction.

5. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 4, characterized in that, In step 3, gravity or magnetic forward modeling is performed on each random three-dimensional model to obtain the corresponding forward modeling anomaly field data.

6. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 5, characterized in that, The specific steps in step 5 to construct the field separation dataset are as follows: Take the first... One normal modeling anomaly field data Superposition generates composite superposition field As network input, with the first A single, separate target field As the target output Calculate according to the following formula: ； Composite superposition field and separation target field Constructing a field separation dataset The training set and validation set are divided proportionally.

7. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 6, characterized in that, Step 6 utilizes field separation datasets The training ground separation network is specifically as follows: Construct a field separation network, establish a nonlinear mapping from the composite superimposed field to the target separation field through the field separation network, and optimize the weight parameters of the field separation network using the partitioned training set and validation set until the field separation network reaches the preset convergence criterion; The field separation network consists of a field separation estimation network and a field separation sub-network. The field separation estimation network consists of multiple convolutional layers and activation functions, used to extract multi-scale features of the composite superimposed field. The field separation sub-network consists of an encoder for feature extraction and data compression and a decoder for data reconstruction. The calculation formula for the convolutional layer of the field separation estimation network is as follows: ； ； ； In the formula, , Indicates the current convolutional layer Input data, This represents the output of the current convolutional layer after performing a convolution operation on the input data; This represents the weight values ​​of the current convolutional layer. Indicates the number of channels for the input data; , These represent the width and depth of the current convolutional layer, respectively. and This represents the stride and padding values ​​used in the convolution operation on the input data. This represents the bias coefficient of the current convolutional layer; Represents a convolutional layer Size, It can be represented as: ； In the formula, This represents the activation function applied during the current model training.

8. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 7, characterized in that, Step 7 specifically involves: inputting the collected measured gravity and magnetic exploration data into the trained field separation network, outputting the separated field data, and calculating the difference between the measured gravity and magnetic exploration data and the separated field data to obtain the remaining field data according to the following formula: ； In the formula, For the remaining field data, The collected measured gravity and magnetic exploration data, This is the data for the separation field.

9. The source localization method based on intelligent gravity and magnetic field separation and Eulerian deconvolution according to claim 7, characterized in that, In step 9, based on the field separation results, a joint Euler deconvolution method is used for solving the problem, specifically as follows: A joint Euler deconvolution method for gravity and magnetic anomaly data is proposed by calculating the vertical derivative. The formula is based on the Euler homogeneous equation. ； in, and These are the coordinates of the observation point and the center of the geological body, respectively. To construct the index, and These are background field and gravity / magnetic anomaly data, respectively. Transformed into a vector product form: ； To reduce interference from the choice of the construction exponent N and the background field B, second-order derivatives are taken in the x, y, and z directions to derive the joint Euler deconvolution formula for gravity gradient data or magnetic gradient data: ； Transformed into a vector product form: ； The equations are solved using the sliding window method, with each sliding window providing the corresponding solution. This is the preliminary result of the source localization. The number of solutions.

Images