A mineralization prediction method based on rare data synthesized from mineralization spatial structure

Abstract

This invention relates to a mineralization prediction method based on rare data synthesized from mineralization spatial structure. The method involves collecting and preprocessing geological data based on prediction requirements; constructing a mineralization prediction model based on the geological background of the study area, determining prediction variables, and creating prediction layers; performing mineralization spatial structure analysis, and synthesizing a preset number of rare data prediction variables based on the prediction variables of the study area to generate several training datasets; training the constructed mineralization prediction model using these training datasets to predict the probability of mineralization in the study area; evaluating the performance of the mineralization prediction model, and delineating prospecting target areas based on the prediction results of the best-performing model. By employing the mineralization prediction method disclosed in this invention, kernel density analysis of the spatial distribution of known ore deposits can synthesize a certain number of positive samples while taking into account geological context, which is beneficial for training the mineralization prediction model and improves the accuracy of prospecting prediction.

Description

Technical Field

[0001] This invention belongs to the field of geology and mineral resources, specifically relating to a mineralization prediction method based on rare data synthesized from mineralization spatial structure. Background Technology

[0002] Because of their ability to handle geological variables with complex nonlinear relationships and to extract deeper mineral exploration information, machine learning / deep learning algorithms are widely used in mineralization prediction. Among them, supervised machine learning / deep learning algorithms, guided by similarity analogy theory, predict undiscovered mineral deposits by mining the correlation between geological data and the locations of discovered mineral deposits.

[0003] In mineralization prediction, the study area is typically divided into a series of grids of equal size. Ore-bearing grid cells are often selected as positive samples, while ore-free cells are usually selected as negative samples, thus mineralization prediction can be viewed as a binary classification problem. However, since mineralization is a rare event, the number of known mineral deposits (i.e., ore-bearing cells) within the study area is usually small. Therefore, when using supervised machine learning / deep learning for mineralization prediction, there is a challenge of insufficient training samples (especially positive samples available for training), leading to insufficient model learning ability, poor generalization ability, and consequently affecting prediction accuracy.

[0004] Currently, in supervised machine learning / deep learning, "synthesizing" a certain number of positive samples is one of the main ways to compensate for insufficient sample quantity. Among techniques for synthesizing positive samples, SMOTE (Synthetic Minority Over-sampling Technique), image affine transformations (rotation, scaling, translation), image processing (such as contrast transformation), and adding random noise are commonly used. However, mineralization is a singular process in which a large amount of mineral-forming materials are massively enriched in a relatively small space; mineralization itself exhibits spatial heterogeneity. At the same time, since the formation of ore deposits is often controlled by specific geological factors (such as strata and structures), mineralization often has a certain directionality. The spatial heterogeneity and directionality of mineralization indicate that it has a specific spatial structure. Although the SMOTE method can synthesize a certain number of positive samples in the feature space, it only considers the properties of the ore-bearing unit itself, ignoring the spatial relationship and property characteristics between the ore-bearing unit and its neighboring units, i.e., it does not consider the spatial structure of mineralization. Image radiometric transformation and image processing methods, which use images (i.e., combinations of ore-bearing units and their neighboring units) as training samples, consider the spatial structure of mineralization. However, these transformations may alter the geological context of the original data, thus affecting prediction results. Meanwhile, adding noise to training images, while capable of synthesizing a large number of positive samples and taking into account the spatial structure of mineralization, also increases uncertainty due to the added noise. Therefore, how to synthesize rare positive samples while respecting the geological context of mineralization structure has become a crucial technical problem that urgently needs to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a mineralization prediction method based on rare data synthesized from mineralization spatial structure. This method can synthesize a predetermined number of positive samples while taking into account geological context, thereby improving the generalization ability and prediction accuracy of the mineralization prediction model, thus enhancing the accuracy of mineralization prediction and reducing exploration risks and costs.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a mineralization prediction method based on rare data synthesized from mineralization spatial structure, comprising the following steps:

[0007] S1. Geological data collection and preprocessing based on prediction needs;

[0008] S2. Based on the geological background of the area to be studied, construct a mineralization prediction model, determine the prediction variables, and create a prediction layer;

[0009] S3. Conduct mineralization spatial structure analysis, combine the predictor variables of the area to be studied to synthesize a preset number of rare data predictor variables, and then generate several training datasets.

[0010] S4. Several training datasets are used to train the constructed mineralization prediction model. Based on the created prediction layer, the mineralization probability of the study area is predicted. The performance of the mineralization prediction model is evaluated. Based on the prediction results of the better-performing mineralization prediction model, the mineral exploration target area is delineated.

[0011] Furthermore, the prediction requirements mentioned in step S1 include the mineralization prediction accuracy, the predicted mineral type, and the required deposit type.

[0012] Furthermore, the geological data collected in step S1 includes geological maps of the area to be studied, mineral deposit distribution maps, geochemical data, geophysical data, etc.

[0013] Furthermore, step S2 includes analyzing the geological background of the area under study, summarizing the metallogenic geological conditions, conducting research on typical ore deposits, combining previous analyses of the metallogenic regularity of the area under study, identifying the key geological elements controlling the formation of ore deposits and important prospecting indicators, thereby constructing a metallogenic prediction model.

[0014] Furthermore, in step S2, based on the preprocessed geological data, predictive elements that can indicate key geological features and important mineral exploration indicators are used as predictive variables.

[0015] Furthermore, the predicted variables in step S2 include Carboniferous-Permian strata, NE and NW trending structures, Yanshanian intrusive bodies, aeromagnetic anomalies, and geochemical anomalies related to the predicted elements. Step S2 includes dividing the study area into a series of grid cells of the same size according to the prediction accuracy; performing distance analysis on the ore-controlling strata, ore-controlling structures, and ore-controlling intrusive bodies in the study area to obtain corresponding predicted variable layers; performing factor analysis on the geochemical data to extract factor scores of geochemical element combinations related to mineralization and performing local singularity analysis to obtain predicted variable layers of geochemical anomalies related to mineralization; performing local singularity analysis on the aeromagnetic data to obtain predicted variable layers indicating aeromagnetic anomalies; and storing the information of each layer in the grid cells.

[0016] Furthermore, step S3 includes the following sub-steps:

[0017] S31. Determine the number of synthetic training samples and the corresponding analysis window size as needed;

[0018] S32. Perform kernel density analysis on the first known ore deposit in the study area to obtain the kernel density of the first known ore deposit;

[0019] S33. Taking the location of the first known ore deposit as the center, the original first predicted variable value of the predicted variable layer in the analysis window of the preset size is weighted and summed with the kernel density of the first known ore deposit after normalization. The weighted sum is used as the final first predicted variable at the location of the first known ore deposit, that is, the first predicted variable of the synthetic rare data. The first predicted variable is replaced with the Nth predicted variable, and the weighted sum is performed in the same way until all the predicted variables of the synthetic rare data are obtained.

[0020] S34. Replace the first known mineral deposit in steps S32-S33 with the Nth known mineral deposit, repeat steps S32-S33 to calculate the predictor variables of the rare data at the locations of other known mineral deposits, and combine the predicted variables of the rare data of each known mineral deposit to form a positive sample of the expected multiple.

[0021] S35. Randomly select the same number of units as the positive samples in areas with poor mineralization conditions, and use them as negative samples. Based on the combination of the generated positive and negative samples, obtain a training dataset with a preset sample size.

[0022] Furthermore, in step S33, the predictor variables for rare data are synthesized according to the following formula.

[0023]

[0024] Where i is the analysis window size, i is an odd number, and Y l u For the l-th predictor variable Y at position u in the analysis window l The value of K u To analyze the kernel density K value of the ore deposit at position u in the analysis window.

[0025] Furthermore, in step S35, negative samples are randomly selected a predetermined number of times from the same number of negative samples as the positive samples, and the selected negative samples are combined with positive samples respectively to obtain different training datasets.

[0026] Furthermore, in step S4, the average value of the prediction results of the better-performing mineralization prediction model is calculated, which is the mineralization probability map of the target mineral in the area to be studied. The mineralization probability is classified by the natural discontinuity method, and the high-value areas of mineralization probability are delineated as mineral exploration target areas.

[0027] The advantages of this invention are as follows: Compared with the positive sample synthesis methods in the prior art, the mineralization prediction method based on mineralization spatial structure and rare data disclosed in this invention can synthesize a certain number of positive samples while taking into account the mineralization spatial structure by performing kernel density analysis on the spatial distribution of known ore deposits. This improves the prediction accuracy and generalization ability of the mineralization prediction model, thereby reducing exploration risks. It provides a new method and approach for mineralization prediction based on rare data. Attached Figure Description

[0028] Figure 1 This is the regional geological map of the Taqian-Zhuxi area as described in Embodiment 1 of the present invention;

[0029] Figure 2 This is a schematic diagram of kernel density estimation shown in the mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in Embodiment 1 of the present invention.

[0030] Figure 3 The image shown is a neighborhood map (3×3) of the ore deposit kernel density analysis based on pixel units, as described in the mineralization spatial structure synthesis rare data method of Embodiment 1 of the present invention.

[0031] Figure 4 This is a typical ore deposit kernel density analysis neighborhood diagram (3×3) shown in the mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in Embodiment 1 of the present invention.

[0032] Figure 5 This is a general schematic diagram of the synthetic rare data prediction variables shown in the mineralization prediction method based on mineralization spatial structure synthetic rare data according to Embodiment 1 of the present invention.

[0033] Figure 6 This is a schematic diagram illustrating the process of synthesizing rare data prediction variables in a mineralization prediction method based on mineralization spatial structure as described in Embodiment 1 of the present invention.

[0034] Figure 7 The AUC values ​​of the prediction results of each model after training the mineralization prediction model with different training datasets are shown in the mineralization prediction method based on mineralization spatial structure synthetic rare data described in Embodiment 1 of the present invention.

[0035] Figure 8 This is a mineralization probability map of W-Cu polymetallic deposits in the Taqian-Zhuxi area generated using a mineralization prediction method based on rare data synthesized from mineralization spatial structure, as described in Embodiment 1 of this invention. Detailed Implementation

[0036] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0037] Example 1

[0038] In this embodiment, the W-Cu polymetallic deposit in the Taqian-Zhuxi area of ​​Jiangxi Province is used as an example to illustrate the mineralization prediction method based on rare data synthesized from mineralization spatial structure disclosed in this invention.

[0039] The Taqian-Zhuxi area is tectonically located in the Jiangxi section of the Jiangnan Orogenic Belt, at the junction of the Yangtze and Cathaysia plates, on the northwest side of the deep fault in northeastern Jiangxi. It is one of the most important Cu-Au-Pb-Zn-Ag polymetallic mineral clusters in South China. The area has relatively complete strata exposure, exhibiting a typical binary structure: a basement composed of the Neoproterozoic Wannian Group and Shuangqiaoshan Group, and a sedimentary cover composed of Devonian, Carboniferous, Permian, Triassic, Jurassic, Cretaceous, Paleogene, and Quaternary strata. Influenced by NE-trending regional thrust tectonics, the Neoproterozoic Wannian Group superimposed on the Upper Triassic strata. The area's structures are mainly NE, NW, and near-EW trending. The concealed intrusive bodies revealed by boreholes are primarily granite and granite porphyry, followed by granodiorite, diorite porphyry, and lamprophyre. Within the region, several super-large W-Cu deposits have been discovered, including the Taqian porphyry tungsten-molybdenum deposit and the crescent-shaped vein-like Cu-Pb-Zn deposit, indicating exceptionally favorable mineralization geological conditions. Therefore, systematically conducting mineralization prediction and utilizing machine learning algorithms to mine deeper mineral exploration information is of great significance for mineral exploration in this area.

[0040] This embodiment provides a mineralization prediction method based on rare data synthesized from mineralization spatial structure. The method includes the following steps:

[0041] S1. Geological data collection and preprocessing based on prediction needs.

[0042] Based on the accuracy requirements of mineralization prediction and the needs of actual predicted mineral types and deposit types, geological data from the geological, geochemical, geophysical, and remote sensing aspects of the area under study are systematically collected. The scale of the collected geological data can be selected as 1:50,000, 1:200,000, etc. Geological data preprocessing includes unifying the data coordinate system, organizing and cleaning the data, removing missing values, and replacing data below the detection limit.

[0043] like Figure 1 As shown, in this embodiment, 1:50,000 regional geological maps (including strata, structures, and igneous rocks), mineral deposit distribution maps, 1:50,000 geochemical data of stream sediments (including 17 elements such as Au, Ag, Cu, Pb, Zn, W, Sn, Mo, Bi, Cd, Hg, Cr, Co, Ni, Sb, As, and Ba) and 1:50,000 aeromagnetic data were collected for the Taqian-Zhuxi area.

[0044] S2. Based on the geological background of the area to be studied, construct a mineralization prediction model and predictive variables, and create a predictive variable layer.

[0045] The geological background of the area under study is analyzed, the metallogenic geological conditions are systematically summarized, typical mineral deposits are studied, and the key geological elements controlling the formation of mineral deposits and important prospecting indicators are identified in combination with previous analyses of the metallogenic regularity of the area, thereby constructing a metallogenic prediction model.

[0046] Based on the prediction accuracy, the study area is divided into a series of grid cells of equal size. On this basis, according to the mineralization prediction model, predictive variables are constructed using existing geological data. Specifically, based on preprocessed geological data, predictive elements that can indicate key geological features and important mineral exploration indicators are used as predictive variables.

[0047] Predictor variable layers are the representation of predictor variables, presented in the form of layers. When performing model calculations, the variables need to be converted into variable layers that can be used for GIS calculations.

[0048] Distance analysis was used to analyze the ore-controlling strata, ore-controlling structures, and ore-controlling intrusive bodies in the study area, generating corresponding predictive variable layers. Factor analysis was performed on geochemical data to extract factor scores of geochemical element combinations related to mineralization, and local singularity analysis was conducted to obtain predictive variable layers for geochemical anomalies related to mineralization. Local singularity analysis was also performed on aeromagnetic data to obtain predictive variable layers indicating aeromagnetic anomalies. The information from each layer was stored in grid cells.

[0049] Following the previous example, based on previous analyses of the metallogenic geological conditions and metallogenic regularities in the Taqian-Zhuxi area, the main ore-bearing strata are Carboniferous-Permian carbonate rocks. These strata contain high levels of ore-forming elements such as W, Cu, Pb, and Zn, providing an important material basis for mineralization. The main structural feature in the area is the NE-trending thrust-nappe structure. This thrust-nappe structure allows for inter-layer thrusting and detachment of the ore-bearing strata, forming structurally weak zones, which are important spaces for magma emplacement and ore body hosting. The rocks closely related to mineralization in this area are mainly peraluminous granites, with mineralization ages concentrated around 150 Ma. S and Pb isotope analysis indicates that the ore-forming materials mainly originate from crustal remelted granites and surrounding rocks. Based on exploration experience in this area, the magnetic rocks are mainly various types of skarn. Positive and negative facies magnetic anomalies distributed within or at the edges of Late Paleozoic carbonaceous tectonic rocks are favorable locations for skarn-type mineral exploration and are important indicators of mineralization. Meanwhile, surface anomalies of W, Cu, Pb, and Zn provide good indications of concealed ore bodies and are important geochemical indicators of mineralization. As shown in Table 1, based on these findings, a preliminary metallogenic prediction model and predictive variables related to W-Cu polymetallic mineralization were constructed for this region.

[0050] Table 1. Metallogenic prediction models related to W-Cu polymetallic mineralization in the Taqian-Zhuxi area.

[0051]

[0052] As shown in Table 1, the mineralization prediction model has seven predictive variables, namely the Y-type strata of the Carboniferous-Permian system. 1 , NE, NW direction construction Y 2 Yanshanian intrusive body Y 3 Aeromagnetic anomaly Y 4 Geochemical anomalies related to W, Y 5 Geochemical anomalies Y associated with Pb-Zn 6 and the geochemical anomaly Y associated with Cu 7 .

[0053] To meet the required prediction accuracy, the study area was divided into a series of identically sized grid cells, serving as the baseline for the range and grid size of each prediction layer. Based on the aforementioned mineralization prediction model, Carboniferous-Permian acid rock strata were selected from the 1:50,000 geological map. Using inverse distance ratio, a prediction variable, Carboniferous-Permian Y... 1 Distance prediction layer. The NE and NW-oriented tectonic Y-axis prediction variables were selected from the 1:50,000 geological map. 2 Using the inverse distance ratio, a predictive layer of distance to the fault was constructed. Similarly, the predictive variable Y, an intermediate-acidic intrusive rock mass from the Yanshanian period, was selected from the 1:50,000 geological map. 3By utilizing the inverse distance ratio, a predictive layer for distances to Yanshanian intermediate-acidic intrusive intrusive bodies was constructed. Local singularity analysis was performed on the 1:50,000 scale aeromagnetic poles to obtain the predictive variable, aeromagnetic anomaly Y. 4 The layer in question. Factor analysis was performed on 17 geochemical elements to screen for factors related to W-Cu polymetallic mineralization: F1 (W-Mo-As), F2 (Pb-Zn-Cd), and F3 (Cu-Co-Ba), representing the geochemical element combinations associated with W, Pb-Zn, and Cu mineralization, respectively. Local singularity analysis was performed on F1, F2, and F3 to obtain the geochemical anomaly Y related to W as a predictor variable. 5 Geochemical anomalies Y associated with Pb-Zn 6 and the geochemical anomaly Y associated with Cu 7 The prediction layer in which it is located.

[0054] S3. Perform mineralization spatial structure analysis, combine the predictor variables of the area to be studied to synthesize a preset number of rare data predictor variables, and then generate several training datasets.

[0055] Since mineral deposits are formed randomly through mineralization, mineralization is often the result of multi-factor coupling and critical transitions, thus exhibiting chaotic effects. Their spatial distribution displays both randomness and statistical regularity. Therefore, the spatial distribution pattern of mineral deposits can be studied as a realization of a random process using spatial point pattern analysis. Kernel density analysis is an effective method for analyzing the spatial heterogeneity of spatial point patterns. For example... Figure 2 As shown, kernel density analysis estimates the local intensity of a process at each location in space given a bandwidth (i.e., search radius).

[0056] Commonly used kernel functions include the Gaussian kernel function and the bivariate quadratic kernel function. Taking the bivariate quadratic kernel function as an example, the kernel function can be expressed as:

[0057]

[0058] In the formula, K(u) is the kernel density estimate at u, ||ux i ||For u to x i The distance at point h is the bandwidth (or search radius). The choice of bandwidth h has a significant impact on the estimation of kernel density. When the bandwidth h is large, the kernel function is smoother, and it better reflects the changes in the region while ignoring the variability at the local scale. When the bandwidth h is small, the kernel function can better reflect the variability at the local scale. The Silverman rule is often used to determine the bandwidth. According to this rule, the optimal bandwidth can be expressed by the following formula:

[0059]

[0060] In the formula, IQR is the interquartile range, σ is the standard deviation of the analyzed sample, and N is the total number of spatial point processes.

[0061] Estimating the kernel density of the spatial distribution of known mineral deposits can, to some extent, reflect the spatial structure of mineralization (such as its spatial heterogeneity and directionality). On the one hand, areas closer to the known deposit have higher kernel density and stronger mineralization. On the other hand, since the distribution of mineral deposits is usually controlled by specific geological factors, its spatial distribution exhibits a certain directionality. The kernel density of a mineral deposit also reflects, to some extent, the directionality of its distribution and the spatial directionality of mineralization. Figure 3 As shown, mineralization prediction based on pixel units can be viewed as a special case where the kernel density of the ore-bearing unit is 1, while the kernel density of other adjacent units is 0. This is a special case where the bandwidth infinitely approaches 0. The kernel density in the general case is as follows: Figure 4 As shown.

[0062] Based on the required number of positive samples, determine the size of the analysis window needed for synthesizing the positive samples, centered on the ore-bearing unit. For example, if one-time positive samples are needed, a 3×3 analysis window is selected; if two-time positive samples are needed, 3×3 and 5×5 analysis windows are selected respectively; if three-time positive samples are needed, 3×3, 5×5, and 7×7 analysis windows are selected respectively. That is, when n times the number of positive samples is needed, the number of analysis windows required for synthesizing the positive samples is also n, and the required analysis window sizes are 3×3, 5×5, ..., (2m+1)×(2m+1), ..., (2n+1)×(2n+1) (where m is a natural number from 1 to n).

[0063] Assuming the analysis window size is i (where i is an odd number), then there are i predictor variable layers within the window. 2 Each pixel unit. Y l u For the l-th predictor variable Y at position u in the analysis window l The value of K u To analyze the value of the ore deposit kernel density K at position u within the analysis window, the predictor variable in the synthesized rare data is determined under this window size. The value is:

[0064]

[0065] That is, by combining the kernel density of the ore deposit, the values ​​within a preset window range in the predictor variable layer are weighted and summed to obtain the composite predictor variable value, which is the predictor variable in the rare data. The composite value.

[0066] like Figure 5 As shown, similarly, the synthesized values ​​of the predictor variables for other rare data can be obtained under the same window size condition. Repeating the above steps, window analysis is performed on the location of each ore-bearing unit to obtain the predictor variables for all rare data at the ore-bearing unit location. Combining the predictor variables of rare data at all ore-bearing units under the same analysis window size generates one times the number of positive samples. One times the number of positive samples can be obtained for each analysis using an analysis window of a different size. Therefore, the required number of positive samples can be synthesized as needed. Simultaneously, negative samples, with the same number as the positive samples, are randomly selected from areas with poor ore-forming geological conditions, and together with the positive samples, constitute the training dataset.

[0067] Kernel density analysis is performed on known ore deposits to obtain their kernel density maps. The analysis window is selected based on the number of training samples, using the location of the ore deposit cell as the center. For example, if a single positive sample size is required, a 3×3 analysis window is selected.

[0068] like Figure 6 As shown, taking the synthesis method of predictor variable Y as an example, the sample synthesis process is briefly described. The yellow grid represents the prediction layer Y, and the pixel value of each small grid represents the value of the predictor variable Y at that location. The light blue grid represents the ore deposit kernel density, and the red grid in the layer represents the grid where the ore deposit is located, based on the high value area of ​​the kernel density. The ore deposit kernel density can not only reflect the strength of mineralization, but also the directionality of mineralization (as shown in the figure, the kernel density value is higher in the NE direction, indicating that the mineralization is distributed in the NE direction), that is, it characterizes the spatial structure of mineralization. In the sample synthesis process, a 3×3 window is used for analysis. The new predictor variable value is the weighted sum of the original predictor variable value in the analysis window and the normalized weight of the ore deposit kernel density, that is, the predictor variable of rare data, which is expressed by the formula:

[0069]

[0070] Similarly, this method can be used to calculate the composite value of the predictor variables for other rare data at the location of the deposit within the same size analysis window. This process is repeated for other deposit locations, calculating the composite values ​​of the predictor variables for each rare data point, and finally combining them to form a single positive sample.

[0071] In this embodiment of the invention, an example is given where the area under study contains 13 known mineral deposits, and positive samples of 1x (13), 2x (26), and 3x (39) quantities are synthesized respectively. The original values ​​of each predicted variable at each mineral-bearing unit constitute 13 positive samples (Novel1); the original values ​​and the synthesized 1x positive samples constitute 26 positive samples (Novel2); the original values ​​and the synthesized 2x positive samples constitute 39 positive samples (Novel3); and the original values ​​and the synthesized 3x positive samples constitute 52 positive samples (Novel4).

[0072] Then, according to the principle of negative sample selection (i.e., randomly selecting the same number of units as the positive samples from areas with poor mineralization conditions), a corresponding number of 52 negative samples were selected. From these 52 negative samples, 13 negative samples were randomly selected, repeated 4 times, and combined with Novel1 to obtain four training datasets with a sample size of 26 (13 positive samples + 13 negative samples), namely Novel11, Novel12, Novel13, and Novel14; from these 52 negative samples, 26 negative samples were randomly selected, repeated 4 times, and combined with Novel2 to obtain four training datasets with a sample size of 52 (26 positive samples + 26 negative samples), namely Novel21. Novel22, Novel23, and Novel24 were selected as training datasets. From these 52 negative samples, 39 negative samples were randomly selected and repeated four times. These were then combined with Novel3 to obtain four training datasets with a total of 78 samples (39 positive samples + 39 negative samples): Novel31, Novel32, Novel33, and Novel34. The 52 negative samples were then combined with Novel4 to obtain a training dataset Novel41 with a total of 104 samples (52 positive samples + 52 negative samples).

[0073] S4. Based on the created prediction layer and several training datasets, perform mineralization prediction and evaluate the mineralization prediction model based on the prediction results.

[0074] Supervised machine learning algorithms are used to train the constructed mineralization prediction models on the training dataset. The trained mineralization prediction models are then used to make predictions for the entire study area. The mineralization probability value of each grid cell in the study area is calculated, mineralization prospective areas are delineated, and the trained mineralization prediction models are evaluated using ROC curves and AUC values.

[0075] In this embodiment, the supervised machine learning algorithms include decision trees, support vector machines, random forests, and adaptive augmentation machine learning algorithms.

[0076] Continuing from the previous example, this embodiment uses the random forest algorithm to learn Novel11, Novel12, Novel13, and Novel14 respectively. Figure 7 As shown, four mineral-bearing probability layers for the study area were obtained. ROC curves for each of the four mineral-bearing probability layers were plotted, and their average AUC values ​​were calculated. The above steps were repeated to calculate the average AUC values ​​for the four training datasets with different amounts of data. According to the model evaluation results, the average AUC value was the highest (0.997) when the number of synthesized positive samples was twice that of the original dataset. Figure 8 As shown, the average of the prediction results obtained from this training dataset is used to obtain the W-Cu mineralization probability map for the Taqian-Zhuxi area. Based on this, the mineralization probability is classified using the natural discontinuity method, and the W-Cu polymetallic mineral exploration target area can be delineated according to the high-value areas of mineralization probability.

[0077] In this embodiment, the example is that the random forest algorithm is used to learn Novel11, Novel12, Novel13 and Novel14 respectively. In fact, the random forest algorithm can also be used to learn Novel21, Novel22, Novel23 and Novel24. There is no limitation on this.

[0078] As can be seen from the above embodiments, the mineralization prediction method disclosed in this invention, which synthesizes rare data based on mineralization spatial structure, constructs a mineralization prediction model and predictive variables based on the geological background of the area under study, preprocesses the prediction layer, performs mineralization spatial structure analysis, and synthesizes rare data predictive variables by combining the predictive variables of the area under study, thereby generating several training datasets. This allows for the synthesis of a predetermined number of positive samples while taking into account the geological context, improving the generalization ability and prediction accuracy of the mineralization prediction model. This, in turn, improves the accuracy and efficiency of mineralization prediction, reduces exploration risks and costs, and provides important basis and guidance for mineral exploration.

[0079] The method described in this invention is not limited to the embodiments described in the specific implementation. Other implementation methods derived by those skilled in the art based on the technical solution of this invention also fall within the scope of technical innovation of this invention.

Claims

1. A mineralization prediction method based on rare data synthesized from mineralization spatial structure, comprising the following steps: S1. Geological data collection and preprocessing based on prediction needs; S2. Based on the geological background of the area to be studied, construct a mineralization prediction model, determine the prediction variables, and create a prediction variable layer; S3. Conduct mineralization spatial structure analysis, combine the predictor variables of the area to be studied to synthesize a preset number of rare data predictor variables, and then generate several training datasets. S4. Several training datasets are used to train the constructed mineralization prediction model. Based on the created prediction variable layer, the mineralization probability of the study area is predicted, and the performance of the mineralization prediction model is evaluated. Based on the prediction results of the better-performing mineralization prediction model, the mineral exploration target area is delineated. Step S3 includes the following sub-steps: S31. Determine the appropriate analysis window size based on the number of training samples to be synthesized; S32. Perform kernel density analysis on the first known ore deposit in the study area to obtain the kernel density of the first known ore deposit; S33. Taking the location of the first known ore deposit as the center, the original first predicted variable value of the predicted variable layer in the analysis window of the preset size is weighted and summed with the kernel density of the first known ore deposit after normalization. The weighted sum is used as the final first predicted variable at the location of the first known ore deposit, that is, the first predicted variable of the synthetic rare data. The first predicted variable is replaced with the Nth predicted variable, and the weighted sum is performed in the same way until all the predicted variables of the synthetic rare data are obtained. S34. Replace the first known mineral deposit in steps S32-S33 with the Nth known mineral deposit, repeat steps S32-S33 to calculate the predictor variables of the rare data at the locations of other known mineral deposits, and combine the predicted variables of the rare data of each known mineral deposit to form a positive sample of the expected multiple. S35. Randomly select the same number of units as the positive samples in areas with poor mineralization conditions, and use them as negative samples. Based on the combination of the generated positive and negative samples, a training dataset with a preset sample size is obtained. In step S33, the predictor variables for rare data are synthesized according to the following formula. : Where i is the analysis window size, i is an odd number, and Y u l For the l-th predictor variable Y at position u in the analysis window l The value of K u To analyze the kernel density K value of the ore deposit at position u in the analysis window.

2. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: The prediction requirements mentioned in step S1 include the mineralization prediction accuracy, the predicted mineral type, and the required deposit type.

3. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: The geological data collected in step S1 includes geological maps of the area to be studied, mineral deposit distribution maps, geochemical data, and geophysical data.

4. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: Step S2 includes analyzing the geological background of the area under study, summarizing the metallogenic geological conditions, conducting research on typical ore deposits, combining previous analyses of the metallogenic regularity of the area under study, identifying the key geological elements controlling the formation of ore deposits and important prospecting indicators, thereby constructing a metallogenic prediction model.

5. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: In step S2, based on the preprocessed geological data, predictive elements that can indicate key geological features and important mineral exploration indicators are used as predictive variables.

6. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 5, characterized in that: The predicted variables in step S2 include Carboniferous-Permian strata, NE and NW trending structures, Yanshanian intrusive bodies, aeromagnetic anomalies, and mineralization-related geochemical element anomalies. Step S2 includes dividing the study area into a series of grid cells of the same size according to the prediction accuracy; performing distance analysis on the ore-controlling strata, ore-controlling structures, and ore-controlling intrusive bodies in the study area to obtain corresponding predicted variable layers; performing factor analysis on the geochemical data to extract factor scores of mineralization-related geochemical element combinations and performing local singularity analysis to obtain predicted variable layers of mineralization-related geochemical element anomalies; performing local singularity analysis on the aeromagnetic data to obtain predicted variable layers indicating aeromagnetic anomalies; and storing the information of each layer in the grid cells.

7. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: In step S35, negative samples are randomly selected a predetermined number of times from the same number of negative samples as the positive samples, and the selected negative samples are combined with positive samples to obtain different training datasets.

8. The mineralization prediction method based on rare data synthesized from mineralization spatial structure as described in claim 1, characterized in that: In step S4, the average value of the prediction results of the metallogenic prediction model with better performance is used to obtain the metallogenic probability map of the target mineral in the area to be studied. The metallogenic probability is classified by the natural discontinuity method, and the high-value areas of metallogenic probability are delineated as mineral exploration target areas.

Images