A method and system for ore-prospecting calibration of elemental geochemical data screening

Abstract

The application discloses a kind of ore-prospecting calibration methods and systems of element geochemical data screening, and the application relates to the technical field of geological exploration, obtains the element geochemical original sampling data in research area, carries out format standardization processing to the element geochemical original sampling data, the dimension of different element concentration is unified, is arranged as structured data set, eliminates the format difference of different data sources, the multidimensional three-dimensional screening mechanism including statistics, space, element association is designed for element geochemical data, can effectively eliminate instrument noise, error sampling and irrelevant interference data in original data, substantially improve the purity of input data, effectively solve the serious interference problem caused by invalid original data input to subsequent deep learning prediction model.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, specifically to a mineral exploration calibration method and system for screening elemental geochemical data. Background Technology

[0002] With the application of deep learning technology in the field of geological prospecting, existing technologies mostly focus on the algorithm optimization of back-end prediction models, hoping to improve the accuracy of prospecting prediction by improving model structure and optimizing training strategies. However, in geoscientific data processing, the quality of the front-end raw data often determines the upper limit of model performance. Existing methods generally ignore the quality problems of raw elemental geochemical data caused by environmental interference, sampling equipment errors and human operation errors.

[0003] In existing technologies, for example, Chinese patent document CN119903897B discloses a method and device for optimizing gold ore target areas based on elemental geochemical anomalies. This scheme uses a converter network combined with a self-distillation model architecture to detect geochemical anomalies in order to optimize and screen gold ore target areas. The feature of this technical solution is that it constructs a complex end-to-end network to process geochemical samples. However, this technical solution directly inputs the original elemental geochemical sampling data into the model for training, without performing any pre-screening and correction for sampling noise, operational errors, spatial misalignment, and anomalous interference points in the original data. This inevitably leads to the model being severely interfered with by a large amount of invalid and erroneous data during the training process. Since geochemical sampling data generally exhibits sparse spatial distribution, the noise in the original data cannot be filtered out by the model itself, but is further amplified by the feature extraction mechanism of the complex model. This lack of control over the quality of the data source ultimately severely limits the accuracy of anomaly detection and target area selection, and cannot meet the urgent need for high-quality standardized data in current high-precision gold ore exploration. Summary of the Invention

[0004] The technical problem this invention aims to solve is that in existing gold prospecting technologies, the raw elemental geochemical data generally suffers from quality defects such as sampling noise, operational errors, abnormal interference, and batch system bias. Existing technologies do not design a pre-processing mechanism to address these issues and directly use the raw data for back-end model training, resulting in poor input data quality for subsequent prospecting models. Consequently, the accuracy of prospecting prediction and target area selection is low, failing to meet the needs of high-precision gold prospecting.

[0005] To address the aforementioned technical problems, this invention provides a mineral exploration calibration method and system for screening elemental geochemical data, which improves the quality of input data from the data source and provides standardized, high-quality data for subsequent mineral exploration prediction.

[0006] This invention provides a mineral exploration calibration method for screening elemental geochemical data, comprising the following steps:

[0007] Step S1: Raw data acquisition and preprocessing;

[0008] Raw elemental geochemical sampling data of the study area were obtained, and the raw elemental geochemical sampling data were subjected to preliminary format standardization processing to unify the dimensions of different element concentrations and organize them into a standard structured dataset to eliminate format differences between different data sources.

[0009] Step S2: Multi-dimensional data filtering;

[0010] For noisy, invalid, and interfering data in the structured dataset, a comprehensive screening is performed from three dimensions: statistical, spatial, and element correlation, to remove unqualified data points and obtain valid data points.

[0011] In the statistical dimension screening: statistical distribution analysis is performed on the concentration data of each element. Based on the regional background distribution characteristics of the elements, extreme outliers that deviate from the statistical distribution are removed, and noise data caused by operational errors and instrument errors in the sampling process is eliminated.

[0012] In spatial dimension filtering: based on the spatial coordinates of the sampling points, the neighborhood sampling density of each sampling point is calculated, isolated sampling points with abnormal spatial distribution are removed, and invalid data caused by sampling position deviation and sampling point misalignment are excluded;

[0013] In the element association dimension screening: based on the element symbiotic combination law of gold mineralization, the element association characteristics of each sampling point are analyzed, abnormal data points whose element combinations do not conform to the gold mineralization association law are removed, and interference data unrelated to gold mineralization are excluded.

[0014] Step S3: Mineral exploration calibration processing;

[0015] For the selected valid data points, a triple standardization calibration is performed, including background value calibration, anomaly threshold calibration, and mineralization correlation calibration, to eliminate systematic errors and unify the mineral exploration judgment criteria.

[0016] In background value calibration: combining the geological background and crustal element abundance of the study area, the regional background value of each element is calibrated, and systematic errors caused by different sampling batches and different sampling personnel are corrected to eliminate data deviations between batches.

[0017] In the anomaly threshold calibration: based on the elemental anomaly characteristics of known gold deposits in the study area, the anomaly judgment threshold of each mineralized element is calibrated, an anomaly judgment standard adapted to the current study area is established, and the scale of anomaly identification is unified.

[0018] In mineralization correlation calibration: the calibrated data is correlated and matched with the elemental characteristics of known gold deposits in the study area to establish a mapping relationship between elemental characteristics and mineralization potential, providing a standardized feature mapping benchmark for subsequent mineral exploration prediction and generating standardized high-quality data.

[0019] Step S4: Mineral exploration prediction processing;

[0020] The standardized, high-quality data is input into a preset mineral exploration prediction model to identify and optimize gold ore target areas, and the final mineral exploration results are output.

[0021] Furthermore, in step S2, the specific process of statistical dimension filtering is as follows: for the first... Calculate the mean and standard deviation of the concentration of all sampling points in the structured dataset using the given elements;

[0022] The formula for calculating the average concentration is:

[0023] ;

[0024] The formula for calculating standard deviation is:

[0025] ;

[0026] In the formula, For the first The average concentration of each element This represents the total number of sampling points. For the first The sampling point of the nth sampling point Concentration values ​​of the elements, For the first The standard deviation of the concentration of each element;

[0027] Calculate each sampling point for the th Standard score of elements The calculation formula is:

[0028] ;

[0029] when At that time, the judgment of the first Several sampling points were identified as extreme outliers and were removed. This is the preset statistical anomaly threshold.

[0030] Further, in step S2, the specific process of spatial dimension filtering is as follows: obtaining the first... Latitude and longitude coordinates of each sampling point and the Latitude and longitude coordinates of each sampling point The actual spatial distance between two points is calculated using the spherical distance formula. :

[0031] ;

[0032] In the formula, The average radius of the Earth;

[0033] Set the distance to the neighborhood radius Calculate the first Neighborhood sampling density of each sampling point :

[0034] ;

[0035] In the formula, This is an indicator function that takes the value 1 when the condition is met, and 0 otherwise.

[0036] like Then determine the first The sampling points were isolated and were removed. This is the preset lower limit threshold for density.

[0037] Furthermore, in step S2, the specific process of element association dimension filtering is as follows: setting the main ore-forming elements that indicate gold ore formation. With associated mineral elements A local window dataset is constructed centered on each sampling point, and the local correlation coefficient is calculated. :

[0038] ;

[0039] In the formula, This represents the number of sampling points within the local window. and Each is the first in the local window Major ore-forming elements at each sampling point Concentration and associated mineral elements concentration, and These are the average concentrations of the two elements within a local window, respectively.

[0040] like If the element combination at the central sampling point does not conform to the mineralization law, it will be discarded. This is the preset correlation coefficient threshold.

[0041] Further, in step S3, the background value calibration correction process is as follows:

[0042] ;

[0043] In the formula, For the corrected first The sampling point of the nth sampling point The standard concentration of each element, The first data point in the batch to which this sampling point belongs. Batch background mean of the element, To determine the first region based on the abundance of crustal elements in the region The standard area background value of each element.

[0044] Furthermore, in step S3, the mineralization correlation calibration includes calculating the mineralization potential score for each sampling point. :

[0045] ;

[0046] In the formula, This is the preset total number of ore-forming elements related to mineralization. For the first The preset weights of elements in mineralization association, The first step determined in the abnormal threshold calibration process The threshold for anomaly detection of a certain element.

[0047] This invention also provides a mineral exploration calibration system for screening elemental geochemical data, comprising:

[0048] The data preprocessing module is used to acquire raw elemental geochemical data, complete the data format standardization and dimensional unification processing, and output a structured raw dataset.

[0049] The multi-dimensional data filtering module is used to perform data cleaning and contains three sub-units: the statistical filtering unit, which is used to remove outlier noise points in the statistical dimension;

[0050] Spatial filtering unit, used to remove isolated invalid points in spatial dimensions;

[0051] The element association filtering unit is used to remove interfering data in the element association dimension;

[0052] The mineral exploration calibration module is used to perform unified and standardized data processing. It contains three sub-units: the background value calibration unit, which is used to perform regional background value calibration and batch system error correction;

[0053] Anomaly threshold calibration unit is used to perform anomaly determination threshold calibration for region adaptation;

[0054] The mineralization correlation calibration unit is used to perform calibration of the mapping relationship between elemental characteristics and mineralization potential;

[0055] The mineral exploration prediction module is used to identify and predict gold ore target areas based on calibrated standardized data, and output the final mineral exploration results.

[0056] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0057] (1) The present invention designs a multi-dimensional three-dimensional screening mechanism for elemental geochemical data, which includes statistics, space and elemental correlation. It can effectively remove instrument noise, erroneous sampling and irrelevant interference data in the original data, greatly improve the purity of the input data, and effectively solve the problem of serious interference caused by invalid original data input to the subsequent deep learning prediction model.

[0058] (2) This invention corrects the systematic errors caused by different sampling batches, different operators and regional geological changes from the source by designing a triple calibration mechanism of background value, anomaly threshold and mineralization correlation. It establishes a unified anomaly mapping benchmark that is adapted to specific exploration areas, and provides a guarantee for ensuring the scientific nature of data analysis scale from the source.

[0059] (3) Significantly improve the accuracy of high-precision gold mine target area selection and system generalization ability: After pre-screening and standardization calibration, the noise limitation of the subsequent model during training is greatly reduced and the convergence speed is significantly increased. The standardized input data not only improves the accuracy of gold mine target area prediction, but also, due to the calibration mechanism's adaptation to the current geological background characteristics, the overall mineral exploration system exhibits more excellent generalization and promotion capabilities when dealing with different geological structural areas and complex sampling methods. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method of the present invention;

[0061] Figure 2 This is a system architecture diagram of the present invention;

[0062] Figure 3 This is a concentration distribution diagram of the present invention;

[0063] Figure 4 This is a graph showing the prediction results of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0065] It should be noted that, in the specification of this invention, unless otherwise explicitly specified and limited, the connection relationship between various technical features and the interaction order of modules are understood in accordance with the conventional logic in the fields of geological exploration and computer data processing. The purpose of the element concentration, distance calculation and mathematical modeling formulas involved in the text is to enable the technical essence of this invention to be specifically realized through computer programming and mathematical derivation, and to ensure that the solution of this invention has complete operability and implementation value in actual complex geological prospecting projects.

[0066] As attached Figure 1-4 As shown, the core of this invention lies in constructing an end-to-end elemental geochemical data screening and calibration system, which changes the simple and crude approach of directly throwing coarse raw data into a black-box deep learning model in traditional geological prediction. This invention, through a series of rigorous mathematical statistics and geological geochemical rules for preprocessing, enables any nonlinear prediction model to run on a highly clean and unbiased standardized data base, thereby approaching the limit of mineral exploration prediction accuracy.

[0067] Example 1

[0068] As attached Figures 1-3 As shown, this embodiment takes a gold mining area as the first target research area and provides a mineral exploration calibration method for screening elemental geochemical data in a specific application scenario. The geological and mineralization background of this area is complex. After many years of exploration work, elemental geochemical data from multiple batches of stream sediment sampling points have been collected. Under this background, the data quality is significantly affected by sampling time, analytical instruments and human factors.

[0069] 1. Raw data acquisition and preprocessing

[0070] In this embodiment, raw elemental geochemical sampling data from a total of 1,200 stream sediment sampling points in the study area were obtained through a geological big data interface. These data included concentration data of 10 mineralization-related elements, including Au (gold), As (arsenic), Sb (antimony), Hg (mercury), Ag (silver), Cu (copper), Pb (lead), Zn (zinc), W (tungsten), and Mo (molybdenum). The corresponding latitude and longitude coordinates of the sampling points and sampling batch labels were also recorded. In addition, the dataset clearly identified the locations of 20 known gold mineralization deposits to facilitate subsequent anomaly feature extraction and model validation.

[0071] After acquiring the raw data, due to the chaotic format of the raw data provided by different laboratories, the concentration units of some elements were in ppm, some in ppb, and some even in mass percentage. This invention performs preliminary format standardization processing. Through a preset dimension conversion mapping table, the concentration units of all elements are strictly unified to μg / g (i.e., ppm level, while sufficient floating-point precision is retained for Au element to reflect low concentration background). After processing, the messy report text is organized into a standardized two-dimensional structured dataset matrix that is easy for computers to read, eliminating the chaotic format and missing null values ​​in the raw data (filling in a very small number of null values ​​using regional mean interpolation).

[0072] 2. Multi-dimensional data filtering

[0073] Although the original dataset has been cleaned, it still contains a lot of systematic and random noise. This embodiment performs a comprehensive screening and filtering of noise, invalid and interference data in the original data from three dimensions.

[0074] First, perform statistical dimension filtering, targeting the first... Calculate the average concentration of a given element (e.g., Au) across 1200 sampling points in a structured dataset. and standard deviation Since the abundance of geochemical elements typically follows a log-normal distribution within a region, this embodiment calculates the standard score for each sampling point for each element after taking the logarithm. For the calculation formula, please refer to the invention content section:

[0075] ;

[0076] This embodiment sets a statistical anomaly threshold. =3.0. When the absolute value of the standard score of a key element at a sampling point exceeds 3.0, that is, it deviates from the mean by 3 times the standard deviation, it means that the value is extremely unreasonable under natural geological conditions. It is highly likely that it is an extreme abnormal noise point caused by instrument failure or human error in decimal point recording. The computer system automatically judges the sampling point as an extreme abnormal point and removes it as a whole. In this statistical dimension screening, a total of 32 extreme noise points caused by instrument error and operation error were accurately removed.

[0077] Secondly, spatial dimension filtering is performed. Geochemical sampling points need to have a certain degree of spatial representativeness. Isolated sampling points cannot form an effective geochemical halo, and this is usually due to coordinate input errors. This embodiment uses the latitude and longitude coordinates of each sampling point. The Haversine formula is used to calculate the actual spatial distance between two points. :

[0078] ;

[0079] In the formula, the average radius of the Earth The value is 6371 km, and this embodiment sets the distance to the neighborhood radius. =1.0km, that is, taking each sampling point as the center, the number of neighboring sampling points within a 1 km radius is calculated as the neighborhood sampling density. Set a lower limit threshold for density. =5. If there are fewer than 5 other sampling points within a 1 km neighborhood of a certain sampling point, the spatial distribution of that point is considered too isolated or the coordinates are distorted. It is then judged as an invalid data point and removed. In this spatial dimension screening, a total of 18 invalid data points caused by misaligned sampling locations or abnormal coordinate records were removed.

[0080] Finally, elemental correlation dimension screening was performed. Based on the specific elemental symbiotic combination patterns in gold mineralization in this study area, geological research indicates that Au mineralization in this region is usually accompanied by strong synchronous enrichment of As (arsenic) and Sb (antimony), meaning that Au, As, and Sb elements have a significant positive correlation in the evolution of ore-forming fluids. This embodiment sets the main ore-forming element as... Au, associated mineral element For As and Sb, construct a local window containing 15 adjacent sampling points around each sampling point, and calculate the correlation coefficient within this local space. and Set the correlation coefficient threshold =0.3. If the Au concentration in a certain area is very high, but its correlation coefficient with As and Sb is less than 0.3, it indicates that the Au anomaly is not caused by hydrothermal mineralization, but is most likely caused by external pollution (such as modern industrial pollution or human gold mining residue). The computer determines that such data does not conform to the correlation law of gold mineralization and removes it. A total of 25 pollution interference data points that are not related to mineralization were removed in this correlation dimension screening.

[0081] After the above three layers of rigorous physical and geological logic filtering, the original 1200 sampling points were reduced to the remaining 1125 high-confidence valid data points. Invalid and noisy data in the original data were fully analyzed and removed, achieving a leapfrog improvement in the purity of the input data.

[0082] 3. Mineral exploration calibration processing

[0083] For the aforementioned 1125 valid data points, this embodiment further performs triple standardization calibration to completely eliminate systematic errors caused by different batch sampling and to unify the mineral exploration judgment criteria that machine learning algorithms can identify.

[0084] First, background value calibration was performed. Since the 1125 data points were derived from multiple batches of tests conducted by different engineering teams over three years, baseline drift in the spectral data from different laboratories resulted in a significant systematic bias. This embodiment combines the latest geological background rock spectral sampling survey and regional crustal element abundance data for the region to accurately calibrate the standard regional background values ​​for elements in this area. (For example, the standard background value for Au is calibrated to be 0.002 μg / g, and for As it is calibrated to be 2.5 μg / g.) Subsequently, for each batch of data, the batch mean background value of the corresponding element within that batch is calculated. The correction formula is used:

[0085] ;

[0086] The concentration of each element at each sampling point was offset reset and corrected one by one. Based on this, the concentration baseline deviation caused by instrument differences between different sampling batches was completely eliminated, and the data homogenization was achieved.

[0087] Next, an anomaly threshold calibration is performed. Traditional anomaly thresholds often use globally fixed empirical values ​​(such as the mean plus twice the standard deviation), which are prone to failure in complex geological regions. This embodiment fully utilizes information from 20 known gold deposits in the region, statistically analyzes the distribution patterns of calibrated and corrected concentration data from these 20 deposits and their adjacent areas, and extracts the lower quartile as the lower limit for anomaly detection. Thus, an Au anomaly detection threshold suitable for this specific region is calibrated. The abnormal threshold of As is 0.05 μg / g. The abnormal threshold of Sb is 10 μg / g. With a threshold of 0.8 μg / g, this dynamically generated threshold replaces the universal fixed threshold, greatly enhancing the adaptability of the anomaly identification standard to the current regional structure.

[0088] Finally, mineralization association calibration (potential mapping) is performed. In order to provide a more intuitive and comprehensive label guidance feature for the backend deep learning prediction model, this embodiment maps the calibrated data to a single scalar feature—mineralization potential score. The calculation formula is:

[0089] ;

[0090] For gold mines, the setting =Three key elements (Au, As, Sb), with weights assigned based on mineralization theories. =0.6, =0.25, =0.15. This calibration process calculates the relative enrichment exceeding the threshold at each point, assigns weights, and sums the results to generate a mineralization potential score. The higher the value, the greater the overlap between the anomalous combination of elements in the region and the known gold mineralization characteristics, and the stronger the probability of mineralization. This establishes a standardized and continuous feature mapping benchmark matrix for the backend model.

[0091] 4. Mineral Exploration Prediction Processing

[0092] As attached Figure 4 As shown, after completing the rigorous noise reduction and standardization calibration process described above, this embodiment inputs the 1125 high-quality standardized structural data containing multi-dimensional calibration features into the preset mineral exploration prediction model.

[0093] The mineral exploration prediction model used in this embodiment is an improved deep residual network (ResNet) model. Compared with a simple feedforward neural network, the residual network introduces a skip connection mechanism in the residual block. It can effectively solve the gradient vanishing and gradient explosion problems that occur in deep network training. In specific implementation, the multidimensional calibration features and neighborhood feature tensors of each sampling point are transformed into a one-dimensional and two-dimensional hybrid tensor format input. The model constructs a combination of convolutional layers containing four stages of feature extraction, uses Leaky ReLU as the activation function to maintain the ability to learn weak negative values, and configures the Adam optimization algorithm to adaptively adjust the learning rate.

[0094] Because the underlying data of the input residual network is extremely pure and has a unified physical meaning, the loss function (such as binary cross entropy loss) of the model converges extremely quickly during the training process. The training convergence speed is improved by more than 30% compared with the unprocessed version. The model calculates the predicted probability of the existence of gold mines in each grid space through forward inference.

[0095] Finally, based on the continuous probability distribution map output by the model, the system used a high-value contiguous clustering algorithm to delineate and output four high-confidence gold prospecting target areas with significant anomalies within the entire region.

[0096] Implementation Verification Results: To verify the reliability of the technical approach of this invention, the subsequent exploration team deployed a large number of trenching and deep drilling projects for the four prospecting target areas output by this embodiment. The actual engineering results showed that obvious gold-bearing ferruginous mineralization and quartz vein alteration zones were found underground in all four target areas. Notably, the Au grade obtained from core analysis of two of the target areas had reached the boundary requirements for industrial mining. The geological mineralization verification accuracy of the target areas selected by this system reached 100%. In contrast, if the original 1200 sampled data were directly input into the same residual network for prediction, due to the large amount of noise and batch bias in the data, the model could only barely delineate three fuzzy target areas, and subsequent exploration only found slight mineralization in one of them, with an accuracy of only 33%. Through the introduction of pre-screening and multiple calibration, the target area selection accuracy of this invention has achieved a revolutionary improvement, thoroughly verifying the core logic that "high-quality data input determines high-quality prediction output".

[0097] Example 2

[0098] To further demonstrate that the mineral exploration calibration method and system for screening elemental geochemical data proposed in this invention has a strong generalization ability and can adapt to different geological mineralization types without being constrained by specific environments, this embodiment selects a large-scale deep fault orogenic belt area as the second target study area for implementation of the method.

[0099] The ore-forming fluids of orogenic gold deposits typically migrate within ductile-brittle shear zones, and their ore-forming element assemblage is more complex than that of Example 1. Furthermore, the soil geochemical background is severely affected by intense tectonic weathering. A total of 2,800 soil geochemical sampling points were deployed and collected in this area, and elemental analysis was performed on various tracer elements, including Au, Ag, As, Sb, Bi, Hg, Cu, Pb, and Zn. Fifteen small veins have been identified in this area.

[0100] In step S1, facing a much larger dataset of 2800 sampling points, the data preprocessing module reads the original CSV file provided by the surveying department and extracts multiple chemical analysis indicators in a unified manner. In order to cope with the drastic fluctuations in the element content in the soil medium, this embodiment first performs a scale normalization operation based on logarithmic transformation (Log10) on elements with a large concentration range (such as Cu, Pb, Zn) to force all variables to be suppressed within the standard distribution range, and then organizes and outputs a brand-new two-dimensional feature matrix.

[0101] In step S2, given the strong heterogeneity of the soil data, the screening parameters were adaptively adjusted accordingly. In the statistical dimension screening, calculations were still performed. Standard scores are used, but due to frequent geological activity in orogenic belts, real anomalies may be extremely significant. To prevent the misclassification of real mineral-related anomalies, this embodiment will use statistical anomaly thresholds. The threshold was relaxed to 3.5 standard deviations, meaning points deviating more from extremes were removed. Ultimately, 45 points with serious analytical errors were eliminated. In the spatial dimension screening, considering that orogenic belts often extend linearly along fault zones, and that sampling points might not have been placed in some areas due to fault obstruction, this embodiment still calculated the spherical distance and set the lower density threshold. The density was lowered to 3, allowing for the existence of slightly lower-density linear edge points. This step eliminated 22 isolated points. In the elemental correlation dimension screening, in addition to the classic Au-As-Sb indicator element assemblage for orogenic gold deposits, a chalcophile element group was added. In this embodiment, two sets of composite correlation windows were constructed: the first set detected the positive correlation between Au and As and Sb, and the second set detected the moderate positive correlation between Au and Cu and Pb. When calculating the correlation matrix locally, a threshold was set for the joint correlation coefficient between the principal element Au and the two sets of elements. =0.25. If neither of these conditions is met, the data is considered a weathering enrichment artifact and is removed. A total of 41 data points were removed, resulting in 2692 clean and valid data points.

[0102] In step S3, the uplift of fault blocks in the orogenic belt varies greatly. In the background value calibration stage, this embodiment not only performs calibration based on the overall batch, but also introduces stratigraphic zonal calibration. That is, data points are associated with different parent rock stratigraphic blocks through GIS spatial mapping, and the background values ​​of specific areas of different strata (such as schist areas and granite areas) are calculated separately. To conduct stratigraphic-scale analysis This correction method is more precise than single-system batch correction, overcoming the interference of false elemental anomalies caused by lithological differences. In anomaly threshold calibration, it utilizes information from 15 ore bodies within the region and accurately captures the anomaly determination thresholds of different fault hanging walls and footwalls through non-parametric statistics (such as constructing cumulative frequency curves). In the calculation of potential score for mineralization correlation calibration At that time, expand The values ​​contain Ag and Bi elements, and the weight vector is set to... =[0.5,0.15,0.15,0.1,0.1], precisely mapping the exclusive metallogenic potential feature matrix of orogenic gold deposits.

[0103] In step S4, due to the abundant data and increased feature dimensions of the orogenic belt, this embodiment uses a deeper one-dimensional fully convolutional neural network (1D-FCN) fused with a self-attention mechanism to perform mineral exploration prediction processing. The 2692 data points with finely reconstructed calibration features from the pre-output are fed into the model. Through the model's deep nonlinear approximation and the allocation of attention weights between feature channels, the prediction model finally outputs a gridded smooth output of the mineralization probability within the entire orogenic fault zone. Five target areas highly coupled with the fault zone strike were successfully delineated. In the subsequent actual trenching project, strong ductile-brittle shear accompanied by sulfide alteration zones were found in all five target areas. Due to the triple standardization calibration at the front end, various complex background interferences have been fully removed. Even in the extremely difficult field of mineral exploration in soil-covered orogenic belts, this system has demonstrated extremely strong adaptability and technological advancement.

[0104] In summary, the mineral exploration calibration method and system for elemental geochemical data screening provided by this invention, based on rigorous mathematical logic and profound geological and geochemical principles, not only constructs a complete pre-processing data denoising and calibration chain, but also achieves a comprehensive upgrade of the deep learning geoscience data processing workflow. This technical solution has milestone practical guiding significance for promoting high-precision, intelligent deep mineral exploration and data-driven mineral exploration technologies.

[0105] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A mineral exploration calibration method for screening elemental geochemical data, characterized in that, include: Raw data acquisition and preprocessing: Obtain raw elemental geochemical sampling data within the study area, standardize the format of the raw elemental geochemical sampling data, unify the dimensions of different element concentrations, organize it into a structured dataset, and eliminate format differences between different data sources; Multi-dimensional data filtering: For noisy data, invalid data, and interference data in the structured dataset, perform the following filtering in sequence to obtain valid data points: Extreme outliers in the statistical dimension are eliminated by calculating the standard scores of specific element concentrations at each sampling point. The spherical distance is calculated based on the coordinates of each sampling point, and isolated points with abnormal spatial distribution are eliminated by statistically analyzing the neighborhood sampling density. A local window is constructed based on the preset main ore-forming elements and associated ore-forming elements. The local correlation coefficient is calculated, and interference points that do not conform to the ore-forming correlation law are eliminated. Mineral exploration calibration processing: For the valid data points, background value calibration, anomaly threshold calibration and mineralization correlation calibration are performed in sequence to eliminate system errors, unify mineral exploration judgment standards, and generate standardized high-quality data; Mineral exploration prediction processing: The standardized high-quality data is input into a preset mineral exploration prediction model to identify and select gold ore target areas, and output mineral exploration results.

2. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process for filtering the statistical dimensions is as follows: For each element, calculate the mean concentration and standard deviation of concentration for all sampling points in the structured dataset; Based on the concentration mean and concentration standard deviation, calculate the standard score for each sampling point for that element. The standard score is the quotient obtained by dividing the difference between the element concentration value and the concentration mean at that sampling point by the concentration standard deviation. When the absolute value of the standard score of a sampling point exceeds the preset statistical anomaly threshold, the sampling point is determined to be an extreme outlier and is removed.

3. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process for spatial dimension filtering is as follows: Based on the latitude and longitude coordinates of each sampling point, the actual spatial distance between any two sampling points is calculated using the spherical distance formula. Set a neighborhood radius, and with each sampling point as the center, count the number of other sampling points within the neighborhood radius, which is used as the neighborhood sampling density of that sampling point; If the sampling density of a sampling point's neighborhood is lower than a preset lower density threshold, then the sampling point is determined to be an isolated sampling point and is removed.

4. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process for filtering by the element association dimension is as follows: Define the primary and associated ore-forming elements in gold deposits; A local window dataset is constructed centered on each sampling point; Within the local window dataset, the local correlation coefficient between the main ore-forming element and the associated ore-forming element is calculated. The local correlation coefficient is obtained based on the ratio of the sum of the products of the deviations of the concentration values ​​of each sampling point of the two elements within the local window from their respective mean values ​​within the window to the square root of the product of the sum of the squares of the deviations of the two elements. If the local correlation coefficient is lower than the preset correlation coefficient threshold, the element combination of the central sampling point is determined to be inconsistent with the gold mineralization correlation law and is removed.

5. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process for background value calibration is as follows: Based on the geological background and crustal element abundance of the study area, standard regional background values ​​for each element were determined. For each batch of data, calculate the batch background mean of each element in that batch; The calibrated concentration of each sampling point is obtained by subtracting the batch background mean of the corresponding element from the element concentration of that batch, and then adding the standard area background value of the corresponding element. This process eliminates data deviations between batches.

6. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process for anomaly threshold calibration is as follows: Based on the elemental anomaly characteristics of known gold deposits in the study area, the distribution patterns of the concentration data of each ore-forming element in the known gold deposits and their adjacent areas after background value calibration and correction were statistically analyzed. Based on the distribution pattern, the corresponding statistical feature values ​​are extracted as the anomaly judgment thresholds for each ore-forming element, and an anomaly judgment standard adapted to the current study area is established to unify the anomaly identification scale.

7. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The specific process of the mineralization correlation calibration is as follows: The data after background value calibration and anomaly threshold calibration are correlated and matched with the elemental characteristics of known gold deposits in the study area. For each sampling point, the relative enrichment degree of each ore-forming element is calculated when its calibrated concentration exceeds its corresponding anomaly judgment threshold. The larger value between the relative enrichment degree and zero is taken, and the sum is performed according to the preset weights of each ore-forming element to obtain the ore-forming potential score of the sampling point. The higher the mineralization potential score, the greater the overlap between the anomalous combination of elements at the sampling point and the known gold mineralization characteristics, and the stronger the probability of mineralization. This establishes a mapping relationship between elemental characteristics and mineralization potential, providing a standardized feature mapping benchmark for subsequent mineral exploration prediction.

8. The mineral exploration calibration method for screening elemental geochemical data according to claim 1, characterized in that, The preset mineral exploration prediction model is a deep residual network model that introduces a residual block jump connection mechanism; the specific process of identifying and predicting gold ore target areas includes: converting the multidimensional calibration features of each sampling point into a one-dimensional and two-dimensional hybrid tensor format and inputting it into the deep residual network model, outputting a continuous probability distribution map of the existence of gold ore in each grid space, and delineating the gold ore target area through a clustering algorithm.

9. A mineral exploration calibration system for screening elemental geochemical data, characterized in that, include: The data preprocessing module is used to acquire raw elemental geochemical sampling data, complete the data format standardization and dimensional unification processing, and output a structured raw dataset. The multi-dimensional data filtering module includes statistical filtering units, spatial filtering units, and element association filtering units; The statistical screening unit is used to calculate the concentration mean and concentration standard deviation of each element, and to remove extreme abnormal noise points in the statistical dimension based on the comparison results of the absolute value of the standard score of each sampling point with the preset statistical anomaly threshold. The spatial filtering unit is used to calculate the neighborhood sampling density of each sampling point, and based on the comparison result of the neighborhood sampling density and the preset density lower limit threshold, to remove isolated invalid points in the spatial dimension. The element association filtering unit is used to calculate the local correlation coefficient between the main ore-forming element and the associated ore-forming element within the local window of each sampling point, and to remove interfering data in the element association dimension based on the comparison result of the local correlation coefficient and the preset correlation coefficient threshold. The mineral exploration calibration module includes a background value calibration unit, an anomaly threshold calibration unit, and a mineralization correlation calibration unit; The background value calibration unit is used to perform regional standard background value calibration and to perform batch systematic error correction on each sampling point; The anomaly threshold calibration unit is used to calibrate and adapt the anomaly judgment thresholds of each metallogenic element in the current study area based on the element distribution patterns of known gold deposits. The mineralization correlation calibration unit is used to calculate the mineralization potential score of each sampling point and establish the mapping relationship between elemental characteristics and mineralization potential; The mineral exploration prediction module is used to identify and predict gold ore target areas based on the standardized high-quality data output by the mineral exploration calibration module, and output mineral exploration results.

Images