Mine geological resource reserve management method

Abstract

The invention relates to the technical field of mining planning and management, in particular to a mine geological resource reserve management method. The method comprises the following steps: collecting mine geological multi-source data, and carrying out multi-source data preprocessing to obtain fused geological data; geological semantic feature extraction is carried out on the fused geological data, semantic reconstruction is carried out, and semantic geological units are obtained; constructing a variation function by using the fused geological data; performing reserve probability calculation according to the variation function to obtain unit probability data; performing Bayesian inference according to the unit probability data, and performing unit reserve probability generation to obtain a unit reserve probability; mining unit construction is conducted on the semantic geological units, and mining units are obtained; and according to the mining unit and the unit reserve probability, constructing a mining optimization objective function to obtain a comprehensive objective function. The change of mine resources is accurately reflected through the probability reserve assessment method, the resource utilization rate is improved, and the production cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mine mining planning and management, and in particular to a mine geological resource reserve management method. Background Art

[0002] The existing mine geological resource reserve management methods usually construct static geological models, which cannot reflect the changes in geological conditions during mining and lack a dynamic update mechanism for mine resource reserve management. As mining progresses, new geological information is constantly revealed, and the static model cannot absorb this new information in a timely manner, resulting in a decrease in model accuracy, affecting the accuracy of reserve assessment and mining planning.

[0003] Existing methods mostly use deterministic models for reserve assessment, such as the block model method, which regards the grade of each block as a certain value. However, there is a lot of uncertainty in geological data itself, such as sparse distribution of drilling data and measurement errors. These uncertainties have not been fully considered in traditional methods, affecting the accuracy of resource reserve management.

[0004] In summary, the problems of the existing methods lacking a dynamic update mechanism and the single reserve assessment method need to be solved urgently. Summary of the invention

[0005] Based on this, it is necessary to provide a method for managing mining geological resource reserves to solve at least one of the above technical problems.

[0006] To achieve the above object, a method for managing the reserves of mining geological resources comprises the following steps: Step S1: Collect multi-source data of mine geology, and perform multi-source data preprocessing to obtain fused geological data; extract geological semantic features from the fused geological data, and perform semantic reconstruction to obtain semantic geological units; Step S2: construct a variogram using the fused geological data; calculate the reserve probability based on the variogram to obtain unit probability data; perform Bayesian inference based on the unit probability data and generate unit reserve probability to obtain unit reserve probability; Step S3: construct mining units for semantic geological units to obtain mining units; construct mining optimization objective functions according to mining units and unit reserve probabilities to obtain comprehensive objective functions; construct constraint functions according to mining units to obtain constraint functions; perform multi-objective optimization solutions for mining units according to comprehensive objective functions and constraint functions, and evaluate mining plans to obtain target mining plans; Step S4: simulate the mining process according to the target mining plan to obtain a mining performance data set; perform mining difference analysis on the mining performance data set to obtain a difference analysis report; and modify the geological model according to the difference analysis report and the semantic geological unit to obtain a modified geological model.

[0007] The present invention integrates multi-source geological data and extracts key geological semantic features to construct semantic geological units. First, geological data from different sources, such as drilling data, geological maps, geophysical data, etc., are obtained and preprocessed, including data cleaning, format conversion, coordinate unification, etc., to eliminate data differences and form fused geological data. Then, geological semantic features, such as lithology, mineral composition, structural structure, etc., are extracted from the fused data set, and these features are used to semantically reconstruct the geological space, divide semantic geological units with similar geological attributes, and finally form semantic geological units. Based on the semantic geological unit, a variogram model analysis is performed to quantify the spatial variability and correlation of geological attributes. Then, a reserve probability model is constructed using the variogram, such as a probability model based on a Bayesian network, to express the relationship between geological attributes, spatial correlation and reserves in a probabilistic form. Finally, a Bayesian inference method is used to combine prior information and a unit probability model to generate a unit reserve probability, and a probabilistic estimate of the reserves of each semantic geological unit is performed. According to the semantic geological unit and the mining technical parameters, a mining unit is constructed to divide the ore body into exploitable units. Then, based on the mining unit and the unit reserve probability, a mining optimization objective function is constructed, such as maximizing the net present value, minimizing the mining cost, etc., and the mining constraints are converted into mathematical expressions to form a constraint function. Finally, a multi-objective optimization algorithm, such as a genetic algorithm, a simulated annealing algorithm, etc., is used to solve the mining unit for multi-objective optimization, and the obtained mining plan is evaluated to finally determine the target mining plan. A mining simulation environment is constructed according to the target mining plan, the semantic geological unit and the mining unit. Then, the mining process is simulated in the simulation environment, and mining performance data, such as output, grade, cost, etc., are collected. Next, the simulated mining performance data is compared and analyzed with the actual mining data to find out the differences, and the parameters that need to be corrected in the geological model are determined according to the difference analysis report. Finally, a suitable correction method is selected according to the correction target, the model parameters of the semantic geological unit are adjusted, and the model is verified and evaluated, and finally a corrected geological model set is generated. Therefore, the present invention provides a mine geological resource reserve management method, which effectively solves the problems of the lack of dynamic update mechanism and single reserve evaluation method in the existing geological resource reserve management method by constructing a closed-loop system with dynamic feedback and adopting a probabilistic reserve evaluation method. Accurately reflect changes in mining resources, improve resource utilization and reduce production costs.

[0008] Preferably, step S1 comprises the following steps: Step S11: collecting multi-source data of mine geology; performing multi-source data preprocessing on the multi-source data of mine geology to obtain fused geological data; Step S12: constructing an initial three-dimensional geological model for the fused geological data, and performing preliminary grid division to obtain an initial geological grid model; Step S13: extracting geological semantic features from the fused geological data according to the initial geological grid model to obtain a geological feature vector set; Step S14: performing geological semantic segmentation on the initial geological grid model according to the geological feature vector set to obtain an initial semantic unit set; Step S15: Optimize the initial semantic unit set into semantic geological units to obtain semantic geological units.

[0009] The present invention obtains and pre-processes multi-source data of mine geology to form fused geological data, eliminates the differences in data sources and formats, provides a unified and standardized data basis for subsequent geological modeling and analysis, and improves the consistency and reliability of data. An initial three-dimensional geological model is constructed based on the fused geological data and grid division is performed, the geometric form and spatial distribution of the ore body and surrounding rock are digitized, and an initial geological grid model that is convenient for computer processing and analysis is formed, which lays a foundation for the subsequent extraction and segmentation of geological semantic features. Geological semantic features are extracted from the initial geological grid model, and these features are quantified into a geological feature vector set, so as to realize the digital expression of geological information and provide necessary input data for the subsequent geological semantic segmentation. The initial geological grid model is subjected to geological semantic segmentation using the geological feature vector set, and the ore body is divided into different initial semantic units, which preliminarily realizes the classification and identification of the geological body, and provides a basic unit for the subsequent reserve estimation and mining planning. By optimizing the initial semantic unit set, such as merging similar units, splitting complex units, adjusting boundaries, etc., the semantic geological unit is made more consistent with the geological law and the actual mining situation, the accuracy and practicality of the geological model are improved, and a more reliable geological unit division result is provided for the subsequent reserve management.

[0010] Preferably, step S2 comprises the following steps: Step S21: determining key geological attributes of semantic geological units; Step S22: performing reserve probability calculation on the semantic geological unit based on the Bayesian network structure according to the variation function to obtain unit probability data; Step S23: performing Bayesian inference based on the unit probability data and updating parameters to obtain updated probability data; Step S24: generating reserve probability distribution for semantic geological units according to the updated probability data to obtain unit reserve distribution; Step S25: Generate unit reserve probability for unit reserve distribution to obtain unit reserve probability.

[0011] The present invention constructs a reserve probability model of Bayesian network structure based on variogram and semantic geological unit, expresses the complex relationship between geological attributes, spatial correlation and reserves in the form of probability, and provides an effective method for more comprehensive and accurate assessment of reserve uncertainty. The unit probability data is updated with Bayesian inference and newly added exploration data to obtain updated probability data, so that the reserve probability model can be continuously learned and improved, improves the prediction accuracy and adaptability of the model, and makes the reserve assessment results closer to the actual situation. The reserve probability distribution of each semantic geological unit is generated according to the updated probability data, and the reserves are expressed in the form of probability distribution rather than a single deterministic value, which more comprehensively reflects the uncertainty of the reserves and provides more reliable information for subsequent mining decisions. By sorting and normalizing the unit reserve distribution, the unit reserve probability is generated, and the reserve uncertainty information of all units is stored in a structured form, which provides a convenient data interface and more standardized data input for subsequent mining optimization, and improves the efficiency and accuracy of mining planning.

[0012] Preferably, step S3 comprises the following steps: Step S31: obtaining mining technical parameters and mining constraints; constructing mining units for semantic geological units according to the mining technical parameters to obtain mining units; Step S32: constructing a mining optimization objective function according to the mining units and the unit reserve probabilities to obtain a comprehensive objective function; Step S33: converting the mining constraint conditions into mathematical expressions, and constructing constraint functions according to the mining units and mining technical parameters to obtain constraint functions; Step S34: taking the comprehensive objective function and the constraint function as inputs of the optimization algorithm, performing multi-objective optimization on the mining unit based on the multi-objective optimization algorithm, and obtaining a mining sequence solution set; Step S35: generating a mining plan for the mining sequence solution set to obtain a mining plan; Step S36: Evaluate the mining plan and perform Pareto optimal solution analysis to obtain a target mining plan.

[0013] The present invention obtains mining technical parameters and constraints of the mine, and constructs mining units for semantic geological units based on these parameters, thereby obtaining mining units that are more in line with the actual mining situation, laying the foundation for subsequent mining optimization and making the optimization results more practical. The mining optimization objective function is constructed according to the mining unit and unit reserve probability, and multiple mining objectives of the mine (such as reserve maximization, grade balance, geological structure stability, and mining efficiency) are integrated into a comprehensive objective function, realizing the simultaneous optimization of multiple objectives, so that the mining plan is more in line with the overall interests of the mine. The mining constraints are converted into mathematical expressions, and the constraint function is constructed, and various limiting factors in actual mining are incorporated into the optimization model, which ensures the feasibility of the mining plan and avoids violations of safety regulations or exceeding the production capacity of the mine. The mining unit is optimized and solved using a multi-objective optimization algorithm, and the Pareto optimal solution set is obtained, providing a series of mining plans that take into account multiple objectives, providing decision makers with more choices, and making the decision-making process more flexible and scientific. The mining sequence solution set is converted into a specific mining plan, and the mining plan is refined, including mining sequence, time arrangement, resource allocation, etc., so that the mining plan is more operational and instructive, and is convenient for actual implementation. By evaluating the mining plan and analyzing the Pareto optimal solution, the final target mining plan was obtained, which comprehensively considered multiple objectives and constraints, achieved the optimal allocation and sustainable mining of mine resources, and maximized the economic and social benefits of the mine. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a schematic diagram of the steps of a method for managing the reserves of geological resources in a mine.

[0015] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0016] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0017] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0018] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0019] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic diagram of the steps of the mine geological resource reserve management method of the present invention. In this example, the mine geological resource reserve management method includes the following steps: Step S1: Collect multi-source data of mine geology, and perform multi-source data preprocessing to obtain fused geological data; extract geological semantic features from the fused geological data, and perform semantic reconstruction to obtain semantic geological units; In an embodiment of the present invention, first, the borehole data (lithology, grade, coordinates, etc.), geophysical data (electrical data, etc.), remote sensing data, existing geological maps and reports, etc. in the mining area are comprehensively collected. These multi-source heterogeneous data are preprocessed, including data cleaning (processing missing values, outliers), format conversion (unified data format), coordinate system conversion and alignment (unified to UTM coordinate system). Then, using the grade data in the borehole data, the Kriging interpolation method is used to generate a three-dimensional grade distribution map to form fused geological data. Next, based on the fused geological data and borehole data, an implicit modeling method is used (radial basis function interpolation to construct the stratigraphic interface, and moving surface fitting to construct the fault plane) to construct the initial three-dimensional geological model, and regular grid division is performed (for example, 5m*10m). 5m*(A 5m cubic unit), an initial geological grid model is obtained. Subsequently, the geometric features (coordinates of the center point, volume, shape index, distances to interfaces and faults), property features (average grade, main lithology, density), topological features (adjacency matrix), and geological structure features (distance to faults, degree of fault influence, electrical anomaly features) of each grid unit are extracted, and these features are combined into a feature vector. Finally, based on the set of geological feature vectors, the K-means clustering algorithm is used for geological semantic segmentation, and optimization (treatment of isolated units, unit merging, splitting, boundary adjustment) is performed according to geological prior knowledge (such as stratigraphic contact relationships) to obtain semantic geological units.

[0020] Step S2: Construct a variogram using the fused geological data; calculate the reserve probability according to the variogram to obtain unit probability data; perform Bayesian inference based on the unit probability data and generate the unit reserve probability to obtain the unit reserve probability. In the embodiment of the present invention, the borehole data corresponding to the key geological attributes in the fused geological data (FGD) is used for variogram analysis. Taking the copper grade in a copper deposit as an example: 1. Experimental variogram calculation: Select a semantic geological unit (such as a certain ore body unit), and extract the copper grade data of all boreholes within this unit from the FGD. Calculate the experimental variogram values at different lag distances (h). . The specific calculation formula is: , where N(h) is the number of sample point pairs with a lag distance of h, Z(xi) is the grade value at the position xi, and Z(xi + h) is the grade value at the position at a distance h from xi. Calculate the experimental variogram values in multiple directions (such as 0°, 45°, 90°, 135°) and at different lag distances, and draw the experimental variogram graph. 2. Theoretical variogram model fitting: According to the shape of the experimental variogram graph, select an appropriate theoretical variogram model for fitting. Commonly used theoretical variogram models include the spherical model, exponential model, Gaussian model, etc. For example, if the experimental variogram graph shows an obvious "sill effect", the spherical model can be selected for fitting. The formula of the spherical model is: , when h < a; , when h>=a. Where C0 is the nugget constant, C is the arch height, and a is the range. Using optimization algorithms such as the least squares method, determine the parameters of the theoretical variogram model (C0, C, a) so that the theoretical variogram curve fits the experimental variogram points best. 3. Anisotropy analysis: If the experimental variogram shows obvious differences in different directions, it means that the grade spatial distribution is anisotropic. It is necessary to determine the main range direction (the direction with the smallest variability) and the secondary range direction (the direction with the largest variability), and calculate the variogram parameters in different directions respectively. Repeat the above variogram analysis process for each key geological attribute of each semantic geological unit. Finally, the variogram (VFMS) is obtained, which contains the variogram models and their parameters of all semantic geological units and all key geological attributes. For each semantic geological unit, determine the key geological attributes (such as copper grade and ore body thickness) that need to be probabilistically evaluated. Then, the experimental variogram value of each key geological attribute is calculated by using the borehole data in the fused geological data, and the appropriate theoretical variogram model (such as the spherical model) is selected for fitting, and the parameters of the variogram model (nugget constant, arch height, and range) are determined to obtain the variogram. Next, based on the variogram and semantic geological units, the Bayesian network structure is constructed, the attribute nodes (key geological attributes), reserve nodes, and hidden nodes (such as hidden faults) are defined, and the directed edges between the nodes are determined. The node prior probability distribution is determined based on historical data and expert experience, and the conditional probability table of the attribute node is constructed using the variogram model and the Kriging interpolation method. The conditional probability distribution of the reserve node is then derived through Monte Carlo simulation to obtain the unit probability data. Subsequently, if there is new exploration data, the Gibbs sampling algorithm in the Markov Chain Monte Carlo (MCMC) method is used for Bayesian inference, the model parameters are updated, and the updated probability data is obtained (if there is no new data, the prior distribution is used directly). Finally, the updated probability data is used for random simulation to generate the probability distribution of the reserves of each semantic geological unit, and it is organized and normalized into the unit reserve probability.

[0021] Step S3: construct mining units for semantic geological units to obtain mining units; construct mining optimization objective functions according to mining units and unit reserve probabilities to obtain comprehensive objective functions; construct constraint functions according to mining units to obtain constraint functions; perform multi-objective optimization solutions for mining units according to comprehensive objective functions and constraint functions, and evaluate mining plans to obtain target mining plans; In an embodiment of the present invention, the semantic geological unit is further divided into smaller mining units according to the mining technical parameters (bucket capacity of the shovel loader, size of the blasting block, minimum working space, etc.) to obtain the mining unit. Then, the single objective function of mining optimization (maximum reserves, grade balance, geological structure stability, mining efficiency) is defined and standardized. Next, the weight of each single objective function is determined by the fuzzy comprehensive evaluation method, and the standardized single objective function is weighted and summed to obtain the comprehensive objective function. At the same time, the mining constraints of the mine (maximum slope angle, minimum mining width, maximum mining depth, safety distance, equipment capacity, etc.) are converted into mathematical inequalities or equations, and the penalty function is constructed by the external point method to obtain the constraint function. Subsequently, the comprehensive objective function and the constraint function are used as input, and the non-dominated sorting genetic algorithm II (NSGA-II) is used to solve the multi-objective optimization of the mining unit to obtain a set of Pareto optimal mining sequence solutions. Finally, the mining sequence solution set is converted into specific mining plans (including mining time step division, working face layout, equipment configuration, production plan, etc.), and these mining plans are evaluated (adding economic evaluation indicators such as net present value), and the Pareto front is analyzed using tools such as parallel coordinate graphs. Combined with manager preferences, a target mining plan is selected.

[0022] Step S4: simulating the mining process according to the target mining plan to obtain a mining performance data set; performing mining difference analysis on the mining performance data set to obtain a difference analysis report; and correcting the geological model according to the difference analysis report and the semantic geological unit to obtain a corrected geological model; In the embodiment of the present invention, according to the target mining scheme, semantic geological unit and mining unit, the event-based discrete event simulation method is used to construct a mining simulation environment (including a three-dimensional geometric model, an equipment model, event definition and logical modeling). Then, the simulation environment is used to simulate the mining process of the target mining scheme, and the various data generated during the simulation process (such as the mining time, output, grade of each mining unit, the working time and utilization rate of the equipment, etc.) are recorded to obtain the original simulation data record. Then, the original simulation data is processed and integrated (cleaned, converted, and counted) to generate a mining performance data set. Subsequently, the actual mining data of the mine is obtained, and compared and analyzed with the mining performance data set to find out the differences in output, grade, equipment utilization, etc., and the reasons for the differences are analyzed to generate a difference analysis report. Then, according to the difference analysis report, the geological model parameters that need to be corrected and their correction directions are determined to obtain a model correction target set. Then, according to the model correction target, a suitable correction method (such as inverse distance weighted interpolation, direct modification method, buffer method, etc.) is selected to adjust the model parameters of the semantic geological unit to obtain an adjusted semantic unit set. Finally, the model is verified and evaluated based on the adjusted semantic unit set (compared with the actual mining data), a model verification report is generated, and the final revised geological model is formed based on the verification results.

[0023] Preferably, step S1 comprises the following steps: Step S11: collecting multi-source data of mine geology; performing multi-source data preprocessing on the multi-source data of mine geology to obtain fused geological data; Step S12: constructing an initial three-dimensional geological model for the fused geological data, and performing preliminary grid division to obtain an initial geological grid model; Step S13: extracting geological semantic features from the fused geological data according to the initial geological grid model to obtain a geological feature vector set; Step S14: performing geological semantic segmentation on the initial geological grid model according to the geological feature vector set to obtain an initial semantic unit set; Step S15: Optimize the initial semantic unit set into semantic geological units to obtain semantic geological units.

[0024] In an embodiment of the present invention, first, all available geological data in the mining area are comprehensively collected. These data include: drilling data, geophysical data, remote sensing data, existing geological maps of the mine, geological reports, exploration reports and other document materials. The collected multi-source heterogeneous geological data are preprocessed. Data cleaning: Check whether all data have missing values, abnormal values ​​(such as grade values ​​outside the reasonable range, coordinate errors, etc.) and duplicate values. Data format conversion: Convert data from different sources and formats into a standard format. Coordinate system conversion and alignment: Unify all data into the same coordinate system (such as UTM projection coordinate system) and perform spatial alignment. Data fusion (preliminary): For geochemical analysis data (element grade) in the drilling data, the Kriging interpolation method is used to generate a three-dimensional grade distribution map of the mining area based on the drilling sample point data. A fused geological data (FGD) containing all preprocessed geological data is formed.

[0025] Using the information such as drilling data and geological maps in the fused geological data (FGD), the implicit modeling method is used to construct the initial three-dimensional geological model of the deposit. 1. Stratum interface modeling: Based on the lithology layering information in the drill core logging data, the radial basis function (RBF) interpolation algorithm is used to construct the three-dimensional surface of each stratigraphic interface. The radial basis function interpolation uses the stratigraphic interface points exposed by the drill hole as control points, and generates a continuous and smooth stratigraphic interface surface by calculating the influence of each control point on any point in space. 2. Fault modeling: According to the fault information (fault location, occurrence) exposed in the drilling data and the fault line on the geological map, the moving surface fitting algorithm is used to construct the three-dimensional surface of the fault plane. The moving surface fitting algorithm adjusts the surface morphology to make it best fit the known fault data points. 3. Model combination: Boolean operations (such as cutting and merging) are performed on each stratigraphic interface and fault plane to form a closed three-dimensional geological model, which represents the spatial distribution of the ore body. Then, the initial three-dimensional geological model is regularly meshed. The 3D block grid model is used to divide the ore body and surrounding rock space into a series of cubic units (or cuboid units) of the same size. The size of the grid unit is determined according to the minimum unit size of mining and the calculation accuracy requirements. For example, it is set to 5m* 5m *5m. Finally, the initial geological grid model (IGMM) is obtained.

[0026] Based on the initial geological grid model (IGMM) and fused geological data (FGD), the geological semantic features of each grid cell are extracted and a set of geological feature vectors is constructed. 1. Geometric feature extraction: For each grid cell, its center point coordinates (x, y, z), volume, shape index (such as aspect ratio, sphericity) and distance to the nearest stratigraphic interface and fault plane are calculated. 2. Attribute feature extraction: Using the grade data in the fused geological data (FGD), the average grade of each grid cell is calculated using the inverse distance weighted interpolation method. For lithology, the proportion of drill hole samples of each lithology type in each grid cell is counted, and the lithology type with the largest proportion is taken as the main lithology of the grid cell. The average density of the grid cell is calculated (estimated based on the lithology and its corresponding density value). 3. Topological feature extraction: The adjacency relationship between each grid cell and its adjacent grid cells (upper, lower, left, right, front, and back) is calculated to construct an adjacency matrix. 4. Geological structural features: For each grid cell, its distance to the nearest fault and the degree of influence of the fault on the grid cell are calculated (estimated based on the fault occurrence and rock mass strength). By using the geophysical data (taking electrical data as an example) in the fused geological data, a convolutional neural network (CNN) is used to input the profile of the electrical data to train a convolutional neural network model that can identify electrical anomaly patterns related to mineralization. The trained CNN model is used to extract the deep geological features corresponding to each grid cell. The geometric features, attribute features, topological features, and geological structure features of each grid cell are combined into a feature vector. Finally, a geological feature vector set (GFVS) is obtained, which contains the geological semantic feature information of all grid cells.

[0027] Based on the geological feature vector set (GFVS) and the initial geological grid model (IGMM), the K-means clustering algorithm is used to perform geological semantic segmentation on the grid cells. Data standardization: Standardize the various features in the geological feature vector set (GFVS) to eliminate the influence of dimension and numerical range differences. Clustering: Use the standardized feature vector as the input of the K-means algorithm and set the number of clusters K (determined according to the geological complexity of the deposit and the number of geological unit types, for example, K=10). Boundary optimization: Use geological prior knowledge (such as stratigraphic contact relationships and fault cutting relationships) to optimize the clustering results. Isolated unit processing: For isolated units (units with no similar units around them) or too small units that appear in the clustering results, merge them into the adjacent units with the most similar characteristics.

[0028] Optimize the initial semantic unit set (ISUS) to generate semantic geological units (SGUS) that are more in line with geological laws and mining practices. Unit merging: Merge units based on set rules. For example, if two adjacent semantic units have similar lithology, grade, and geological structure characteristics, and the shape of the merged unit is more regular, then merge the two units into one unit. Unit splitting: Split semantic units with large internal attribute changes. For example, if the grade variation range within a semantic unit exceeds the preset threshold (for example, the grade standard deviation is greater than 10%), split it into multiple sub-units with more uniform grades based on the grade distribution. Boundary adjustment: Adjust the boundary of the semantic unit according to geological rules. For example, if the intersection line between the boundary of a semantic unit and the fault plane is not smooth, smooth the boundary according to the strike and dip of the fault plane to make it more consistent with the actual distribution of the fault. Attribute review: For each optimized semantic unit, recalculate its comprehensive geological attributes (such as average grade, main lithology, density, etc.) to ensure the accuracy of attribute assignment. Manual inspection and adjustment: Geological engineers manually inspect the optimized semantic unit set and manually adjust unreasonable unit divisions based on geological expertise and mining experience.

[0029] Step S21: determining key geological attributes of semantic geological units; Step S22: performing reserve probability calculation on the semantic geological unit based on the Bayesian network structure according to the variation function to obtain unit probability data; Step S23: performing Bayesian inference based on the unit probability data and updating parameters to obtain updated probability data; Step S24: generating reserve probability distribution for semantic geological units according to the updated probability data to obtain unit reserve distribution; Step S25: Generate unit reserve probability for unit reserve distribution to obtain unit reserve probability.

[0030] In the embodiment of the present invention, based on the semantic geological unit (SGUS), for each semantic geological unit, the key geological attributes that need to be probabilistically evaluated are determined. Based on the variogram (VFMS) and the semantic geological unit (SGUS), a Bayesian network structure is constructed for reserve probability calculation. 1. Node definition: For each semantic geological unit, an "attribute node" is created for each key geological attribute (such as copper grade, ore body thickness) to represent the probability distribution of the attribute. A "reserve node" is created to represent the reserve probability distribution of the unit. 2. Edge definition: A directed edge is added from each "attribute node" to the "reserve node" to represent the impact of the attribute on the reserve. According to geological knowledge, it is determined whether there is a dependency relationship between different attributes. If so, a directed edge is added between the corresponding "attribute nodes". For example, if a high-grade ore body is usually accompanied by a larger thickness, a directed edge is added between the "grade node" and the "thickness node". 3. Prior distribution determination: For each "attribute node", its prior probability distribution is determined based on historical data (such as previous exploration data, similar ore deposit data) and expert experience. For example, for copper grade, it can be assumed that it obeys lognormal distribution, and the parameters (mean and variance) of lognormal distribution can be estimated based on historical data. For "reserve node", its range can be preliminarily estimated based on experience, and it is assumed to obey uniform distribution. 4. Likelihood function construction: The conditional probability distribution of each "attribute node" is constructed by using the variogram model in the variogram (VFMS) combined with the Kriging interpolation method. Specifically, for each semantic geological unit, the attribute values ​​and their estimated variances of the unsampled positions in the unit are estimated using the Kriging interpolation method based on the existing drill hole data (from FGD) and the variogram model. The mean estimated by Kriging is used as the expected value of the "attribute node", and the variance estimated by Kriging is used as the uncertainty measure of the "attribute node". According to the reserve calculation formula (such as the volume method: reserves = volume × density × grade), the functional relationship between the "reserve node" and other "attribute nodes" is established, and the conditional probability distribution of the "attribute node" is derived in combination with the conditional probability distribution of the "attribute node". 5. Bayesian network construction: Combine nodes, edges, prior distributions and likelihood functions into a complete Bayesian network. Repeat the above Bayesian network construction process for each semantic geological unit. Finally, the unit probability data (UPMS) is obtained. This set contains the Bayesian network models of all semantic geological units for reserve probability assessment.

[0031] Based on the unit probability data (UPMS) and the newly added exploration data (if any), Bayesian inference is performed to update the model parameters. 1. Data preparation: If there is newly added exploration data (such as the grade and thickness data of the new drill hole), it is organized into a format corresponding to the Bayesian network node. 2. Bayesian inference: Select the Gibbs sampling algorithm in the Markov Chain Monte Carlo (MCMC) method for Bayesian inference. The Gibbs sampling algorithm approximates the posterior distribution of the entire network by iteratively extracting samples from the fully conditional posterior distribution of each node. The specific steps are as follows: (1) Initialization: Set an initial value for each node in the Bayesian network (randomly sampled from the prior distribution). (2) Iterative sampling: For each node, fix the values ​​of other nodes, and calculate the fully conditional posterior distribution of the node based on the conditional probability distribution of the node (determined by the Bayesian network structure and the conditional probability table) and the observed data (if there is new data). Extract a new value from the fully conditional posterior distribution and update the value of the node. Repeat this process until the preset number of iterations is reached. (3) Burning period: Discard the sampling results of the initial stage (burning period) because the sampling results of the initial stage may not have converged to a stable distribution. (4) Sample collection: Collect samples from the sampling results after the burning period to form posterior samples. 3. Parameter update: Use the posterior samples to estimate the posterior probability distribution of each node. For the "reserve node", calculate the expected value, variance, confidence interval and other statistics of the reserves based on its posterior samples. Replace the corresponding prior distribution in the unit probability data (UPMS) with the updated posterior distribution and update the model parameters.

[0032] If there is no new exploration data, the Bayesian inference step is skipped and the prior distribution is used directly to generate the subsequent reserve probability distribution. Finally, the updated probability data (UPMS') is obtained, which contains the Bayesian network model updated (or not updated) by Bayesian inference.

[0033] Based on the updated probability data (UPMS') and semantic geological units (SGUS), the reserve probability distribution of each semantic geological unit is generated.

[0034] Random simulation: For each semantic geological unit, a random simulation is performed using its corresponding Bayesian network model (from UPMS'). Specifically, a set of random values ​​is extracted from the posterior distribution (or prior distribution if Bayesian inference is not performed) of each node in the Bayesian network, and the reserve value is calculated according to the reserve calculation formula (such as the volumetric method: reserve = volume × density × grade). This process is repeated many times (such as 10,000 times) to obtain a series of simulated values ​​of the unit's reserves.

[0035] Probability distribution estimation: Based on the simulated reserve values, a reserve histogram or kernel density estimation curve is drawn to obtain the probability distribution of reserves. Calculate the expected value, variance, P10 (value with a 10% probability of exceeding), P50 (median), P90 (value with a 90% probability of exceeding) and other statistics of reserves to describe the uncertainty of reserves.

[0036] Repeat the above random simulation and probability distribution estimation process for each semantic geological unit. Finally, the unit reserve distribution (URDS) is obtained, which contains the reserve probability distribution information (histogram, kernel density estimation curve, statistics, etc.) of all semantic geological units.

[0037] The unit reserve distribution (URDS) is collated and normalized to generate the unit reserve probability (URPS).

[0038] Data organization: Organize the data in the unit reserve distribution (URDS) to ensure that the reserve probability distribution information of each semantic geological unit is stored in a unified format. For example, the reserve probability distribution of each unit is represented as a table containing multiple quantiles (such as P10, P20, ..., P90) and their corresponding reserve values, or stored as parameters of the probability density function (PDF) (if the probability distribution conforms to a known distribution, such as normal distribution, lognormal distribution, etc.).

[0039] Data normalization: Normalize the reserve data, for example, unify the reserve unit into tons and the grade unit into grams / ton or percentage.

[0040] Quality check: Check the integrity and consistency of the data to ensure that each semantic geological unit has corresponding reserve probability distribution information and that there are no errors or omissions in the data.

[0041] Set generation: The reserve probability distribution information (normalized) of all semantic geological units is combined into a set to form a unit reserve probability distribution (URPS). This set stores the reserve uncertainty information of all units in a structured form, providing input for subsequent mining sequence optimization. URPS can be stored in the form of database tables, XML files, or JSON files.

[0042] Finally, the unit reserve probability set (URPS) is obtained, which stores the reserve probability distribution information of all semantic geological units in a standardized and structured form, providing key input for subsequent mining planning.

[0043] Preferably, step S22 includes the following steps: Step S221: define the attribute of the semantic geological unit - the reserve node; define the hidden node according to the fused geological data; Step S222: determining the Bayesian network structure of the attribute-reserve nodes and hidden nodes to obtain a Bayesian network structure diagram; Step S223: Acquire historical geological data; determine the prior probability distribution of the Bayesian network structure diagram according to the historical geological data to obtain the node prior probability; Step S224: determining the spatial correlation of geological attributes according to the variation function, and constructing a likelihood function according to the Bayesian network structure diagram and the semantic geological unit to obtain a conditional probability table; Step S225: Perform Bayesian inference on the Bayesian network structure diagram according to the node prior probability and the conditional probability table to obtain the node posterior probability; Step S226: extract the posterior probability distribution of the reserve node from the node posterior probability, and generate a unit probability model according to the Bayesian network structure diagram to obtain unit probability data.

[0044] In an embodiment of the present invention, based on the semantic geological unit (SGUS), a node of the Bayesian network is defined for each semantic geological unit. 1. Attribute node definition: For each semantic geological unit, its key geological attributes (from step S21) are defined as "attribute nodes". For example, for a copper ore body unit, its copper grade, ore body thickness, density and other attributes are defined as "copper grade node", "thickness node" and "density node" respectively. 2. Reserve node definition: For each semantic geological unit, a "reserve node" is created to represent the reserve of the unit. All "attribute nodes" and "reserve nodes" of all semantic geological units are combined into a set, namely, attribute-reserve node (ASNS). Then, based on the fused geological data (FGD) and geological prior knowledge, it is determined whether hidden nodes need to be introduced. Hidden nodes represent geological factors that are not directly observed but have an impact on reserves. For example, if there is a hidden fault in the mining area that is not fully exposed by drilling, and the fault may affect the grade and continuity of the ore body, a "hidden fault node" can be created. The value of a hidden node can be discrete (such as "existence" or "non-existence") or continuous (such as the distance from the fault to the ore body). All defined hidden nodes are combined into a set, namely the hidden nodes (HNS).

[0045] Based on the attribute-reserve nodes (ASNS) and hidden nodes (HNS), determine the structure of the Bayesian network, that is, the directed edges between nodes. 1. Edges between attribute nodes and reserve nodes: Add a directed edge from each "attribute node" to its corresponding "reserve node" to indicate the direct impact of the attribute on the reserve. For example, add directed edges from the "copper grade node", "thickness node", and "density node" to the "reserve node" respectively. 2. Edges between attribute nodes: Based on geological knowledge and data analysis, determine whether there is a dependency relationship between different attributes. If so, add directed edges between the corresponding "attribute nodes". For example, if high-grade ore bodies are usually accompanied by greater thickness, add a directed edge between the "copper grade node" and the "thickness node" (the direction can be determined based on the causal relationship, such as from the "grade node" to the "thickness node"). 3. Edges between hidden nodes and other nodes: If there are hidden nodes, determine the relationship between the hidden nodes and other nodes based on geological knowledge. For example, if the presence or absence of a "hidden fault node" affects the grade and continuity of the ore body, add directed edges from the "hidden fault node" to the "copper grade node" and "thickness node". The direction of the edge represents a causal relationship or a conditional dependency. All nodes and directed edges are combined into a directed acyclic graph (DAG), namely the Bayesian network structure graph (BNSG). This graph represents the dependency between variables and provides a basis for subsequent probabilistic inference.

[0046] Collect historical geological data of the mining area and its surrounding areas, including previous exploration data, information on similar deposits, geological reports, scientific research literature, etc. These data are used to determine the prior probability distribution of nodes in the Bayesian network. 1. Prior distribution of attribute nodes: For each "attribute node", determine its prior probability distribution type and parameters based on historical data and expert experience. For example, for the copper grade node, it can be assumed that it obeys a lognormal distribution, and use historical data (such as grade data from previous drill holes) to estimate the parameters (mean and variance) of the lognormal distribution. If historical data is insufficient, you can refer to data from similar deposits or consult geological experts. For discrete attribute nodes (such as lithology), the frequency of different lithology types is statistically calculated based on historical data as its prior probability. 2. Prior distribution of reserve nodes: For each "reserve node", preliminarily estimate its reserve range based on historical data and expert experience, and select a suitable probability distribution to represent its uncertainty. For example, it can be assumed that the reserves obey a uniform distribution or a triangular distribution, and the parameters of the distribution (lower limit, upper limit, most likely value) can be determined based on experience. 3. Hidden node prior distribution: For each "hidden node", its prior probability distribution is determined based on geological knowledge and expert experience. For example, for a "hidden fault node", its probability of existence can be estimated based on the regional tectonic background and geophysical data. The prior probability distributions of all nodes are combined into a set, namely the node prior probability (NPPS). This set provides initial information for Bayesian inference.

[0047] Based on the variogram (VFMS), Bayesian network structure graph (BNSG) and semantic geological unit (SGUS), the likelihood function of the Bayesian network, i.e., the conditional probability table, is constructed. 1. Determination of spatial correlation of geological attributes: For each key geological attribute of each semantic geological unit, the corresponding variogram model is selected from the variogram (VFMS). The model describes the variability and correlation of the attribute in space. 2. Kriging estimation: For each semantic geological unit, the attribute values ​​and their estimated variances of the unsampled locations in the unit are estimated using the Kriging interpolation method based on the existing borehole data (from FGD) and the variogram model. The Kriging interpolation method takes into account spatial correlation and can provide unbiased optimal estimates. 3. Construction of attribute node conditional probability table: For each "attribute node" in the Bayesian network, its conditional probability table (CPT) is constructed based on the results of Kriging estimation. If the node has no parent node, its CPT is its own probability distribution (which can be assumed to obey a normal distribution or other distribution based on the mean and variance estimated by Kriging). If the node has a parent node, its CPT represents the probability distribution of the node given different values ​​of the parent node. For example, if the "copper grade node" has a parent node "hidden fault node", it is necessary to calculate the probability distribution of the "copper grade node" in the presence and absence of hidden faults. 4. Derivation of conditional probability distribution of reserve nodes: For each "reserve node", its conditional probability distribution is derived according to the reserve calculation formula (such as volume method) and the conditional probability table of its parent node (attribute node). This can be achieved through Monte Carlo simulation: randomly select a set of values ​​from the conditional probability distribution of each parent node, calculate the reserve value according to the reserve calculation formula, repeat multiple times, and obtain a series of simulated values ​​of reserves, thereby estimating the conditional probability distribution of reserves. The conditional probability tables of all nodes are combined into a set, namely the conditional probability table (CPTS). This set represents the probability of the child node value given the parent node value, and is the core component of the Bayesian network.

[0048] Based on the node prior probability (NPPS), conditional probability table (CPTS) and Bayesian network structure graph (BNSG), Bayesian inference is performed to calculate the posterior probability distribution of each node. Select a suitable Bayesian inference algorithm. Since this is a static Bayesian network with a limited number of nodes and states, an exact inference algorithm such as variable elimination can be used. The basic idea of ​​the variable elimination method is to decompose the joint probability distribution into a series of product of conditional probabilities by eliminating non-evidence nodes (non-observation nodes) in the network one by one, so as to calculate the posterior probability of the target node (the node for which the posterior probability needs to be calculated). The specific steps include: 1. Determine the order of variable elimination. A good elimination order can reduce the computational complexity. 2. For each variable to be eliminated, multiply all the factors (conditional probability table) in which it is located to obtain a new factor. 3. Sum (or integrate) the variable in the new factor and eliminate it. 4. Repeat steps 2 and 3 until all non-evidence nodes are eliminated. 5. Multiply the remaining factors and normalize them to get the posterior probability distribution of the target node. If the network structure is complex and the number of nodes and states is large, an approximate inference algorithm can be used, such as variational inference or Markov chain Monte Carlo (MCMC) method. The posterior probability distribution of each node (including attribute nodes, reserve nodes, and hidden nodes) is calculated. The posterior probability distributions of all nodes are combined into a set, namely the node posterior probability (NPPS'').

[0049] Based on the node posterior probability (NPPS'') and the Bayesian network structure graph (BNSG), the reserve probability model of each semantic geological unit is extracted. 1. Extraction of reserve node posterior probability distribution: Extract the posterior probability distribution of the "reserve node" corresponding to each semantic geological unit from the node posterior probability (NPPS''). This distribution reflects the updated understanding of the reserves of the unit after considering geological prior knowledge, attribute data and Bayesian network structure. 2. Unit probability model generation: Process the posterior probability distribution of the "reserve node" of each semantic geological unit to generate the probability model of the unit. Specifically, it includes: calculating the expected value, variance, confidence interval and other statistics of the reserves to quantitatively describe the uncertainty of the reserves. Plot the probability density function (PDF) or cumulative distribution function (CDF) curve of the reserves to visualize the probability distribution of the reserves. If the posterior probability distribution conforms to a known distribution (such as normal distribution, lognormal distribution, etc.), its distribution parameters can be estimated. 3. Model set generation: The probability models of all semantic geological units (including statistics, probability distribution curves or distribution parameters) are combined into a set, namely the unit probability data (UPMS). This set stores the reserve probability models of all units in a structured form, providing key input for subsequent mining planning. Finally, the unit probability data (UPMS) is obtained, which contains the reserve probability models of all semantic geological units and provides a basis for the subsequent mining sequence optimization.

[0050] Preferably, step S224 is specifically as follows: Extract the geological attributes of semantic geological units, calculate the correlation coefficients of geological attributes between different spatial positions according to the variogram, and obtain the spatial correlation of attributes; Kriging estimation is performed based on attribute spatial correlation and semantic geological units to obtain attribute node estimation values; According to the estimated values ​​of the attribute nodes, the attribute node conditional probability table is constructed for the Bayesian network structure graph to obtain the attribute node conditional probability table; The conditional probability table of the attribute node is used to derive the conditional probability distribution of the reserve node and obtain the conditional probability table.

[0051] In an embodiment of the present invention, based on the semantic geological unit (SGUS), the key geological attribute data of each semantic geological unit is extracted. For example, for a copper ore body unit, the copper grade values ​​of all the drill hole data inside it are extracted. Then, according to the variogram (VFMS), a variogram model corresponding to the key geological attributes of each semantic geological unit is selected. For example, if the copper grade variogram model of the copper ore body unit recorded in the VFMS is a spherical model with the parameters: nugget constant C0=0.01, arch height C=0.05, and range a=100 meters, the model is used to calculate the correlation coefficient of copper grade between different spatial positions. The specific calculation process is as follows: For any two position points xi and xj in the copper ore body unit, calculate the distance h=||xi-xj|| between them. According to the spherical model formula, the variogram value between the two locations is calculated as γ(h)=0.01+0.05×[1.5(h / 100)-0.5(h / 100)^3] (when h<100 meters) or γ(h)=0.01+0.05=0.06 (when h>=100 meters). Then, the correlation coefficient between the two locations is calculated as ρ(h)=(C0+C-γ(h)) / (C0+C)=1-γ(h) / 0.06. Repeat the above calculation process for each key geological attribute of each semantic geological unit. Finally, the attribute spatial correlation (ASCM) is obtained, which contains the correlation coefficient calculation model (i.e., the variogram model and its parameters) between each semantic geological unit and each key geological attribute at different spatial locations.

[0052] Based on attribute spatial correlation (ASCM) and semantic geological units (SGUS), Kriging estimation is performed for each key geological attribute of each semantic geological unit. Take ordinary Kriging as an example: for a copper ore body unit, estimate the copper grade value of an unsampled position x0 inside it. 1. Determine the neighborhood: According to the range of the variation function model of the copper grade of the unit in ASCM (for example, 100 meters), determine the neighborhood range of x0, and select the drill data points with known copper grade values ​​within the range (assuming that there are n drill data points, their positions are x1, x2,..., xn, and the corresponding copper grade values ​​are Z(x1), Z(x2),..., Z(xn)). 2. Construct the Kriging equation group: According to the principle of ordinary Kriging, construct the Kriging equation group: Σλi×C(xi,xj)+μ=C(x0,xj)(j=1,2,...,n); Σλi=1; Where λi is the weight coefficient, C(xi, xj) is the covariance between positions xi and xj (calculated according to the variogram model in ASCM, C(h)=C0+C-γ(h)), and μ is the Lagrange multiplier. 3. Solve the Kriging equations: Solve the above equations using matrix operations to obtain the weight coefficients λi and the Lagrange multiplier μ. 4. Calculate the estimated value and variance: Calculate the copper grade estimate Z×(x0)=Σλi×Z(xi) at x0. Calculate the estimated variance σ^2(x0)=Σλi×C(x0, xi)+μ. For each key geological attribute of each semantic geological unit, repeat the above Kriging estimation process at all unsampled locations within it (which can be gridded at a certain interval). Finally, the attribute node estimate value (AEVS) is obtained, which contains the Kriging estimate and estimated variance of each semantic geological unit and each key geological attribute at each unsampled location.

[0053] Based on the attribute node estimate value (AEVS) and the Bayesian network structure diagram (BNSG), the conditional probability table (CPT) of the "attribute node" in the Bayesian network is constructed. For a "copper grade node" of a copper ore body unit, if the node has no parent node, its CPT is its own probability distribution. According to the Kriging estimate and estimated variance of the copper grade of the unit in the AEVS, it is assumed that the copper grade follows a normal distribution, and the Kriging estimate is used as the mean of the normal distribution, and the Kriging estimate variance is used as the variance of the normal distribution, so as to determine the CPT of the node. If the "copper grade node" has a parent node (such as a "hidden fault node"), its CPT represents the probability distribution of the "copper grade node" given different values ​​of the parent node. For example, if the "hidden fault node" has two values: "existence" and "non-existence", it is necessary to calculate the probability distribution of the "copper grade node" in the presence and absence of hidden faults (also assuming a normal distribution, and determine the parameters of the normal distribution based on the Kriging estimation results in different situations in AEVS). For other "attribute nodes" (such as thickness nodes and density nodes), repeat the above CPT construction process. Finally, the attribute node conditional probability table (ACPTS) is obtained, which contains the conditional probability tables of all "attribute nodes" in the Bayesian network.

[0054] Based on the attribute node conditional probability table (ACPTS), the conditional probability distribution of the "reserve node" in the Bayesian network is derived. For a "reserve node" of a copper ore body unit, its reserves are determined by the value of its parent node (attribute node, such as "copper grade node", "thickness node", "density node") and the reserve calculation formula (such as volume method: reserves = volume × density × grade). The Monte Carlo simulation method is used to derive the conditional probability distribution of the "reserve node": 1. From the conditional probability table of each "attribute node" in ACPTS, according to the value of its parent node (if there is a parent node), a set of attribute values ​​(such as a set of copper grade values, thickness values ​​and density values) are randomly extracted. 2. According to the reserve calculation formula, the reserve value is calculated using the extracted attribute values. 3. Repeat steps 1 and 2 multiple times (such as 10,000 times) to obtain a series of simulated values ​​of reserves. 4. Based on the simulated reserve values, a histogram or kernel density estimation curve of the reserves is drawn to obtain the conditional probability distribution of the "reserve node". The reserve value can also be divided into several intervals, and the frequency of each interval can be counted to obtain a conditional probability table in discrete form.

[0055] For the "reserve nodes" of other semantic geological units, the above conditional probability distribution derivation process is repeated. Finally, the conditional probability table (CPTS) is obtained, which contains the conditional probability tables of all nodes (including attribute nodes and reserve nodes) in the Bayesian network.

[0056] Preferably, step S23 includes the following steps: Step S231: acquiring newly added exploration data; integrating observation data of the newly added exploration data to obtain integrated observation data; Step S232: performing Gibbs sampling on the Bayesian network structure diagram according to the conditional probability table and the integrated observation data to obtain a posterior sample; Step S233: Calculate the posterior distribution according to the posterior samples to obtain the node posterior probability; Step S234: Use the node posterior probability to update the graph parameters of the Bayesian network structure graph to obtain updated probability data.

[0057] In the embodiment of the present invention, the newly generated exploration data after the mine reserves probability assessment is obtained, including the newly drilled borehole data (lithology, grade, thickness, etc.), the actual geological data (lithology, grade, structure, etc.) revealed by the newly mined mining face, and the new geophysical exploration data. The newly added exploration data is preprocessed and integrated to correspond to the nodes in the Bayesian network, and the integrated observation data (IOD) is obtained.

[0058] Based on the conditional probability table (CPTS), integrated observation data (IOD) and Bayesian network structure graph (BNSG), the Gibbs sampling algorithm is used for Bayesian inference to generate posterior samples. Gibbs sampling is a Markov chain Monte Carlo (MCMC) method that approximates the joint posterior distribution by iteratively sampling from the fully conditional posterior distribution of each variable. The specific steps are as follows: 1. Initialization: Set an initial value for each node in the Bayesian network. It can be randomly drawn from the prior distribution of the node, or set to any reasonable value. 2. Iterative sampling: Sample each node in turn in a predetermined order (or random order). For the node to be sampled currently (assuming it is node Xi): fix the values ​​of all other nodes (referred to as Xi). According to the Bayesian network structure and the conditional probability table (CPTS), determine the fully conditional posterior distribution of node Xi, that is, P(Xi|Xi, IOD). This distribution only depends on the nodes and observation data IOD in Xi's Markov Blanket. Extract a new value from P(Xi|Xi, IOD) as the updated value of node Xi. 3. Repeat step 2 to perform a round of sampling on all nodes, which is called an iteration. 4. Repeat multiple iterations (such as 10,000 times). 5. Burning period: Discard the sampling results in the initial stage (such as the first 1,000 iterations) because these samples may not have converged to a stable distribution. 6. Sample collection: From the iterations after the burning period, collect the values ​​of all nodes at regular intervals (such as every 10 iterations) to form a sample. Combine all collected samples into a set, namely the posterior sample set (PSS).

[0059] Based on the posterior sample (PSS), the posterior probability distribution of each node in the Bayesian network is calculated. For each node: Continuous nodes (such as grade, thickness, reserves): Use the sample value of the node in the posterior sample (PSS) to estimate its probability density function (PDF) using the kernel density estimation (KDE) method. KDE is a non-parametric method that places a kernel function (such as a Gaussian kernel function) at each sample point and then superimposes all kernel functions to obtain an estimate of the probability density function. Statistics such as sample mean, variance, and quantile can also be calculated as features of the posterior distribution. Discrete nodes (such as lithology, whether a hidden fault exists): Count the frequency of each value of the node in the posterior sample (PSS) as an estimate of its posterior probability. For example, if a "hidden fault node" has two values, "existence" and "non-existence", and if "existence" appears 600 times and "non-existence" appears 400 times in 1000 posterior samples, the posterior probability of "existence" is estimated to be 0.6 and the posterior probability of "non-existence" is estimated to be 0.4. The posterior probability distributions (PDFs or discrete probabilities) of all nodes are combined into a set, namely the node posterior probability (NPPS'').

[0060] Based on the node posterior probability (NPPS''), update the parameters of the Bayesian network model to obtain updated probability data. The specific operations include: 1. Update the conditional probability table: replace the corresponding conditional probability distribution in the conditional probability table (CPTS) with the posterior probability distribution in the node posterior probability (NPPS''). For example, if the posterior distribution of the "copper grade node" is the normal distribution N(μ', σ'^2), then update the conditional probability table of the node in the CPTS to the normal distribution. If the posterior probability of the "hidden fault node" is P(existence) = 0.6 and P(non-existence) = 0.4, then update the conditional probability table of the node in the CPTS to these two probability values. 2. Update the reserve probability distribution: For the "reserve node" of each semantic geological unit, its posterior probability distribution is already included in the NPPS''. According to the new posterior distribution, recalculate the expected value, variance, confidence interval and other statistics of the reserves, and update the probability density function (PDF) or cumulative distribution function (CDF) of the reserves. 3. Update model set: Integrate the updated conditional probability table and reserve probability distribution information into the unit probability data (UPMS) to form updated probability data (UPMS'). Finally, the updated probability data (UPMS') is obtained.

[0061] Preferably, step S3 comprises the following steps: Step S31: obtaining mining technical parameters and mining constraints; constructing mining units for semantic geological units according to the mining technical parameters to obtain mining units; Step S32: constructing a mining optimization objective function according to the mining units and the unit reserve probabilities to obtain a comprehensive objective function; Step S33: converting the mining constraint conditions into mathematical expressions, and constructing constraint functions according to the mining units and mining technical parameters to obtain constraint functions; Step S34: taking the comprehensive objective function and the constraint function as inputs of the optimization algorithm, performing multi-objective optimization on the mining unit based on the multi-objective optimization algorithm, and obtaining a mining sequence solution set; Step S35: generating a mining plan for the mining sequence solution set to obtain a mining plan; Step S36: Evaluate the mining plan and perform Pareto optimal solution analysis to obtain a target mining plan.

[0062] In an embodiment of the present invention, the mining technical parameters and mining constraints actually used in the mine are obtained. The mining technical parameters include: the model, rated bucket capacity, loading capacity, and average loading time of the shovel loader; the model, load capacity, average driving speed, and unloading time of the transport truck; blasting parameters, such as the type of explosives, the charge of a single hole, the design of the blasting network, and the maximum amount of explosives in one section; the advancement speed of the mining face, etc. The mining constraints include: the maximum slope angle (such as 45 degrees), the minimum mining width (such as 20 meters), the maximum mining depth (such as -500 meters above sea level), the safety distance (such as more than 50 meters from the fault), etc. According to the mining technical parameters of the mine, the semantic geological unit (SGUS) is further divided into smaller mining units suitable for actual mining. The specific operations include: 1. Unit refinement: According to the bucket capacity of the shovel loader and the size of the blasting block segment, the larger semantic geological unit (such as the entire ore body) is further divided into smaller blocks. For example, if the bucket capacity of the shovel loader is 10 cubic meters and the block size of the blasting design is 5 meters × 5 meters × 5 meters, the semantic geological unit can be refined according to the size of 5 meters × 5 meters × 5 meters. 2. Unit merging: For semantic geological units that are too small (such as isolated, thin interlayers), merge them with adjacent units according to the minimum operating space requirements of the mining equipment. For example, if the minimum operating width of the shovel loader is 3 meters, merge the thin interlayers with a width of less than 3 meters with the ore body units of its upper or lower wall. 3. Unit adjustment: According to the advancement direction and shape of the mining face, adjust the boundaries of the mining unit to make it more consistent with the actual mining geometry. For example, adjust the boundaries of the mining unit to an inclined plane parallel to the mining face instead of a vertical plane. 4. Unit numbering: Uniquely number all the divided mining units. Finally, the mining unit (MUS) is obtained.

[0063] Based on the mining unit (MUS) and unit reserve probability (URPS), the objective function of mining sequence optimization is constructed. The objective function needs to comprehensively consider multiple goals of mining, such as reserve maximization, grade balance, geological structure stability, mining efficiency, etc. Specific operations include: 1. Define a single objective function: Reserve maximization (f1): For a given mining sequence, calculate the sum of the expected reserves of all mining units in it. The expected reserves can be calculated based on the reserve probability distribution of each unit in the unit reserve probability (URPS). The objective function f1 is the inverse of the total reserves (because the optimization algorithm usually minimizes the objective function). Grade balance (f2): For a given mining sequence, calculate the variance of the expected grade (based on the URPS) of all mining units in it. The objective function f2 is this variance. Geological structural stability (f3): According to the geological structural information (such as faults, fractures, etc.) in the semantic geological unit (SGUS), the geological structural stability of each mining unit is evaluated. A stability score function can be defined, for example, the farther away from the fault and the better the rock integrity, the higher the score. The objective function f3 is the inverse of the average stability score of all units in the mining sequence. Mining efficiency (f4): According to the geometric characteristics of the mining unit (such as location, volume) and the mining technology parameters of the mine (such as the efficiency of shovel loaders and haul trucks), the mining time and transportation time of each mining unit are estimated. The objective function f4 may be the total production time of the production sequence, the total transportation time, or a weighted sum thereof.

[0064] 2. Standardization of objective functions: Since different objective functions have different dimensions and value ranges, they need to be standardized to make them comparable. The range standardization method can be used to map the value of each objective function to the interval [0, 1].

[0065] 3. Weight determination: Determine the weight of each objective function to reflect its importance in the comprehensive objective function. The weight can be determined by using the analytic hierarchy process (AHP) or expert scoring method.

[0066] 4. Construction of comprehensive objective function: Multiply the standardized single objective function by the corresponding weight, and then add them together to obtain the comprehensive objective function (COF). For example: COF = w1 × f1' + w2 × f2' + w3 × f3' + w4 × f4', where w1, w2, w3, and w4 are the weights of the four objective functions, and f1', f2', f3', and f4' are the standardized objective function values.

[0067] Convert the mining constraints of the mine into mathematical expressions and construct constraint functions. The specific operations include: 1. Classification of constraints: Mining constraints are divided into geometric constraints, production constraints and geological constraints.

[0068] Geometric constraints: such as maximum slope angle, minimum mining width, and maximum mining depth. Production constraints: such as the loading capacity of shovel loaders and the transportation capacity of transport trucks. Geological constraints: such as the safe distance from faults.

[0069] 2. Mathematical expression of constraints: Convert each constraint into a mathematical inequality or equation.

[0070] Maximum slope angle constraint: Assume that mining units i and j are adjacent in the vertical direction, with i on top and j on the bottom. Let their height difference be Δh, horizontal distance be Δd, and maximum slope angle be α. Then the constraint can be expressed as: arctan(Δh / Δd)<=α. Minimum mining width constraint: Let the width of mining unit i be wi, and the minimum mining width be Wmin. Then the constraint can be expressed as: wi>=Wmin. Maximum mining depth constraint: Let the bottom elevation of mining unit i be zi, and the maximum mining depth be Zmax. Then the constraint can be expressed as: zi>=Zmax. Loading capacity constraint of shovel loader: Let the total number of mining units mined in each time period be S, and the maximum loading capacity of the shovel loader be Qmax. Then the constraint can be expressed as: Σqi<=Qmax(i∈S), where qi is the amount of ore in mining unit i. Transport truck transportation capacity constraint: Let the total number of mining units transported in each time period be S, and the maximum transportation capacity of the transport truck be Tmax. The constraint condition can be expressed as: Σqi<=Tmax(i∈S). Safety distance constraint from the fault: Let the shortest distance from mining unit i to the fault be di, and the safety distance be Dsafe. The constraint condition can be expressed as: di>=Dsafe.

[0071] 3. Constraint function construction: Use the exterior point penalty function method to convert the constraint conditions into penalty terms of the objective function. For the inequality constraint g(x)<=0, the penalty term is P(x)=α×max(0,g(x))^2, where α is the penalty factor. For the equality constraint h(x)=0, the penalty term is P(x)=β×h(x)^2, where β is the penalty factor. Add up all the penalty terms to get a total penalty function as the constraint function.

[0072] The comprehensive objective function constructed in step S32 and the constraint function constructed in step S33 are used as inputs and solved using the NSGA-II algorithm. Set the population size to 100, the number of iterations to 500, the crossover probability to 0.9, and the mutation probability to 0.1. The NSGA-II algorithm first initializes a population containing 100 random mining sequences. Each mining sequence represents the mining order of a mining unit. The algorithm sorts the individuals in the population by non-dominated sorting and crowding distance calculation, selects excellent individuals for crossover and mutation operations, and generates a new mining sequence. The crossover operation exchanges some fragments of the two parent mining sequences to generate two new child sequences. The mutation operation randomly changes the mining order of certain units in the mining sequence. After 500 iterations, a set of 100 Pareto optimal solutions is finally output, namely the mining sequence solution set. Each Pareto optimal solution represents a mining sequence that cannot be dominated by other solutions in terms of the comprehensive objective function. For example, one solution may perform best in terms of reserve maximization, while another solution may perform best in terms of geological structural stability.

[0073] Traverse each mining sequence in the mining sequence solution set obtained in step S34. Take one of the mining sequences as an example, which defines the mining order of 1000 mining units. According to the volume of each mining unit and the preset blasting parameters (for example, the single-hole charge is 2kg / m³, and the explosive type is emulsion explosive), calculate the amount of explosives and blasting time required for each mining unit. Assume that the average blasting time for each mining unit is 2 hours. Combined with the parameters of the existing shovel loaders in the mine (for example, 3 shovel loaders with a bucket capacity of 10m³, the average loading time is 5 minutes / bucket) and transport trucks (for example, 5 trucks with a load capacity of 50 tons, the average transportation time is 30 minutes / trip), arrange the operation plan of the shovel loaders and transport trucks. For example, divide 1000 mining units into 100 batches, each batch contains 10 mining units. The mining process of each batch is as follows: First, blasting operation is carried out, which takes 2 hours. Then, three shovel loaders load the ore at the same time. Each shovel loader needs to load about 34 buckets of ore (total volume of 10 units / bucket capacity), which takes about 170 minutes. Five transport trucks transport the ore in a cycle, and each truck transports 50 tons of ore each time. The number of transportations is calculated based on the ore density and volume of each mining unit. Assuming that each truck needs to transport 5 times on average, the total transportation time is about 150 minutes. Therefore, the mining time for each batch is about 2+170 / 60+150 / 60=7.33 hours. The mining plans of 100 batches are arranged in sequence to form a complete mining plan, including the specific mining time of each mining unit, the required equipment and personnel arrangements, blasting parameters, transportation routes and other information.

[0074] Each mining scheme generated in step S35 is evaluated. The net present value (NPV) and the geological structure stability score are used as examples for explanation. The net present value of each mining scheme is calculated based on the ore quantity, grade, mining cost (e.g., explosive cost, equipment depreciation, labor cost, etc.) and metal price of each mining unit in each mining scheme. The geological structure stability score of each mining unit is calculated based on factors such as the distance between each mining unit and the fault and the quality of the rock mass, and the scores of all mining units are averaged as the geological structure stability score of the mining scheme. The net present value and geological structure stability score of all mining schemes are plotted in a two-dimensional coordinate system to form a Pareto frontier graph. The points on the Pareto frontier graph represent the mining schemes corresponding to the Pareto optimal solution. By analyzing the Pareto frontier graph, a mining scheme that strikes a balance between different objectives can be found. For example, a certain scheme has the highest net present value, but a lower geological structure stability score; while another scheme has a slightly lower net present value, but a higher geological structure stability score. The final target mining scheme is selected based on the actual situation and risk preference of the mine. For example, if the mine focuses on economic benefits, it can choose the option with the highest net present value; if the mine focuses on safe production, it can choose the option with the highest geological structure stability score.

[0075] Preferably, step S32 includes the following steps: Step S321: defining a single objective function according to the mining unit and the unit reserve probability to obtain a single objective function, wherein the single objective function includes a reserve maximization objective, a grade balance objective, a geological structure stability objective, and a mining efficiency objective; Step S322: performing single objective function standardization on the single objective function to obtain a standardized single objective function; Step S323: determining the target weight based on fuzzy comprehensive evaluation according to the standardized single objective function to obtain a target weight vector; Step S324: construct a comprehensive objective function based on the standardized single objective function and the objective weight vector to obtain a comprehensive objective function.

[0076] In the embodiment of the present invention, four single objective functions are defined according to the mining unit (MUS) and the unit reserve probability (URPS). The total number of mining units is 50, numbered 1-50. The reserve probability distribution of each mining unit is known, expressed in discrete form, including three quantiles of P10, P50, and P90 and their corresponding reserve values ​​(unit: tons). The copper grade probability distribution of each mining unit is known, expressed in discrete form, including three quantiles of P10, P50, and P90 and their corresponding copper grade values ​​(unit: %). The distance from each unit to the fault is known (unit: meter). 1. Reserve maximization objective (f1): For a given mining sequence (an arrangement of 50 mining units), calculate its total expected reserves. The expected reserves of each mining unit are the reserves value corresponding to its P50 quantile. The total expected reserves of the mining sequence are the sum of the expected reserves of all mining units. The objective function f1 is the inverse of the total expected reserves (because the optimization algorithm usually minimizes the objective function). 2. Grade balance target (f2): For a given mining sequence, calculate the variance of the copper grade of the mining. First, calculate the expected copper grade of each mining unit as the copper grade value corresponding to its P50 quantile. Then, calculate the variance of the expected copper grade of all mining units in the mining sequence. The objective function f2 is this variance. 3. Geological structural stability target (f3): For a given mining sequence, calculate its average distance to the fault. The distance from each mining unit to the fault is known. The average distance to the fault of the mining sequence is the average of the distances to the fault of all mining units. The objective function f3 is the inverse of the average distance to the fault. 4. Mining efficiency target (f4): For a given mining sequence, calculate its total mining time. Assume that the shovel loader can process 100 cubic meters of ore per hour and the haul truck can transport 50 cubic meters of ore per hour. The volume of each mining unit is known (calculated based on its geometry and size). The mining time of a mining unit is its volume divided by the shovel loader processing speed. The transportation time of a mining unit is its volume divided by the transportation speed of the transport truck. The total mining time of the mining sequence is the sum of the mining time and transportation time of all mining units. The objective function f4 is the total mining time. Finally, a single objective function is obtained, which includes f1, f2, f3 and f4.

[0077] The single objective function obtained in step S321 is standardized. The range standardization method is adopted. Assume that the following data are obtained by calculating a large number of randomly generated mining sequences: the maximum value (-f1_min, i.e. the minimum value of the total expected reserves) of the reserve maximization target (f1) is -10 million tons, and the minimum value (-f1_max, i.e. the maximum value of the total expected reserves) is -6 million tons. The maximum value (f2_max) of the grade balance target (f2) is 0.1, and the minimum value (f2_min) is 0.01. The maximum value (-f3_min) of the geological structure stability target (f3) is -50 meters, and the minimum value (-f3_max) is -200 meters. The maximum value (f4_max) of the mining efficiency target (f4) is 50,000 hours, and the minimum value (f4_min) is 30,000 hours.

[0078] For a given mining sequence, the four single objective function values ​​are: f1 = -8 million tons, f2 = 0.05, f3 = -100 meters, f4 = 40,000 hours. The standardized single objective function value is: f1'=(f1-f1_min) / (f1_max-f1_min)=(-800-(-1000)) / (-600-(-1000))=200 / 400=0.5; f2'=(f2-f2_min) / (f2_max-f2_min)=(0.05-0.01) / (0.1-0.01)=0.04 / 0.09=0.444; f3'=(f3-f3_min) / (f3_max-f3_min)=(-100-(-50)) / (-200-(-50))=-50 / -150=0.333; f4'=(f4-f4_min) / (f4_max-f4_min)=(40000-30000) / (50000-30000)=10000 / 20000=0.5; Finally, a standardized single objective function is obtained, including f1', f2', f3', and f4'.

[0079] According to the standardized single objective function, the objective weight vector is determined by using the fuzzy comprehensive evaluation method.

[0080] 1. Establish a factor set: The factor set U = {u1, u2, u3, u4}, which corresponds to the four goals of reserve maximization, grade balance, geological structure stability, and mining efficiency.

[0081] 2. Establish a comment set: comment set V = {v1, v2, v3, v4, v5} = {very important, important, general, not very important, very unimportant}.

[0082] 3. Establish a fuzzy relationship matrix: Invite 5 experts to evaluate the importance of the four goals. Each expert gives the membership of each goal relative to each comment in the comment set. For example, for goal u1 (maximizing reserves), an expert gives the following membership: (0.6, 0.3, 0.1, 0, 0), which means that the expert believes that the goal of maximizing reserves has a membership of 0.6 for "very important", 0.3 for "important", 0.1 for "general", and 0 for "not very important" and "very unimportant". The evaluation results of the 5 experts are combined to obtain the fuzzy relationship matrix R: R = | 0.6 0.3 0.1 0.0 0.0 | (u1) | 0.2 0.4 0.3 0.1 0.0 | (u2) | 0.3 0.4 0.2 0.1 0.0 | (u3) | 0.1 0.3 0.4 0.2 0.0 | (u4) Each row represents a goal and each column represents a rating.

[0083] Determine the weight vector: Normalize the fuzzy relationship matrix R so that the sum of the elements in each row is 1. Then, calculate the weight of each target. The weight calculation formula is: wi=(Σrij) / (ΣΣrij), where rij is the element in the i-th row and j-th column of the fuzzy relationship matrix R.

[0084] R_normalized = | 0.6 0.3 0.1 0.0 0.0 | | 0.2 0.4 0.3 0.1 0.0 | | 0.3 0.4 0.2 0.1 0.0 | | 0.1 0.3 0.4 0.2 0.0 | The calculated weight vector W=(0.4, 0.2, 0.2, 0.2).

[0085] According to the standardized single objective function and the target weight vector, a comprehensive objective function is constructed. Assume that the standardized single objective function value of a mining sequence is: f1'=0.5, f2'=0.444, f3'=0.333, f4'=0.5. The target weight vector is W=(0.4, 0.2, 0.2, 0.2).

[0086] Then the comprehensive objective function value is: COF=w1×f1'+w2×f2'+w3×f3'+w4×f4'=0.4×0.5+0.2×0.444+0.2×0.333+0.2×0.5=0.2+0.0888+0.0666+0.1=0.4554; The comprehensive objective function value (COF) of this mining sequence is 0.4554.

[0087] The comprehensive objective function (COF) is a weighted sum of the four standardized single objective function values. The weight reflects the relative importance of each single objective function in the comprehensive evaluation. By adjusting the weight, the trade-off relationship between different objectives can be changed, thereby affecting the final optimization result. In practical applications, the weight needs to be determined according to the preferences and decision-making goals of mine managers. The final constructed comprehensive objective function (COF) is a scalar function that transforms the multi-objective optimization problem into a single-objective optimization problem and provides an objective function for subsequent mining sequence optimization algorithms (such as NSGA-II).

[0088] Preferably, step S33 includes the following steps: Step S331: classifying the mining constraints to obtain a classified constraint set, wherein the classified constraint set includes a geometric constraint set, a production constraint set, and a geological constraint set; Step S332: performing constraint mathematical expression on the classification constraint set to obtain a constraint expression; Step S333: selecting an exterior point penalty function for the constraint expression and determining a penalty factor to obtain a penalty function; Step S334: Generate a constraint function for the penalty function to obtain a constraint function.

[0089] In the embodiment of the present invention, the constraints of mining are classified. There are 50 mining units, numbered 1-50. 1. Geometric constraints: Maximum slope angle constraint: The final slope angle formed by mining shall not exceed 45 degrees. Minimum mining width constraint: The mining width of each mining unit shall not be less than 20 meters. Maximum mining depth constraint: The mining depth shall not exceed the elevation of -500 meters. 2. Production constraints: Shovel loader capacity constraint: In each time period (for example, one year), the total amount of ore mined by the shovel loader shall not exceed its maximum processing capacity. Assume that the mine has a type A shovel loader with a maximum annual processing capacity of 1 million tons. Transport truck capacity constraint: In each time period (for example, one year), the total amount of ore transported by the transport truck shall not exceed its maximum transportation capacity. Assume that the mine has two type B transport trucks, each with a maximum annual transportation capacity of 500,000 tons. 3. Geological constraints: Safety distance constraint: The distance from each mining unit to the fault shall not be less than 50 meters. Finally, a classified constraint set is obtained, including a geometric constraint set, a production constraint set, and a geological constraint set.

[0090] The classification constraint set obtained in step S331 is converted into a mathematical expression. The mining unit is represented by i, i=1, 2, ..., 50.

[0091] Geometric constraints: Maximum slope angle constraint: For any two adjacent mining units i and j in the vertical direction (assuming i is on top and j is on the bottom), let the elevation difference between their center points be Δh_ij and the horizontal distance be Δd_ij. Then the constraint expression is: arctan(Δh_ij / Δd_ij)≤45°. Minimum mining width constraint: Let the width of mining unit i be w_i. Then the constraint expression is: w_i≥20 meters. Maximum mining depth constraint: Let the elevation of the center point of mining unit i be z_i. Then the constraint expression is: z_i≥-500 meters.

[0092] Production constraints: Shovel loader capacity constraint: Let the total mining units mined in the tth year be S_t, and the ore volume of mining unit i be q_i (unit: tons). Then the constraint expression is: Σq_i≤1 million tons (i∈S_t). Transport truck capacity constraint: Let the total mining units transported in the tth year be S_t, and the ore volume of mining unit i be q_i (unit: tons). Then the constraint expression is: Σq_i≤2×500,000 tons=1 million tons (i∈S_t).

[0093] Geological constraints: Safety distance constraint: Let the distance from mining unit i to the fault be d_i. Then the constraint expression is: d_i ≥ 50 meters.

[0094] Finally, we get the constraint expression, which contains the mathematical expression of all constraint conditions.

[0095] For the constraint expression obtained in step S332, the penalty function is constructed using the exterior point method.

[0096] 1. Convert inequality constraints to standard form: Convert all inequality constraints to the form of g(x)≤0.

[0097] Maximum slope angle constraint: arctan(Δh_ij / Δd_ij)-45°≤0. Minimum mining width constraint: 20-w_i≤0. Maximum mining depth constraint: -500-z_i≤0. Loader capacity constraint: Σq_i-1 million≤0(i∈S_t). Transport truck capacity constraint: Σq_i-1 million≤0(i∈S_t). Safety distance constraint: 50-d_i≤0.

[0098] 2. Select the penalty function: For each inequality constraint g(x)≤0, select the penalty function P(x)=α×max(0,g(x))^2, where α is the penalty factor.

[0099] 3. Determine the penalty factor: The value of the penalty factor α needs to be large enough to ensure that sufficient penalties are imposed on solutions that violate the constraints. Based on experience and experiments, the penalty factor α of all constraints is uniformly set to 1000.

[0100] 4. Construct penalty function: Maximum slope angle constraint penalty function: P1=1000×max(0,arctan(Δh_ij / Δd_ij)-45°)^2. Minimum mining width constraint penalty function: P2=1000×max(0,20-w_i)^2. Maximum mining depth constraint penalty function: P3=1000×max(0,-500-z_i)^2. Loader capacity constraint penalty function: P4=1000×max(0,Σq_i-1 million)^2(i∈S_t). Transport truck capacity constraint penalty function: P5=1000×max(0,Σq_i-1 million)^2(i∈S_t). Safety distance constraint penalty function: P6=1000×max(0,50-d_i)^2. Finally, the penalty function is obtained, including P1, P2, P3, P4, P5, and P6.

[0101] The penalty functions obtained in step S333 are combined to generate a constraint function. The constraint function consists of a total penalty function, which is the sum of all individual constraint penalty functions. For a given mining sequence, the penalty function values ​​corresponding to all constraint violations in the sequence are calculated. For example, if in a mining sequence, two mining units violate the maximum slope angle constraint, three mining units violate the minimum mining width constraint, and the mining volume in the first year exceeds the shovel loader capacity constraint, then the total penalty function value is: Penalty=2×P1+3×P2+P4; where P1, P2, and P4 are the penalty function values ​​of the corresponding constraints defined in step S333, respectively. The constraint function contains only one function, namely the total penalty function Penalty. This function takes all constraint violations into account and represents the degree of violation of the constraint by the mining sequence through a scalar value. In the optimization algorithm, the penalty function value is added to the objective function value, thereby converting the constrained optimization problem into an unconstrained optimization problem. Finally, the constraint function {Penalty} is obtained. This set contains only one total penalty function.

[0102] Preferably, step S4 comprises the following steps: Step S41: constructing a simulation environment according to the target mining plan, semantic geological units and mining units to obtain a mining simulation environment; Step S42: using the mining simulation environment to simulate the mining process of the target mining plan to obtain mining process simulation data; collecting and recording the mining process simulation data to obtain original simulation data records; Step S43: Processing and integrating the original simulation data records to obtain processed simulation data; generating a mining performance data set for the processed simulation data to obtain a mining performance data set; Step S44: obtaining actual mining data; performing mining difference analysis on the actual mining data and the mining performance data set to obtain a difference analysis report; determining the geological model parameter correction target according to the difference analysis report to obtain a model correction target set; Step S45: selecting a correction method according to the model correction target set to obtain a model correction method set; adjusting model parameters of the semantic geological unit using the model correction target set and the model correction method set to obtain an adjusted semantic unit set; Step S46: Perform model verification and evaluation based on the adjusted semantic unit set to obtain a model verification report; generate a revised geological model based on the model verification report and the adjusted semantic unit set to obtain a revised geological model.

[0103] In an embodiment of the present invention, a mining simulation environment is constructed based on a target mining scheme, a semantic geological unit (SGUS) and a mining unit (MUS). An ore body model is established using three-dimensional mining simulation software, and each semantic unit in the SGUS is imported into the simulation software, and its corresponding attributes, such as lithology, grade, density, etc., are assigned. The mining unit division scheme defined in the target mining scheme is imported into the simulation software, and the ore body model is subdivided into mining units defined in the MUS. Existing mining equipment in the mine is added to the simulation environment, such as three shovel loaders with a bucket capacity of 10m³ and five transport trucks with a load capacity of 50 tons, and their operating parameters, such as shoveling speed, transportation speed, fuel consumption, etc., are set. Boundary conditions of the simulation environment are set, such as terrain, groundwater level, climate, etc. In the mining simulation environment constructed in step S41, the mining process is simulated according to the mining sequence and parameters specified in the target mining scheme. The running trajectory, loading and unloading operations, and advancement process of the mining face of each shovel loader and transport truck are simulated. During the simulation process, the location, status (e.g., working, idle, maintenance), amount of ore loaded, amount of ore transported, fuel consumption, and other data of each device are recorded every hour. These data are stored as original simulation data records, for example, in CSV file format. The original simulation data records obtained in step S42 are processed and integrated. For example, the ore loading volume of each shovel loader in a month is accumulated to calculate the monthly output of each shovel loader. The ore transportation volume and fuel consumption of each transport truck in a month are calculated to evaluate the transportation efficiency. Based on the processed simulation data, a mining performance data set is generated. The mining performance data set includes the following indicators: monthly ore production, monthly stripping volume, average grade, average mining cost, equipment utilization, fuel consumption, etc. The actual mining data of the mine is obtained, such as monthly ore production, monthly stripping volume, average grade, average mining cost, etc. The actual mining data is compared and analyzed with the mining performance data set generated in step S43, and the difference between the two is calculated. For example, the difference between the actual monthly production and the simulated monthly production is compared to analyze the cause of the difference. According to the difference analysis report, determine the geological model parameters that need to be corrected. For example, if the average grade of actual mining is lower than the simulation result, the estimated value of the grade in the geological model needs to be corrected. The parameters that need to be corrected and their target values ​​are recorded in the model correction target set. For example, add the "average grade of ore body A" to the model correction target set, and set its target value to the average grade of actual mining. According to the model correction target set obtained in step S44, select an appropriate model correction method. For example, if the model correction goal is to increase the average grade of ore body A, the Kriging interpolation method can be used to re-interpolate the grade model of ore body A using actual mining data. Use the selected model correction method to adjust the relevant parameters in the semantic geological unit (SGUS).For example, using the Kriging interpolation method, the grade value of each semantic unit in ore body A is recalculated based on the actual mining data and the variation function model. The semantic unit set adjusted in step S45 is verified and evaluated. The adjusted semantic unit set is imported into the mine mining simulation environment constructed in step S41, the mining process is re-simulated, and a new mining performance data set is generated. The new mining performance data set is compared with the actual mining data to evaluate the effect of the model correction. The verification results are recorded in the model verification report. Based on the model verification report and the adjusted semantic unit set, a revised geological model is generated. The revised geological model contains the updated semantic unit set and other relevant geological information, such as faults, lithology distribution, etc.

[0104] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0105] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for managing reserves of geological resources in a mine, characterized in that: The following steps are involved: Step S1: Collect multi-source data of mine geology and pre-process the multi-source data to obtain fused geological data; Extract geological semantic features from the fused geological data and perform semantic reconstruction to obtain semantic geological units; Step S2: construct a variogram using the fused geological data; calculate the reserve probability based on the variogram to obtain unit probability data; perform Bayesian inference based on the unit probability data and generate unit reserve probability to obtain unit reserve probability; Step S3: construct mining units for semantic geological units to obtain mining units; construct mining optimization objective functions according to mining units and unit reserve probabilities to obtain comprehensive objective functions; construct constraint functions according to mining units to obtain constraint functions; perform multi-objective optimization solutions for mining units according to comprehensive objective functions and constraint functions, and evaluate mining plans to obtain target mining plans; Step S4: simulating the mining process according to the target mining plan to obtain a mining performance data set; performing mining difference analysis on the mining performance data set to obtain a difference analysis report; The geological model is revised according to the difference analysis report and semantic geological units to obtain a revised geological model.

2. The method for managing reserves of geological resources in mines according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting multi-source data of mine geology; performing multi-source data preprocessing on the multi-source data of mine geology to obtain fused geological data; Step S12: constructing an initial three-dimensional geological model for the fused geological data, and performing preliminary grid division to obtain an initial geological grid model; Step S13: extracting geological semantic features from the fused geological data according to the initial geological grid model to obtain a geological feature vector set; Step S14: performing geological semantic segmentation on the initial geological grid model according to the geological feature vector set to obtain an initial semantic unit set; Step S15: Optimize the initial semantic unit set into semantic geological units to obtain semantic geological units.

3. The method for managing reserves of geological resources in mines according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: determining key geological attributes of the semantic geological unit; Step S22: performing reserve probability calculation on the semantic geological unit based on the Bayesian network structure according to the variation function to obtain unit probability data; Step S23: performing Bayesian inference based on the unit probability data and updating parameters to obtain updated probability data; Step S24: generating reserve probability distribution for semantic geological units according to the updated probability data to obtain unit reserve distribution; Step S25: Generate unit reserve probability for unit reserve distribution to obtain unit reserve probability.

4. The method for managing reserves of geological resources in mines according to claim 3, characterized in that: Step S22 includes the following steps: Step S221: define the attribute of the semantic geological unit - the reserve node; define the hidden node according to the fused geological data; Step S222: determining the Bayesian network structure of the attribute-reserve nodes and hidden nodes to obtain a Bayesian network structure diagram; Step S223: Acquire historical geological data; determine the prior probability distribution of the Bayesian network structure diagram according to the historical geological data to obtain the node prior probability; Step S224: determining the spatial correlation of geological attributes according to the variation function, and constructing a likelihood function according to the Bayesian network structure diagram and the semantic geological unit to obtain a conditional probability table; Step S225: Perform Bayesian inference on the Bayesian network structure diagram according to the node prior probability and the conditional probability table to obtain the node posterior probability; Step S226: extract the posterior probability distribution of the reserve node from the node posterior probability, and generate a unit probability model according to the Bayesian network structure diagram to obtain unit probability data.

5. The method for managing reserves of geological resources in mines according to claim 4, characterized in that: Step S224 is specifically as follows: Extract the geological attributes of semantic geological units, calculate the correlation coefficients of geological attributes between different spatial positions according to the variogram, and obtain the spatial correlation of attributes; Kriging estimation is performed based on attribute spatial correlation and semantic geological units to obtain attribute node estimation values; According to the estimated values ​​of the attribute nodes, the attribute node conditional probability table is constructed for the Bayesian network structure graph to obtain the attribute node conditional probability table; The conditional probability table of the attribute node is used to derive the conditional probability distribution of the reserve node and obtain the conditional probability table.

6. The method for managing the reserves of geological resources in a mine according to claim 4, characterized in that: Step S23 includes the following steps: Step S231: acquiring newly added exploration data; integrating observation data of the newly added exploration data to obtain integrated observation data; Step S232: performing Gibbs sampling on the Bayesian network structure diagram according to the conditional probability table and the integrated observation data to obtain a posterior sample; Step S233: Calculate the posterior distribution according to the posterior samples to obtain the node posterior probability; Step S234: Use the node posterior probability to update the graph parameters of the Bayesian network structure graph to obtain updated probability data.

7. The method for managing reserves of geological resources in mines according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: obtaining mining technical parameters and mining constraints; constructing mining units for semantic geological units according to the mining technical parameters to obtain mining units; Step S32: constructing a mining optimization objective function according to the mining units and the unit reserve probabilities to obtain a comprehensive objective function; Step S33: converting the mining constraint conditions into mathematical expressions, and constructing constraint functions according to the mining units and mining technical parameters to obtain constraint functions; Step S34: taking the comprehensive objective function and the constraint function as inputs of the optimization algorithm, performing multi-objective optimization on the mining unit based on the multi-objective optimization algorithm, and obtaining a mining sequence solution set; Step S35: generating a mining plan for the mining sequence solution set to obtain a mining plan; Step S36: Evaluate the mining plan and perform Pareto optimal solution analysis to obtain a target mining plan.

8. The method for managing reserves of geological resources in mines according to claim 7, characterized in that: Step S32 includes the following steps: Step S321: defining a single objective function according to the mining unit and the unit reserve probability to obtain a single objective function, wherein the single objective function includes a reserve maximization objective, a grade balance objective, a geological structure stability objective, and a mining efficiency objective; Step S322: performing single objective function standardization on the single objective function to obtain a standardized single objective function; Step S323: determining the target weight based on fuzzy comprehensive evaluation according to the standardized single objective function to obtain a target weight vector; Step S324: construct a comprehensive objective function based on the standardized single objective function and the objective weight vector to obtain a comprehensive objective function.

9. The method for managing reserves of geological resources in mines according to claim 7, characterized in that: Step S33 includes the following steps: Step S331: classifying the mining constraints to obtain a classified constraint set, wherein the classified constraint set includes a geometric constraint set, a production constraint set, and a geological constraint set; Step S332: performing constraint mathematical expression on the classification constraint set to obtain a constraint expression; Step S333: selecting an exterior point penalty function for the constraint expression and determining a penalty factor to obtain a penalty function; Step S334: Generate a constraint function for the penalty function to obtain a constraint function.

10. The method for managing reserves of geological resources in mines according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: constructing a simulation environment according to the target mining plan, semantic geological units and mining units to obtain a mining simulation environment; Step S42: using the mining simulation environment to simulate the mining process of the target mining plan to obtain mining process simulation data; collecting and recording the mining process simulation data to obtain original simulation data records; Step S43: Processing and integrating the original simulation data records to obtain processed simulation data; generating a mining performance data set for the processed simulation data to obtain a mining performance data set; Step S44: obtaining actual mining data; performing mining difference analysis on the actual mining data and the mining performance data set to obtain a difference analysis report; determining the geological model parameter correction target according to the difference analysis report to obtain a model correction target set; Step S45: selecting a correction method according to the model correction target set to obtain a model correction method set; adjusting model parameters of the semantic geological unit using the model correction target set and the model correction method set to obtain an adjusted semantic unit set; Step S46: Perform model verification and evaluation based on the adjusted semantic unit set to obtain a model verification report; generate a revised geological model based on the model verification report and the adjusted semantic unit set to obtain a revised geological model.

Images