Gold polymetallic ore cluster analysis method of multi-property parameters and application

Abstract

The application relates to the field of noble metal mineral exploration data processing, and discloses a gold polymetallic ore concentration area multi-property parameter clustering analysis method and application.The method comprises the following steps: collecting the density, magnetic susceptibility, resistivity, polarization rate and seismic wave velocity multi-dimensional physical property parameters of ores and surrounding rocks to construct an original physical property data set; performing outlier rejection, range standardization and PCA feature dimension reduction under geological constraints on the original physical property data to obtain a multi-property feature data set; introducing gold polymetallic ore-forming geological prior information, selecting seed samples with clear geological attributes to form initial seed clusters, establishing a weighted distance measurement criterion based on the ability of physical property parameter differentiation, and constructing an improved K-Means+ clustering algorithm; and then verifying and calibrating the clustering results by using a density peak clustering algorithm.The application can improve the consistency between the multi-property data clustering results and ore-forming geological understanding, and improve the efficiency of ore body delineation and resource evaluation.

Description

Technical Field

[0001] This invention relates to the field of precious metal mineral exploration data processing technology, and more specifically, to a method and application of multi-physical parameter clustering analysis for gold polymetallic mineral clusters. Background Technology

[0002] Gold polymetallic ore clusters typically possess complex metallogenic geological backgrounds. Ore bodies, alteration zones, host rocks, and ores of different grades may exhibit differences in physical properties such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity. However, these differences are often not directly determined by a single parameter but rather reflected by the combined effects of multiple physical properties. In traditional mineral exploration, geologists usually make comprehensive judgments about mineralized bodies based on geological mapping, sample testing, geophysical interpretation, and borehole data. This method is highly empirical and can fully utilize professional geological knowledge. However, with the increase in data volume and the proliferation of multi-source data in ore clusters, manual comprehensive interpretation is easily affected by single indicators, local anomalies, and differences in the experience of interpreters.

[0003] With the continuous expansion of Earth science data, machine learning methods have been gradually applied to mineral, petrological, and geochemical data mining, metallogenic potential evaluation, and ore deposit type identification. In 2025, Wang Zhiyu, in his article "Application and Prospect of Machine Learning in Mineral, Petrological, and Geochemical Big Data Mining" published in the *Journal of Chengdu University of Technology*, pointed out that, as... Figure 2 As shown, machine learning modeling generally includes data collection and preprocessing, feature engineering, model selection and training, model evaluation and validation, model optimization and tuning, and result interpretation and application. This paper also explains that PCA can extract principal components and remove redundant information through linear transformation, K-means can achieve data partitioning by minimizing intra-cluster distance, and semi-supervised learning can improve model performance by utilizing both limited labels and abundant unlabeled data. These studies demonstrate that dimensionality reduction, clustering, semi-supervised information guidance, and model validation have become important methods in Earth science data processing.

[0004] Existing research on geochemical anomaly identification also indicates that geological processes and ore-forming systems are complex, and exploration anomalies often exhibit weak geochemical characteristics, multiple overlapping distribution patterns, and nonlinear features. Traditional methods based on a unified threshold or normal distribution assumptions are easily limited. In 2025, Lin Zijian published "Extraction and Geological Significance of Geochemical Weak Anomaly Information from Xiaoshiqiao Novel Lithium Deposit on the Western Margin of the Yangtze River" in the journal *Geophysical and Geochemical Computational Technology*. Addressing the problem of identifying weak geochemical anomalies in the Xiaoshiqiao novel lithium deposit, this paper points out that self-organizing mapping suffers from weak interpretability and low convergence efficiency, and that k-means++ is sensitive to initial cluster centers, such as... Figure 3As shown, a hybrid clustering method combining SOM and k-means++ is proposed for anomaly identification. This type of research demonstrates the application value of clustering algorithms in mineral anomaly identification, but there is still room for improvement in the selection of initial centers, consistency of geological interpretation, and anomaly boundary identification using ordinary K-means or K-means++. In the fusion of mineral prediction and exploration data, existing research is gradually shifting from processing single geochemical data to comprehensive analysis of multi-source information such as geology, geophysics, geochemistry, and remote sensing. Related studies indicate that multi-source remote sensing data combined with geological, geophysical, and geochemical data can be used to acquire information on ore-controlling structures, rock and mineral identification, alteration mineral distribution, and surface geochemical anomalies. Furthermore, research on the application of PCA-KM joint analysis in comprehensive radioactive anomaly identification shows that the comprehensive anomalies extracted by PCA-KM can reflect the known uranium mineralization distribution and reduce the ambiguity of comprehensive interpretation.

[0005] However, when existing clustering algorithms are directly used for multi-physical parameter analysis of gold polymetallic deposits, they lack prior geological information to guide them. The initial cluster centers are prone to deviating from the true geological categories, and the differences in the contribution of different physical parameters to mineralization identification cannot be fully reflected in the distance metric. This results in insufficient matching between the clustering results and the actual geological attributes of mineralized bodies, surrounding rocks, and ores of different grades. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of existing technologies, this invention provides a multi-physical parameter clustering analysis method and application for gold polymetallic ore clusters. This method involves collecting multi-dimensional physical property parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity from the ore and surrounding rocks to construct an original physical property dataset. The original physical property dataset is then preprocessed to obtain a multi-physical property feature dataset. Prior geological information about gold polymetallic ore formation is transformed into seed samples, initial seed clusters, and initial cluster centers. Furthermore, the differences in the contributions of parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity to mineralization identification are converted into a weighted distance metric. An improved K-Means+ clustering algorithm is constructed, incorporating prior geological information about gold polymetallic ore formation into the clustering process. This allows the clustering results to more accurately correspond to the actual geological properties of mineralized bodies, surrounding rocks, and ores of different grades, thereby addressing the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for cluster analysis of multiple physical property parameters in a gold polymetallic ore cluster includes the following steps: Step 1: Collect multidimensional physical property parameters of ore and surrounding rock in the gold polymetallic ore cluster area to construct the original physical property dataset.

[0008] Step 2: Preprocess the original physical property dataset to obtain a multi-physical property feature dataset.

[0009] Step 3: Introduce prior geological information on gold polymetallic deposits, select samples with clear geological attributes from the multi-physical property dataset as seed samples, and determine the initial cluster centers based on the seed samples.

[0010] Step 4: Establish a weighted distance metric and construct an improved K-Means+ clustering algorithm based on the initial cluster centers and the weighted distance metric.

[0011] Step 5: Use the improved K-Means+ clustering algorithm to perform iterative clustering analysis on the multi-physical property dataset to obtain mineralization body categories, surrounding rock categories, and ore categories of different grades.

[0012] Step 6: Verify and calibrate the mineralization type, the surrounding rock type, and the types of ore of different grades.

[0013] It should be noted that the different grade ore categories in this invention are subcategories of mineralized body categories. They are formed by further classifying samples within a mineralized body category based on the differences in multiple physical properties after the mineralized body category has been identified. They are not primary categories that are independent of or parallel to the mineralized body category.

[0014] As a further aspect of this invention, step 1 involves collecting multidimensional physical property parameters of ore and surrounding rock from the gold polymetallic ore cluster area to construct an original physical property dataset. This includes the following specific content: Within the gold polymetallic ore cluster area, for ore samples and surrounding rock samples that have been collected or tested, the system systematically organizes their multidimensional physical property parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity, and treats each sample or measuring point as an independent data unit to form an original physical property data record. For samples with clear geological attributes, their geological attribute information is recorded simultaneously. For samples whose geological attributes are not yet clear, a marker to be identified is retained so that subsequent cluster analysis can be used for category determination, establishing a correspondence between the physical property characteristics of the sample and its spatial location and geological background. Among these parameters, density is used to reflect the quality differences of ore or surrounding rock, magnetic susceptibility is used to reflect the differences in magnetic response, resistivity and polarizability are used to reflect electrical differences, and seismic wave velocity is used to reflect the elastic wave propagation characteristics of the rock mass. Through the above unified recording method, scattered geological sample data and physical property test results can be organized into an original physical property dataset with a consistent structure and complete fields.

[0015] As a further aspect of this invention, step 2 involves preprocessing the original physical property dataset to obtain a multi-physical property feature dataset, including the following specific contents: After completing the construction of the original physical property dataset, outliers are removed from physical property parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity. The removal targets mainly include data with significant deviations caused by sampling errors, testing errors, data entry errors, or local non-representative factors. During the processing, the distribution range of a single physical property parameter can be checked individually, or the combination relationship between multiple physical property parameters can be combined to determine whether a sample is abnormal. Locally anomalous samples with real geological significance are not directly deleted, but are retained and their causes are marked to avoid mistakenly deleting data that may reflect mineralization characteristics. Subsequently, the physical property data after outlier removal is subjected to range standardization processing, converting physical property parameters with different units, dimensions, and numerical ranges to a unified numerical scale. This allows density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity to participate in subsequent analysis under the same comparison conditions, avoiding the unreasonable dominance of clustering results by a parameter with a large numerical range in distance calculations.

[0016] Based on this, further PCA feature dimensionality reduction processing under geological constraints is carried out. That is, combining the differences in physical property responses of known mineralized samples, typical surrounding rock samples, and ore samples of different grades, principal component extraction is performed on the standardized multi-physical property parameters, and the principal components that can characterize the core physical property features of mineralization are retained. In this way, while reducing redundant information and parameter correlation, the main information related to the identification of mineralized bodies, surrounding rocks, and ore of different grades is retained as much as possible, and finally a multi-physical property feature dataset suitable for input into the improved K-Means+ clustering algorithm is formed.

[0017] As a further aspect of this invention, step 3 involves introducing prior geological information about gold polymetallic deposits. Samples with clear geological attributes are selected from the multi-physical property dataset as seed samples, and initial cluster centers are determined based on these seed samples. This includes the following specific details: The prior geological information about gold polymetallic deposits includes existing geological exploration data, borehole exposure data, known mineralized sample test results, and typical surrounding rock sample test results. Based on this prior information, a small number of samples with clear geological attributes are selected from the preprocessed multi-physical property dataset as seed samples, and these seed samples are divided into mineralized body seed samples, surrounding rock seed samples, and ore seed samples of different grades. Then, initial seed clusters are formed according to the geological attributes of the seed samples, and the center position of each initial seed cluster in the reduced-dimensional feature space is calculated. This center position is used as the initial cluster center for the improved K-Means+ clustering algorithm.

[0018] As a further aspect of this invention, step 4 involves establishing a weighted distance metric and constructing an improved K-Means+ clustering algorithm based on the initial cluster centers and the weighted distance metric. This includes the following specific steps: using the multi-physical property dataset as the computational object and the already formed initial cluster centers as the reference object, so that the data to be processed simultaneously possesses standardized physical property expression and geological attribute guidance. Subsequently, using the mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades as statistical benchmarks, the distribution of values ​​for density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component dimensionality reduction features of various seed samples are analyzed. The degree of concentration of samples within the same geological category, the degree of separation between different geological categories, the gradient change characteristics between ores of different grades, and the consistency relationship between each physical property parameter and known geological attribute records are analyzed to obtain the contribution judgment results of each physical property parameter to mineralization identification, surrounding rock differentiation, and grade subdivision. Based on this, the contribution levels of each physical property parameter or principal component feature are classified. Features that can stably characterize the difference between the mineralized body and the surrounding rock are classified as high-contribution features; features that can support the subdivision of ores of different grades are classified as medium-high-contribution features; features mainly used to supplement background lithological information are classified as auxiliary features; and features that are easily affected by local non-representative anomalies or test fluctuations and have limited contribution to mineralization identification are classified as low-contribution features. A contribution level table for physical property parameters is thus formed. The weight relationship of each feature in distance calculation is determined according to the contribution level table, so that high-contribution features play a dominant role in judging the proximity of samples to cluster centers, medium-high-contribution features participate in the detailed identification of ore grade categories, auxiliary features are used to supplement comprehensive physical property background information, and low-contribution features participate in the calculation in a weakly constrained form. This avoids the dominance of clustering results by parameters with significant numerical fluctuations but limited geological indicative significance. Finally, the above weight relationship is embedded in the distance calculation process between the samples to be clustered and each cluster center, forming a weighted distance metric. An improved K-Means+ clustering algorithm is constructed based on the initial cluster centers and the weighted distance metric.

[0019] As a further aspect of this invention, an improved K-Means+ clustering algorithm is used to perform iterative clustering analysis on a multi-physical property feature dataset to obtain mineralization categories, surrounding rock categories, and ore categories of different grades. This includes the following specific steps: When performing iterative clustering analysis using the improved K-Means+ clustering algorithm, the initial cluster centers formed by the mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades are first used as the starting point for classification, and a weighted distance metric is used as the basis for judging sample similarity. Subsequently, each sample in the multi-physical property feature dataset is compared with each cluster center using a weighted distance metric. The system compares samples and assigns them to the category with the closest comprehensive physical properties. After completing one round of sample assignment, it updates the mineralization category center, the surrounding rock category center, and the ore category centers of different grades by combining the density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component characteristics of the samples in each category. The sample assignment and center update are repeated until the cluster center changes are stable, the category division results are stable, or the preset number of iterations is reached. Finally, the mineralization category and the surrounding rock category are obtained, and the samples in the mineralization category are further divided into high-grade ore category, medium-grade ore category, and low-grade ore category.

[0020] As a further aspect of this invention, step 6 verifies and calibrates the mineralization body category, surrounding rock category, and ore categories of different grades, including the following specific content: The mineralization body category, surrounding rock category, and ore categories of different grades output by the improved K-Means+ clustering algorithm are used as input results. Density peak clustering algorithm is used for verification, and the local density and relative distance of each sample are calculated in the multi-property feature space to identify core samples, boundary samples, and outlier samples. For samples that are consistent with the cluster center and have a stable distribution, their original category labels are retained. For samples located at category boundaries or with abnormal distribution, calibration is performed based on their density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, principal component characteristics, and spatial location. Subsequently, the calibrated category labels are assigned back to the sampling points, borehole points, or profile control points. The ore body distribution range is determined based on the spatial continuity of the mineralization body category, grade is classified according to the spatial combination relationship of different grade ore categories, and the exploration target area is delineated based on the surrounding rock boundary and abnormal continuity.

[0021] As a further aspect of this invention, an application of multi-physical parameter clustering analysis in a gold polymetallic ore cluster includes: collecting multi-dimensional physical parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity of the ore and surrounding rock to form an original physical parameter dataset; after outlier removal, range standardization, and PCA feature dimensionality reduction under geological constraints, introducing prior geological information on mineralization to determine seed samples and initial cluster centers, and constructing an improved K-Means+ clustering algorithm based on a weighted distance metric; finally, verifying and calibrating through density peak clustering to achieve ore body identification, grade classification, and target area delineation.

[0022] The technical effects and advantages of this invention, a multi-physical parameter clustering analysis method for gold polymetallic ore clusters and its application, are as follows: This invention collects multi-dimensional physical parameters of ore and surrounding rock to form a dataset with a unified structure; it performs outlier removal, range standardization, and PCA dimensionality reduction under geological constraints to reduce the impact of erroneous data and dimensional differences; it introduces prior information on metallogenic geology to determine seed samples and initial cluster centers; it also establishes a weighted distance metric based on the identification contribution of physical parameters to make the clustering results closer to the actual geological properties, and further uses density peak clustering for verification and calibration to improve the reliability of ore body identification, grade classification, and target area delineation. Attached Figure Description

[0023] Figure 1 This is a flowchart of a multi-physical parameter clustering analysis method for gold polymetallic mineral clusters according to the present invention.

[0024] Figure 2 A schematic diagram of the general process for modeling machine learning with existing technologies.

[0025] Figure 3 The flowchart is for the existing SOM-k-means++ algorithm.

[0026] Figure 4 This is a line graph comparing the multi-property characteristic responses of various seed samples in this invention.

[0027] Figure 5 This is a scatter plot of the initial sample allocation in the multi-property feature space of this invention.

[0028] Figure 6 This is a schematic diagram of the sample allocation area of ​​the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1

[0031] like Figure 1 As shown, this invention provides a multi-physical parameter clustering analysis method for gold polymetallic ore clusters, used for mineralization identification, surrounding rock differentiation, grade classification, and exploration target delineation based on multi-physical parameters of ore and surrounding rock. The method includes the following steps: Step 1: Collect multidimensional physical property parameters of ore and surrounding rock in the gold polymetallic ore cluster area to construct the original physical property dataset.

[0032] Step 2: Preprocess the original physical property dataset to obtain a multi-physical property feature dataset.

[0033] Step 3: Introduce prior geological information on gold polymetallic deposits, select samples with clear geological attributes from the multi-physical property dataset as seed samples, and determine the initial cluster centers based on the seed samples.

[0034] Step 4: Establish a weighted distance metric and construct an improved K-Means+ clustering algorithm based on the initial cluster centers and the weighted distance metric.

[0035] Step 5: Use the improved K-Means+ clustering algorithm to perform iterative clustering analysis on the multi-physical property dataset to obtain mineralization body categories, surrounding rock categories, and ore categories of different grades.

[0036] Step 6: Verify and calibrate the mineralization type, the surrounding rock type, and the types of ore of different grades.

[0037] In this embodiment, multiple physical property parameters were first collected from ore samples and surrounding rock samples within the gold polymetallic ore cluster. The collected samples included known mineralized samples, ore samples of different grades, surrounding rock samples near the ore body, and typical surrounding rock samples far from the ore body. A corresponding data record was established for each sample, including sample number, sampling point coordinates, borehole number, borehole depth, sample type, density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity.

[0038] Density parameters reflect the change in mass per unit volume of the sample. Magnetic susceptibility parameters reflect the sample's responsiveness to an applied magnetic field. Resistivity parameters reflect the sample's conductivity. Polarizability parameters reflect the sample's ability to produce a polarization response under an electric field. Seismic wave velocity parameters reflect the propagation speed of elastic waves in the ore or surrounding rock. These parameters reflect sample properties from different physical perspectives and collectively constitute the comprehensive physical property characteristics of mineralized bodies, surrounding rocks, and ores of different grades. In terms of data table structure, the original physical property dataset can adopt the field format shown in Table 1: Table 1 Field Format

[0039] In this embodiment, outlier removal is first performed on the original physical property data. This outlier removal includes single-parameter anomaly detection and multi-parameter combination anomaly detection. Single-parameter anomaly detection involves statistically analyzing the distribution of density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity to identify data points that significantly deviate from the reasonable distribution range of these parameters. Samples with a clearly abnormal single physical property parameter but normal other parameters can be marked as samples to be reviewed. If the review results indicate that the anomaly originates from instrument error, data entry error, or sample contamination, the outlier is removed. If the anomaly is related to known mineralization phenomena, tectonic fracture zones, or special lithology, the sample is retained and the cause of the anomaly is marked. Multi-parameter combination anomaly detection involves simultaneously examining the combination relationships between multiple physical property parameters. When a sample deviates from the normal combination characteristics of adjacent or similar samples in multiple parameters, it is determined whether the sample belongs to a local anomaly point, a data entry error point, or a special mineralization point. For samples whose determination cannot be clearly made, their original records can be retained and processed as boundary samples or outlier samples in subsequent density peak clustering verification. The purpose of outlier removal is to reduce the interference of obviously erroneous data on cluster center and weighted distance calculations, while avoiding the wrong deletion of geologically significant real outlier samples.

[0040] After outlier removal, range standardization is used to standardize each physical property parameter. Since density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity have different units and significantly different numerical ranges, directly calculating the distance between samples could lead to parameters with large numerical ranges having an excessively strong influence on the clustering results. Therefore, this embodiment transforms each physical property parameter to a uniform numerical range. Range standardization can be performed as follows: for any physical property parameter, obtain its maximum and minimum values ​​in the sample set, and convert the parameter value for each sample into a standardized value. After standardization, all physical property parameters are within a uniform range, allowing each parameter to participate in PCA dimensionality reduction at the same scale. Range standardization reduces the impact of dimensional differences on the clustering results.

[0041] PCA feature reduction is performed on the multi-physical property parameters after range standardization. The purpose of PCA feature reduction is to extract comprehensive features that can represent the main direction of data change and reduce the correlation and redundant information between parameters. In this embodiment, PCA feature reduction is not simply based on mathematical variance to select principal components, but is performed under geological constraints. The geological constraints refer to combining the differences in physical property responses of known mineralized samples, typical surrounding rock samples, and ore samples of different grades when selecting principal components to determine whether the retained principal components can reflect the core physical property characteristics of mineralization. If a principal component has a high variance contribution, but mainly reflects lithological background changes or local disturbances unrelated to mineralization, its weight in subsequent clustering can be reduced. If a principal component can effectively distinguish between mineralized bodies and surrounding rocks, it is used as an important feature input into the improved K-Means+ clustering. In this embodiment, the cumulative contribution rate of the principal components retained by PCA feature reduction can be set to 80% to 95%. When the data dimension of the mining area is low and the sample quality is high, fewer principal components can be retained; when the geological background of the mining area is complex and there is a strong correlation between physical parameters, the cumulative contribution rate of principal components can be increased to retain more complete mineralization identification information.

[0042] In this embodiment, a set of prior geological information for gold polymetallic deposits is established to guide cluster analysis. This prior geological information mainly originates from existing geological exploration data, borehole exposure data, known mineralized sample test results, and typical surrounding rock sample test results. Existing geological exploration data reflects the known distribution of mineralized zones, surrounding rock types, and mineralization anomaly locations within the ore cluster area. Borehole exposure data reflects the actual exposure of ore bodies, surrounding rocks, and transitional areas at different depths. Known mineralized sample test results characterize the typical response characteristics of mineralized bodies in multiple physical property parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity. Typical surrounding rock sample test results characterize the physical property response characteristics of unmineralized or weakly mineralized background rocks. Subsequently, the above prior information is correlated with a multi-physical property feature dataset after outlier removal, range standardization, and PCA feature dimensionality reduction under geological constraints. A small number of samples with clear geological attributes and representative physical property characteristics are selected as seed samples. These seed samples are not randomly selected but are representative of the overall geological characteristics. The table lists samples with known mineralization, known surrounding rock, or ore characteristics of different grades. After the seed samples are determined, they are divided into mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades according to their corresponding geological attributes. Among them, mineralization seed samples are used to guide the algorithm to identify samples with similar physical properties to known mineralization, surrounding rock seed samples are used to guide the algorithm to identify samples with similar physical properties to background surrounding rocks, and ore seed samples of different grades are used to provide a category basis for subsequent ore grade classification. Then, seed samples with the same geological attributes are grouped into the same initial seed cluster, and the center position of each initial seed cluster is calculated in the dimensionality-reduced multi-physical property feature space. This center position can reflect the comprehensive physical property characteristics of the corresponding category of samples. Finally, the center position of each initial seed cluster is used as the initial cluster center of the improved K-Means+ clustering algorithm, so that the subsequent clustering process no longer completely depends on random initial centers, but starts from known mineralization patterns, typical surrounding rock characteristics, and ore characteristics of different grades for sample allocation, thereby improving the consistency between the clustering results and the actual geological attributes.

[0043] In this embodiment, the multi-physical property dataset is used as the computational object, and the already formed initial cluster centers are used as reference objects, so that the data to be processed simultaneously possesses standardized physical property expression and geological attribute guidance basis; subsequently, using the mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades as statistical benchmarks, the value distribution of each type of seed sample in terms of density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component dimensionality reduction features is sorted out, and the concentration degree of samples within the same geological category, the separation degree between different geological categories, the gradient change characteristics between different grades of ore, and the consistency relationship between each physical property parameter and known geological attribute records are analyzed, thereby obtaining the contribution judgment results of each physical property parameter to mineralization identification, surrounding rock differentiation, and grade subdivision, such as... Figure 4 The figure shows the response differences of seed samples from different geological categories in multiple physical property features. Based on this, the contribution levels of each physical property parameter or principal component feature are classified. Features that can stably characterize the difference between mineralized bodies and surrounding rocks are classified as high-contribution features; features that can support the subdivision of ores of different grades are classified as medium-high-contribution features; features mainly used to supplement background lithological information are classified as auxiliary features; and features easily affected by local non-representative anomalies or test fluctuations and whose contribution to mineralization identification is limited are classified as low-contribution features. A physical property parameter contribution level table is thus formed. The weight relationship of each feature in distance calculation is determined according to the physical property parameter contribution level table. This ensures that high-contribution features play a dominant role in judging the proximity of samples to cluster centers, that medium-high-contribution features participate in the refined identification of ore grade categories, that auxiliary features are used to supplement comprehensive physical property background information, and that low-contribution features participate in calculations with weak constraints. This avoids single parameters with significant numerical fluctuations but limited geological indicative significance dominating the clustering results. Finally, the aforementioned weighting relationships are embedded into the distance calculation process between the sample to be clustered and each cluster center, forming a weighted distance metric. This ensures that the proximity of each sample to the mineralization center, the surrounding rock center, and the centers of ores of different grades is no longer determined by the simple equivalence of each physical property parameter, but rather by the mineralization identification contribution, geological attribute orientation, and comprehensive physical property similarity of each parameter. When calculating the weighted distance between the sample to be clustered and any cluster center, the differences between the sample and the cluster center in density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component characteristics are calculated respectively. Each difference value is multiplied by its corresponding feature weight and then summed to obtain the weighted distance of the sample relative to the cluster center. The smaller the weighted distance, the closer the sample is to the comprehensive physical property characteristics of the corresponding category.

[0044] Furthermore, an improved K-Means+ clustering algorithm is constructed based on the initial cluster centers and the weighted distance metric. This includes: reading the multi-physical property feature dataset after outlier removal, range standardization, and PCA feature dimensionality reduction under geological constraints; reading the initial cluster centers formed by mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades; and reading the established weighted distance metric. The multi-physical property feature dataset serves as the set of samples to be clustered, the initial cluster centers serve as the starting classification benchmark for the algorithm, and the weighted distance metric serves as the calculation rule for judging the similarity between samples and cluster centers.

[0045] Furthermore, based on the geological origin of the initial cluster centers, the cluster categories were initially set as mineralization categories, host rock categories, and ore categories of different grades, and each initial cluster center was bound to its corresponding geological attribute. After completing the initial category binding and the first round of sample allocation, through... Figure 5 This displays the initial cluster distribution of samples in the reduced-dimensional feature space. The initial cluster centers for mineralized bodies represent the comprehensive physical properties of known mineralized bodies, the initial cluster centers for surrounding rocks represent the comprehensive physical properties of typical surrounding rocks, and the initial cluster centers for ores of different grades represent the differences in comprehensive physical properties between ores of different grades.

[0046] Furthermore, a weighted similarity judgment rule is established, specifically: the weighted distance metric is embedded into the sample allocation process of the improved K-Means+ clustering algorithm. When comparing the proximity of the sample to be clustered to each cluster center, the judgment is no longer made in the manner that all physical property parameters participate equally. Instead, the weighted processing is performed based on the contribution of density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component characteristics to distinguish mineralized bodies, surrounding rocks, and ores of different grades.

[0047] Furthermore, using each initial cluster center as a comparison object and the weighted similarity judgment rule as the calculation basis, the degree of closeness between each sample to be clustered in the multi-physical property data set and the initial cluster center of the mineralized body, the initial cluster center of the surrounding rock, and the initial cluster center of the ore of different grades is compared one by one, and each sample to be clustered is assigned to the category with the closest comprehensive physical property characteristics.

[0048] Furthermore, after completing the first round of sample allocation, all samples were extracted from different mineralization categories, host rock categories, and ore categories of different grades, such as... Figure 6The diagram shows the sample allocation area. Based on the multi-physical property characteristics and weighted distance metrics of samples within each category, the mineralization category center, the surrounding rock category center, and the ore category centers for different grades are updated. Specifically, according to the category labels already obtained for each sample, samples assigned to the mineralization category are grouped into the mineralization sample set, samples assigned to the surrounding rock category are grouped into the surrounding rock sample set, and samples assigned to different grade ore categories are grouped into the corresponding grade ore sample sets. Subsequently, within each sample set, the density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component characteristics of each sample are read, and then... The established weighted distance metric uses physical properties that contribute highly to mineralization identification as the primary basis for center updates, and physical properties that contribute less as secondary basis. This ensures that the updated centers for mineralization body categories, surrounding rock categories, and different grade ore categories reflect the comprehensive physical property distribution characteristics of all samples within their respective categories. The centers for mineralization body categories are jointly determined by the weighted comprehensive physical property characteristics of each sample in the mineralization body sample set; the centers for surrounding rock categories are jointly determined by the weighted comprehensive physical property characteristics of each sample in the surrounding rock sample set; and the centers for different grade ore categories are jointly determined by the weighted comprehensive physical property characteristics of each sample in the corresponding grade ore sample set. When updating cluster centers, the standardized values ​​of density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component feature dimensions of each sample within the same category are used, combined with the corresponding weights of each feature in the weighted distance metric, to determine the new comprehensive physical property center for that category. This ensures that the updated cluster centers simultaneously reflect the overall physical property distribution of samples within the category and the differences in the contribution of different physical properties to mineralization identification. During the center update process, mineralization seed samples, surrounding rock seed samples, and ore seed samples of different grades continue to participate in center determination as guiding samples for category direction. This ensures that the updated cluster centers still correspond to the comprehensive physical properties of known mineralizations, typical surrounding rocks, and ores of different grades, and prevents the category centers from deviating from existing geological attributes during the iteration process.

[0049] Furthermore, the updated cluster centers replace the previous cluster centers as the clustering benchmark for the next round of sample category allocation; at the same time, the weighted similarity judgment rule remains unchanged, so that the next round of sample allocation inherits the results of the previous round of sample allocation and center update, and continues to be constrained by the differences in the contribution of physical property parameters, thereby realizing the continuous transfer between the cluster center update results and the sample redistribution process.

[0050] Furthermore, based on the updated clustering benchmark, all samples in the multi-physical property feature dataset are reclassified, and the cluster centers of each category are updated again after the reclassification is completed. Then, the process of "weighted distance judgment, sample category allocation, category center update, and updating the center as the next round of clustering benchmark" is repeated continuously, so that mineralized body categories, surrounding rock categories, and different grade ore categories gradually form stable groups in the multi-physical property feature space.

[0051] Furthermore, after each round of center update, the status of cluster center position change, sample category division, and iteration count are determined; when the cluster center position change reaches the preset stability condition, or the sample category division result remains stable, or the preset iteration count is reached, the iteration stops and the final clustering result is output.

[0052] In this embodiment, when performing iterative clustering analysis on a multi-property feature dataset using the improved K-Means+ clustering algorithm, the initial cluster centers of mineralized bodies, surrounding rocks, and ores of different grades, formed in the preceding steps, are first read and used as the starting point for this round of clustering analysis. Simultaneously, the established weighted distance metric is invoked, and density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component features are used to determine sample similarity based on their contribution to mineralization identification and grade differentiation. Subsequently, each sample to be clustered in the multi-property feature dataset is read sequentially, and a weighted distance comparison is performed between the sample and the cluster centers of mineralized bodies, surrounding rocks, and ores of different grades. Based on the comparison results, the sample is assigned to the category with the closest comprehensive physical property characteristics, thus completing one round of sample category assignment. After completing this round of allocation, all samples within the mineralization body category, the surrounding rock category, and the ore categories of different grades are extracted. Combining the multi-physical property characteristics of the samples within each category with the weighted distance metric, the centers of the mineralization body category, the surrounding rock category, and the ore categories of different grades are updated to reflect the comprehensive physical property distribution characteristics of the samples within the corresponding category. Then, the updated center of each category is used as the cluster center for the next iteration, and all samples are again compared using weighted distance and assigned to a category. This sample allocation and center update process is repeated until the cluster center changes reach a stable condition, the category division results remain stable, or the preset number of iterations is reached. Finally, the mineralization body category, the surrounding rock category, and the ore categories of different grades are output.

[0053] In this embodiment, a density peak clustering algorithm is used to assist in the verification of the improved K-Means+ clustering results. The density peak clustering algorithm includes a local density calculation unit, a relative distance calculation unit, and a boundary sample identification unit. The local density calculation unit calculates the neighborhood density of each sample in the multi-property feature space. Samples with higher local density are typically located in the core region of a certain category. The relative distance calculation unit calculates the distance of a sample relative to samples with higher density. If a sample has both high local density and a large relative distance, it may represent a category center. The boundary sample identification unit identifies samples located between two categories that have low local density or abnormal distance distribution. When the high-density center identified by density peak clustering coincides with the category center of the improved K-Means+ clustering, it indicates that the category has high stability. When density peak clustering finds some samples at the boundary or outlier positions, these samples are marked as samples to be calibrated. For samples to be calibrated, category correction is performed by combining their weighted distance, neighboring sample categories, spatial location, and known geological attributes. Through this verification and calibration step, this embodiment can reduce misclassifications at class boundaries caused by improved K-Means+ clustering, thereby improving the reliability of mineralization boundary identification.

[0054] Due to differences in geological background, sample size, physical property distribution, and mineralization type among different gold polymetallic ore clusters, this embodiment allows setting parameter ranges based on actual data conditions, as shown in Table 2. Table 2. Feasible parameter configuration methods

[0055] In this embodiment, a multi-parameter clustering analysis of gold polymetallic ore clusters is applied. Density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity of ore and surrounding rock samples within the gold polymetallic ore clusters are collected, and sample number, spatial location, borehole depth, profile location, and sample type are recorded simultaneously to form an original physical property dataset. Outlier removal is performed on the original physical property data to eliminate significant deviations caused by sampling errors, testing errors, or data entry errors. Then, range standardization is used to transform physical property parameters of different dimensions to a uniform scale. Finally, PCA feature dimensionality reduction under geological constraints is employed to extract principal components that characterize the core physical properties of mineralization. Based on existing geological exploration data, borehole exposure data, test results of known mineralized samples, and test results of typical surrounding rock samples, seed samples of mineralized bodies, seed samples of surrounding rocks, and seed samples of ores of different grades are selected, and the initial seed cluster centers are calculated as the initial cluster centers. Then, a weighted distance metric is established based on the ability of each physical property parameter to distinguish mineralized bodies, surrounding rocks, and ores of different grades, and the samples are allocated and the cluster centers are updated until the category classification is stable. Density peak clustering is used to identify core samples, boundary samples, and outlier samples, and the calibrated category results are assigned back to the geological space to complete the ore body identification, grade classification, and exploration target area delineation.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0057] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for cluster analysis of multiple physical property parameters in a gold polymetallic ore cluster, characterized in that, Includes the following steps: Step 1: Collect multi-dimensional physical property parameters of ore and surrounding rock in the gold polymetallic ore cluster area to construct the original physical property dataset; Step 2: Preprocess the original physical property dataset to obtain a multi-physical property feature dataset; Step 3: Introduce prior geological information on gold polymetallic deposit formation, select samples with clear geological attributes from the multi-physical property dataset as seed samples, and determine the initial cluster centers based on the seed samples; Step 4: Establish a weighted distance metric and construct an improved K-Means+ clustering algorithm based on the initial cluster centers and the weighted distance metric. Step 5: Use the improved K-Means+ clustering algorithm to perform iterative clustering analysis on the multi-physical property dataset to obtain mineralization body categories, surrounding rock categories, and ore categories of different grades; Step 6: Verify and calibrate the mineralization type, the surrounding rock type, and the types of ore of different grades; The construction of the improved K-Means+ clustering algorithm includes: initially setting the clustering categories as mineralization categories, surrounding rock categories, and ore categories of different grades; Establish weighted similarity judgment rules, complete sample allocation, extract all samples within mineralization body category, surrounding rock category, and ore category of different grades, and update the mineralization body category center, surrounding rock category center, and ore category center of different grades by combining the multi-physical property characteristics of samples within each category and the weighted distance metric. Then, classify all samples in the multi-physical property data set, update the cluster centers of each category again, and output the final clustering results.

2. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 1, characterized in that... The extraction of all samples within the mineralization body category, the surrounding rock category, and the different grade ore categories, combined with the multi-physical property characteristics and weighted distance metric of samples within each category, and the subsequent updating of the mineralization body category center, the surrounding rock category center, and the different grade ore category centers, includes: according to the category labels already obtained for each sample, gathering samples assigned to the mineralization body category into the mineralization body sample set, gathering samples assigned to the surrounding rock category into the surrounding rock sample set, and gathering samples assigned to different grade ore categories into the corresponding grade ore sample sets, within each sample set... The density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity, and principal component characteristics of each sample are read. According to the established weighted distance metric, the physical properties that contribute highly to mineralization identification are used as important criteria for center updating, while the physical properties that contribute less are used as auxiliary criteria. The center of mineralization category is jointly determined by the weighted comprehensive physical properties of each sample in the mineralization sample set. The center of surrounding rock category is jointly determined by the weighted comprehensive physical properties of each sample in the surrounding rock sample set. The center of ore category for different grades is jointly determined by the weighted comprehensive physical properties of each sample in the corresponding grade ore sample set.

3. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 2, characterized in that, The formation of the weighted distance metric includes: based on the multi-property feature dataset and the initial cluster centers, statistically analyzing the distribution of various seed samples in terms of density, magnetic susceptibility, resistivity, polarizability, seismic wave velocity and principal component characteristics, analyzing intra-class concentration, inter-class separation, grade gradient and geological attribute consistency, and classifying contribution levels and determining weights accordingly.

4. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 1, characterized in that, The original physical property dataset uses each sample or measuring point as an independent data unit. Each original physical property data record includes sample number, sampling point coordinates, borehole number, borehole depth, profile location, sample type, density test value, magnetic susceptibility test value, resistivity test value, polarizability test value, seismic wave velocity test value, and geological attribute marker field. The geological attribute marker field is used to record the mineralization marker, surrounding rock marker, high-grade ore marker, medium-grade ore marker, or low-grade ore marker for known samples. Samples with unclear geological attributes are marked as unidentified samples in this field.

5. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 1, characterized in that, The preprocessing of the original physical property dataset includes: outlier removal, range standardization, and PCA feature dimensionality reduction under geological constraints.

6. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 1, characterized in that, The prior geological information for the formation of the gold polymetallic deposit includes existing geological exploration data, borehole exposure data, test results of known mineralized samples, and test results of typical surrounding rock samples. The seed samples include mineralized body seed samples, surrounding rock seed samples, and ore seed samples of different grades.

7. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 1, characterized in that, The density peak clustering algorithm is used to verify and calibrate the mineralization type, the surrounding rock type, and the ore types of different grades.

8. The method for cluster analysis of multiple physical property parameters in gold polymetallic ore clusters according to claim 7, characterized in that, The density peak clustering algorithm is used to calculate the local density and relative distance of each sample in the multi-physical property feature space, and to identify core samples, boundary samples and outliers based on the local density and relative distance. For core samples, the category label output by the improved K-Means+ clustering algorithm is retained. For boundary samples and outliers, the category is calibrated by combining weighted distance, neighboring sample category, spatial location and geological attribute information. The calibrated category label is then assigned back to the sampling point, borehole point or profile control point. The ore body distribution range is determined according to the spatial continuity of mineralization category points. The grade is graded according to the spatial combination relationship of ore category points of different grades. The exploration target area is delineated by combining the surrounding rock boundary and anomaly continuity.

9. An application of multi-physical parameter clustering analysis for gold polymetallic ore clusters, characterized in that, The multi-physical property parameter clustering analysis method for gold polymetallic ore clusters as described in any one of claims 1-8 is applied to the following: by collecting multi-dimensional physical property parameters such as density, magnetic susceptibility, resistivity, polarizability, and seismic wave velocity of ore and surrounding rock, an original physical property dataset is formed; after outlier removal, range standardization, and PCA feature dimensionality reduction under geological constraints, prior geological information on metallogenic formation is introduced to determine seed samples and initial cluster centers, and an improved K-Means+ clustering algorithm is constructed by combining a weighted distance metric criterion; finally, density peak clustering is used for verification and calibration to achieve ore body identification, grade classification, and target area delineation.

Images