A dynamic path optimization prediction method, system, device and medium for mineral exploration drilling arrangement

Abstract

This application relates to a dynamic path optimization prediction method, system, equipment, and medium for borehole layout in mineral exploration. The method includes: constructing a 3D mesh model based on geoscientific data and initializing the model, drill bit position, and cost; dividing confidence regions based on the model and constructing a weighted directed graph; generating a global borehole sequence using an improved A* algorithm, and then combining this with a fast exploration random tree guided by the confidence regions to generate a drilling trajectory sequence; simulating drilling feedback based on this sequence, dynamically updating the model and cost, iterating until a termination condition is met, and outputting the final borehole layout scheme. This method, through dynamic iterative path optimization and multi-source geoscientific data modeling, significantly improves the accuracy, economy, and exploration efficiency of borehole layout in mineral exploration.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to a dynamic path optimization prediction method, system, equipment and medium for borehole layout in mineral exploration. Background Technology

[0002] Mineral exploration is an indispensable foundational task in the early stages of mineral resource development and utilization. The scientific and rational nature of the borehole layout plan directly determines the operational efficiency, economic input costs, and the accuracy of exploration control over the spatial morphology of ore bodies and resource reserves. Currently, the mineral exploration industry generally adopts a regular grid borehole layout or manual layout based on the experience of geologists. All borehole locations and construction sequences are determined at once in the early stages of exploration, and subsequent drilling operations are carried out using a static and fixed borehole layout planning model.

[0003] On the one hand, existing traditional borehole layout planning is a one-time static planning mechanism, which does not establish the technical logic of dynamic feedback of drilling information and iterative updating of geological cognitive models during the exploration process. The real geological information such as stratigraphic lithology, mineralization development characteristics, and spatial geological variations revealed by drilled boreholes cannot be used to optimize and adjust the borehole locations in subsequent unexplored areas. Once the initial geological a priori model is determined, it is no longer updated. This not only wastes drilling measurement information resources, but also easily leads to insufficient borehole layout in areas with high mineralization potential and ineffective drilling in areas without mineralization potential. The overall borehole layout decision is always in a suboptimal state.

[0004] On the other hand, mineral exploration areas are large-scale three-dimensional spatial domains. After discretizing the exploration area into a three-dimensional grid, the number of grid cells can typically reach hundreds of thousands or even millions. Optimizing borehole layout paths is essentially a high-dimensional combinatorial optimization problem. However, existing methods such as conventional dynamic programming and enumeration optimization exhibit exponential growth in computational complexity as the number of boreholes increases, making it difficult to solve for the globally optimal path within a reasonable computation time. This results in significant computational bottlenecks in high-dimensional search spaces, making them unsuitable for the intelligent borehole planning requirements of complex, large-scale exploration areas. While heuristic search algorithms and fast random tree-based path planning algorithms have been widely applied in path navigation, existing technologies often directly transfer general algorithms to mineral exploration scenarios without considering geological characteristics such as mineralization probability, geological prediction variance, and confidence region classification. Classical path algorithms lack geological uncertainty guidance mechanisms, cannot distinguish the search priorities of high-mineralization confidence regions from low-mineralization confidence regions, easily generate a large number of redundant nodes during the search process, have low path optimization convergence efficiency, and are poorly adapted to the complex geological constraints of three-dimensional space in mineral exploration.

[0005] In addition, some existing studies have attempted to introduce artificial intelligence algorithms such as reinforcement learning to optimize drilling decisions. However, these algorithms rely on a large number of on-site interactive training samples to support model convergence. Mineral exploration field measurement data has a long acquisition cycle, high economic cost, and scarce effective samples, making it difficult to meet the conditions for algorithm training and implementation. They remain only at the theoretical simulation level and cannot be applied to actual exploration engineering projects. Summary of the Invention

[0006] Therefore, it is necessary to provide a dynamic path optimization prediction method, system, equipment and medium for the layout of boreholes in mineral exploration to address the above-mentioned technical problems, aiming to improve the dynamic adaptability of the borehole layout scheme to geological uncertainties and the global optimization efficiency.

[0007] Firstly, this application provides a dynamic path optimization prediction method for borehole layout in mineral exploration, including:

[0008] S1. Based on the original geological data from multiple sources, construct an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell; use the initial three-dimensional grid model as the current model, initialize the current drill bit position as the exploration area entry point, initialize the set of drilled hole positions as an empty set, and initialize the cumulative total cost to zero.

[0009] S2. Based on the mineralization probability and prediction variance in the current model, divide the grid cells into regions to obtain the division results and generate a confidence region label matrix; determine the target region set based on the set of drilled borehole locations; and perform hierarchical sparsification processing on the grid cells based on the confidence region label matrix to construct a weighted directed graph.

[0010] S3. Starting from the current drill bit position and targeting the target region set, the improved A* search algorithm is used to perform path optimization on the weighted directed graph to generate a global borehole position sequence. The first borehole position is taken from the global borehole position sequence as the current target point. Based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, the fast exploration random tree algorithm guided by the confidence region is executed to generate a drilling trajectory sequence from the current drill bit position to the current target point.

[0011] S4. Based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, generate simulated drilling feedback data according to the preset drilling simulation rules; update the current model based on the simulated drilling feedback data to obtain the updated model, and add the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulate the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain the updated result; based on the updated result, determine whether the preset termination condition is met. If not, use the updated model and the simulated final hole position corresponding to the current target point as the current model and current drill bit position for the next iteration, and re-execute S2 to S3; otherwise, output a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model.

[0012] Secondly, this application also provides a dynamic path optimization prediction system for mineral exploration borehole layout, including:

[0013] The initialization module is used to execute S1. S1 includes: constructing an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell based on the multi-source geological raw data; using the initial three-dimensional grid model as the current model, initializing the current drill bit position as the exploration area entry point, initializing the set of drilled hole positions as an empty set, and initializing the cumulative total cost to zero.

[0014] The region partitioning and graph construction module is used to execute S2. S2 includes: partitioning the grid cells into regions based on the mineralization probability and prediction variance in the current model, obtaining the partitioning results, and generating a confidence region label matrix; determining the target region set based on the set of drilled borehole locations; and performing hierarchical sparsification processing on the grid cells based on the confidence region label matrix to construct a weighted directed graph.

[0015] The path optimization and trajectory generation module is used to execute S3. S3 includes: starting from the current drill bit position and targeting the target region set, performing path optimization on the weighted directed graph using an improved A* search algorithm to generate a global borehole position sequence; extracting the first borehole position from the global borehole position sequence as the current target point; and based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, executing a fast exploration random tree algorithm guided by the confidence region to generate a drilling trajectory sequence from the current drill bit position to the current target point.

[0016] The drilling simulation and update module is used to execute S4. S4 includes: generating simulated drilling feedback data based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, according to preset drilling simulation rules; updating the current model based on the simulated drilling feedback data to obtain the updated model, adding the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulating the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain the update result; based on the update result, determining whether the preset termination condition is met. If not, the updated model and the simulated final hole position corresponding to the current target point are used as the current model and current drill bit position for the next iteration, and S2 to S3 are re-executed; otherwise, a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model is output.

[0017] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the first aspect.

[0018] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the first aspect.

[0019] The aforementioned dynamic path optimization prediction method, system, equipment, and medium for borehole layout in mineral exploration firstly constructs an initial 3D mesh model using multi-source geoscientific data for iterative updates, solving the problem that traditional static borehole layout models cannot dynamically optimize using drilling feedback information, thus achieving the co-evolution of the geological cognitive model and the borehole layout scheme. Secondly, by using region partitioning and hierarchical sparse mapping driven by mineralization probability and prediction variance, the computational bottleneck problem of full search in high-dimensional mesh space is solved, achieving effective focusing of confidence regions in the exploration area and dimensionality reduction and compression of the search space. Furthermore, by hierarchically coupling an improved A* algorithm with a fast exploration random tree algorithm guided by confidence regions, the optimality of the global borehole sequence and the geological adaptability of local drilling trajectories are taken into account, avoiding redundant searches and suboptimal convergence caused by the lack of geological constraints in general path algorithms. Finally, by simulating drilling feedback-driven model updates and termination condition determination, the dynamic response capability of borehole layout, the accuracy of drilling economic cost control, and the reliability of resource prediction in complex 3D geological scenarios are enhanced. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of a dynamic path optimization prediction method for borehole layout in mineral exploration, provided as an exemplary embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of a dynamic path optimization prediction system for borehole layout in mineral exploration, provided as an exemplary embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] In one embodiment, such as Figure 1 As shown, a dynamic path optimization prediction method for borehole layout in mineral exploration is provided. This embodiment illustrates the method using a terminal as an example. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0025] S1. Based on multi-source geological raw data, construct an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell; use the initial three-dimensional grid model as the current model, initialize the current drill bit position as the exploration area entry point, initialize the set of drilled hole positions as an empty set, and initialize the cumulative total cost to zero.

[0026] Specifically, mineral exploration planning relies on the effective integration of multi-source geoscientific data, including geological mapping, geophysical exploration, and geochemical data. However, these data are discretely distributed and contain redundant noise. Therefore, data cleaning and normalization are necessary to remove outliers and standardize the data format. Kriging interpolation can be used to interpolate the processed discrete data into a 3D grid of a preset resolution, ensuring each grid cell corresponds to a unique spatial coordinate point. A Bayesian probability model is used to calculate the initial mineralization probability of ore bodies within each cell. Combined with interpolation error analysis and geological variograms, an initial prediction variance is constructed, resulting in an initial 3D grid model containing two core parameters. The initial prediction variance characterizes the degree of uncertainty in the mineralization probability. Subsequently, this initial model can be used as the current model for the first round of calculations. A location with convenient transportation and suitable terrain at the exploration area boundary is selected as the initial drill bit position. The set of drilled locations can be initialized as an empty set, indicating no drilling operations have been conducted. The cumulative total cost is set to zero for statistical analysis of subsequent drilling economic input. The above model construction and parameter initialization lay the foundation for subsequent region division, path optimization and iterative updates.

[0027] S2. Based on the mineralization probability and prediction variance in the current model, divide the grid cells into regions to obtain the division results and generate a confidence region label matrix; determine the target region set based on the set of drilled borehole locations; and perform hierarchical sparsification processing on the grid cells based on the confidence region label matrix to construct a weighted directed graph.

[0028] Specifically, the mineralization potential and uncertainty of geological grid units vary significantly. Therefore, based on the mineralization probability and prediction variance of each unit in the current model, grid units can be classified according to preset rules to generate a confidence region label matrix. Each element of this matrix corresponds one-to-one with a grid unit to mark the region type, achieving a structured division of geological space. Subsequently, by combining the set of drilled locations (empty set in the first round) and analyzing the controlled area using a preset detection radius, uncontrolled grid units with mineralization potential can be designated as the target region set, clarifying the core direction of path optimization. Furthermore, to address the high-dimensional computational bottleneck of the 3D grid, hierarchical sparsity processing can be performed based on the confidence region label matrix. This involves eliminating low-value, non-confidence region units and employing differentiated strategies for different confidence regions. For example, high-confidence regions are simplified, while low-confidence regions retain details. Weighted edges are established between adjacent units or simplified nodes, with weights set based on spatial distance, thus constructing a weighted directed graph. This graph retains key geological structural information while effectively compressing the computational scale, contributing to improved efficiency of subsequent algorithms.

[0029] S3. Starting from the current drill bit position and targeting the target region set, use the improved A* search algorithm to perform path optimization on the weighted directed graph to generate a global borehole position sequence. Take the first borehole position from the global borehole position sequence as the current target point. Based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, execute the fast exploration random tree algorithm guided by the confidence region to generate a drilling trajectory sequence from the current drill bit position to the current target point.

[0030] Specifically, S3 allows for collaborative path planning using a global optimum-local feasibility approach, balancing the overall rationality of borehole layout with construction feasibility. For example, at the global level, starting from the current drill bit position and targeting a set of target areas, an improved A* search algorithm can be invoked to optimize on a weighted directed graph. Illustratively, this algorithm can balance path length and geological potential using a heuristic function, generating a priority-ordered sequence of global borehole positions. Path length represents economic cost, while geological potential can be quantified using mineralization probability and uncertainty. Therefore, this path optimization ensures both the economy and the targeted nature of the overall layout. At the local level, the first borehole position in the sequence can be taken as the current target point. Based on the current drill bit position, the current model prediction variance, and the confidence region label matrix (used to provide regional constraints), a fast exploration random tree algorithm guided by the confidence region can be executed. For example, by combining random sampling and regional guidance with an extended path tree, drilling process constraints (such as borehole inclination angle and curvature limitations) and geological obstacles can be avoided to generate a continuous drilling trajectory sequence. This solves the problem that traditional algorithms cannot balance global optimality and local feasibility, ensuring that the trajectory adapts to actual construction needs.

[0031] S4. Based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, generate simulated drilling feedback data according to the preset drilling simulation rules; update the current model based on the simulated drilling feedback data to obtain the updated model, and add the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulate the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain the updated result; based on the updated result, determine whether the preset termination condition is met. If not, use the updated model and the simulated final hole position corresponding to the current target point as the current model and current drill bit position for the next iteration, and re-execute S2 to S3; otherwise, output a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model.

[0032] Specifically, based on the drilling trajectory sequence and the core parameters of the current model, the drilling simulation logic for detecting and verifying geological parameters can be simulated according to preset drilling simulation rules. This generates simulated drilling feedback data containing verification results of geological parameters along the trajectory and uncertainty correction coefficients, thus avoiding the high costs and long cycles of actual drilling. Subsequently, the current model can be updated based on the feedback data. For example, the variance of high-probability areas in the trajectory verification can be reduced, and the probability values ​​of low-probability areas can be corrected to obtain an updated model. Simultaneously, the simulated final borehole position can be added to the set of drilled boreholes, and the simulated drilling cost corresponding to the trajectory can be accumulated to the total cumulative cost, forming an updated result. The simulated drilling cost can be calculated based on trajectory length, construction difficulty, etc. Based on the updated result, it can be determined whether the termination conditions are met, such as whether the total cost reaches the budget limit, whether the target area is fully covered, and whether the model uncertainty is below the corresponding threshold. If not, the updated model can be used as the new current model, the simulated final borehole position as the new drill bit position, and the process can return to S2-S3 for re-execution, performing multiple iterations. If the conditions are met, a borehole layout scheme containing the set of boreholes drilled, the trajectory sequence of each round, and the updated model can be output. This scheme reduces exploration planning costs and is scientific, economical, and feasible.

[0033] The aforementioned method first constructs a 3D grid model based on multi-source geological raw data and initializes its parameters, establishing a standardized geological spatial representation system that includes mineralization probability and prediction variance, laying a data foundation for subsequent path optimization. Second, it divides the grid cells into confidence regions based on mineralization probability and variance, and constructs a weighted directed graph through hierarchical sparsification, effectively compressing the high-dimensional grid search space. Furthermore, it integrates an improved A* search algorithm for global borehole path optimization, combined with a fast exploration random tree algorithm guided by confidence regions to generate local drilling trajectories. This approach balances global borehole optimization with local drilling process constraints, solving the problems of insufficient geological scenario adaptability and numerous redundant search nodes in general path algorithms. Finally, it generates simulated drilling feedback data based on preset drilling simulation rules, iteratively updates the geological model, and performs iterative replanning. This closed-loop optimization iteration can be completed without relying on actual on-site drilling operations, effectively improving the intelligence level, planning efficiency, and economic applicability of mineral exploration borehole layout.

[0034] In one embodiment, the partitioning result may include unconfidence regions and confidence regions, with the confidence regions including low-confidence regions and high-confidence regions.

[0035] Indicatively, in the dynamic path optimization prediction method S2 for borehole layout in mineral exploration provided in this application, based on the mineralization probability and prediction variance in the current model, the grid cells are divided into regions to obtain the division results and generate a confidence region label matrix; based on the set of borehole locations, the target region set is determined, which may include:

[0036] Read the mineralization probability and prediction variance of each grid cell from the current model, and set the low mineralization probability threshold, high mineralization probability threshold, and variance threshold;

[0037] For each grid cell, perform a region partitioning operation. If the mineralization probability is less than the low mineralization probability threshold, the grid cell is marked as a non-confidence region. If the mineralization probability is greater than or equal to the high mineralization probability threshold and the prediction variance is less than or equal to the variance threshold, the grid cell is marked as a high confidence region. All grid cells except those marked as non-confidence and high confidence regions are marked as low confidence regions, thus obtaining the partitioning result.

[0038] Based on the segmentation results, generate a confidence region label matrix;

[0039] Based on the set of drilled locations and the preset detection radius, the controlled grid cells are determined; based on the controlled grid cells, the uncontrolled grid cells in the current model with a mineralization probability not lower than the low mineralization probability threshold are summarized to form the target area set.

[0040] Specifically, we can first extract two core parameters from the currently constructed 3D mesh model: the mineralization probability and the prediction variance for each mesh cell. The mineralization probability quantifies the likelihood of a mineral body existing within the mesh cell, while the prediction variance characterizes the degree of uncertainty surrounding that probability. A larger variance indicates a more ambiguous understanding of the area based on existing geological data, necessitating further exploration to supplement information. Subsequently, we can obtain three key thresholds: a low mineralization probability threshold, a high mineralization probability threshold, and a variance threshold. These thresholds can be set based on the geological background of the exploration area, historical exploration data, and exploration objectives. For example, the low mineralization probability threshold can be set by referring to the statistical results of mineralization probabilities in known non-mineralized areas within the region to ensure effective differentiation of areas without mineralization potential. The high mineralization probability threshold can be set by combining the distribution characteristics of the mineralization probabilities of discovered mineralized bodies. The variance threshold can be set based on data acquisition accuracy, the error level of the interpolation algorithm, and the reliability requirements of geological interpretation, used to define the acceptable range of uncertainty. Furthermore, these thresholds must satisfy the constraint that the low mineralization probability threshold is less than the high mineralization probability threshold and the variance threshold is greater than 0. The mathematical expression for this is:

[0041]

[0042] in, This indicates a low probability threshold for mineralization. This represents the threshold for a high probability of mineralization. This represents the variance threshold.

[0043] Based on the aforementioned threshold, a region division operation can be performed on each grid cell. For example, for any given grid cell, if its mineralization probability is less than the low mineralization probability threshold... If the probability of mineralization within the grid cell is extremely low, it is not worth further exploration and can be marked as a non-confidence area; if its mineralization probability is greater than or equal to the high mineralization probability threshold... And the prediction variance is less than or equal to the variance threshold. This indicates that the grid cell not only has outstanding mineralization potential, but also has low uncertainty in the existing geological data regarding its understanding, belonging to an area with clear exploration value, and can be marked as a high-confidence area; apart from the above two cases, other grid cells, i.e., those with a mineralization probability between the low mineralization probability threshold, High mineralization probability threshold Between, or with a mineralization probability not lower than the high mineralization probability threshold. Value, but the predicted variance is greater than the variance threshold. Grid cells that are not fully defined can be marked as low-confidence areas, indicating that while these areas have some mineralization potential, the existing information is highly uncertain and requires further verification and data supplementation through drilling. Using the above judgment logic, the regional division of all grid cells can be completed, yielding the final division result.

[0044] Based on the above region division results, a confidence region label matrix can be further generated. The dimension of this matrix is ​​consistent with the spatial dimension of the 3D mesh model. Each element in the matrix corresponds one-to-one with each mesh cell in the 3D mesh, and the element value is a preset label code used to clarify the region type of the corresponding mesh cell. For example, when using numerical encoding, 0 represents a non-confidence region, 1 represents a low-confidence region, and 2 represents a high-confidence region. This matrix can quickly index the region attributes of any mesh cell, providing efficient region information query support for subsequent hierarchical sparsity processing and path optimization algorithms.

[0045] Specifically, when determining the target area set, the controlled grid cells can be identified first based on the set of drilled borehole locations and a preset detection radius. The preset detection radius can be set by comprehensively considering the drilling equipment's detection depth, the effective influence range of core sampling, and the continuity characteristics of the geological body, ensuring that the drilled borehole locations effectively control the geological information of surrounding grid cells. For example, a spherical space can be constructed with each drilled borehole location as the center and the preset detection radius as the radius. All grid cells covered by this spherical space are considered controlled grid cells, indicating that the geological information of these grid cells has been verified through previous drilling and does not require repeated exploration. Subsequently, uncontrolled grid cells with a mineralization probability not lower than the low mineralization probability threshold can be selected from the current 3D grid model. The grid cells are aggregated to form a target area set, focusing on areas that have not yet been explored and verified but have potential mineralization value. This avoids redundant investment in already controlled areas and ensures the mineralization potential of the target area set, thus clarifying the core target direction for subsequent path optimization. Furthermore, the target area set can ultimately be output as a set of spatial coordinates of the grid cells for easy invocation by the path optimization algorithm.

[0046] In one embodiment, the dynamic path optimization prediction method S2 for mineral exploration borehole layout provided in this application, which constructs a weighted directed graph by performing hierarchical sparsification of grid cells based on the confidence region label matrix, may include:

[0047] Based on the confidence region label matrix, delete the grid cells labeled as non-confidence regions;

[0048] For grid cells labeled as low confidence regions, retain the original spatial coordinates of the grid cells labeled as low confidence regions, and establish a first edge with a first weight between adjacent grid cells labeled as low confidence regions. The first weight is the Euclidean distance between the two grid cells labeled as low confidence regions.

[0049] For a grid cell labeled as a high confidence region, eight spatially adjacent grid cells labeled as high confidence regions are merged into a super node. The geometric center of the eight grid cells labeled as high confidence regions is used as the representative coordinate of the super node. A second edge with a second weight is established between adjacent super nodes. The second weight is calculated by multiplying the Euclidean distance between the representative coordinates of the two super nodes by a preset coarsening factor.

[0050] Establish cross-region edges between grid cells labeled as low-confidence regions and their adjacent super nodes. The weight of the cross-region edge is the Euclidean distance between the grid cell labeled as a low-confidence region and its adjacent super node.

[0051] A weighted directed graph is constructed based on grid cells labeled as low-confidence regions, super nodes, first edges, second edges, and cross-region edges.

[0052] Specifically, grid cells in non-confidence regions have been identified as areas with extremely low mineralization probability and no exploration value through previous regional division. Their spatial information has no positive significance for path optimization. Therefore, by traversing all grid cells in the 3D grid model and matching the corresponding label codes in the confidence region label matrix one by one, if the code represents a non-confidence region, the grid cell can be removed from the unit set for subsequent processing. Only grid cells in confidence regions (low-confidence regions and high-confidence regions) are retained to participate in subsequent sparsification processing and graph construction, reducing the number of nodes in subsequent graph construction and reducing computational redundancy.

[0053] Furthermore, refined characterization processing can be performed on grid cells labeled as low-confidence regions. These low-confidence regions, representing areas with some mineralization potential but high uncertainty, need to retain their original spatial details to ensure the accuracy of path optimization. Therefore, the original spatial coordinates of these grid cells can be directly preserved, meaning the coordinate values ​​are consistent with the preset spatial coordinate system in the 3D mesh model to ensure the accuracy of node positions. Subsequently, a first edge with a first weight is established between adjacent low-confidence region grid cells. The adjacency relationship can be determined based on 3D spatial neighborhood rules. For example, if the absolute value of the coordinate difference between two low-confidence region grid cells on the x, y, and z axes does not exceed 1, satisfying the adjacency condition of sharing faces, edges, or vertices in space, then they are considered adjacent cells. The first weight is the Euclidean distance between two adjacent low-confidence region grid cells, calculated using the following formula:

[0054]

[0055] in, Indicates the first weight. The original spatial coordinates of the first low-confidence region grid cell. The first weight represents the original spatial coordinates of the second low-confidence region grid cell. This first weight directly quantifies the spatial distance between two adjacent low-confidence region cells, providing a basic cost reference for path optimization.

[0056] Specifically, coarsening can be performed on grid cells labeled as high-confidence regions. These high-confidence regions have a high probability of mineralization and low uncertainty, with relatively clear geological information. Therefore, it's unnecessary to retain excessively high spatial resolution. Coarsening can significantly reduce the number of nodes and improve computational efficiency. For example, an 8-neighborhood merging rule can be used to merge eight spatially adjacent high-confidence region grid cells into a single supernode. An 8-neighborhood is defined as eight high-confidence region grid cells centered on a given high-confidence region grid cell, whose x-axis ±1, y-axis ±1, and z-axis ±1 directions coincide (if any high-confidence region grid cells exist in that direction). These form a 3×3×3 cubic spatial cell group, which is then combined into a single supernode. The representative coordinates of this supernode can be obtained by calculating the geometric centers of the eight high-confidence region grid cells, using the following formula:

[0057]

[0058] in, These are the representative coordinates of the supernode. The original spatial coordinates of the k-th high-confidence region mesh cell participating in the merging are given. A second edge with a second weight can be established between adjacent supernodes. The criterion for determining adjacent supernodes is that the cubic spatial cell groups corresponding to the two supernodes are adjacent in 3D space (i.e., sharing faces, edges, or vertices in space). The second weight is calculated by multiplying the Euclidean distance between the coordinates representing the two supernodes by a preset coarsening factor, and its calculation formula is as follows:

[0059]

[0060] in, Indicates the second weight. This is a preset coarsening factor, with a value greater than 1. Its magnitude is calibrated based on the grid density and exploration accuracy requirements of high-confidence areas, and is used to balance computational efficiency and the accuracy of path cost estimation. and These are the representative coordinates of two adjacent supernodes.

[0061] Specifically, cross-region edges can be established between low-confidence region grid cells and adjacent super nodes to achieve path connectivity between different confidence regions. The adjacency criteria for low-confidence region grid cells and super nodes can be that the spatial distance between the original spatial coordinates of the low-confidence region grid cell and the representative coordinates of the super node is less than a preset neighborhood threshold, or that any high-confidence region grid cell in the cubic spatial cell group corresponding to the low-confidence region grid cell and the super node satisfies the three-dimensional spatial adjacency condition. The weight of the cross-region edge is the Euclidean distance between the original spatial coordinates of the low-confidence region grid cell and the representative coordinates of the adjacent super node. This weight directly reflects the spatial cost between nodes in different confidence regions, ensuring the accuracy of cost calculation for cross-region paths.

[0062] Based on the above process, nodes and their corresponding edges are obtained, and a weighted directed graph can be constructed. The node set of this weighted directed graph includes all low-confidence region grid cells that retain their original coordinates and all merged super nodes. The edge set includes the first edge between low-confidence region cells, the second edge between super nodes, and cross-region edges between low-confidence region cells and super nodes. Each edge corresponds to a unique weight value. Through the above hierarchical sparsity construction strategy, the weighted directed graph not only preserves the spatial details of low-confidence regions, ensuring the accuracy of path planning in high-uncertainty regions, but also significantly reduces the number of nodes and edges through coarsening of high-confidence regions, effectively improving the solution efficiency of subsequent path optimization algorithms. Furthermore, it achieves seamless connectivity between different confidence regions through cross-region edges, ensuring the integrity and feasibility of the global path.

[0063] In one embodiment, the improved A* search algorithm employs an improved cost function, in which the total cost can be calculated based on the actual cost and the heuristic function value, and the heuristic function value includes an uncertainty guiding term based on the entropy of mineralization probability information.

[0064] Specifically, in the dynamic path optimization prediction method S3 for borehole layout in mineral exploration provided in this application, starting from the current drill bit position and targeting the set of target areas, an improved A* search algorithm is used to perform path optimization on a weighted directed graph to generate a global borehole position sequence, which may include:

[0065] S201: Read the mineralization probability of each grid cell from the current model; for each grid cell in the weighted directed graph, calculate the corresponding information entropy based on the mineralization probability of each grid cell in the weighted directed graph. The information entropy is used to characterize the degree of geological uncertainty at the grid cell.

[0066] Specifically, the mineralization probability of each grid cell can be read from the current 3D grid model. This probability is the probability of ore body existence within the cell obtained based on the fusion of multi-source geoscientific data, and its value distribution directly reflects the mineralization potential characteristics of the cell. For each grid cell in the weighted directed graph, the corresponding information entropy is calculated based on its mineralization probability. The calculation of information entropy follows the quantification logic of uncertainty in information theory. That is, the closer the mineralization probability is to 0.5, the larger the information entropy value, indicating higher geological uncertainty of the cell and more ambiguous understanding of its mineralization potential based on existing data. Conversely, when the mineralization probability approaches 0 or 1, the information entropy value approaches 0, indicating lower geological uncertainty. The calculated information entropy can serve as the core guiding term of the subsequent heuristic function, used to prioritize expansion towards high-uncertainty areas in path optimization, thereby reducing the overall information uncertainty of the exploration.

[0067] S202: Based on the proportion of the cumulative total cost to the preset budget limit, dynamically adjust the ratio of the first weight coefficient and the second weight coefficient; for each grid cell in the weighted directed graph, calculate the estimated distance from each grid cell in the weighted directed graph to the target region set.

[0068] Specifically, the ratio of cumulative total cost to the preset budget ceiling reflects the resource consumption level at the current exploration stage. When this ratio is low, it indicates sufficient budget, and priority can be given to reducing geological uncertainty; therefore, the proportion of the second weighting coefficient can be increased. When this ratio is high, the remaining budget is limited, and priority should be given to ensuring the economic efficiency of the path; therefore, the proportion of the first weighting coefficient can be increased. By dynamically adjusting the ratio of the two, a balance can be achieved between cost and uncertainty reduction in path optimization. For each grid cell in the weighted directed graph, its estimated distance to the target region set can be defined as the shortest Euclidean distance between the spatial coordinates of that cell and the spatial coordinates of all cells in the target region set. This distance can be calculated based on the spatial coordinate system of the 3D grid model to ensure the accuracy of the estimated distance and provide a basic quantitative value for the path cost of the heuristic function.

[0069] S203: Calculate the first result by multiplying the first weight coefficient by the estimated distance, obtain the second result based on the second weight coefficient and the information entropy, and calculate the heuristic function value of each grid cell in the weighted directed graph based on the first and second results.

[0070] Specifically, through heuristic functions, the economic efficiency and information gain of a path can be balanced during the path optimization process. The core logic is to combine the estimated distance with the information entropy weighted by the credibility of multi-source data. The first result, representing the path's economic cost tendency, is obtained by multiplying the estimated distance by a dynamically adjusted first weight coefficient. The second result, representing the path's uncertainty reduction tendency, is obtained by multiplying the information entropy and the weighted sum of the credibility of multi-source data by a dynamically adjusted second weight coefficient. The heuristic function value can be defined as the first result minus the second result. When the information entropy of a unit is higher and the weighted sum of the credibility of multi-source data is higher, the second result is larger and the heuristic function value is smaller. This can guide the algorithm to prioritize expanding to that unit, thus balancing cost and information gain in path planning.

[0071] S204: Starting from the grid cell where the current drill bit is located, initialize the open set and the closed set, and add the starting point to the open set; initialize the actual cost of each grid cell in the weighted directed graph to a preset maximum value, and only set the actual cost of the starting point to zero; calculate the total cost of the starting point according to the cost function; where the total cost of the starting point is equal to the sum of the actual cost of the starting point and the heuristic function value of the starting point.

[0072] Specifically, the open set is used to store nodes to be expanded, and the closed set is used to store nodes that have been expanded to avoid duplicate processing. The starting point can be the grid cell where the current drill bit is located. The spatial coordinates of this cell correspond to the real-time position of the drill bit in the current 3D mesh model. This cell is added to the open set, and the actual cost of all grid cells in the weighted directed graph is initialized to a preset maximum value. This indicates that initially, all cells except the starting point are unreachable, and only the actual cost of the starting point is set to zero, representing that the path cost from the starting point to itself is zero. Illustratively, the cost function used in the improved A* search algorithm can be:

[0073]

[0074] in, For grid cells The total cost is the core optimization objective of path optimization; From the starting point to the grid cell The actual cost is calculated based on the actual cost of the parent node, edge weights, and a geological difficulty correction term. For grid cells In the parent node of the search tree, ,parent For nodes The weight of the edge with respect to the parent node corresponds to the distance or coarsening distance of the edge in the weighted directed graph. This is a geological difficulty correction factor, calibrated based on rock mechanics data from the exploration area. For grid cells The rock compressive strength was obtained from engineering geological parameters in multi-source geoscience data. For grid cells The borehole design inclination angle at that location is a constraint parameter for the drilling process. This is a heuristic function, and its definition is consistent with the steps described above. , These are the heuristic function weight coefficients, namely the first and second weight coefficients mentioned in the steps. For grid cells Set up in the target area The estimated distance, i.e., the shortest Euclidean distance, For grid cells The entropy of mineralization probability information. To increase the number of multi-source geoscientific data types, For the first The reliability weight of the data source satisfies , For the first Data source in grid cell The credibility of the value ranges from 0 to 1. This is a penalty term used to suppress low-value paths and circumvent high-confidence region boundaries. , , The penalty coefficient is... For grid cells The probability of mineralization ranges from 0 to 1. This is an indicator function that takes the value 1 when the condition within the parentheses is true, and 0 otherwise. This represents the boundary mesh set for the high-confidence region. Specifically, the total cost of the starting point can be equal to the sum of the actual cost of the starting point, the heuristic function value, and the penalty term. Since the actual cost of the starting point is zero, its total cost is mainly determined by the heuristic function value and the penalty term, providing an initial cost benchmark for subsequent node expansion.

[0075] S205: Based on open sets, closed sets, and the actual and total costs of each grid cell in the weighted directed graph, repeatedly perform node expansion operations. Stop the node expansion operation when the current grid cell belongs to the target region set. The node expansion operation includes selecting the current grid cell, processing neighboring grid cells, calculating the trial actual cost, updating the parameters of neighboring grid cells, and maintaining open sets.

[0076] Specifically, during the node expansion operation, the grid cell with the lowest total cost can be selected from the open set as the current grid cell. This cell represents the node with the greatest path advantage at the current stage. It is then removed from the open set and moved into the closed set, and marked as a processed node. Subsequently, all neighboring grid cells of the current grid cell in the weighted directed graph are traversed. If a neighboring cell is already in the closed set, it is skipped to avoid redundant expansion. If a neighboring cell has not been processed, the trial actual cost of reaching that neighboring cell from the starting point via the current grid cell is calculated. The calculation of the trial actual cost can be based on the actual cost... The calculation logic combines the actual cost of the current grid cell, the weights of the edges between the current cell and its neighbors, and the sum of geological difficulty correction terms. A penalty term is also introduced, based on the probability of mineralization; the lower the probability of mineralization, the larger the penalty term, to suppress expansion into cells with low mineralization potential. If the trial cost is less than the current actual cost of a neighboring cell, the parent node pointer of the neighboring cell is updated to the current grid cell, and its actual cost is updated to the trial cost. Then, the total cost of the neighboring cells is calculated according to the cost function. If a neighboring cell is not in the open set, it can be added to the open set. This node expansion operation is repeated until the current grid cell belongs to the target area set, at which point expansion stops, ensuring the path has reached the target exploration area.

[0077] S206: Starting from the current grid cell, backtrack to the starting point, and output the grid cells on the backtracking path in the order from the starting point to the target area set to obtain the global borehole location sequence.

[0078] Specifically, starting from the current grid cell (i.e., the cell within the target area set), the process traces back along the parent node pointer of each cell to the starting point, recording all grid cells along the traversal path. These cells are then rearranged in the order from the starting point to the target area set to obtain the global borehole location sequence. Each cell in this sequence corresponds to a candidate borehole location, ordered according to the priority determined by path optimization. This provides a sequence of target points for subsequent local trajectory planning, ensuring that the overall borehole layout considers economy, mineral potential, and information gain, while also avoiding the problem of path planning being out of sync with actual exploration needs.

[0079] In one embodiment, the node expansion operation in step S205 above, based on the open set, the closed set, and the actual cost and total cost of each grid cell in the weighted directed graph, is repeatedly performed. When the current grid cell belongs to the target region set, the node expansion operation is stopped. This may include the following steps:

[0080] S301: Select the grid cell with the minimum total cost from the open set and determine the grid cell with the minimum total cost as the current grid cell.

[0081] Specifically, the open set can be stored using a priority queue data structure, with elements sorted in ascending order of total cost. The core principle of this selection logic is based on the optimality criterion of the A* algorithm. The cell with the lowest total cost represents the optimal candidate path node from the starting point to the target region at the current stage. Prioritizing the expansion of this cell ensures that the final generated path is globally optimal and avoids redundant computation of invalid paths. During the selection process, if multiple grid cells with the same total cost exist in the open set, their heuristic function values ​​can be further compared, prioritizing the cell with the smaller heuristic function value to accelerate convergence to the target region.

[0082] S302: Move the current grid cell out of the open set and into the closed set.

[0083] Specifically, the closing set can be stored using a hash table data structure, with the spatial coordinates of the grid cell as the unique key. Moving the current grid cell into the closing set marks that the cell has completed the expansion process, avoiding repeated access to it during subsequent node expansions, thereby reducing the algorithm's time complexity. After the removal operation, the opening set can automatically adjust the priority queue's sorting structure to ensure that the cell with the minimum total cost can still be quickly located during the next selection, maintaining the algorithm's efficiency.

[0084] S303: For each neighboring grid cell of the current grid cell in the weighted directed graph, if the neighboring grid cell is already in the closed set, skip the neighboring grid cell; otherwise, calculate the trial actual cost of reaching the neighboring grid cell from the starting point through the current grid cell. The trial actual cost is calculated based on the actual cost of the current grid cell, the weight of the edge from the current grid cell to the neighboring grid cell, and the penalty term. The penalty term is constructed based on the preset penalty coefficient and the mineralization probability of the neighboring grid cell.

[0085] Specifically, neighboring grid cells are determined based on edge connections in the weighted directed graph. All grid cells or supernodes directly connected to the current grid cell via a first edge, second edge, or cross-region edge are considered neighboring grid cells. A hash table is used to check if a neighboring cell is in the closed set. If it already exists, it is skipped to avoid duplicate calculations; if it does not exist, the calculation of the trial actual cost is initiated. The calculation of the trial actual cost integrates the path base cost and the geological value penalty, and its mathematical expression can be:

[0086]

[0087] in, To start from the current grid cell Reaching neighboring grid cells The actual cost of testing; For the current grid cell The actual cost, i.e., from the starting point to The optimal path cost; For the current grid cell To neighboring grid cells The weight of the edges between them can be either Euclidean distance or coarsened distance (depending on the type of edge). For neighboring grid cells The corresponding penalty term is the same as that in the above embodiment. Through this formula, the lower the probability of mineralization in a neighboring cell, the larger the penalty term, and the higher the actual cost of exploration. This inhibits the algorithm from expanding into areas with low mineralization potential, ensuring the targeted nature of path planning.

[0088] S304: If the trial actual cost is less than the current actual cost of the neighboring grid cell, then update the parent node pointer of the neighboring grid cell to the current grid cell, update the current actual cost of the neighboring grid cell to the trial actual cost, and calculate the total cost of the neighboring grid cell according to the cost function.

[0089] Specifically, the current actual cost of a neighboring grid cell is initialized to a preset maximum value, indicating that the reachable path cost of that cell is unknown in the initial state. When the trial actual cost is less than the current actual cost, it means that the path to that neighboring cell via the current grid cell is better, and relevant parameters need to be updated. Among these updates, the parent node pointer is used to record the node connection relationship of the optimal path, providing a basis for subsequent path backtracking; the update of the actual cost ensures that the cost data of the neighboring cell always reflects the currently known optimal path cost. This update operation can be performed by directly overwriting the parameter values ​​corresponding to the neighboring cell to ensure the real-time performance and accuracy of the data.

[0090] Schematic, the total cost can be calculated following the cost function calculation logic of the improved A* algorithm in the above embodiments, and if the current actual cost of updating the neighboring grid cells is the trial actual cost, the trial actual cost incorporates a penalty term in S303. Therefore, when calculating the total cost of neighboring grid cells, a penalty term can be omitted to avoid duplicate calculations. During the calculation, all parameters use the latest grid cell data to ensure that the total cost accurately reflects the comprehensive advantages of neighboring cells as path nodes, providing a reliable basis for the subsequent sorting and selection of open sets.

[0091] S305: If a neighboring grid cell is not in the open set, add the neighboring grid cell to the open set.

[0092] Specifically, neighboring units can be added to the open set through a priority queue insertion operation. During insertion, the units are automatically sorted according to their total cost, maintaining the queue order from smallest to largest total cost. This S306 allows unprocessed optimal candidate nodes to be included in the set to be expanded, ensuring that potential optimal path nodes are continuously expanded in subsequent iterations, avoiding the omission of better paths. If a neighboring unit is already in the open set, it does not need to be added again; only its sorted position in the queue needs to be maintained.

[0093] S306: Return to steps S301 to S305 until the current grid cell belongs to the target region set, triggering the node expansion operation to stop.

[0094] Specifically, after each iteration, the cell with the minimum total cost is selected again from the open set as the new current grid cell, and the above expansion process is repeated. Illustratively, determining whether the current grid cell belongs to the target region set can be achieved by querying whether the cell's spatial coordinates are within the coordinate range of the target region set. The target region set is stored as a coordinate set, supporting fast lookup. When it is determined that the current grid cell belongs to the target region set, it indicates that the optimal path from the starting point to the target region has been found, and the node expansion operation can be stopped to avoid invalid subsequent iterations. This ensures that the algorithm terminates promptly after finding the optimal path, balancing solution efficiency and path quality.

[0095] In one embodiment, in the dynamic path optimization prediction method S3 for borehole layout in mineral exploration provided by this application, the first borehole position is extracted from the global borehole position sequence as the current target point. Based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, a fast exploration random tree algorithm guided by the confidence region is executed to generate a drilling trajectory sequence from the current drill bit position to the current target point, which may include:

[0096] S401: Extract the first borehole position from the global borehole position sequence and determine the borehole position as the current target point; use the current drill bit position as the root node to initialize the initial tree.

[0097] Specifically, the global borehole location sequence is sorted according to path optimization priority. The first borehole location is the optimal exploration target for the current stage, and its spatial coordinates can be directly used as the coordinates of the current target point. These coordinates are consistent with the spatial coordinate system of the 3D mesh model. The initial tree can be stored using an adjacency list data structure. The coordinates of the root node completely correspond to the current drill bit position. The root node has no parent node pointer and only records its own spatial coordinates and the type of the confidence region it belongs to. Initialization of the initial tree can also include setting node storage attributes. Each node must contain three core pieces of information: spatial coordinates, parent node index, and the label of the confidence region it belongs to. This provides data support for subsequent sampling expansion and path backtracking, ensuring the integrity and traceability of the tree structure.

[0098] S402: Read the prediction variance for each grid cell from the current model.

[0099] Predictive variance characterizes the uncertainty of mineralization probability. The larger the value, the more ambiguous the existing geological data's understanding of the mineralization potential of that grid cell, requiring further verification through drilling. During the reading process, a mapping relationship between predictive variance and the spatial coordinates of the grid cell can be established, ensuring that the variance data of each grid cell can be quickly indexed via coordinates. This provides real-time data support for subsequent non-uniform sampling and gradient direction calculation. This mapping relationship can be stored in the form of a hash table or array.

[0100] S403: Set the target bias probability, and use the difference between 1 and the target bias probability as the remaining probability.

[0101] Specifically, by setting a target bias probability, the convergence speed and exploration capability of the algorithm can be further balanced. Its value can be dynamically calibrated based on the spatial distance between the current target point and the current drill bit position, and the geological complexity. For example, when the distance between the two is large or the geological structure is simple, a higher target bias probability is used to accelerate trajectory convergence to the target point. Conversely, when the distance is small or the geological complexity is high, such as multiple faults or densely packed high-uncertainty areas, a lower target bias probability is used to increase the proportion of random exploration and avoid getting trapped in local optima. The target bias probability and the remaining probability satisfy a constraint that their sum is 1, i.e., target bias probability ∈ (0,1), and remaining probability = 1 - target bias probability, to ensure the normalization of the probability distribution.

[0102] S404: Based on the target bias probability and the remaining probability, perform a sampling expansion operation to generate new nodes. When the distance between the new node and the current target point is less than a preset distance threshold, stop the iterative sampling expansion operation; otherwise, repeat the sampling expansion operation. The sampling expansion operation includes:

[0103] S4041: Generate random sampling points according to the probability selection rule. The probability selection rule is as follows: In each iteration, the current target point is selected independently as a random sampling point with the target bias probability, or random sampling is performed within the grid cell belonging to the confidence region with the remaining probability, based on the non-uniform probability density function to obtain random sampling points; where the non-uniform probability density function is constructed based on the prediction variance of the grid cell.

[0104] S4042: Find the node closest to the random sampling point in the initial tree and determine the node closest to the random sampling point as the nearest node;

[0105] S4043: Calculate the random direction unit vector from the nearest node to the random sampling point, and calculate the gradient direction unit vector of the prediction variance at the nearest node;

[0106] S4044: The random direction unit vector and the gradient direction unit vector are weighted and synthesized according to a preset guiding coefficient to obtain the final expansion direction;

[0107] S4045: Determine the expansion step size based on the confidence region type to which the nearest node belongs. If the nearest node is located in a high confidence region, a preset large step size is used as the expansion step size; if the nearest node is located in a low confidence region, a preset small step size is used as the expansion step size.

[0108] S4046: Add the product of the expansion step size and the final expansion direction to the nearest node to obtain the position of the new node;

[0109] S4047: If the connection from the nearest node to the new node satisfies the preset drilling process constraints and does not collide with the preset geological obstacles, then add the new node to the initial tree.

[0110] Specifically, the non-uniform probability density function is constructed based on the prediction variance of grid cells. The core logic is that grid cells with larger prediction variances have a higher probability of being sampled, thus guiding the trajectory to extend into areas of high uncertainty and improving the information gain of the exploration. For example, the mathematical expression of the non-uniform probability density function is:

[0111]

[0112] in, For the spatial coordinates of the grid cell Sampling probability density at location, coordinates The prediction variance corresponding to the grid cell, The set of all confidence region grid cells. This is the sum of the prediction variances of all grid cells within the confidence region, used for normalization to ensure that the sum of the probability densities is 1. During sampling, the coordinates of random sampling points can be generated based on this probability density function using roulette wheel sampling or inverse transform sampling to ensure that the sampling results meet the requirements of a non-uniform distribution.

[0113] Furthermore, to improve search efficiency, the nodes of the initial tree can be stored using a spatial index structure, such as a KD-tree or an octree. During the search for the nearest node, the candidate node set can be quickly located using the spatial index, and then the Euclidean distance between the candidate node and the random sampling point is calculated. The node with the smallest distance is selected as the nearest node. Based on this nearest node, its random direction unit vector can be obtained by normalizing the difference vector between the coordinates of the random sampling point and the coordinates of the nearest node. The gradient direction of the prediction variance represents the direction in which uncertainty increases the fastest at this node. The gradient can be calculated using the central difference method, that is, by obtaining the prediction variance of the grid cells adjacent to the nearest node, calculating the partial derivatives in the x, y, and z directions in three-dimensional space to form a gradient vector, and then normalizing the gradient vector to obtain the gradient direction unit vector. This vector provides geological information guidance for subsequent expansion directions.

[0114] Specifically, a preset guidance coefficient is used to adjust the weight ratio of random exploration and geological information guidance. Its value is dynamically calibrated based on the current initial tree expansion scale. For example, in the initial stage, the guidance coefficient is smaller to prioritize random exploration capabilities and cover a wider geological space. However, as the tree structure expands, the guidance coefficient can gradually increase to strengthen gradient direction guidance, focusing the expansion towards areas of high uncertainty. Subsequently, the random direction unit vector and the gradient direction unit vector can be weighted and synthesized according to the preset guidance coefficient. The mathematical expression for the weighted synthesis is:

[0115]

[0116] in, The final extension direction unit vector, The preset guiding coefficient (value range 0-1) A unit vector with a random direction. This is a unit vector representing the gradient direction. After synthesis, it can also be used to... The vector is normalized again to ensure it is a unit vector and to guarantee the consistency of the expansion direction.

[0117] Specifically, since the geological and mineralization information in high-confidence areas is clear and has low uncertainty, a large step size can be used to improve trajectory generation efficiency without excessively detailed exploration. Conversely, in low-confidence areas with high uncertainty, a small step size can be used to improve sampling accuracy and ensure that the trajectory accurately captures high-value geological areas. Furthermore, the preset large and small step sizes can be calibrated based on the resolution of the 3D mesh model and the precision requirements of drilling operations, while satisfying the constraints that the large step size is greater than the small step size and both are greater than the side length of the mesh cells. This avoids computational redundancy due to an excessively small step size or a coarse trajectory due to an excessively large step size. Based on the determined step size, the coordinates of new nodes can be calculated according to the spatial vector superposition rule. That is, the coordinates of the new node are equal to the coordinates of the nearest node plus the product of the expansion step size and the unit vector of the final expansion direction. This ensures that the new node expands along the preset direction, and the spatial distance to the nearest node is equal to the expansion step size. This calculation process can be directly implemented through coordinate operations to ensure the accuracy of the new node's position.

[0118] Furthermore, during the connection detection process from the nearest node to the new node, preset drilling process constraints can include borehole inclination constraints and curvature constraints. The inclination constraint limits the angle range between the borehole and the vertical or horizontal direction to avoid exceeding the mechanical operating capabilities of the drilling equipment. The curvature constraint requires that the radius of curvature of the connection line be no less than a preset minimum value to prevent excessive curvature of the trajectory from causing the drilling tool to jam or be damaged. Collision detection is achieved by determining the spatial intersection of the connection line from the nearest node to the new node with preset geological obstacles. Geological obstacles are stored in the form of a set of three-dimensional spatial coordinates or a geometric model. If the connection line does not intersect with an obstacle, a collision is determined to be absent. After the constraints are met, the coordinates of the new node, the parent node index (pointing to the nearest node), and the label of its confidence region can be entered into the initial tree to complete node expansion.

[0119] To illustrate, the above sampling expansion operation is repeated until the Euclidean distance between the new node and the current target point is less than a preset distance threshold, ensuring that the spatial deviation between the new node and the target point is within the allowable range. At this point, the iterative sampling expansion can be stopped. This preset distance threshold can be set based on the positioning accuracy requirements of the drilling operation.

[0120] S405: Connect the new node with the current target point. If the connection is successful, backtrack from the current target point along the initial tree to the root node to obtain the initial path node sequence.

[0121] Specifically, connectivity can be achieved by constructing a straight line connecting the new node to the current target point. Then, it is checked whether the connection satisfies the preset drilling process constraints and the no-collision condition. If so, connectivity is considered successful. The backtracking process can be achieved by traversing the parent node indices of the nodes. Starting from the current target point, the parent node pointers are used to find the next higher-level node sequentially until the root node (the current drill bit position) is reached. All nodes traversed during the backtracking process are recorded in the order of "current target point → root node". This sequence is then reversed to obtain the initial path node sequence of "root node → current target point", which is a discrete set of nodes.

[0122] S406: Perform cubic B-spline interpolation smoothing on the initial path node sequence to generate the drilling trajectory sequence.

[0123] Specifically, cubic B-spline interpolation can transform discrete node sequences into continuous, smooth curves, meeting the continuity requirements of mechanical motion during drilling operations and avoiding construction difficulties caused by abrupt trajectory changes. During the interpolation process, the initial path node sequence can be used as control points for the B-spline curve. By constructing cubic B-spline basis functions, continuous coordinate points on the curve are calculated. For example, the parameterization method of the interpolation nodes (such as uniform parameterization or cumulative chord length parameterization) can be determined first to ensure the smoothness of the curve in densely populated node regions. Subsequently, the spatial coordinates corresponding to any parameter are calculated based on the basis functions, generating a sufficiently dense continuous trajectory point. Finally, the obtained drilling trajectory sequence is output as an array of coordinate points. The density of the trajectory points can be set based on the control precision of the drilling operation to ensure that the curvature of each trajectory segment meets the process constraints, thereby obtaining a feasible and smooth drilling path from the current drill bit position to the current target point.

[0124] In one embodiment, the dynamic path optimization prediction method S4 for borehole layout in mineral exploration provided by this application can simulate the drilling trajectory sequence based on the mineralization probability and prediction variance in the current model, according to preset drilling simulation rules. These preset drilling simulation rules are constructed based on the geological information acquisition logic of actual drilling. For example, the drilling trajectory sequence can be discretized according to a preset sampling interval. The sampling interval setting needs to match the resolution of the 3D grid model and the actual accuracy requirements of drilling data acquisition, ensuring that each sampling point corresponds to a unique grid cell in the 3D grid. For each sampling point, the mineralization probability and prediction variance of the corresponding grid cell in the current model are extracted and used as the model prediction value. Subsequently, a verification value is generated based on a preset error simulation mechanism. This error simulation mechanism needs to consider factors such as the detection accuracy and core recovery rate of actual drilling, and correct the model prediction value by introducing a random error term (following a normal distribution with a preset variance), making the verification value closer to the actual drilling detection results. Finally, simulated drilling feedback data can be output, which can include two parts: first, the geological parameter verification results at each sampling point along the trajectory, i.e., the corrected mineralization probability verification value; second, the uncertainty correction coefficient, which is calculated based on the deviation between the model prediction value and the verification value. The larger the deviation, the larger the correction coefficient, indicating that the uncertainty of the model in that area needs to be significantly adjusted. In addition, the feedback data can also include auxiliary information such as the spatial coordinates of the sampling points and the type of confidence region to which they belong, so as to provide accurate spatial positioning and parameter correlation basis for subsequent model updates.

[0125] Furthermore, based on simulated drilling feedback data, the mineralization probability and prediction variance of grid cells can be dynamically corrected, enabling iterative accumulation of geological information. For example, for updating the mineralization probability, a Bayesian posterior probability fusion algorithm can be used. This algorithm uses the mineralization probability in the current model as the prior probability and the validation values ​​from the simulated drilling feedback data as the observation data. By fusing the prior and observation information, the posterior mineralization probability is obtained. The core formula is:

[0126]

[0127] in, This represents the updated posterior probability of mineralization. This represents the prior mineralization probability before the update (taken from the current model). Let be the likelihood function, representing the observed verification value given a prior probability of mineralization. The probability and likelihood function can be set based on the error distribution characteristics of drilling detection, such as the normal distribution likelihood function. For updating the prediction variance, the principle of "higher verification accuracy, smaller variance" can be followed. If the uncertainty correction coefficient of a certain grid cell in the feedback data is small (i.e., the deviation between the model prediction and the verification value is small), the prediction variance of that cell is multiplied by the correction coefficient to reduce it. If the correction coefficient is large, the prediction variance is appropriately increased or the original variance is maintained to reflect the model's cognitive deficiencies in that area. During the update process, parameters can be adjusted for grid cells along the drilling trajectory and within the preset influence range. The influence range can be set based on the continuity characteristics of the geological body and the effective radiation range of the drilling data to ensure the spatial correlation and rationality of the model update, ultimately forming an updated model that includes the updated mineralization probability and prediction variance.

[0128] Specifically, the simulated final borehole location is the coordinate of the endpoint of the drilling trajectory sequence. These coordinates, verified by drilling process constraints, serve as the basis for final borehole location in actual construction. When added to the set of drilled borehole locations, the construction parameters of that borehole, such as trajectory length, final borehole depth, and iteration number, can be recorded simultaneously. The set of drilled borehole locations can be stored using a spatial index structure, supporting fast querying and range retrieval, providing a basis for determining the subsequent target area set. The simulated drilling cost can be calculated considering three main factors: trajectory length, geological difficulty, and process constraints. The trajectory length cost is the product of the basic unit length cost and the total trajectory length. The geological difficulty cost is obtained by weighted summation of parameters such as rock compressive strength and geological complexity of each grid unit along the trajectory. The process constraint cost is set according to the degree to which the trajectory meets dip angle and curvature constraints. If the trajectory repeatedly approaches the constraint threshold, additional process adjustment costs are required. The above costs are accumulated to obtain the single-hole simulated drilling cost, which is included in the cumulative total cost. The cumulative total cost reflects the economic investment in exploration in real time, providing data support for determining subsequent budget constraint-based termination conditions.

[0129] Finally, the update results can be used to determine whether the preset termination conditions are met. These preset termination conditions can include three core judgment dimensions, and meeting any one of them constitutes termination: The first is a budget constraint condition: when the cumulative total cost reaches or exceeds the preset budget limit, the iteration terminates to ensure that exploration investment is controlled within an economically acceptable range. The second is a target coverage condition: when the proportion of uncontrolled grid cells in the target area set is lower than the preset coverage threshold, the iteration terminates, indicating that the core exploration target has been achieved. The third is an uncertainty constraint condition: when the mean prediction variance of all confidence area grid cells in the updated model is lower than the preset variance threshold, the iteration terminates, indicating that the existing geological data is sufficient to support exploration decisions and further drilling is unnecessary. If none of the above termination conditions are met, the next iteration can be started. This involves using the updated model as the new current model, ensuring that subsequent area division and path optimization are based on the latest geological information, and using the simulated final borehole position as the new current drill bit position, achieving continuous advancement of borehole layout and avoiding repeated positioning and path redundancy. If the termination conditions are met, a complete borehole layout plan can be output. In addition to the set of borehole locations, the sequence of drilling trajectories for each round, and the updated model, the plan can also include auxiliary information such as a cost analysis report and a heat map of model uncertainty distribution, so as to provide comprehensive technical support for actual exploration and construction.

[0130] Based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, simulated drilling feedback data is generated according to preset drilling simulation rules. Based on the simulated drilling feedback data, the current model is updated to obtain an updated model. The simulated final hole position corresponding to the current target point is added to the set of drilled hole positions. The simulated drilling cost corresponding to the drilling trajectory sequence is accumulated to the total cumulative cost to obtain the updated result. Based on the updated result, it is determined whether the preset termination condition is met. If not, the updated model and the simulated final hole position corresponding to the current target point are used as the current model and current drill bit position for the next iteration, and S2 to S3 are re-executed. Otherwise, a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model is output.

[0131] Based on the same inventive concept, this application also provides a dynamic path optimization prediction system for mineral exploration borehole layout, used to implement the dynamic path optimization prediction method for mineral exploration borehole layout described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of a dynamic path optimization prediction system for mineral exploration borehole layout provided below can be found in the limitations of the dynamic path optimization prediction method for mineral exploration borehole layout described above, and will not be repeated here.

[0132] In one exemplary embodiment, such as Figure 2As shown, a dynamic path optimization prediction system 500 for mineral exploration borehole layout is provided, including:

[0133] Initialization module 501 is used to execute S1. S1 includes: constructing an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell based on multi-source geological raw data; using the initial three-dimensional grid model as the current model, initializing the current drill bit position as the exploration area entry point, initializing the set of drilled hole positions as an empty set, and initializing the cumulative total cost to zero.

[0134] The region partitioning and graph construction module 502 is used to execute S2. S2 includes: partitioning the grid cells into regions based on the mineralization probability and prediction variance in the current model, obtaining the partitioning results, and generating a confidence region label matrix; determining the target region set based on the set of drilled borehole locations; and performing hierarchical sparsification processing on the grid cells based on the confidence region label matrix to construct a weighted directed graph.

[0135] The path optimization and trajectory generation module 503 is used to execute S3. S3 includes: starting from the current drill bit position and targeting the target region set, performing path optimization on the weighted directed graph using an improved A* search algorithm to generate a global borehole position sequence; extracting the first borehole position from the global borehole position sequence as the current target point; and executing a fast exploration random tree algorithm based on the confidence region based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix to generate a drilling trajectory sequence from the current drill bit position to the current target point.

[0136] The drilling simulation and update module 504 is used to execute S4. S4 includes: generating simulated drilling feedback data based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, and according to the preset drilling simulation rules; updating the current model based on the simulated drilling feedback data to obtain the updated model, adding the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulating the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain the update result; based on the update result, determining whether the preset termination condition is met. If not, the updated model and the simulated final hole position corresponding to the current target point are used as the current model and the current drill bit position for the next iteration, and S2 to S3 are re-executed; otherwise, a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model is output.

[0137] In one exemplary embodiment, the present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the dynamic path optimization prediction method for borehole layout in mineral exploration according to this application. A multi-core processor is preferred to improve the system's parallel processing capability. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of data and computational tasks.

[0138] In one exemplary embodiment, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of a dynamic path optimization prediction method for borehole layout in mineral exploration according to the present application. The computer-readable storage medium may include: a read-only memory, a random access memory, a solid-state drive, or an optical disk, etc.

[0139] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A dynamic path optimization prediction method for borehole layout in mineral exploration, characterized in that, The method includes: S1. Based on multi-source geological raw data, construct an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell; use the initial three-dimensional grid model as the current model, initialize the current drill bit position as the exploration area entry point, initialize the set of drilled hole positions as an empty set, and initialize the cumulative total cost to zero. S2. Based on the mineralization probability and prediction variance in the current model, the grid cells are divided into regions to obtain the division results and generate a confidence region label matrix; the target region set is determined according to the set of drilled borehole locations; the grid cells are subjected to hierarchical sparsification processing according to the confidence region label matrix to construct a weighted directed graph. S3. Starting from the current drill bit position and targeting the target region set, an improved A* search algorithm is used to perform path optimization on the weighted directed graph to generate a global borehole position sequence. The first borehole position is taken from the global borehole position sequence as the current target point. Based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, a fast exploration random tree algorithm guided by the confidence region is executed to generate a drilling trajectory sequence from the current drill bit position to the current target point. S4. Based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, generate simulated drilling feedback data according to preset drilling simulation rules; update the current model according to the simulated drilling feedback data to obtain an updated model, and add the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulate the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain an updated result; based on the updated result, determine whether a preset termination condition is met. If not, use the updated model and the simulated final hole position corresponding to the current target point as the current model and current drill bit position for the next iteration, and re-execute S2 to S3; otherwise, output a drilling layout scheme containing the set of drilled hole positions, the drilling trajectory sequence, and the updated model.

2. The method according to claim 1, characterized in that, The partitioning results include non-confidence regions and confidence regions; the confidence regions include low-confidence regions and high-confidence regions; In step S2, based on the mineralization probability and prediction variance in the current model, the grid cells are divided into regions to obtain the division results and generate a confidence region label matrix. Based on the set of drilled borehole locations, a target area set is determined, including: Read the mineralization probability and prediction variance of each grid cell from the current model, and set a low mineralization probability threshold, a high mineralization probability threshold, and a variance threshold; For each of the grid cells, a region partitioning operation is performed. If the mineralization probability is less than the low mineralization probability threshold, the grid cell is marked as the non-confidence region. If the mineralization probability is greater than or equal to the high mineralization probability threshold, and the prediction variance is less than or equal to the variance threshold, the grid cell is marked as the high confidence region. The grid cells other than those marked as the non-confidence region and the high confidence region are marked as the low confidence region, thus obtaining the partitioning result. Based on the segmentation results, the confidence region label matrix is ​​generated; Based on the set of drilled locations and the preset detection radius, the controlled grid cells are determined; based on the controlled grid cells, the uncontrolled grid cells in the current model with a mineralization probability not lower than the low mineralization probability threshold are summarized to form the target area set.

3. The method according to claim 1, characterized in that, In step S2, based on the confidence region label matrix, the grid cells are subjected to hierarchical sparsity processing to construct a weighted directed graph, including: Based on the confidence region label matrix, delete the grid cells labeled as non-confidence regions; For the grid cells labeled as low-confidence regions, the original spatial coordinates of the grid cells labeled as low-confidence regions are retained, and a first edge with a first weight is established between adjacent grid cells labeled as low-confidence regions. The first weight is the Euclidean distance between two grid cells labeled as low-confidence regions. For the grid cells labeled as high confidence regions, eight spatially adjacent grid cells labeled as high confidence regions are merged into a super node. The geometric center of the eight grid cells labeled as high confidence regions is used as the representative coordinate of the super node. A second edge with a second weight is established between adjacent super nodes. The second weight is calculated by multiplying the Euclidean distance between the representative coordinates of the two super nodes by a preset coarsening factor. A cross-region edge is established between the grid cell labeled as a low-confidence region and the adjacent super node, and the weight of the cross-region edge is the Euclidean distance between the grid cell labeled as a low-confidence region and the adjacent super node. The weighted directed graph is constructed based on the grid cells labeled as low-confidence regions, the super nodes, the first edge, the second edge, and the cross-region edge.

4. The method according to claim 1, characterized in that, The improved A* search algorithm employs an improved cost function, in which the total cost equals the sum of the actual cost and the heuristic function value, and the heuristic function value includes an uncertainty guiding term based on the entropy of mineralization probability information. In step S3, starting from the current drill bit position and targeting the set of target regions, an improved A* search algorithm is used to perform path optimization on the weighted directed graph to generate a global borehole position sequence, including: S201: Read the mineralization probability of each grid cell from the current model; for each grid cell in the weighted directed graph, calculate the corresponding information entropy based on the mineralization probability of each grid cell in the weighted directed graph, the information entropy being used to characterize the degree of geological uncertainty at the grid cell; S202: Based on the proportion of the cumulative total cost to the preset budget limit, dynamically adjust the ratio of the first weight coefficient and the second weight coefficient; for each grid cell in the weighted directed graph, calculate the estimated distance from each grid cell in the weighted directed graph to the target region set; S203: Calculate the first result by multiplying the first weight coefficient by the estimated distance, obtain the second result based on the second weight coefficient and the information entropy, and calculate the heuristic function value of each grid cell in the weighted directed graph based on the first result and the second result; S204: Using the grid cell where the current drill bit position is located as the starting point, initialize the open set and the closed set, and add the starting point to the open set; initialize the actual cost of each grid cell in the weighted directed graph to a preset maximum value, and only set the actual cost of the starting point to zero; calculate the total cost of the starting point according to the cost function; wherein, the total cost of the starting point is equal to the sum of the actual cost of the starting point and the heuristic function value of the starting point; S205: Based on the open set, the closed set, and the actual cost and total cost of each grid cell in the weighted directed graph, repeatedly perform the node expansion operation. When the current grid cell belongs to the target region set, stop the node expansion operation. The node expansion operation includes selecting the current grid cell, processing neighboring grid cells, calculating the trial actual cost, updating the parameters of neighboring grid cells, and maintaining the open set. S206: Starting from the current grid cell, backtracking to the starting point, output the grid cells on the backtracking path in sequence from the starting point to the target area set to obtain the global borehole location sequence.

5. The method according to claim 4, characterized in that, In step S205, based on the open set, the closed set, and the actual and total costs of each grid cell in the weighted directed graph, the node expansion operation is repeatedly executed. When the current grid cell belongs to the target region set, the node expansion operation is stopped, including the following steps: S301: Select the grid cell with the minimum total cost from the open set, and determine the grid cell with the minimum total cost as the current grid cell; S302: Remove the current mesh cell from the open set and move it into the closed set; S303: For each neighboring grid cell of the current grid cell in the weighted directed graph, if the neighboring grid cell is already in the closed set, skip the neighboring grid cell; otherwise, calculate the trial actual cost of reaching the neighboring grid cell from the starting point via the current grid cell. The trial actual cost is calculated based on the actual cost of the current grid cell, the weight of the edge from the current grid cell to the neighboring grid cell, and a penalty term; the penalty term is constructed based on a preset penalty coefficient and the mineralization probability of the neighboring grid cell. S304: If the actual cost of the trial is less than the current actual cost of the neighboring grid cell, then update the parent node pointer of the neighboring grid cell to the current grid cell, update the current actual cost of the neighboring grid cell to the actual cost of the trial, and calculate the total cost of the neighboring grid cell according to the cost function; S305: If the neighboring grid cell is not in the open set, then add the neighboring grid cell to the open set; S306: Return to steps S301 to S305 until the current grid cell belongs to the target region set, triggering the node expansion operation to stop.

6. The method according to claim 2, characterized in that, In step S3, the first borehole position is extracted from the global borehole position sequence as the current target point. Based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, a fast exploration random tree algorithm guided by the confidence region is executed to generate a drilling trajectory sequence from the current drill bit position to the current target point, including: S401: Extract the first borehole position from the global borehole position sequence and determine the borehole position as the current target point; initialize the initial tree with the current drill bit position as the root node; S402: Read the prediction variance of each grid cell from the current model; S403: Set the target bias probability, and use the difference between 1 and the target bias probability as the remaining probability; S404: Based on the target bias probability and the remaining probability, perform a sampling expansion operation to generate a new node. When the distance between the new node and the current target point is less than a preset distance threshold, stop the iterative sampling expansion operation; otherwise, repeat the sampling expansion operation. S405: Connect the new node with the current target point. If the connection is successful, backtrack from the current target point along the initial tree to the root node to obtain the initial path node sequence. S406: Perform cubic B-spline interpolation smoothing on the initial path node sequence to generate the drilling trajectory sequence; The sampling expansion operation includes: S4041: Generate random sampling points according to the probability selection rule, wherein the probability selection rule is: in each iteration, the current target point is independently selected as a random sampling point based on the target bias probability, or random sampling is performed within the grid cell range belonging to the confidence region based on the remaining probability, according to the non-uniform probability density function; wherein, the non-uniform probability density function is constructed based on the prediction variance of the grid cell; S4042: Find the node closest to the random sampling point in the initial tree, and determine the node closest to the random sampling point as the nearest node; S4043: Calculate the random direction unit vector from the nearest node to the random sampling point, and calculate the gradient direction unit vector of the prediction variance at the nearest node; S4044: The random direction unit vector and the gradient direction unit vector are weighted and synthesized according to a preset guiding coefficient to obtain the final expansion direction; S4045: Determine the expansion step size based on the confidence region type to which the nearest node belongs. If the nearest node is located in the high confidence region, a preset large step size is used as the expansion step size; if the nearest node is located in the low confidence region, a preset small step size is used as the expansion step size. S4046: The product of the expansion step size and the final expansion direction is added to the nearest node to obtain the position of the new node; S4047: If the connection from the nearest node to the new node satisfies the preset drilling process constraints and does not collide with the preset geological obstacles, then the new node is added to the initial tree.

7. The method according to claim 5, characterized in that, The mathematical expression for the cost function is: in, For grid cells The total cost; From the starting point to the grid cell The actual cost mentioned above; For grid cells The parent node in the A* search tree; For nodes The weight of the edge between the node and its parent node; , All are geological difficulty correction factors; For grid cells The compressive strength of the rock; For grid cells Drilling design inclination angle at the location; It is a heuristic function; , These are all heuristic function weight coefficients, namely the first weight coefficient and the second weight coefficient; For grid cells Set up in the target area The estimated distance is the shortest Euclidean distance; For grid cells The information entropy of the mineralization probability; The number of multi-source geoscience data types, including geology, geophysics, and geochemistry; For the first The reliability weights of the data source satisfy the following conditions: ; For the first Data source in grid cell The credibility of the value ranges from 0 to 1. For the aforementioned penalty item; , , All are penalty coefficients; For grid cells The mineralization probability ranges from 0 to 1. This is an indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. This is the set of boundary grids for the high-confidence region.

8. A dynamic path optimization prediction system for borehole layout in mineral exploration, characterized in that, The system includes: An initialization module is used to execute S1, which includes: constructing an initial three-dimensional grid model containing the initial mineralization probability and initial prediction variance of each grid cell based on multi-source geoscience raw data; using the initial three-dimensional grid model as the current model, initializing the current drill bit position as the exploration area entry point, initializing the set of drilled hole positions as an empty set, and initializing the cumulative total cost to zero. The region partitioning and graph construction module is used to execute S2, which includes: partitioning the grid cells into regions based on the mineralization probability and prediction variance in the current model, obtaining the partitioning results, and generating a confidence region label matrix; determining the target region set based on the set of drilled borehole locations; and performing hierarchical sparsification processing on the grid cells based on the confidence region label matrix to construct a weighted directed graph. The path optimization and trajectory generation module is used to execute S3, which includes: starting from the current drill bit position and targeting the target region set, performing path optimization on the weighted directed graph using an improved A* search algorithm to generate a global borehole position sequence; extracting the first borehole position from the global borehole position sequence as the current target point; and based on the current drill bit position, the prediction variance in the current model, and the confidence region label matrix, executing a fast exploration random tree algorithm guided by the confidence region to generate a drilling trajectory sequence from the current drill bit position to the current target point. The drilling simulation and update module is used to execute S4, which includes: generating simulated drilling feedback data based on the drilling trajectory sequence, combined with the mineralization probability and prediction variance in the current model, according to preset drilling simulation rules; updating the current model according to the simulated drilling feedback data to obtain an updated model, adding the simulated final hole position corresponding to the current target point to the set of drilled hole positions, and accumulating the simulated drilling cost corresponding to the drilling trajectory sequence to the cumulative total cost to obtain an update result; based on the update result, determining whether a preset termination condition is met; if not, using the updated model and the simulated final hole position corresponding to the current target point as the current model and current drill bit position for the next iteration, and re-executing S2 to S3; otherwise, outputting a drilling layout scheme including the set of drilled hole positions, the drilling trajectory sequence, and the updated model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images