Deep prospecting method and system based on electromagnetic sounding data fusion and imaging

Abstract

The present invention discloses a deep prospecting method and system based on electromagnetic sounding data fusion and imaging, which relates to the technical field of deep prospecting. The method and system comprise the following steps: using an electromagnetic source to excite a broadband electromagnetic signal to obtain an electromagnetic response tensor, and performing a differential geometry transformation on the electromagnetic response tensor to obtain an extended response tensor field; performing feature mapping on the extended response tensor field, and performing cluster analysis on the mapping results to obtain a number of electrical mutation areas; using a preset geometric reconstruction algorithm to extract target area coordinate sets of the several electrical mutation areas, and using a greedy algorithm to generate a number of initial detection range boundaries; constructing a multi-objective optimization function, using the NSGA-I algorithm to optimize and screen the several initial detection range boundaries to obtain an optimal detection range, and performing deep prospecting. The method solves the problem that traditional methods cannot accurately construct target area contours, resulting in an overly large or overly small detection range, and thus reducing the accuracy of identifying deep ore bodies.

Description

Technical Field

[0001] The present invention relates to the technical field of deep prospecting, and more particularly to a deep prospecting method and system based on electromagnetic sounding data fusion and imaging. Background Art

[0002] Against the backdrop of increasingly limited mineral resources, deep prospecting has become a crucial area of ​​mineral exploration. However, as exploration depths continue to increase, significant bottlenecks are emerging in data analysis accuracy, imaging depth, and anomaly identification capabilities. Especially under complex geological conditions, electromagnetic sounding methods are susceptible to background noise, electrical interference, and nonlinear data distribution. This can lead to unclear anomaly identification and inaccurate target location, severely restricting the efficiency and success rate of deep prospecting.

[0003] Traditional methods lack an automatic target extraction mechanism. Especially when multiple anomalies overlap or their boundaries are blurred, the target area outline cannot be accurately constructed, resulting in a detection range that is too large or too small, which in turn reduces the accuracy of deep ore body identification.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a deep mineral exploration method and system based on electromagnetic sounding data fusion and imaging. By extracting the electrical mutation target area and optimizing the detection boundary, it solves the problem that the traditional method cannot accurately construct the target area contour, resulting in a detection range that is too large or too small, and thus reducing the accuracy of deep ore body identification.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A deep prospecting method based on electromagnetic sounding data fusion and imaging includes the following steps: using an electromagnetic source to excite a broadband electromagnetic signal to obtain an electromagnetic response tensor, and performing a differential geometric transformation on the electromagnetic response tensor to obtain an extended response tensor field; performing feature mapping on the extended response tensor field, and performing cluster analysis on the mapping results to obtain a number of electrical mutation areas; using a preset geometric reconstruction algorithm to extract target area coordinate sets of the several electrical mutation areas, and using a greedy algorithm to generate a number of initial detection range boundaries; constructing a multi-objective optimization function, and using the NSGA-I algorithm to optimize and screen the several initial detection range boundaries to obtain the optimal detection range, and then conducting deep prospecting.

[0008] In a preferred embodiment, the electromagnetic source is used to excite a broadband electromagnetic signal, an electromagnetic response tensor is obtained, and a differential geometry transformation is performed on the electromagnetic response tensor to obtain an extended response tensor field. Specifically, the method comprises: deploying a multi-band electromagnetic emission source array to excite a broadband electromagnetic pulse sequence according to a preset timing; collecting electromagnetic field component data through distributed receiving nodes and constructing an electromagnetic response tensor; calculating the covariant derivative of the electromagnetic response tensor to obtain a Riemann connection coefficient matrix; and performing a curvature transformation on the electromagnetic response tensor based on the Riemann connection coefficient matrix to obtain an extended response tensor field.

[0009] In a preferred embodiment, the extended response tensor field is subjected to feature mapping, specifically: performing eigenmode decomposition on the extended response tensor field to obtain several modal components; calculating the geodesic distance between each spatial point tensor in the several modal components to construct a distance matrix; constructing an adjacency graph based on the distance matrix; using spectral graph theory, calculating the Laplace matrix of the adjacency graph and performing eigendecomposition to obtain several eigenvectors.

[0010] In a preferred embodiment, the mapping results are clustered and analyzed to obtain several electrical mutation areas, specifically: a first feature space is constructed based on several feature vectors, and the spatial points are mapped into the first feature space to obtain a feature mapping point set; the feature mapping point set is modeled using a Gaussian mixture model to obtain several multivariate Gaussian distribution components; a clustering algorithm is used to perform cluster analysis on the several multivariate Gaussian distribution components to obtain several dense clusters; each dense cluster is reversely mapped back to a preset three-dimensional space to obtain an electrical mutation area.

[0011] In a preferred embodiment, the preset geometric reconstruction algorithm is used to extract the target coordinate sets of several electrical mutation areas, specifically: extracting the boundary point sets of several electrical mutation areas, and calculating the electrical gradient tensor of each boundary point; based on the electrical gradient tensor, using an adaptive threshold method to screen the boundary points to obtain several first boundary points; performing three-dimensional Delaunay triangulation on the several first boundary points to construct a triangular mesh model, wherein the triangular mesh model includes several tetrahedrons; taking each tetrahedron as a graph node, and judging whether several tetrahedrons share a triangular face, if two tetrahedrons share a triangular face, constructing a first-class edge between the corresponding graph nodes to construct an electrical mutation graph model; using an improved Louvain algorithm to perform graph clustering analysis on the electrical mutation graph model to obtain the target coordinate set of the electrical mutation area.

[0012] In a preferred embodiment, the improved Louvain algorithm is used to perform graph clustering analysis on the electrical mutation graph model to obtain the target coordinate set of the electrical mutation area, specifically: each graph node of the electrical mutation graph model is initialized as an independent community to obtain a first community set; each graph node is traversed, and the modularity gain of the independent community corresponding to the adjacent graph node is calculated; the graph nodes are moved in the first community set according to the modularity gain until the preset iteration condition is met, and the movement is stopped to obtain a second community set; the tetrahedron set corresponding to each community in the second community set is extracted, and the centroid coordinate point set of all tetrahedrons in the tetrahedron set is calculated to obtain the centroid coordinate point set of each community; the principal component analysis method is used to determine the best fitting plane for the centroid coordinate point set of each community; a three-dimensional surface function is constructed based on the best fitting plane, and data analysis is performed to identify ridges and branch points; the three-dimensional coordinate set of ridges and branch points is used as the target coordinate set of the electrical mutation area.

[0013] In a preferred embodiment, the greedy algorithm is used to generate several initial detection range boundaries, specifically: merging the target coordinate sets of all electrical mutation areas to construct a first target coordinate point cloud; extracting the ridges and branch points in the first target coordinate point cloud to construct an initial seed point set; initializing the first mineralized area with each seed point in the initial seed point set as the starting point; calculating the regional boundary contour energy of the mineralized first area, and iteratively expanding the regional boundary according to the contour energy to obtain an initial contour surface; discretizing the initial contour surface to obtain a three-dimensional boundary surface; and decomposing the three-dimensional boundary surface into components to obtain several initial detection range boundaries.

[0014] In a preferred embodiment, the multi-objective optimization function is constructed, and the NSGA-I algorithm is used to optimize and screen several initial detection range boundaries to obtain the optimal detection range. Specifically, the initial detection range boundaries are discretized to obtain a parameterized curve set; a multi-objective optimization function is constructed, and the NSGA-I algorithm is used to perform non-dominated sorting on the parameterized curve set, and a Pareto optimal boundary solution set is generated through an adaptive crossover mutation operation; the super volume occupancy ratio of the Pareto optimal boundary solution set is calculated, and a solution set quality evaluation matrix is ​​constructed based on the super volume occupancy ratio; the Pareto optimal boundary solution set is screened based on the solution set quality evaluation matrix to obtain the optimal detection range.

[0015] The technical effects and advantages of the deep prospecting method and system based on electromagnetic sounding data fusion and imaging of the present invention are as follows:

[0016] 1. The present invention can effectively extract electrical mutation target areas in deep geology by introducing key technologies such as differential geometry analysis, characteristic mapping of electromagnetic response tensor fields, graph structure reconstruction and multi-objective boundary optimization, and accurately generate the optimal detection boundary based on the greedy algorithm and NSGA-I optimization strategy, thereby solving the problem of setting the detection range too large or too small due to fuzzy target boundaries and inaccurate target area construction in traditional methods. This method not only improves the ability to identify electrical anomalies in complex geological structures, but also realizes the automatic construction of target area boundaries through high-dimensional feature space clustering and graph theory analysis, significantly improving the accuracy and robustness of target area construction. At the same time, through the introduction of multi-objective optimization functions, comprehensive consideration of multiple indicators such as coverage, boundary compactness and recognition reliability is taken into account to further improve the consistency between the detection range and the actual mineralization distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flow chart of the deep prospecting method based on electromagnetic sounding data fusion and imaging according to the present invention.

[0018] Figure 2 It is a structural schematic diagram of the deep prospecting system based on electromagnetic sounding data fusion and imaging of the present invention. DETAILED DESCRIPTION

[0019] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1, Figure 1 The present invention provides a deep prospecting method based on electromagnetic sounding data fusion and imaging, which includes the following steps:

[0021] S1, using an electromagnetic source to excite a broadband electromagnetic signal, obtaining an electromagnetic response tensor, and performing a differential geometric transformation on the electromagnetic response tensor to obtain an extended response tensor field;

[0022] In this example, an electromagnetic source is used to excite a broadband electromagnetic signal to obtain an electromagnetic response tensor, and a differential geometric transformation is performed on the electromagnetic response tensor to obtain an extended response tensor field, specifically:

[0023] Deploy a multi-band electromagnetic emission source array to stimulate a broadband electromagnetic pulse sequence according to a preset time sequence;

[0024] Collect electromagnetic field component data through distributed receiving nodes and construct electromagnetic response tensors;

[0025] Calculate the covariant derivatives of the electromagnetic response tensor to obtain the Riemann connection coefficient matrix;

[0026] The electromagnetic response tensor is subjected to curvature transformation based on the Riemann connection coefficient matrix to obtain the extended response tensor field.

[0027] For example, in a test area (approximately 5 km × 5 km) below the mineralization belt, an electromagnetic excitation source array and a distributed receiving node network are deployed, with the following specific parameters:

[0028] Excitation source type: three-component dipole antenna; excitation frequency range: 10 Hz–10 kHz, covering the low- to medium-frequency electromagnetic range; excitation sequence: sequential multi-frequency pulses, with 200 Hz intervals between each frequency band; number of receiving nodes: 144 (one every 500 m), arranged in a 12×12 grid; sampling rate: 10 kHz, sampling duration: 10 seconds; data collection: Each node collects six electromagnetic field components: Ex, Ey, Ez, Hx, Hy, and Hz. The collected data are Fourier transformed by frequency to construct a complex frequency-domain electromagnetic response tensor. Each receiving node generates a tensor for each of the ten frequency bands, for a total of 144 × 10 electromagnetic response tensors.

[0029] It should be noted that the Riemann connection coefficient matrix is ​​the core tool for implementing geometric transformations of the electromagnetic response tensor. It is used to describe the translation and variation of the electromagnetic tensor field in space. Specifically, the electromagnetic response tensor exhibits a certain variation pattern in space. This variation is not simply a simple derivative relationship but is also affected by the coupling between tensors and the spatial structure. Therefore, it is necessary to introduce the Riemann connection coefficients in differential geometry to characterize the covariant derivatives of the tensor along different directions. The Riemann connection coefficients are calculated from the local rate of change of the tensor field. They can quantify the geometric relationship between points in space, that is, the path of change of the tensor in non-Euclidean space. In this invention, the covariant derivatives are first calculated based on the rate of change of the electromagnetic response tensor in the three-dimensional space (x, y, z) directions, and the Riemann connection coefficient matrix corresponding to each spatial point is constructed based on this. Subsequently, these connection coefficients are further used to construct the curvature tensor, that is, to calculate the degree of "curvature" of the electromagnetic tensor field in space. This curvature tensor reflects the intrinsic variation characteristics of the local electromagnetic structure. Finally, the main eigenvalues ​​of the curvature tensor or its norm are extracted as the new extended response tensor field, forming a high-order transformation result of the original tensor field.

[0030] S2, feature mapping the extended response tensor field and clustering the mapping results to obtain several electrical mutation regions;

[0031] In this example, the extended response tensor field is feature mapped as follows:

[0032] Perform eigenmode decomposition on the extended response tensor field to obtain several modal components;

[0033] Calculate the geodesic distance between each spatial point tensor in several modal components and construct a distance matrix;

[0034] Construct an adjacency graph based on the distance matrix;

[0035] Using spectral graph theory, the Laplacian matrix of the adjacency graph is calculated and eigendecomposition is performed to obtain several eigenvectors.

[0036] In this example, the geodesic distance between each spatial point tensor in several modal components is calculated. The specific formula is as follows:

[0037]

[0038] in, is the geodesic distance between space point i and space point j in the modal tensor field, is the number of modal components, and is the tensor at spatial point i and spatial point j under mode k, is the Frobenius norm.

[0039] It should be noted that the geodesic distance is an important indicator for measuring the relative differences between different modal components in the extended response tensor field in the high-dimensional tensor space. Specifically, after obtaining a number of physically meaningful modal components by performing intrinsic mode decomposition on the extended response tensor field, it is necessary to further evaluate the similarity of these modes at different spatial points. Since these modal components essentially constitute a high-dimensional tensor structure, the traditional Euclidean distance cannot effectively characterize their intrinsic geometric differences. Therefore, the geodesic distance is introduced to describe the shortest path between tensors on the tensor manifold, that is, to calculate their true relative positions on the nonlinear manifold while maintaining the continuity of the tensor's geometric structure. This distance measurement method is more in line with the physical reality of tensor changes, making subsequent analysis more accurate.

[0040] After obtaining the geodesic distances between all spatial points, constructing a distance matrix becomes a key step in converting modal information into a graph structure. The distance matrix is ​​a symmetric matrix. Each element in the matrix represents the geodesic distance between any two points. It comprehensively describes the geometric relationship between all spatial points in the tensor field. Through this matrix, local dense or mutation areas can be effectively identified, providing basic data support for subsequent graph construction and clustering analysis. When further constructing an adjacency graph based on this matrix, the values ​​in the distance matrix can be converted into edge weights by setting a distance threshold or a Gaussian kernel function, thereby forming a weighted undirected graph. Each node in the graph represents a spatial point, and the edge weight reflects the strength of the electrical connection between the nodes.

[0041] Based on the constructed adjacency graph, spectral graph theory is used to extract its potential structural features. The specific method is to first calculate the Laplace matrix of the adjacency graph, which is defined as the difference between the node degree matrix and the adjacency matrix, representing the structural features of the graph and its local connectivity. Subsequently, the Laplace matrix is ​​eigendecomposed to obtain several eigenvectors. These eigenvectors are low-dimensional embedded representations of the graph structure, which retain the topological relationship and local density characteristics between spatial points. In the present invention, these eigenvectors serve as input for subsequent clustering analysis, which helps to identify the mutation area of ​​the electrical structure in high-dimensional space, thereby improving the accuracy and robustness of prospecting feature extraction.

[0042] In this example, cluster analysis was performed on the mapping results to obtain several electrical mutation regions, specifically:

[0043] Constructing a first feature space based on a number of feature vectors, and mapping spatial points into the first feature space to obtain a feature mapping point set;

[0044] The Gaussian mixture model is used to model the feature mapping point set to obtain several multivariate Gaussian distribution components;

[0045] Clustering algorithm is used to perform cluster analysis on several multivariate Gaussian distribution components to obtain several dense clusters;

[0046] Each dense cluster is reverse-mapped back to the preset three-dimensional space to obtain the electrical mutation area.

[0047] It should be noted that the characteristic mapping point set refers to the set of eigenvectors obtained by processing the extended electromagnetic response tensor at different spatial locations underground through spectral graph theory. These eigenvectors preserve the differences and connectivity between electrical structures. Specifically, after completing the modal decomposition and geodesic distance calculation of the extended response tensor field, an adjacency graph is constructed and its Laplace matrix is ​​calculated. Multiple low-dimensional eigenvectors are then obtained through eigendecomposition. Each spatial point has a new coordinate representation in this feature space, forming the so-called characteristic mapping point set.

[0048] In order to identify possible areas of mineralized anomalies, these feature map point sets are then input into a Gaussian mixture model for modeling. The Gaussian mixture model assumes that the data in the entire feature map space is composed of multiple potential Gaussian distributions, each of which represents a geological electrical structure, such as the edge of an ore body, a fault zone, or a resistivity mutation zone. By iteratively fitting the feature data using the expectation maximization algorithm, the Gaussian mixture model can identify the center, shape, and range of these potential distributions, thereby dividing the complex underground electrical distribution into multiple dense clusters with soft boundaries. This clustering process not only considers the positional relationship of the points, but also identifies fine-grained mineralized structures hidden in the high-dimensional feature space, and has strong adaptability to nonlinear distributions.

[0049] Finally, in order to convert the electrical structure clustering results identified by the model into actual geological spatial locations, this method reversely maps each dense cluster back to the preset three-dimensional space. Since the feature mapping point set is mapped from real-space points, each point after clustering still retains its original geographic coordinates. Through this correspondence, the point set of each Gaussian cluster can be restored to its distribution position in the underground space, and then those areas with clear shape, size and electrical mutation characteristics in space can be identified. These areas are potential mineralized anomaly zones in deep prospecting, which may correspond to deep ore bodies, fault zones or enrichment areas, and become the core goals of subsequent target area modeling and boundary optimization.

[0050] S3, using a preset geometric reconstruction algorithm to extract target coordinate sets of several electrical mutation areas, and using a greedy algorithm to generate several initial detection range boundaries;

[0051] In this example, a preset geometric reconstruction algorithm is used to extract target coordinate sets of several electrical mutation regions, specifically:

[0052] Extracting a set of boundary points of several electrical mutation regions and calculating the electrical gradient tensor of each boundary point;

[0053] Based on the electrical gradient tensor, the boundary points are screened using the adaptive threshold method to obtain a number of first boundary points;

[0054] Performing three-dimensional Delaunay triangulation on the plurality of first boundary points to construct a triangular mesh model, wherein the triangular mesh model includes a plurality of tetrahedrons;

[0055] Each tetrahedron is used as a graph node, and it is determined whether several tetrahedrons share a triangular face. If two tetrahedrons share a triangular face, a first-class edge is constructed between the corresponding graph nodes to construct an electrical mutation graph model.

[0056] The improved Louvain algorithm was used to perform graph clustering analysis on the electrical mutation map model to obtain the target coordinate set of the electrical mutation area.

[0057] It's important to note that in electromagnetic imaging processing for deep mineral exploration, 3D Delaunay triangulation is a key step in reconstructing regional geometry from spatial point clouds. This method constructs a non-overlapping spatial grid structure composed of tetrahedrons based on a set of boundary points selected for significant linear gradients. The core principle is that the circumscribed sphere of any constructed tetrahedron must not contain points from other point sets, thus ensuring the stability and geometric rationality of the triangulation.

[0058] After the Delaunay decomposition is completed, the resulting triangular mesh model consists of a series of tetrahedral units, each consisting of four boundary points. This mesh not only represents the geometric outline of the mutation region in space but also provides the basic structure for subsequent graph modeling. Each tetrahedron can be abstracted as a node in the graph, and the geometric adjacency relationship between tetrahedrons is represented by their shared triangular faces (patches). To further construct the electrical mutation graph model, it is necessary to determine whether any two tetrahedrons share a triangular face. Specifically, given two tetrahedrons, if there are three points such that they form the same triangular face (that is, the set of points is completely consistent and the order is irrelevant), then the two tetrahedrons are said to share this triangular face. Sharing a triangular face indicates that the two tetrahedrons are directly connected and adjacent in the three-dimensional structure. This topological adjacency relationship constitutes the most basic edge connection in the graph. According to this rule, if two tetrahedrons meet the condition of sharing triangular faces, a first-class edge is established between their corresponding graph nodes, representing continuous or homologous structural units in the electrical mutation region, providing high-precision spatial connectivity information for subsequent graph-based clustering analysis (such as the improved Louvain algorithm).

[0059] In this example, the improved Louvain algorithm is used to perform graph clustering analysis on the electrical mutation map model to obtain the target coordinate set of the electrical mutation area, specifically:

[0060] Initialize each graph node of the electrical mutation graph model as an independent community to obtain the first community set;

[0061] Traverse each graph node and calculate the independent community modularity gain corresponding to moving to the adjacent graph node;

[0062] In the first community set, the graph nodes are moved to the community according to the modularity gain until the preset iteration condition is met, and the movement is stopped to obtain the second community set;

[0063] Extract the tetrahedron set corresponding to each community in the second community set, and calculate the centroid coordinate point set of all tetrahedrons in the tetrahedron set to obtain the centroid coordinate point set of each community;

[0064] The principal component analysis method is used to determine the best fitting plane for each community's centroid coordinate point set;

[0065] Construct a three-dimensional surface function based on the best-fit plane, perform data analysis, and identify ridges and branch points;

[0066] The three-dimensional coordinate set of the ridge line and branch point is used as the target coordinate set of the electrical mutation area.

[0067] It should be noted that in the process of deep prospecting, in order to achieve accurate clustering and identification of the electrical mutation graph model, it is necessary to iteratively optimize the community movement of the graph nodes based on the modularity gain on the first community set initially divided. The specific steps are: first, each tetrahedral graph node is initialized as an independent community to form the initial first community set; then, each node in the graph is traversed, and each node is tried to be moved to the adjacent community in turn, and the modularity gain that each move may bring is calculated according to the modularity gain formula; if a target community can bring the maximum and positive modularity gain, the node is moved to that community; this process is repeated for all nodes in the graph, and when no positive gain can be obtained for all nodes, it is judged that this round of iteration has converged; then, each formed community is regarded as a new super node, the compressed graph structure is reconstructed, and the above community movement operation is repeated; the entire process terminates after meeting the preset iteration termination conditions (such as the maximum number of iterations, the modularity improvement threshold, or the node no longer changes), thus forming the final second community set.

[0068] Furthermore, the best-fit plane is determined by performing principal component analysis (PCA) on the centroid coordinates of all tetrahedrons within a community, extracting the two main directions in which the data most significantly varies in three-dimensional space. This allows the identification of a two-dimensional plane that minimizes the sum of the squared distances from the point to the plane. This plane most accurately reflects the primary spatial distribution characteristics of the community structure. Based on this best-fit plane, a three-dimensional surface function is further constructed. This involves using mathematical interpolation or fitting methods (such as spline curves or polynomial fitting) to transform the distribution of the centroid point set in three-dimensional space into a continuous, smooth surface expression. This surface can capture local geometric variations and reveal the spatial fluctuations and structural complexity of the electrical mutation region. By performing data analysis on this three-dimensional surface function, such as calculating its gradient, curvature, and its higher-order derivatives, ridges and branch points on the surface are identified. These characteristic points represent the "skeleton" structure where the surface varies most significantly and are key spatial markers of electrical anomalies. Ultimately, the three-dimensional coordinate set of ridges and branch points is used as the target coordinate set for the electrical mutation region, which not only accurately depicts the core spatial extent of the ore body but also provides a scientific basis for subsequent exploration and drilling positioning.

[0069] In this example, a greedy algorithm is used to generate several initial detection range boundaries, specifically:

[0070] Merge the target coordinate sets of all electrical mutation areas to construct the first target coordinate point cloud;

[0071] Extract the ridges and branch points in the first target area coordinate point cloud to construct an initial seed point set;

[0072] Initialize the first mineralized area with each seed point in the initial seed point set as the starting point;

[0073] Calculating the regional boundary contour energy of the first mineralized region, and iteratively expanding the regional boundary according to the contour energy to obtain an initial contour surface;

[0074] Discretize the initial contour surface to obtain a three-dimensional boundary surface;

[0075] The three-dimensional boundary surface is decomposed into components to obtain several initial detection range boundaries.

[0076] In this example, the energy calculation formula of the region boundary contour is as follows:

[0077]

[0078] in, is the region boundary contour energy, is the contour line on the boundary surface of the first mineralized area, and is the preset weight coefficient, is the curvature value at point s on the boundary contour line, is the arc length parameter on the contour line.

[0079] It should be noted that the first target area coordinate point cloud refers to the three-dimensional spatial point set formed by merging the target area coordinate sets of all electrical mutation areas. These points accurately reflect the spatial range and morphological characteristics of the potential ore body distribution and serve as the basic data for subsequent exploration and analysis. Based on the first target area point cloud, a representative initial seed point set is constructed by extracting ridges and branch points—key spatial points with significant curvature changes and connectivity features. These points serve as the starting mark for the potential core area of ​​the ore body and help guide the initial delineation of the exploration range.

[0080] The first mineralized area is the area where the ore body is most likely to exist and the electrical mutation is most significant, with each initial seed point as the center, combined with the surrounding spatial distribution characteristics and electrical information. This area serves as the starting point for detection and modeling. After obtaining the initial mineralized area, the contour energy of its boundary is calculated, that is, the smoothness and complexity of the boundary surface are evaluated, and the boundary shape is continuously optimized using an iterative expansion method to obtain an initial contour surface; then, the three-dimensional boundary surface is decomposed into components, usually using mathematical decomposition techniques (such as principal component analysis, Fourier decomposition or waveform decomposition, etc.) to divide the complex three-dimensional surface into several independent and easy-to-analyze local surfaces or boundary fragments. These decomposed fragments constitute several initial detection range boundaries with physical and geometric significance, providing diverse and reasonable spatial boundary candidates for subsequent optimization screening and precise prospecting.

[0081] S4, construct a multi-objective optimization function, use the NSGA-I algorithm to optimize and screen several initial detection range boundaries, obtain the optimal detection range, and conduct deep prospecting.

[0082] In this example, a multi-objective optimization function is constructed, and the NSGA-I algorithm is used to optimize and screen several initial detection range boundaries to obtain the optimal detection range, specifically:

[0083] Discretize the initial detection range boundary to obtain a set of parameterized curves;

[0084] A multi-objective optimization function is constructed, the NSGA-I algorithm is used to perform non-dominated sorting on the parameterized curve set, and the Pareto optimal boundary solution set is generated through adaptive crossover and mutation operations.

[0085] Calculate the hypervolume occupancy ratio of the Pareto optimal frontier solution set and construct the solution set quality evaluation matrix based on the hypervolume occupancy ratio;

[0086] The Pareto optimal boundary solution set is screened based on the solution set quality evaluation matrix to obtain the optimal detection range.

[0087] In this example, the multi-objective optimization function is constructed as follows:

[0088] Define the first optimization goal: maximize the average electrical anomaly intensity within the initial detection range boundary (obtained by claim 5), which is calculated based on the principal component modulus of the extended response tensor field (obtained by claim 2)

[0089] The second optimization objective is to minimize the surface area to volume ratio of the initial detection range boundary to constrain the compactness of the morphology.

[0090] Define the third optimization objective: maximize the spatial correlation between the initial detection range boundary and known geological structures (such as faults and rock mass boundaries), measured by calculating the inverse of the shortest distance from the boundary surface to the known structure;

[0091] Define the fourth optimization goal: minimize the overlapping volume between the initial detection range boundaries to avoid redundant detection;

[0092] Combining the above four objectives, a multi-objective optimization function of the Pareto optimal frontier is established.

[0093] In this example, deep prospecting is carried out, specifically:

[0094] First, based on the three-dimensional spatial model constructed with the optimal boundary, high-resolution electromagnetic sounding profiles and ground or borehole excitation observation points are laid out in the area to improve the density and accuracy of data acquisition; at the same time, combined with geological survey data, the structural distribution, lithological characteristics and previous mineral exploration anomalies of the area are compared and verified. Subsequently, a targeted inversion algorithm is used to perform imaging analysis on the collected high-density electromagnetic data to identify the electrical properties, morphological characteristics and spatial extension direction of deep anomalies. Combined with the structural control information such as the target area coordinates, ridge lines and branch points extracted in the early stage, the anomalies are three-dimensionally modeled and graded to determine the targets with the greatest mineralization potential. Finally, based on the model results, a directional drilling or ground geophysical exploration density layout plan is formulated to gradually verify the actual mineralization of the target body and achieve the transition from macroscopic identification to detailed exploration.

[0095] It is important to note that the Pareto-optimal boundary solutions obtained through the multi-objective optimization algorithm are scientifically screened using a solution quality assessment matrix to determine the optimal exploration range. The specific process involves first quantifying the performance of each boundary solution based on multiple objective functions (such as boundary accuracy, stability, and exploration cost), and constructing a solution quality assessment matrix that reflects the performance of each solution across various metrics and the trade-offs between them. Subsequently, based on this matrix, all Pareto solutions are ranked using a comprehensive evaluation method (such as weighted scoring, ranking statistics, or sorting based on hypervolume metrics) to select the boundary solution that exhibits the best overall performance while balancing multiple objectives. The resulting optimal exploration range not only ensures the scientific rationality of the boundary and the integrity of spatial coverage, but also takes into account the economic and effectiveness of actual exploration, providing a solid decision-making basis for the precise location of deep ore bodies and subsequent resource development.

[0096] Example 2, Figure 2 The present invention provides a deep prospecting system based on electromagnetic sounding data fusion and imaging, including an electromagnetic excitation module, a mutation recognition module, a boundary generation module, and an optimization decision module:

[0097] The electromagnetic excitation module is used to use an electromagnetic source to excite a broadband electromagnetic signal, obtain an electromagnetic response tensor, and perform differential geometric transformation on the electromagnetic response tensor to obtain an extended response tensor field;

[0098] The mutation identification module is used to perform feature mapping on the extended response tensor field and perform cluster analysis on the mapping results to obtain several electrical mutation regions;

[0099] A boundary generation module is used to extract target coordinate sets of several electrical mutation areas using a preset geometric reconstruction algorithm, and to generate several initial detection range boundaries using a greedy algorithm;

[0100] The optimization decision module is used to construct a multi-objective optimization function, use the NSGA-I algorithm to optimize and screen several initial detection range boundaries, obtain the optimal detection range, and conduct deep mineral exploration.

[0101] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0102] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0103] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0104] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0105] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0106] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A deep prospecting method based on electromagnetic sounding data fusion and imaging, characterized by: The following steps are involved: An electromagnetic source is used to excite a broadband electromagnetic signal to obtain an electromagnetic response tensor, and a differential geometric transformation is performed on the electromagnetic response tensor to obtain an extended response tensor field; The extended response tensor field is feature mapped, and the mapping results are clustered to obtain several electrical mutation regions. A preset geometric reconstruction algorithm is used to extract target coordinate sets of several electrical mutation areas, and a greedy algorithm is used to generate several initial detection range boundaries; A multi-objective optimization function was constructed, and the NSGA-I algorithm was used to optimize and screen several initial detection range boundaries to obtain the optimal detection range and conduct deep mineral exploration.

2. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 1 is characterized in that: The electromagnetic source is used to excite a broadband electromagnetic signal to obtain an electromagnetic response tensor, and a differential geometric transformation is performed on the electromagnetic response tensor to obtain an extended response tensor field, specifically: Deploy a multi-band electromagnetic emission source array to stimulate a broadband electromagnetic pulse sequence according to a preset time sequence; Collect electromagnetic field component data through distributed receiving nodes and construct electromagnetic response tensors; Calculate the covariant derivatives of the electromagnetic response tensor to obtain the Riemann connection coefficient matrix; The electromagnetic response tensor is subjected to curvature transformation based on the Riemann connection coefficient matrix to obtain the extended response tensor field.

3. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 2 is characterized in that: The extended response tensor field is subjected to feature mapping, specifically: Perform eigenmode decomposition on the extended response tensor field to obtain several modal components; Calculate the geodesic distance between each spatial point tensor in several modal components and construct a distance matrix; Construct an adjacency graph based on the distance matrix; Using spectral graph theory, the Laplace matrix of the adjacency graph is calculated and eigendecomposition is performed to obtain several eigenvectors.

4. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 3 is characterized in that: The mapping results are clustered and analyzed to obtain several electrical mutation regions, specifically: Constructing a first feature space based on a number of feature vectors, and mapping spatial points into the first feature space to obtain a feature mapping point set; The Gaussian mixture model is used to model the feature mapping point set to obtain several multivariate Gaussian distribution components; Clustering algorithm is used to perform cluster analysis on several multivariate Gaussian distribution components to obtain several dense clusters; Each dense cluster is reversely mapped back to the preset three-dimensional space to obtain the electrical mutation area.

5. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 4 is characterized in that: The preset geometric reconstruction algorithm is used to extract the target coordinate sets of several electrical mutation areas, specifically: Extracting a set of boundary points of several electrical mutation regions and calculating the electrical gradient tensor of each boundary point; Based on the electrical gradient tensor, the boundary points are screened using the adaptive threshold method to obtain a number of first boundary points; Performing three-dimensional Delaunay triangulation on the plurality of first boundary points to construct a triangular mesh model, wherein the triangular mesh model includes a plurality of tetrahedrons; Each tetrahedron is used as a graph node, and it is determined whether several tetrahedrons share a triangular face. If two tetrahedrons share a triangular face, a first-class edge is constructed between the corresponding graph nodes to construct an electrical mutation graph model. The improved Louvain algorithm was used to perform graph clustering analysis on the electrical mutation map model to obtain the target coordinate set of the electrical mutation area.

6. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 5, characterized in that: The improved Louvain algorithm is used to perform graph cluster analysis on the electrical mutation map model to obtain the target coordinate set of the electrical mutation area, specifically: Initialize each graph node of the electrical mutation graph model as an independent community to obtain the first community set; Traverse each graph node and calculate the independent community modularity gain corresponding to moving to the adjacent graph node; In the first community set, the graph nodes are moved to the community according to the modularity gain until the preset iteration condition is met, and the movement is stopped to obtain the second community set; Extract the tetrahedron set corresponding to each community in the second community set, and calculate the centroid coordinate point set of all tetrahedrons in the tetrahedron set to obtain the centroid coordinate point set of each community; The principal component analysis method is used to determine the best fitting plane for each community's centroid coordinate point set; Construct a three-dimensional surface function based on the best-fit plane, perform data analysis, and identify ridges and branch points; The three-dimensional coordinate set of the ridge line and branch point is used as the target coordinate set of the electrical mutation area.

7. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 6, characterized in that: The greedy algorithm is used to generate several initial detection range boundaries, specifically: Merge the target coordinate sets of all electrical mutation areas to construct the first target coordinate point cloud; Extract the ridges and branch points in the first target area coordinate point cloud to construct an initial seed point set; Initialize the first mineralized area with each seed point in the initial seed point set as the starting point; Calculating the regional boundary contour energy of the first mineralized region, and iteratively expanding the regional boundary according to the contour energy to obtain an initial contour surface; Discretize the initial contour surface to obtain a three-dimensional boundary surface; The three-dimensional boundary surface is decomposed into components to obtain several initial detection range boundaries.

8. The deep prospecting method based on electromagnetic sounding data fusion and imaging according to claim 7, characterized in that: The multi-objective optimization function is constructed and the NSGA-I algorithm is used to optimize and screen several initial detection range boundaries to obtain the optimal detection range, specifically: Discretize the initial detection range boundary to obtain a set of parameterized curves; A multi-objective optimization function is constructed, the NSGA-I algorithm is used to perform non-dominated sorting on the parameterized curve set, and the Pareto optimal boundary solution set is generated through adaptive crossover and mutation operations. Calculate the hypervolume occupancy ratio of the Pareto optimal frontier solution set and construct the solution set quality evaluation matrix based on the hypervolume occupancy ratio; The Pareto optimal boundary solution set is screened based on the solution set quality evaluation matrix to obtain the optimal detection range.

9. A deep prospecting system based on electromagnetic sounding data fusion and imaging, applied to the deep prospecting method based on electromagnetic sounding data fusion and imaging according to any one of claims 1 to 8, characterized in that: It includes electromagnetic excitation module, mutation recognition module, boundary generation module and optimization decision module: The electromagnetic excitation module is used to use an electromagnetic source to excite a broadband electromagnetic signal, obtain an electromagnetic response tensor, and perform differential geometric transformation on the electromagnetic response tensor to obtain an extended response tensor field; The mutation identification module is used to perform feature mapping on the extended response tensor field and perform cluster analysis on the mapping results to obtain several electrical mutation regions; A boundary generation module is used to extract target coordinate sets of several electrical mutation areas using a preset geometric reconstruction algorithm, and to generate several initial detection range boundaries using a greedy algorithm; The optimization decision module is used to construct a multi-objective optimization function, use the NSGA-I algorithm to optimize and screen several initial detection range boundaries, obtain the optimal detection range, and conduct deep mineral exploration.

Images