CN118430692A

CN118430692A - Three-dimensional prospecting prediction method, system, equipment and medium

Info

Publication number: CN118430692A
Application number: CN202410881121.5A
Authority: CN
Inventors: 王斌; 高永宝; 刘向东; 贾嘉辉; 侯聪; 杨晨
Original assignee: Xi'an Mineral Resources Survey Center Of China Geological Survey
Current assignee: Xi'an Mineral Resources Survey Center Of China Geological Survey
Priority date: 2024-07-03
Filing date: 2024-07-03
Publication date: 2024-08-02

Abstract

The application relates to a three-dimensional prospecting prediction method, a system, equipment and a medium, which belong to the technical field of geological exploration, wherein the method comprises the following steps: obtaining geological sample data in a target mining area and carrying out data preprocessing; extracting element abnormal threshold values and determining abnormal geological samples; performing element combination analysis on the abnormal geological sample to determine a primary halation element combination; building a two-dimensional construction superposition corona practical model, and building a three-dimensional original corona model according to the three-dimensional spatial distribution characteristics of each original corona element; training the geochemical characteristics based on a machine learning algorithm, and establishing a mineralization distribution prediction model; fusing the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model to obtain a three-dimensional comprehensive prospecting prediction model; and predicting a prospecting target area according to the three-dimensional comprehensive prospecting prediction model to obtain a prospecting prediction result. The application can improve the working efficiency and accuracy of the prospecting prediction.

Description

Three-dimensional prospecting prediction method, system, equipment and medium

Technical Field

The application relates to the technical field of geological exploration, in particular to a three-dimensional prospecting prediction method, a three-dimensional prospecting prediction system, three-dimensional prospecting prediction equipment and three-dimensional prospecting prediction media.

Background

In the current mineral resource exploration field, along with the increasing exhaustion of the shallow surface layer mineral deposit resources, the exploration and prospecting direction gradually turns to deep and peripheral areas, and the transition brings higher requirements to the prospecting technology, especially in the aspects of accuracy and efficiency of deep prospecting prediction.

Although the existing structure superposition corona theory makes progress of certain prediction in the ore finding technology and can reflect the control of the structure on mineralization, the detail and complexity description of deep mineralization distribution are limited, the accurate qualification of an accurate target area is difficult to meet, and the complex ore deposit with multiple ore veins has great limitation. Meanwhile, the machine learning algorithm has strong potential in the field of mineral resource prediction, mineral evaluation capability is improved by mining mineralization information of high-dimensional data, but challenges are still faced in the aspect of unknown region prediction outside a non-engineering control range.

Therefore, various existing prospecting prediction techniques still have many challenges when facing complex geological environments, especially in deep and peripheral prospecting, mineralization distribution is difficult to predict accurately, and resource waste and low prospecting efficiency are caused.

Disclosure of Invention

In order to improve the working efficiency and accuracy of the prospecting prediction, the application provides a three-dimensional prospecting prediction method, a system, equipment and a medium.

In a first aspect, the present application provides a three-dimensional prospecting prediction method, which adopts the following technical scheme:

A three-dimensional prospecting prediction method, the prediction method comprising:

Obtaining geological sample data in a target mining area and carrying out data preprocessing; the geological sample data comprises soil geochemical sample data, drilling sample data and slot probe sample data;

Performing anomaly identification on the preprocessed geological sample data, extracting element anomaly thresholds and determining an anomaly geological sample;

performing element combination analysis on the abnormal geological sample, and determining a primary corona element combination according to concentration banding characteristics, axial banding sequences and element correlation;

Generating a grade contour map according to the geological sample data, and determining ore body positioning information according to the geometric characteristics of the grade contour map;

Constructing a two-dimensional construction superposition corona practical model based on the element abnormal threshold, the concentration banding feature, the axial banding sequence, the primary corona element combination and the ore body positioning information;

according to the original halo element combination and the drilling sample data, obtaining three-dimensional space distribution characteristics of each original halo element;

based on the two-dimensional construction superposition corona practical model, a three-dimensional original corona model is established according to the three-dimensional spatial distribution characteristics of each original corona element;

extracting geochemical characteristics according to the three-dimensional original vignetting model, training the geochemical characteristics based on a machine learning algorithm, and establishing a mineralization distribution prediction model;

fusing the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model to obtain a three-dimensional comprehensive prospecting prediction model;

And predicting a prospecting target area according to the three-dimensional comprehensive prospecting prediction model to obtain a prospecting prediction result.

By adopting the technical scheme, the three-dimensional original vignetting model is established by utilizing the three-dimensional visualization technology on the basis of the two-dimensional structure superposition vignetting model, the change rule of deep space elements can be reflected more three-dimensionally and objectively, meanwhile, the mineralization distribution prediction model based on machine learning is established, and the comprehensive three-dimensional prospecting prediction model is formed by fusing the structure control information provided by the two-dimensional structure superposition vignetting model, the mineralization space distribution characteristics of the three-dimensional original vignetting model and the mineralization probability distribution of the machine learning prediction model, so that the conversion from single dimension to multiple dimension, single element to multiple element and single method to multiple method is realized, the uncertainty of resource prediction is reduced, and the working efficiency and the accuracy of deep and peripheral prospecting are improved.

Optionally, the step of performing anomaly identification on the preprocessed geological sample data, extracting an element anomaly threshold value and determining an anomaly geological sample includes:

obtaining element distribution characteristics according to the preprocessed geological sample data;

Obtaining fractal dimension characteristics according to the element distribution characteristics, and constructing a concentration-length fractal model;

and extracting element anomaly threshold values from the geological sample data based on the concentration-length fractal model, and determining an anomaly geological sample.

Optionally, performing element combination analysis on the abnormal geological sample, and determining the original vignetting element combination according to the concentration banding features, the axial banding sequences and the element correlation comprises the following steps:

Performing data conversion on the abnormal data subset of the abnormal geological sample;

Generating an element concentration banded graph according to the converted abnormal data subset, and identifying and obtaining concentration banded characteristics and an axial banded sequence;

Performing dimension reduction processing on the converted abnormal data subset, and identifying correlation among elements and element combination to obtain an element correlation analysis result;

Obtaining a primary corona element combination according to the concentration banding characteristics, the axial banding sequence and the element correlation analysis result; wherein the combination of native halo elements comprises elements that spatially exhibit a distribution characteristic of leading edge, near-mine, tail halo.

Optionally, based on the two-dimensional construction superposition corona utility model, the step of establishing a three-dimensional original corona model according to the three-dimensional spatial distribution characteristics of each original corona element comprises the following steps:

extracting a construction control factor based on the two-dimensional construction superposition corona practical model;

According to the three-dimensional space distribution characteristics of each original corona element, a three-dimensional element distribution model is established;

Establishing a three-dimensional geological framework according to the three-dimensional element distribution model;

Integrating the construction control factors in the three-dimensional geological frame and adjusting element distribution to obtain a three-dimensional original vignetting model.

Optionally, training the geochemical feature based on a machine learning algorithm, and the step of establishing the mineralization distribution prediction model includes:

preprocessing the geochemical features;

Performing feature selection on the preprocessed geochemical features based on correlation analysis, and performing standardization processing on the selected feature vectors to obtain an input feature set;

dividing the input feature set into a training set, a verification set and a test set;

Inputting the training set into a pre-constructed machine learning algorithm model for training, optimizing model parameters and calculating a loss function until the loss function meets preset conditions or the number of model iterations reaches preset times, so as to obtain the trained mineralized distribution prediction model;

verifying the mineralization distribution prediction model based on the verification set, evaluating the performance of the mineralization distribution prediction model and adjusting the super parameters of the model;

and testing the prediction capability of the adjusted mineralization distribution prediction model based on the test set.

Optionally, the step of fusing the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model to obtain the three-dimensional comprehensive prospecting prediction model comprises the following steps:

According to the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model, respectively obtaining output characteristics of each model and carrying out standardization treatment;

performing feature selection and dimension reduction processing based on the standardized output features to obtain feature vectors after dimension reduction;

Respectively distributing dynamic weight values for the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model;

carrying out weighted average on the feature vectors subjected to dimension reduction according to the weight values to obtain a preliminary fusion prediction result;

Combining the output characteristics of each model with the preliminarily fused prediction results to obtain a data set and dividing the data set into a training set, a verification set and a test set;

Performing primary training on a pre-built deep neural network model based on a training set, and performing cross verification and model parameter optimization adjustment on the primary trained deep neural network model based on a verification set and a test set to obtain a three-dimensional comprehensive prospecting prediction model; the adjustment of the dynamic weight value is based on the prediction error of the model in the cross verification process, and the weight configuration is optimized by minimizing the prediction error through a gradient descent method.

Optionally, predicting the prospecting target area according to the three-dimensional comprehensive prospecting prediction model, and the step of obtaining the prospecting prediction result includes:

Determining a mining target area according to a model prediction result of the three-dimensional comprehensive mining prediction model;

Receiving the field verification data of the mining target area input by a user;

Judging whether the result accords with a model prediction result according to the field verification data, if so, determining the mining target area as the mining prediction result;

if not, carrying out optimization correction on the three-dimensional comprehensive prospecting prediction model according to the field verification data, and adjusting model parameters until a model prediction result accords with the fed-back field verification data to obtain an optimized three-dimensional comprehensive prospecting prediction model;

And predicting a prospecting target area according to the optimized three-dimensional comprehensive prospecting prediction model, and determining the prospecting target area as a prospecting prediction result.

In a second aspect, the application provides a three-dimensional prospecting prediction system, which adopts the following technical scheme:

a three-dimensional prospecting prediction system, the prediction system comprising:

The acquisition module is used for acquiring geological sample data in the target mining area; the geological sample data comprises soil geochemical sample data, drilling sample data and slot probe sample data;

The preprocessing module is used for preprocessing the data of the geological sample;

the anomaly identification module is used for carrying out anomaly identification on the preprocessed geological sample data, extracting element anomaly threshold values and determining an anomaly geological sample;

the primary corona element combination determining module is used for carrying out element combination analysis on the abnormal geological sample and determining a primary corona element combination according to the concentration banding characteristics, the axial banding sequence and the element correlation;

The ore body positioning information determining module is used for generating a grade contour map according to the geological sample data and determining ore body positioning information according to the geometric characteristics of the grade contour map;

The two-dimensional model construction module is used for constructing a two-dimensional construction superposition corona practical model based on the element abnormal threshold, the concentration banding characteristic, the axial banding sequence, the primary corona element combination and the ore body positioning information;

The element distribution feature generation module is used for obtaining three-dimensional space distribution features of each original vignetting element according to the original vignetting element combination and the drilling sample data;

The three-dimensional model building module is used for building a three-dimensional original halo model according to the three-dimensional spatial distribution characteristics of each original halo element based on the two-dimensional construction superposition halo practical model;

the prediction model construction module is used for extracting geochemical characteristics according to the three-dimensional original vignetting model, training the geochemical characteristics based on a machine learning algorithm and establishing a mineralization distribution prediction model;

The comprehensive prediction model construction module is used for fusing the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model to obtain a three-dimensional comprehensive prospecting prediction model;

And the prospecting prediction result generation module is used for predicting a prospecting target area according to the three-dimensional comprehensive prospecting prediction model to obtain a prospecting prediction result.

In a third aspect, the present application provides a computer device, which adopts the following technical scheme:

a computer device comprising a memory, a processor and a computer program stored on the memory, the processor executing the computer program to perform the steps of the method according to the first aspect.

In a fourth aspect, the present application provides a computer readable storage medium, which adopts the following technical scheme:

a computer readable storage medium storing a computer program capable of being loaded by a processor and executing any one of the methods of the first aspect.

Drawings

FIG. 1 is a schematic flow chart of a three-dimensional prospecting method according to an embodiment of the application.

FIG. 2 is a schematic diagram of a second flow chart of a three-dimensional prospecting method according to one embodiment of the present application.

FIG. 3 is a schematic view of a third flow chart of a three-dimensional prospecting method according to an embodiment of the application.

FIG. 4 is a fourth flow chart of a three-dimensional prospecting method according to one embodiment of the present application.

FIG. 5 is a fifth flow chart of a three-dimensional prospecting method according to one embodiment of the present application.

FIG. 6 is a sixth flow chart of a three-dimensional prospecting method according to one embodiment of the application.

FIG. 7 is a seventh flow chart of a three-dimensional prospecting method according to one embodiment of the present application.

FIG. 8 is a schematic diagram of an eighth flow chart of a three-dimensional prospecting method according to one embodiment of the application.

Fig. 9 is a schematic diagram of a constructed superimposed corona utility model established during an actual study of the present application.

FIG. 10 is a schematic representation of a three-dimensional native halo model established during an actual study of the present application.

Fig. 11 is a three-dimensional probability distribution diagram of a random forest algorithm prediction result obtained in the actual research process.

FIG. 12 is a three-dimensional probability distribution diagram of the support vector machine prediction result obtained in the actual research process of the application.

Fig. 13 is a ROC graph of a random forest and support vector machine of the present application during actual research.

FIG. 14 is a schematic view of a three-dimensional comprehensive prospecting prediction model established in the practical research process of the application.

Detailed Description

The present application will be further described in detail with reference to fig. 1 to 14 and examples in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

Terms related to the present embodiment are explained first.

Primary dizziness: the primary halation refers to an abnormal increase in the content of geochemical elements in the rock surrounding the ore deposit, especially around the ore body. This increase usually occurs in synchronism with the mineralization process, accompanied by the formation of ore bodies, exhibiting a certain spatial morphology and element distribution characteristics reflecting the distribution law of mineralized elements in migration and enrichment in the primary environment. The primary corona is an important cue in the exploration study, indicating the mineralization process and the possible extent of the ore body.

Constructing a superimposed corona theory: the theory of formation overlapping corona is based on the concept that a hydrothermal deposit is strictly controlled by a formation, and structural characteristics (such as faults and folds) are considered to influence the distribution of mineralized fluid, so that the control of the formation on the axial zonation of the primary corona is emphasized, and the spatial distribution of mineralization and ore body prediction under the control of the formation can be understood through structural analysis.

Target area of prospecting: the prospecting target area is a specific area or place with higher prospecting potential, which is predicted by a series of technical means in the mineral exploration process and combines the geological theory and the practical experience, is the object of key research and exploration in the exploration work, and is the position which is considered to be most likely to find mineral resources based on the existing data and analysis.

Next, the present application will be further described in the background.

At present, students at home and abroad use a primary corona prospecting method to obtain important breakthroughs in aspects of searching new ore bodies, increasing mineral reserves and the like due to hot liquid forming minerals, and put forward a structural superposition corona theory after putting forward a primary superposition corona concept, and closely combine the structural superposition corona with a mineral formation rule to determine whether a favorable mineral formation space is mineral or not, so that the geochemistry prospecting thought is clearer, and the accuracy of target prediction is improved.

However, the primary corona prospecting method based on the two-dimensional plane is mainly used for qualitative evaluation of single vein bodies, has great limitation on complex deposits with multiple veins, and cannot grasp the overall situation in an omnibearing and quantitative manner; in order to solve the problem, part of scholars introduce a three-dimensional visualization technology into geological work, and deep prediction is carried out by establishing a three-dimensional geological model of a mining area, but the three-dimensional display modeling process is excessively emphasized or a nationwide original vignetting zoning sequence is directly used, each deposit is neglected to be formed on different geological backgrounds, and the mining area is subjected to a complex geological process, so that the mining area has uniqueness. And with the rapid development of machine learning and artificial intelligence, the machine learning algorithm is introduced into three-dimensional mineral quantitative evaluation, so that high-dimensional mineralization information can be excavated to a greater extent, and the method has important reference significance for developing mineral resource potential evaluation of a hot liquid mineral deposit under construction control, but has limitation in prospect prediction beyond the engineering control range.

Therefore, how to combine the structure control information of the two-dimensional structure superposition corona with the mineralization space distribution rule of the three-dimensional original corona and how to efficiently integrate the information with the prediction capability of the machine learning model so as to realize the omnibearing and high-precision prediction of deep and peripheral prospecting, form a comprehensive technical system capable of comprehensively reflecting the deep prospecting rule and expanding the prediction range, become urgent demands for the development of the current prospecting technology, and are the key technical problems to be solved in the present.

Based on the above, the embodiment of the application discloses a three-dimensional prospecting prediction method.

Referring to fig. 1, a three-dimensional prospecting prediction method includes:

S101, obtaining geological sample data in a target mining area and preprocessing the data; the geological sample data comprises soil geochemical sample data, drilling sample data and slot probe sample data;

The geological sample data of the target mining area can be collected through geological investigation, geophysical prospecting, chemical prospecting, drilling and other means; in some embodiments, the geological sample data may include soil geochemical samples, borehole raw halo samples, trench raw halo samples, and other basic analysis samples.

As an implementation mode of data preprocessing, the data preprocessing step can be to clean the collected data, remove abnormal values, ensure the accuracy and consistency of the data and lay a solid foundation for subsequent analysis; for example, a center log ratio transformation method may be employed to transform data from simplex space to euclidean space to achieve a standardized analysis of different geological background data.

Step S102, carrying out anomaly identification on the preprocessed geological sample data, extracting element anomaly threshold values and determining an anomaly geological sample;

The abnormal threshold is extracted by analyzing the distribution difference of elements in different stratum and sampling media. The method comprises the steps of identifying fractal dimension characteristics in nature by calculating the relation between element concentration and sampling data, defining normal and abnormal limits, applying an abnormal threshold to the preprocessed data, and screening out abnormal geological samples beyond the threshold range, wherein the samples are regarded as areas possibly indicating mineralization activities.

Step S103, performing element combination analysis on the abnormal geological sample, and determining a primary halation element combination according to the concentration banded characteristic, the axial banded sequence and the element correlation;

The element combination analysis is to determine element combinations at different stages in the original vignetting according to concentration zonation characteristics, axial zonation sequences and element correlation through deep research on abnormal samples, such as front vignetting, near vignetting and tail vignetting; these elemental combinations provide key chemical markers for ore body localization as they are able to indicate the extent and intensity of mineralization, as well as the migration path and mineralization process of mineralized fluid.

Specifically, the concentration zoning characteristic refers to a change rule of element concentration in space, the concentration is generally higher near the center of a ore body, the concentration gradually decreases outwards, and the distribution mode is helpful for distinguishing a mineralization center from a peripheral area; the axial zoning sequence refers to orderly change of elements distributed along the axial direction of the ore body, and the relative enrichment degree of different elements at different depths or heights reflects zoning information of hydrothermal activity; element correlation analysis reveals symbiotic relationships between different elements, helping to understand how combinations of elements collectively indicate the mineralization process and the presence of ore bodies.

In one embodiment of the application, the combination of primary corona elements (e.g., as, sb As leading edge corona, au, ag, cu, zn, ni, mo As near-mine corona, pb, mn, co, W, fe As tail corona) provides a direct indication of the mineralization process, which is an indispensable geological information in model construction.

Step S104, generating a grade contour map according to geological sample data, and determining ore body positioning information according to geometric features of the grade contour map;

The ore body positioning information comprises the spatial tendency, the lateral bearing direction and the fluid migration direction of the ore body; by drawing the grade contour map, the spatial distribution characteristics of the ore body, such as spatial tendency and lateral direction, can be revealed, and meanwhile, the fluid migration direction can be deduced according to the trend of the contour map, so that the spatial layout of the ore body can be displayed more intuitively, and a direct visual basis is provided for deep ore finding.

Step S105, constructing a two-dimensional construction superposition corona practical model based on element abnormal threshold values, concentration banding features, axial banding sequences, primary corona element combination and ore body positioning information;

The element anomaly threshold provides a quantification standard for distinguishing mineralized areas from non-mineralized areas in the model, the concentration banded features and the axial banded sequences are helpful for reproducing longitudinal banded features and transverse banded features of mineralized bands in the model, the combination of primitive halation elements provides indispensable geological chemical information for model construction, and ore body positioning information is crucial for reflecting three-dimensional spatial distribution of ore bodies in model construction.

Step S106, according to the combination of the original vignetting elements and the drilling sample data, obtaining three-dimensional space distribution characteristics of each original vignetting element;

In some embodiments, three-dimensional spatial interpolation is performed by using three-dimensional coordinate information of a drilling sample and combining with the combination of original vignetting elements through a three-dimensional empirical Bayesian-Kriging method, so that the spatial distribution characteristics of each original vignetting element can be obtained;

it can be understood that three-dimensional structural features of mineralized elements are displayed by reconstructing three-dimensional spatial distribution diagrams of the elements, so that visual assistance is provided for deep mining.

Step S107, based on a two-dimensional construction superposition corona practical model, establishing a three-dimensional original corona model according to the three-dimensional spatial distribution characteristics of each original corona element;

On the basis of constructing the overlapped corona, a two-dimensional overlapped corona model and three-dimensional spatial distribution characteristics of the elements are combined by analyzing the spatial distribution characteristics of each element and the corona body, so that a finer three-dimensional original corona model can be constructed, the process considers the comprehensive influence of three-dimensional distribution and construction control of mineralized elements, the conversion from the two-dimensional original corona to the three-dimensional original corona prospecting prediction method is realized, and the prospecting prediction accuracy is improved.

Step S108, extracting geochemical features according to the three-dimensional original vignetting model, training the geochemical features based on a machine learning algorithm, and establishing a mineralization distribution prediction model;

In some embodiments, geochemical features closely related to mineralization formation are extracted from the three-dimensional native halo model, which may include specific element concentrations, element combination ratios, anomaly intensities, spatial distribution patterns, and the like.

Specifically, training can be performed through a machine learning algorithm (such as a random forest, a support vector machine and the like), a mineralization distribution rule in historical data is learned, and a mineralization distribution prediction model is constructed, and can predict mineralization possibility of an unknown region according to known geological features.

Step S109, fusing a two-dimensional structure superposition corona practical model, a three-dimensional original corona model and a mineralization distribution prediction model to obtain a three-dimensional comprehensive prospecting prediction model;

specifically, the structure control information provided by the two-dimensional structure superposition corona model, mineralization space distribution characteristics of the three-dimensional original corona model and mineralization probability distribution of the machine learning prediction model are fused to form a comprehensive three-dimensional prospecting prediction model, and the model integrates the advantages of qualitative and quantitative prediction and can evaluate prospecting potential more comprehensively.

And S110, predicting a prospecting target area according to the three-dimensional comprehensive prospecting prediction model to obtain a prospecting prediction result.

The method comprises the steps of utilizing a comprehensive prospecting prediction model, and delineating prospecting target areas through a set threshold value or an optimization algorithm, wherein the target areas are hypersalinity probability areas based on model prediction, and are important points of future prospecting work; the method provides scientific basis for subsequent exploration design and resource evaluation by using the prospecting prediction result, is beneficial to efficiently finding potential mineral resources, and provides scientific support for national resource safety and economic development.

In the embodiment, the three-dimensional original halo model is established by utilizing the three-dimensional visualization technology on the basis of the two-dimensional structure superposition halo model, so that the change rule of deep space elements can be reflected more three-dimensionally and objectively, meanwhile, a mineralization distribution prediction model based on machine learning is established, and a comprehensive three-dimensional prospecting prediction model is formed by fusing structure control information provided by the two-dimensional structure superposition halo model, mineralization space distribution characteristics of the three-dimensional original halo model and mineralization probability distribution of the machine learning prediction model, so that the conversion from single dimension to multiple dimension, single element to multiple element and single method to multiple method is realized, the uncertainty of resource prediction is reduced, and the working efficiency and accuracy of deep and peripheral prospecting are improved.

Referring to fig. 2, as an embodiment of step S102, the step of performing anomaly identification on the preprocessed geological sample data, extracting an element anomaly threshold value, and determining an anomaly geological sample includes:

Step S201, element distribution characteristics are obtained according to the preprocessed geological sample data;

the preprocessed geological sample data comprises element average concentration information, and a calculation formula of the element average concentration is as follows:

；

Wherein Cx is the average concentration of an element, n is the total number of samples, and Ci and Li are the element concentration and length of the i-th sample, respectively; this weighted average reflects the average distribution of the elements throughout the rock volume, rather than the concentration on a single sample only.

It can be understood that the element concentration data can be processed to better reflect the element distribution characteristics of deep rock, and the average concentration of elements is calculated by introducing the sample length as a weight factor, and the calculation process of the element distribution characteristics is an important bridge connecting data preprocessing and model construction.

Step S202, fractal dimension characteristics are obtained according to element distribution characteristics, and a concentration-length fractal model is constructed;

The method and the device for detecting the abnormal conditions of the geological sample are characterized in that the length of the sample is newly introduced as a weight index, the relationship between the element concentration in the geological sample data and the length of the sample is analyzed by utilizing the fractal theory, a concentration-length fractal model is constructed, a traditional abnormal extraction method is optimized, the difference of information carried by samples with different lengths is considered, the challenge caused by the non-uniformity of the sampling length of the drill hole is solved, and the accuracy of abnormal identification is improved.

It should be noted that, generally, the geochemical background follows normal or lognormal distribution, while the geochemical anomaly follows power law distribution, and researches in recent years show that the effect of the traditional method adopted in the complex chemical detection data processing process is not ideal, and the fractal model can quantitatively represent the nonlinear geological anomaly intensity, so that the method becomes an important method for determining and explaining the anomaly threshold. The drilling sampling length is obtained by identifying lithology, mineralization and the like of the drilling by a geological editor, and different lengths are manually divided, so that the drilling sampling length has non-uniformity, but the sampling is uniform and continuous in a certain sample length space, so that the drilling sampling length has the characteristic of discreteness.

Specifically, the expression of the concentration-length fractal model is: and L (> C) ≡C ^−α, namely, a power exponent relation exists between the length L (> C) of the sample with the element concentration higher than C and the concentration C, the power exponent alpha is the embodiment of fractal dimension characteristics, and the value of the power exponent alpha can be obtained by fitting data points, so that the value is helpful for understanding the distribution characteristics of data.

As one embodiment for obtaining fractal dimension characteristics, the specific steps include: constructing a bipartite graph of element concentration and sample length L corresponding to more than a specific concentration C based on the collected geological sample data; and (3) observing the data distribution in the bipartite graph, if the data points are approximately distributed in a straight line, indicating that a power law relation exists, and performing linear fitting on the data points through a least square method or other linear fitting methods to obtain a straight line, wherein the slope of the straight line is a power exponent alpha, and the calculated power exponent alpha is the embodiment of fractal dimension characteristics. Wherein, the specific concentration C may be a background concentration of an element or a threshold value preliminarily determined by other methods, in a double-logarithmic coordinate system, a logarithmic value of the concentration C is taken as a horizontal axis, a logarithmic value of a sample length L exceeding the concentration is taken as a vertical axis, and then all data points are marked on the graph.

It can be understood that, because the samples of the drill holes and the exploratory slots are divided into samples with different lengths according to the detailed observation of the rock core by geological editors and the changes of the color, granularity, material composition, mineralization alteration and the like of the rock, the method of directly obtaining the element average value cannot objectively reflect the real situation of the deep rock. According to the method, the length is used as the weight to be introduced into the process of calculating the element average value, the obtained element average concentration information is necessary input data for constructing a concentration-length fractal model, abnormal values can be more reasonably identified, and more accurate basis is provided for subsequent element combination analysis and prospecting prediction.

And step S203, element abnormality thresholds are extracted for the geological sample data based on the concentration-length fractal model, and abnormal geological samples are determined.

As one embodiment of extracting the element anomaly threshold using a concentration-length fractal model, the specific steps include: sequencing element concentration data in each geological sample data, and calculating the accumulated length of samples with the length L being greater than C at different concentration levels C; three straight lines are fitted through a relation between concentration and length shown by a double-log plot (log-log plot), and the three straight lines correspond to a low background area, a high background area and a high abnormal area respectively; the intersection point of the two sections of straight lines is defined as a critical point of the background and the abnormality, namely an abnormality lower limit; and determining an abnormal critical point of each element according to the fitting result, wherein the abnormal critical point is an abnormal threshold value and represents the boundary between the normal background and mineralization abnormality.

In the embodiment, according to the fractal theory, the length of the sample is combined as a weight index, a C-L fractal model based on the length is provided and is used for extracting element anomaly threshold values, separating geological background and anomaly, determining which geological samples have element contents exceeding the background level, and determining that the abnormal state is considered as an abnormal geological sample and possibly indicating mineralization activity, so that the geological sample containing mineralization information is accurately identified, and a foundation is provided for subsequent ore body positioning.

Referring to fig. 3, as an embodiment of step S103, performing an elemental composition analysis on an abnormal geological sample, and determining a native halo elemental composition based on a concentration banding feature, an axial banding sequence, and an elemental correlation, includes:

Step S301, converting the abnormal data subset of the abnormal geological sample;

wherein the anomaly data subset contains only element concentration data exceeding an anomaly threshold value.

In some embodiments, a center-to-log method may be employed to "open" the closed geochemical data such that the anomaly data subset is converted from a simplex space to a euclidean space; moreover, as the distribution of the earth surface sampling media has heterogeneity, the data of different geological units are leveled, so that the subsequent comprehensive analysis and comparison are facilitated;

Step S302, generating an element concentration banded graph according to the converted abnormal data subset, and identifying and obtaining concentration banded features and an axial banded sequence;

the concentration banded characteristic and the axial banded sequence of the elements are used for defining the spatial distribution characteristic and mineralized flow direction of the elements.

In some embodiments, the element concentration banded graph is generated on the vertical longitudinal projection graph, so that the concentration distribution mode of the element in space can be visually displayed, further, the banded with obvious concentration gradient change is identified, and the concentration banded characteristic is obtained; the improved gligo index method (such as Wang Jianxin and the method proposed in 2007) can be used, and the automatic calculation of the banding index of each element can be realized through programming, and the axial banding sequence can be determined according to the calculation result.

Step S303, performing dimension reduction processing on the converted abnormal data subset, and identifying the correlation among elements and the element combination to obtain an element correlation analysis result;

In some embodiments of the present application, the transformed data may be subjected to a dimension reduction process by type-R factor analysis (e.g., principal component analysis) to identify correlations between elements and combinations of elements. For example, the strength of relationship (load value) of each element to the factor may be demonstrated by a twiddle factor load matrix, where loads with absolute values greater than a preset threshold are considered significant.

Step S304, according to concentration banding characteristics, axial banding sequences and element correlation analysis results, a primary corona element combination is obtained; wherein the combination of native halo elements comprises elements that spatially exhibit a distribution characteristic of leading edge, near-mine, tail halo.

In the above embodiment, in order to identify whether there is a statistical interdependence between different elements in a geological sample, element correlation analysis has important significance for understanding geological processes, mineralization mechanisms and guiding prospecting; and finally, determining the element combination of the original halo by comprehensively judging which elements show the distribution characteristics of the leading edge, the near-mine and the tail halo in space.

Referring to fig. 4, as an embodiment of step S104, the geological sample data further includes geological sampling point coordinates and corresponding grade values; generating a grade contour map according to geological sample data, and determining ore body positioning information according to geometric features of the grade contour map specifically comprises the following steps:

step S401, generating a grade contour map based on geological sampling point coordinates and corresponding grade values;

in some embodiments, computer graphics processing techniques, such as interpolation algorithms (e.g., kriging interpolation, inverse distance weighted interpolation, etc.), may be employed to automatically generate a grade contour map on the vertical longitudinal projection map based on the coordinate locations of the sampling points and the corresponding grade values;

Step S402, based on an image processing algorithm, identifying the long axis direction of the grade contour map according to the geometric characteristics of the grade contour map, and obtaining the spatial tendency of the ore body;

in some embodiments, an image processing algorithm may be utilized, for example, hough transform is used to reveal that the direction in which the grade concentration change is most significant is the long axis direction, and the long axis direction reflects the dominant direction of the concentration trend of the mineralized element, and generally indicates the extending direction of the ore body; the space tendency refers to the included angle between the extending direction of the ore body and the horizontal plane.

Step S403, determining the lateral direction of the ore body according to the vertical direction of the long axis direction;

the lateral direction is the vertical direction of the long axis direction of the grade contour line.

Step S404, a geological structure model is built according to geological structure information and a grade contour map of a target mining area;

The geological structure information comprises, but is not limited to, geological map, fault distribution, fold axial plane, rock stratum inclination angle, inclination and the like of the target mining area, and the information can be obtained through the approaches of field investigation, remote sensing data, drilling records and the like and is digitally input into a computer system.

In one embodiment of the present application, a Geographic Information System (GIS) and three-dimensional geologic modeling software may be used to combine such structural information with a grade contour map or the like to form a comprehensive geologic structural model in which structural features such as faults, folds, etc., typically channels or barriers for fluid migration, are particularly marked.

Step S405, based on the geological structure model, simulating a fluid movement path according to the lateral direction, and determining a fluid migration direction;

specifically, in the computer model, the lateral direction implies the lateral diffusion trend of the mineralized fluid, and is usually used as a lateral rule to restrict the lateral boundary condition of fluid migration, and by combining the fault trend and the fold axial plane in the geological structure model, the possible path of the fluid in the three-dimensional space can be simulated;

It can be understood that the fluid migration direction can be determined by simulating the fluid pressure and speed distribution of different flow directions and searching the fluid movement path which is most in line with the geological structure characteristics and the lateral voltage law.

And step S406, obtaining ore body positioning information according to the spatial tendency, the lateral direction and the fluid migration direction of the ore body.

In the embodiment, the depth fusion based on the geological principle and the modern computing technology aims at more accurately predicting the motion trail of the mineralized fluid, integrates geological structure information and a lateral volt rule, realizes the accurate prediction of the fluid migration direction, and provides scientific basis for the deep positioning of the ore body.

Referring to fig. 5, as an embodiment of step S107, the step of creating a three-dimensional native halo model from the three-dimensional spatial distribution characteristics of each native halo element based on the two-dimensional construction of the superimposed halo utility model includes:

Step S501, extracting construction control factors based on a two-dimensional construction superposition corona practical model;

Wherein, the established two-dimensional structure superposition corona practical model is utilized, and the model is based on the original corona theory, so that the rule of how mineralized elements are distributed under the control of the structure, in particular to the axial banding phenomenon, is revealed. By analyzing and actually overlaying the geological structure with the halo model, it can be identified which structure control factors (such as fracture, fold and fold, fault) have a direct effect on mineralization, and these structure control factors are indicative of mineralization activity and are also key to deep ore body prediction.

Step S502, a three-dimensional element distribution model is established according to the three-dimensional space distribution characteristics of each original corona element;

In some embodiments, the three-dimensional distribution map of the element can be obtained by analyzing the three-dimensional spatial distribution characteristics of the element, such as performing a three-dimensional empirical Bayesian-Kriging interpolation method by using software (such as ArcGIS Pro), wherein the three-dimensional distribution map is formed by processing the original data through a geostatistical method, and can be used for displaying a visual model of the three-dimensional spatial distribution characteristics of the element, namely a three-dimensional element distribution model. The distribution characteristics of each original corona element, such as the spatial positions of the front edge, near ore and tail corona elements, are reflected in the model through concentration zonation and the like, and visual representation is provided for the three-dimensional spatial distribution model of the mineralization characteristics.

Step S503, a three-dimensional geological framework is established according to the three-dimensional element distribution model;

Wherein, geological software can be utilized to build a basic three-dimensional geological framework model, and the framework comprises different rock strata, distribution of construction units and geometric forms of geological structures (such as faults and folds) so as to provide a basic platform for subsequent construction control factor integration.

And step S504, integrating construction control factors and adjusting element distribution in the three-dimensional geological framework to obtain a three-dimensional original vignetting model.

Specifically, in the constructed three-dimensional geological frame, the structural elements, such as faults, folds and the like, are integrated, and the fine simulation of the structural elements is realized by accurately calibrating the parameters of the positions, the trends, the dip angles and the like of the structures and the deformation conditions of the structures to surrounding rock stratum in the geological frame.

Then, after the three-dimensional geological framework integrates the structural elements, adjusting element distribution, considering the influence of the structure on axial zonation of the primary corona, and analyzing how the structural features influence migration and enrichment of elements, such as element enrichment zones caused by mineral fluid flowing along faults or element zonation in the axial direction of folds; and (3) adjusting the element distribution model to match with the structural characteristics, and finally obtaining the three-dimensional original corona model taking structural control into consideration.

In the embodiment, the three-dimensional original corona model can simulate and truly reflect mineralization distribution more comprehensively, comprises deep part, structural control and spatial variation of original corona, provides an accurate target area for prospecting prediction, reduces uncertainty of prospecting, and improves prospecting efficiency.

Referring to fig. 6, as an embodiment of step S108, the step of training the geochemical feature based on the machine learning algorithm, and creating the mineralization distribution prediction model includes:

step S601, preprocessing the geochemical characteristics;

the preprocessing step comprises data cleaning (abnormal values and missing values are removed), standardization (the feature sizes are consistent, model processing is facilitated), normalization (the numerical range is scaled to the same size, and model training is conducted due to the fact that certain feature factor value ranges are too large is avoided).

Step S602, performing feature selection on the preprocessed geochemical features based on correlation analysis, and performing standardization processing on the selected feature vectors to obtain an input feature set;

In some embodiments, based on correlation analysis (e.g., pearson correlation coefficients) and feature importance assessment, geochemical features that are highly correlated with mineralization probability are selected, and non-critical features are culled to reduce dimensionality. And then, carrying out standardization processing (such as minimum-maximum normalization) on the selected features, so that all feature values are mapped to the same interval, leading model learning caused by overlarge range of a certain feature factor value in the machine learning process is avoided, and fair learning of the algorithm on all the features is ensured.

Step S603, dividing the input feature set into a training set, a verification set and a test set;

The processed feature set is randomly divided into a training set, a verification set and a test set, and the proportion is usually 70%, 15% and 15%, so that model learning and evaluation are balanced. The training set is used for model learning, the verification set is used for adjusting model parameters, and the test set is used for independently verifying the generalization capability of the final model, so that the robustness of the model is ensured.

Step S604, inputting a training set into a pre-constructed machine learning algorithm model for training, optimizing model parameters and calculating a loss function until the loss function meets preset conditions or the number of model iterations reaches preset times, so as to obtain a trained mineralization distribution prediction model;

in some embodiments, the machine learning algorithm may employ a random forest, a support vector machine, a neural network, etc., to train the training set input model, and during the training process, model parameters (e.g., tree depth, node number, kernel function, hidden layer number, etc.) are continuously adjusted to optimize the loss function (e.g., mean square error, cross entropy loss). The mineralization distribution prediction model after training can be obtained by iteratively updating parameters through methods such as gradient descent, grid search, random search and the like until the loss function converges or reaches the preset iteration times.

Step S605, verifying the mineralization distribution prediction model based on the verification set, evaluating the performance of the mineralization distribution prediction model and adjusting the super parameters of the model;

The performance of the model can be primarily evaluated based on the verification set, such as accuracy, recall, F1 score, area under ROC curve, and the like, and the performance of the model on unseen data can be analyzed. And according to the evaluation result, fine-tuning the model hyper-parameters such as the depth of the tree, regularization factors and the like, further improving the performance of the model and ensuring the generalization capability of the model.

Step S606, testing the prediction capability of the adjusted mineralization distribution prediction model based on the test set.

In particular, the final model performance test is based on the test set, which is the most rigorous test for the generalization ability of the model, since the test set is not seen by the model at all throughout the process. The evaluation result directly reflects the prediction effect of the model on the real unknown data, verifies the practicability of the model and provides a reliable decision basis for mineralization prediction.

In the embodiment, the accuracy and generalization capability of the mineralization distribution prediction model are ensured through strict feature extraction and model training optimization and comprehensive verification test flow; it can be understood that the three-dimensional prospecting prediction based on machine learning can quantitatively evaluate the potential of deep resources, can efficiently and scientifically mine the interrelationship among data, and provides a scientific and effective technical means for mineral resource exploration.

Referring to fig. 7, as an embodiment of step S109, the step of fusing the two-dimensional structure superimposition corona utility model, the three-dimensional original corona model, and the mineralization distribution prediction model to obtain the three-dimensional comprehensive prospecting prediction model includes:

step S701, superposing a corona practical model, a three-dimensional original corona model and a mineralization distribution prediction model according to a two-dimensional structure, respectively obtaining output characteristics of each model and carrying out standardization treatment;

Wherein, respective output characteristics are extracted from the two-dimensional structure superposition halation practical model, the three-dimensional original halation model and the mineralization distribution prediction model, and the output characteristics may comprise prediction data related to various prospecting such as geology, geochemistry, geophysics and the like. In order to eliminate the influence caused by the difference of dimensions or scales among the features, all the features are standardized, and Z-score standardization or Min-Max scaling is generally adopted to ensure that the average value of each feature is 0, and the standard deviation is 1 or the range is 0 to 1; the method ensures fair comparison of the features in the model fusion process, and avoids excessive leading fusion process of some features with larger natural scale.

Step S702, performing feature selection and dimension reduction processing based on the standardized processed output features to obtain feature vectors after dimension reduction;

In some embodiments, feature selection may eliminate irrelevant or redundant features by correlation analysis, mutual information, recursive feature elimination, etc.; the feature space is further compressed by dimension reduction processing such as Principal Component Analysis (PCA), t-SNE or a self-encoder, and the converted feature vector keeps the maximum variance information of the original data, so that the key mode is captured by the model.

Step S703, respectively distributing dynamic weight values for the two-dimensional structure superposition corona practical model, the three-dimensional original corona model and the mineralization distribution prediction model;

Wherein, in order to ensure that the contribution of each basic model accords with the prediction precision thereof, dynamic weight values are introduced, the weight values are calculated based on the prediction errors of the models on a cross verification set, and the weight obtained by the model with smaller errors is larger. The weights can be automatically adjusted by a gradient descent method, and the optimization configuration of the model performance can be realized by iteratively updating the weights until convergence with the aim of minimizing the overall prediction error.

Step S704, carrying out weighted average on the feature vectors subjected to dimension reduction according to the weight values to obtain a preliminary fusion prediction result;

specifically, the feature vector subjected to the dimension reduction processing is multiplied by the dynamic weight value of each model, and then weighted average is carried out to obtain a primary fusion prediction result.

Step S705, combining the output characteristics of each model with the preliminarily fused prediction results to obtain a data set and dividing the data set into a training set, a verification set and a test set;

The original output characteristics of the models are combined with the preliminarily fused prediction results to form a new data set, and the data set contains rich information, so that not only are the direct output of the single models and the prediction results after comprehensive consideration of the single models, but also more comprehensive input is provided for deep learning.

Step S706, performing primary training on a pre-built deep neural network model based on a training set, and performing cross verification and model parameter optimization adjustment on the primary trained deep neural network model based on a verification set and a test set to obtain a three-dimensional comprehensive prospecting prediction model; the adjustment of the dynamic weight value is based on the prediction error of the model in the cross verification process, and the weight configuration is optimized by minimizing the prediction error through a gradient descent method.

As one embodiment of calculating the prediction error of a model over a cross-validation set, a specific calculation process includes: the entire dataset was divided into k subsets (in k-fold cross-validation), where k-1 subsets were used to train the model and the remaining 1 subset was used to validate the model, and this process was repeated k times. For each iteration, a training set is used to train the model, and then predictions are made on the corresponding verification set, where the model produces a series of predicted results that need to be compared with the actual labels on the verification set to calculate an error index. After finishing k iterations, averaging error indexes on each verification set to obtain an overall prediction error of the model on the cross verification set; for example, using the mean square error (Mean Squared Error, MSE) as the error metric, the overall MSE is the average of all k verified MSEs. Based on the obtained overall prediction error of the model on the cross validation set, the weight of the model can be further calculated or adjusted.

For example, the weights can be reversely allocated according to the magnitude of the error, that is, the model with smaller error is allocated with larger weight, and in the specific calculation, the error can be normalized first to ensure that the sum of the weights is 1, and then the calculation is performed by the following formula:

；

in the above-mentioned method, the step of, Is the weight of the i-th model,Is the average error in cross-validation of the ith model, j is the index through all models, i.e., from 1 to k, k is the number of models,For calculating the sum of the inverse of all model errors.

Specifically, the deep neural network can effectively capture complex nonlinear relations in data due to the multilayer structure and a large number of learnable parameters; based on initial training, cross verification is carried out through a verification set, the fitting condition is monitored, and super parameters such as a network structure, a learning rate and the like are adjusted according to a verification result, so that fine adjustment of the model is realized. And finally, carrying out final evaluation on the optimized model by using a test set, and ensuring the generalization capability and practicability of the model.

In the embodiment, the machine learning algorithm and the original halation prospecting theory are combined, the two-dimensional and three-dimensional prospecting prediction results are synthesized, and the weighted average and depth neural network fusion algorithm is combined, so that not only is the advantage complementation between models realized, but also more potential nonlinear relations are excavated through the depth learning technology, a three-dimensional comprehensive prospecting prediction model is constructed, and the accuracy and reliability of the three-dimensional comprehensive prospecting prediction are improved.

Referring to fig. 8, as an embodiment of step S110, predicting a target area for prospecting according to a three-dimensional comprehensive prospecting prediction model, the step of obtaining a prospecting prediction result includes:

step S801, determining a mining target area according to a model prediction result of a three-dimensional comprehensive mining prediction model;

The method comprises the steps of constructing a three-dimensional comprehensive prospecting prediction model (combining a two-dimensional structure superposition halation model, a three-dimensional original halation model and machine learning prediction), predicting mineralization probability distribution through the model, and identifying areas with high mineralization probability, wherein the areas with high mineralization probability are identified as potential prospecting target areas and are key areas for the next geological investigation and exploration.

Step S802, receiving field verification data of a mining target area input by a user;

It will be appreciated that in order to verify the accuracy of the model predictions, the theoretical predictions must be compared with actual geological survey data, which may include borehole sample analysis, tank search results, geophysical measurements, etc., as input by the user in the field, which directly reflect the actual mineralization.

Step S803, judging whether the prediction result accords with the model prediction result according to the field verification data, if so, jumping to step S804; if not, jumping to step S805;

Step S804, determining a target area for prospecting as a prospecting prediction result;

The mineralization probability predicted by the model is compared with mineralization indexes (such as grade, mineral content and the like) of the field verification data, and if the mineralization strength and the mineralization distribution predicted by the model are matched with the actual values, the model prediction is accurate, so that the model can be confirmed to be a target area for prospecting.

Step S805, optimizing and correcting the three-dimensional comprehensive prospecting prediction model according to the field verification data, and adjusting model parameters until the model prediction result accords with the fed-back field verification data, so as to obtain an optimized three-dimensional comprehensive prospecting prediction model;

If the model prediction does not accord with the field data, the model deviation is indicated, and adjustment is needed. Optimizing model parameters (such as learning rate, regularization parameters, depth, tree depth, dynamic weight values of each model and the like) based on the feedback data; and through iterative training and verification, fine tuning is continuously performed until the model prediction is matched with the field data.

And step S806, predicting a prospecting target area according to the optimized three-dimensional comprehensive prospecting prediction model, and determining the prospecting target area as a prospecting prediction result.

It can be understood that the three-dimensional comprehensive ore-finding prediction model after optimization is used for predicting again, and because the model is corrected according to actual data, the prediction result is closer to actual, and the newly predicted ore-finding target area is based on the optimization model, has higher credibility and can be directly used as the basis of ore-finding decision.

In the embodiment, the preliminary prediction of the prospecting target area is realized based on the constructed three-dimensional comprehensive prospecting prediction model; the model can be dynamically optimized by combining an on-site verification and feedback mechanism, so that the prediction result is ensured to be highly consistent with the actual geological condition, and the accuracy and efficiency of prospecting are greatly improved. Through continuous optimization iteration, the model gradually approaches the actual situation of actual mineralization distribution, blind exploration is reduced, resources are saved, mineral exploration process is accelerated, and the method has important significance for resource development and national mineral safety.

Referring to fig. 9, a schematic diagram of a two-dimensional structure overlapping halo practical model is established in the practical research process of the application, the practical research is based on Longmenshan Xinjia zunjin mine, the mine is determined to be sidewise-v in the north-west direction through the known mine sidewise-v law, the sidewise migration direction of fluid in the south-west-north-east direction is determined, the favorable mineral formation space exists in a deep position is presumed, finally the mine region structure overlapping halo practical model shown in fig. 9 is established, the indication features such as overlapping of the front edge halo and the tail halo, abnormal axial zoning, change of localization parameters and the like exist at the depth of 600 meters are known, and the blind mine body exists in the depth of the mine region is predicted, so that the favorable mineral formation space is determined as a prediction target position, and a scientific basis is provided for the subsequent mine region engineering deployment by using the favorable mineral formation space as an output prediction result of the two-dimensional structure overlapping halo practical model.

Referring to fig. 10, a three-dimensional original corona model is created by spatial distribution feature analysis of each element and corona body on the basis of construction of superimposed corona, the actual study is based on Longmenshan Xinjiajian gold mine, and according to the figure, the original corona is in an irregular ellipsoidal form with lower inclination in the southeast direction and the north west direction, and has obvious three-layer zoning, the top part is a tail corona, then a front edge corona and a near-mine corona, and the deep ore body is indicated to be extended; combining with geological background of the mining area, the leaching structure in the southwest-northwest direction is an important fracture system in the area, controls the distribution of ore bodies in the area, and provides important ore guide channels and ore containing spaces for forming and positioning the ore bodies in the secondary structure; therefore, a larger prospecting space is presumed to exist in the deep part in the southwest direction, which also shows that the three-dimensional primary corona prospecting prediction method established based on the two-dimensional structure superposition corona is feasible, and the innovation of the three-dimensional primary corona prospecting method is realized.

Referring to fig. 11, in the process of researching a machine learning algorithm, obtaining a deep prediction value of the Xinjia chewy gold ore by a random forest algorithm, and obtaining the probability distribution condition of the deep gold ore in the mining area shown in fig. 11 by utilizing three-dimensional spatial interpolation; it can be seen that the high probability regions are mainly distributed in the ZK1101, the ZK002, the ZK003, the ZK005 and the bottom of the ZK802, and the ZK1102 and the ZK1103 are mineralized to some extent but are already at the edge of the ore body, so that the high probability regions are judged to be low probability regions and are consistent with the actual situation; there is a high anomaly in the north-west direction, indicating a great potential for prospecting there, consistent with the construction of superimposed and three-dimensional raw vignetting predictions.

Referring to fig. 12, in the process of researching a machine learning algorithm, a probability distribution diagram of the sinuses obtained by a support vector machine algorithm is that the raw halation contains ores; it can be seen that the prediction high probability area is mainly in a strip distribution inclined from the southeast to the northwest, and is concentrated in the middle of ZK1101, the middle of ZK002 and the middle of ZK003, the middle of ZK005 and the bottom of ZK802, and the ZK1601 and the ZK1103 on the two sides are low probability areas, so that the prediction high probability area is more consistent with the actual ore-seeing situation.

Referring to fig. 13, which is a ROC graph of a random forest and a support vector machine, the steeper the ROC graph of a random forest algorithm model, the closer to the upper left corner, the faster the ROC graph approaches an ideal value 1, the AUC value of the random forest is 0.99, and is larger than the AUC value of the support vector machine, which indicates that the higher the accuracy of the random forest algorithm prediction is; therefore, although the random forest and the support vector machine have better accuracy, the accuracy of the support vector machine is far lower than that of a random forest algorithm, namely the random forest algorithm is more applicable to the Xinjia chewy gold mine in the research of the application.

Referring to fig. 14, a three-dimensional comprehensive prospecting prediction model constructed in the actual research process of the application is shown in the schematic diagram, in late three-fold generation, along with marginal ocean closure, the north China plate collides with the Yangzi plate to form a series of thrust-right walking slip faults from north west to south east, and ductile and brittle deformations such as penetrating facial theory, asymmetric folds, stretching line theory and the like, and simultaneously magma hydrothermal events occur. In the process, deep fluid moves laterally from north to south to north along a construction channel, mineral substances and mineral elements carried by the mineral substances are extracted from surrounding rock, precipitation occurs in a favorable construction space, and a mineral body is finally formed through multi-period superposition. Thus, as can be seen from the illustration, on the structural plane along the north east-south west direction, the deep part towards the south west is the favorable position for searching the blind ore body.

In summary, the analysis of the ore formation prediction elements is carried out in the actual research process, and the multi-dimensional and multi-method ore finding prediction is carried out by utilizing the construction superposition halation, the three-dimensional primary halation and the quantitative evaluation of the three-dimensional mineral resources based on machine learning, and the soil secondary halation, the convolutional neural network and other multi-method ore finding methods, so that the fusion of the primary halation ore finding method, the machine learning algorithm and other multi-method ore finding methods is realized by combining regional ore formation rule recognition. Meanwhile, engineering verification is carried out on the ground surface, a better prospecting clue is found, and the method for predicting and evaluating three-dimensional comprehensive prospecting based on the primary corona theory is further effective in the actual research process, and is particularly good in practicability and market prospect.

The embodiment of the application also discloses a three-dimensional prospecting prediction system.

The primary corona element combination determining module is used for carrying out element combination analysis on the abnormal geological sample and determining a primary corona element combination according to the concentration banding characteristics, the axial banding sequences and the element correlation;

the two-dimensional model construction module is used for constructing a two-dimensional construction superposition corona practical model based on element abnormal threshold values, concentration banding characteristics, axial banding sequences, original corona element combination and ore body positioning information;

The prediction model construction module is used for extracting geochemical characteristics according to the three-dimensional original corona model, training the geochemical characteristics based on a machine learning algorithm and establishing a mineralization distribution prediction model;

The three-dimensional prospecting prediction system provided by the embodiment of the application can realize any one of the three-dimensional prospecting prediction methods, and the specific working process of each module in the three-dimensional prospecting prediction system can refer to the corresponding process in the method embodiment.

In several embodiments provided by the present application, it should be understood that the methods and systems provided may be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, a division of a module is merely a logical function division, and there may be another division manner in actual implementation, for example, multiple modules may be combined or may be integrated into another system, or some features may be omitted or not performed.

The embodiment of the application also discloses computer equipment.

The computer equipment comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, and the processor realizes the three-dimensional prospecting prediction method when executing the computer program.

The embodiment of the application also discloses a computer readable storage medium.

A computer readable storage medium storing a computer program loadable by a processor and performing any of the three-dimensional prospecting methods as described above.

Wherein a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device; program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

In the foregoing embodiments, the descriptions of the embodiments are focused on, and for those portions of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.

The foregoing description of the preferred embodiments of the application is not intended to limit the scope of the application in any way, including the abstract and drawings, in which case any feature disclosed in this specification (including abstract and drawings) may be replaced by alternative features serving the same, equivalent purpose, unless expressly stated otherwise. That is, each feature is one example only of a generic series of equivalent or similar features, unless expressly stated otherwise.

Claims

1. The three-dimensional ore finding prediction method is characterized by comprising the following steps of:

2. The method of claim 1, wherein the step of identifying anomalies in the preprocessed geologic sample data, extracting element anomaly thresholds, and determining anomalous geologic samples comprises:

3. The method of claim 2, wherein the step of performing elemental combination analysis on the abnormal geological sample to determine a combination of primary halation elements based on concentration banding characteristics, axial banding sequences, and elemental correlation comprises:

4. The three-dimensional prospecting prediction method according to claim 1, wherein the step of building a three-dimensional primary corona model from the three-dimensional spatial distribution characteristics of each primary corona element based on the two-dimensional construction superposition corona utility model comprises:

5. The method of claim 1, wherein the step of training the geochemical feature based on a machine learning algorithm to create a mineralization distribution prediction model comprises:

preprocessing the geochemical features;

6. The three-dimensional prospecting prediction method according to claim 1, wherein the step of fusing the two-dimensional structured superposition corona utility model, the three-dimensional original corona model and the mineralization distribution prediction model to obtain the three-dimensional comprehensive prospecting prediction model comprises the steps of:

7. The method for predicting a three-dimensional prospecting according to any one of claims 1 to 6, wherein the step of predicting a prospecting target area according to the three-dimensional comprehensive prospecting prediction model to obtain a prospecting prediction result comprises:

8. A three-dimensional prospecting prediction system, wherein the prediction system comprises:

9. A computer device, characterized by: comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method according to any of claims 1 to 7 when the program is executed.

10. A computer-readable storage medium, characterized by: a computer program being stored which can be loaded by a processor and which performs the method according to any one of claims 1 to 7.