A method for identifying deposit types based on trace elements of sphalerite
By constructing a trace element database of sphalerite and combining it with machine learning methods, using random forest and gradient boosting algorithms, the problem of difficulty in distinguishing ore deposit types in existing technologies has been solved, achieving efficient and accurate ore deposit type identification.
Patent Information
- Application Number
- CN202310533691.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-12
- Publication Date
- 2025-12-23
- Estimated Expiration
- 2043-05-12
AI Technical Summary
Existing methods for identifying trace elements in sphalerite cannot accurately distinguish between different types of deposits, resulting in low exploration efficiency and high costs.
A database of trace elements in sphalerite was constructed. By combining machine learning methods with random forest and gradient boosting algorithms, training and test sets were established. The content of trace elements in sphalerite was used as an indicator to determine the type of deposit for accurate identification.
It has improved the accuracy and efficiency of mineral exploration, solved the problems of high difficulty and high cost in mineral exploration in the deep and peripheral parts of mineral deposits, and enabled accurate differentiation of different types of mineral deposits.
Smart Images
Figure CN116933151B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to a sphalerite trace element-based method for distinguishing deposit types, and belongs to the technical field of deposit prospecting and exploration. BACKGROUND
[0002] Determining the genesis of a deposit is still one of the most critical but most challenging problems in the study of ore deposits. Correctly determining the genesis of a deposit helps better understand regional large-scale mineralization processes and earlier application of deposit models, which can significantly improve exploration efficiency. Different types of deposits are marked by different sources of ore-forming materials, physical and chemical conditions, and ore-forming processes, all of which will significantly affect the trace element composition of minerals. Therefore, mineral trace element chemistry is widely used to determine the genesis of ores, and the most commonly used minerals include garnet, quartz, pyrite, and magnetite. However, the same minerals from different ore genetic types can have similar trace element geochemical properties, making it difficult to determine the deposit type.
[0003] Sphalerite is the most important zinc-bearing ore mineral and is ubiquitous in many types of deposits, including volcanic massive sulfide (VMS), Mississippi Valley-type (MVT), porphyry (Porphyry), epithermal (EPI), sedimentary exhalative (SEDEX), and skarn deposits (Skarn). Sphalerite can accommodate a variety of trace elements through substitution, and the content of these elements can be used to distinguish deposit types. In the past few decades, research has been conducted on the use of sphalerite trace elements to classify deposit types. Traditional methods include using Mn-Fe, Co / Ni-Cd / Fe, Cd / Fe-Mn, Ge-In binary diagrams, and Cd-Mn-1000Ge ternary diagrams to enhance the discrimination of sphalerite trace elements for deposit types. However, due to the similar trace element composition of sphalerite from different deposit types, existing discrimination diagrams cannot accurately distinguish different deposit types.
[0004] Searches have found that so far, there has been no related report on constructing a sphalerite trace element database and constructing sphalerite trace element training and test sets, and combining machine learning methods to accurately and efficiently distinguish deposit types. SUMMARY
[0005] The present application first proposes to construct a sphalerite trace element database and construct sphalerite trace element training and test sets, and combine machine learning methods, and use the widely existing sphalerite trace element content in various types of deposits as a marker to determine the deposit type, which improves the efficiency and accuracy of prospecting for sphalerite-related deposits, and at the same time solves the problem of high cost and difficulty in deep and peripheral prospecting of deposits.
[0006] The present application is a method for accurately distinguishing deposit types based on sphalerite trace elements, which specifically includes the following steps:
[0007] Step one: Establishing sphalerite trace element database
[0008] Collecting sphalerite trace element data from published literature around the world;
[0009] Organizing the collected data, establishing at least 3000 groups of trace element database from six types of deposits;
[0010] The database includes deposit name, deposit location, deposit type, sphalerite trace element content;
[0011] In the database, the deposit name, deposit type, sphalerite trace element content are in a corresponding relationship, that is, a certain type of deposit; Its deposit type and the type and content of sphalerite trace elements in the deposit are in a corresponding relationship.
[0012] The six types of deposits at least include: volcanic massive sulfide (VMS), Mississippi Valley type (MVT), porphyry type (Porphyry), hydrothermal type (EPI), exhalative-sedimentary type (SEDEX) and skarn deposit (Skarn);
[0013] The sphalerite trace elements at least include at least 10 kinds of Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu;
[0014] The database is an important content for subsequent machine learning model learning, and is the key to determine the type of deposit. When establishing the database, a large amount of sphalerite trace element content is required, and these sphalerite trace element contents should be correctly divided into different types of deposits;
[0015] Step two: data preprocessing
[0016] The nearest neighbor method interpolation and center logarithmic ratio transformation are performed on the data of various sphalerite trace elements in the sphalerite trace element database established in step one, so that the data covariance does not change and conforms to the normal distribution. The data of various sphalerite trace elements include the contents of Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu;
[0017] During data preprocessing, the data of the determined type of trace element should be processed under the condition of determining the type of deposit, so that the data of the trace element in the deposit type becomes normally distributed. The data includes content data;
[0018] Step three: Establishing training set and test set
[0019] To train and test Random Forests and Gradient Boosting classifiers, a test set and a training set are built by extracting at least 200, preferably 200-400, data from each deposit type for training the Random Forests and Gradient Boosting classifiers from the sphalerite trace element database; to avoid biasing towards the classes with more data, the same number of data is randomly extracted from each deposit type by a random function; the remaining data is used for testing, obtaining the training and testing classification matrix; the training and testing classification matrix can be used to evaluate the accuracy of the model training, each column represents the predicted deposit type, and the total number of each column represents the number of data predicted as the deposit type; each row represents the real deposit type of the data, and the total number of data in each row represents the number of data instances of the deposit type.
[0020] Step 3: Build machine learning model
[0021] The machine learning model is built using the Random Forest and Gradient Boosting algorithms. Random Forest and Gradient Boosting use bootstrap sampling methods to randomly sample training samples from the sample set to generate decision trees and training subsets. When building a decision tree, each node in the decision tree is optimally segmented; therefore, the quality of node segmentation is very important for creating a decision tree. Model hyperparameter tuning is performed using cross-validation, and when the decision tree is greater than or equal to 4500, preferably 5000, the depth is greater than or equal to 3, and the generated child node appears N times, the splitting is stopped, and this parameter is the optimal parameter of the model. The N is less than or equal to 6;
[0022] The model randomly samples in the original data set to form n different sample data sets, then builds n different decision tree models according to these data sets, and finally obtains the final result according to the voting situation of these decision tree models.
[0023] In the present application, for different data types, the hyperparameters of these algorithms need to be adjusted to achieve optimal results. The key parameters of the machine learning model are: n_estimators = 5000, max_depth = 3, min_samples_split = 6;
[0024] Step 5: Evaluate model reliability
[0025] The receiver operating characteristic curve (ROC) of the random forest and gradient boosting algorithm model is obtained by using the orange software, which is realized by describing the true positive rate and the false positive rate, and the ROC curve is obtained by plotting the true positive rate on the y-axis and the false positive rate on the x-axis, the area under the ROC curve (AUC), which is usually used as a measure of the performance of the classifier. The AUC value ranges from 0 to 1, and the AUC value of a reliable model is greater than 0.5, that is, a model with an AUC value greater than 0.5 is considered a reliable model, and the closer to 1, the more reliable the model.
[0026] Step six discriminates the deposit type
[0027] The content of each trace element in the sphalerite to be judged is obtained, and the reliable machine learning model obtained in step five is used to predict the deposit type, that is, the obtained content set of each trace element in the sphalerite to be judged is used to establish a machine learning model by using a random forest and a gradient boosting algorithm, a classification matrix of trace elements in the sphalerite to be judged is obtained, and the deposit type is judged according to the classification matrix.
[0028] The present application is a method for accurately discriminating the deposit type based on the trace elements of sphalerite, in step two, the missing values refer to the content lower than the detection limit of the testing instrument and the testing work not carried out on individual elements by part of the research, and these missing values may change the mean and variance estimates in the analysis, therefore, the present application uses the k-nearest neighbor method for interpolation, and uses the center logarithmic ratio transformation to make the data conform to the normal distribution.
[0029] The database further includes mineral paragenetic combination parameters; the mineral paragenetic combination parameters are the parameterization of the geological information of mineral paragenesis, that is, the paragenetic minerals existing in the deposit are marked as 1, and the non-existing paragenetic minerals are marked as 0, and the paragenetic minerals include at least one of chalcopyrite, pyrite, galena, arsenopyrite and magnetite.
[0030] As preferred, in the present application, in the data preprocessing,
[0031] The elements Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu with less than 40% missing values of the content of trace elements of sphalerite are selected;
[0032] The missing values refer to the content lower than the detection limit of the testing instrument and the testing work not carried out on individual elements by part of the research;
[0033] For the missing values of the trace elements (Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu) of sphalerite in the database, the nearest neighbor method based on Euclidean distance without changing the covariance of the data set is used for interpolation in the data processing software XLSTAT.
[0034] In order to make sphalerite trace elements meet the normal distribution, the sphalerite trace elements are subjected to center logarithmic ratio conversion in the ioGAS.
[0035] As preferred, the present application is a method for accurately distinguishing deposit types based on sphalerite trace elements, wherein during model training,
[0036] A machine learning model is established using a random forest and a gradient boosting algorithm, and the sphalerite trace elements training set is used for training;
[0037] The random forest and the gradient boosting adopt a bootstrap sampling method to randomly extract training samples from the sample set to generate decision trees and training subsets. When constructing the decision tree, each node in the decision tree is subjected to optimal segmentation; therefore, the quality of node segmentation is very important for creating the decision tree. When the decision tree is greater than or equal to 4500, preferably 5000, and the depth is greater than or equal to 3, the generated child nodes appear N times, the splitting is stopped, and the parameter is the optimal parameter of the model. The N is less than or equal to 6. In the technical development process of the present application, it is found that it is very difficult to judge six types of deposits by the content of sphalerite trace elements, because the sphalerite trace elements of the six types of deposits have similar contents, and only by adjusting the parameters multiple times and increasing the geological conditions for constraint, a higher correct rate can be obtained.
[0038] In the technical development process of the present application, various machine learning methods are also tried for deposit type classification, including random forest, gradient boosting decision tree, artificial neural network, lasso algorithm, support vector machine, k-nearest neighbor (Random Forests, Gradient Boosting, Artificial Neural Networks, Least Absolute Shrinkage and Selection Operator, Support Vector Machines, k-Nearest Neighbors) and other machine learning methods for classification; but it is found that the machine learning model established by the two methods of random forest and gradient boosting decision tree is more reliable and accurate.
[0039] In the technical development process of the present application, especially the VMS deposit and other deposits have similar trace element contents and other types of deposits highly overlap, which makes it difficult to distinguish the VMS deposit from other deposit types. We found that the results were significantly improved by parameterizing the geological information. BRIEF DESCRIPTION OF DRAWINGS
[0040] Figure 1 An implementation process diagram for accurately distinguishing deposit types based on sphalerite trace elements is provided. DETAILED DESCRIPTION
[0041] Embodiment 1
[0042] S1. Data collection:
[0043] 4095 groups of sphalerite trace element data of 86 deposits in the world were collected according to the literatures published in recent years, including 11 epithermal deposits, 27 Mississippi Valley-type deposits, 4 porphyry deposits, 5 exhalative-sedimentary deposits, 26 skarn deposits and 12 volcanic massive sulfide deposits, and the element statistics of each deposit type are shown in Table 1; the advantage of the database is that the data sources are extensive, covering sphalerite geochemical data of global deposits, which ensures that the availability of the model used is not limited to any region.
[0044] The database includes deposit name, deposit location, deposit type, mineral paragenetic association parameters, sphalerite trace element content;
[0045] In the database, the deposit name, deposit type and sphalerite trace element content are in a corresponding relationship, that is, a certain type of deposit; the deposit type and the types and contents of sphalerite trace elements in the deposit are in a corresponding relationship.
[0046] The mineral paragenetic association parameters are the geological information of mineral paragenesis, that is, the coexisting minerals in the deposit are marked as 1 and the non-existing coexisting minerals are marked as 0, and the coexisting minerals include at least one of chalcopyrite, pyrite, galena, arsenopyrite and magnetite;
[0047] The sphalerite trace elements include at least Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu;
[0048] The database is an important content for subsequent machine learning model learning and is the key to determine the deposit type discrimination. When the database is established, a large number of sphalerite trace element contents are required, and these sphalerite trace element contents should be correctly divided into different deposit types;
[0049] In the database, the contents of each trace element in each deposit type are shown in Table 1;
[0050] Table 1 Trace element contents of six deposit types in the database
[0051]
[0052]
[0053] N = number; MIN = minimum value; MAX = maximum value; MEAN = average value;
[0054] S2. Data preprocessing:
[0055] The data of various sphalerite trace elements in the sphalerite trace element database established by S1 are interpolated by the nearest neighbor method and are subjected to central logarithmic ratio conversion, so that the data covariance is not changed and is in accordance with the normal distribution. The data of various sphalerite trace elements include the contents of Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu;
[0056] During data preprocessing, the data of the determined kind of trace elements should be processed under the condition of determining the type of the deposit, so that the data of the trace elements in the deposit type become normally distributed; the data include content data;
[0057] The specific operation is as follows:
[0058] For each type of deposit, the following operations are performed after determining the type of the deposit:
[0059] 1. The elements Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu with less than 40% missing values of sphalerite trace elements are selected;
[0060] 2. For the missing values of sphalerite trace elements (Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu) in the database, the nearest neighbor method based on Euclidean distance is used in the data processing software XLSTAT to interpolate without changing the covariance of the data set;
[0061] 3. In order to make the contents of each element in the contents of sphalerite trace elements (Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu) in accordance with the normal distribution, the sphalerite trace elements are subjected to central logarithmic ratio conversion in ioGAS;
[0062] S3. Establishing a sphalerite trace element test set and a training set
[0063] 1. 300 groups of sphalerite trace element data are randomly selected from each type of deposit in the database after data preprocessing using a random function to establish a training set;
[0064] 2. The remaining sphalerite trace element data of each type of deposit are used to establish a test set;
[0065] S4. Model training
[0066] 1. A machine learning model is established using the random forest and gradient boosting algorithm, and the established training set is used for training;
[0067] 2. Random forest and gradient boosting adopt bootstrap sampling method to randomly sample training samples from the sample set to generate decision trees and training subsets. When building a decision tree, the optimal segmentation is performed for each node in the decision tree; therefore, the quality of node segmentation is very important for creating a decision tree. When the decision tree is 5000, the depth is 3, and the generated child node appears 6 times, the splitting is stopped, and the parameter is the optimal parameter of the model (n_estimators=5000, max_depth=3, min_samples_split=6);
[0068] 3. Obtain the sphalerite trace element test set random forest and gradient boosting deposit type classification matrix (Table 1), and the overall classification accuracy is 93.02% and 92.82%, respectively.
[0069] Table 2 Sphalerite trace element test set classification matrix
[0070]
[0071] S5. Evaluate the reliability of the model
[0072] 1. Evaluate the reliability of the machine learning model using the test set.
[0073] The AUC value obtained by the receiver operating characteristic (ROC) curve is used to evaluate the performance of the two models. The receiver operating characteristic curve of the random forest and gradient boosting algorithm model is obtained using the orange software, and the AUC value of the random forest and gradient boosting identification of the deposit type is 0.989, 0.991, indicating that the two machine learning models have high reliability.
[0074] Example 2
[0075] S1. S2. S3. S4. The steps of Example 1 are the same, and S5 and S6 are specifically described here.
[0076] S5. Obtain the sphalerite trace data set of the Qingshuiyang lead-zinc deposit
[0077] 1. Obtain the sphalerite sample of the Qingshuiyang lead-zinc deposit by field sampling, and make a laser sheet of the sphalerite sample;
[0078] 2. Obtain the sphalerite trace element content set of the Qingshuiyang lead-zinc deposit by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) trace element analysis instrument.
[0079] S6. Deposit type prediction
[0080] A machine learning model is established by using random forest and gradient boosting algorithm to predict the deposit type of the Qingshuitang lead-zinc mine, and the obtained sphalerite trace element set of the Qingshuitang lead-zinc mine is used to establish a machine learning model by using random forest and gradient boosting algorithm to obtain a classification matrix of the Qingshuitang sphalerite trace element (Table 2),
[0081] Table 3 Classification matrix of Qingshuitang sphalerite trace elements
[0082]
[0083] According to Table 2, it is judged that the deposit type is Mississippi Valley type (MVT) deposit.
Claims
1. A method for accurately discriminating deposit types based on trace elements of sphalerite, characterized by; It comprises the following steps: Step one: Establishing sphalerite trace element database Collecting sphalerite trace element data from published literature worldwide; Organizing the collected data to establish a database of at least 3000 sets of trace elements from six types of deposits; The database includes deposit name, deposit location, deposit type, and sphalerite trace element content; In the database, the deposit name, deposit type, and sphalerite trace element content are in a corresponding relationship, i.e., a certain type of deposit, its deposit type, and the types and contents of sphalerite trace elements in that deposit are in a corresponding relationship; The six types of deposits include at least volcanic massive sulfide, Mississippi Valley-type, porphyry, epithermal, exhalative-sedimentary, and skarn deposits; The sphalerite trace elements include at least 10 of Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn, and Cu; Step two: Data preprocessing The sphalerite trace element data in the database established in step one is interpolated by the nearest neighbor method and transformed by the center logarithmic ratio, so that the data covariance does not change and conforms to the normal distribution; the sphalerite trace element data includes the contents of Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn, and Cu; During data preprocessing, the data of the determined type of trace element is processed under the condition of determining the type of deposit, so that the data of the trace element in the deposit type becomes normally distributed; the data includes content data; Step three: Establishing training and test sets To train and test random forest and gradient boosting decision tree classifiers, establish test and training sets, extract at least 200 data for training random forest and gradient boosting classifiers from each type of deposit in the established sphalerite trace element database; to avoid bias towards categories with more data, the same number of data is randomly extracted from each deposit type by a random function; the remaining data is used for testing to obtain the classification matrix for training and testing; Step four: Establishing machine learning model Establish a machine learning model using random forest and gradient boosting algorithms; random forest and gradient boosting use bootstrap sampling method to randomly extract training samples from the sample set to generate decision trees and training subsets; when constructing decision trees, each node in the decision tree is optimally segmented; use cross-validation for model hyperparameter tuning; when the decision tree is greater than or equal to 4500 and the depth is greater than or equal to 3, the generated child nodes appear N times, stop splitting, and this parameter is the optimal parameter for the model; N is less than or equal to 6; The model will randomly sample in the original data set to form n different sample data sets, then build n different decision tree models according to these data sets, and finally obtain the final result according to the voting situation of these decision tree models; Step five: Evaluate model reliability The receiver operating characteristic curve, namely the ROC curve, of the random forest and gradient boosting algorithm model is obtained by describing the true positive rate and the false positive rate, the ROC curve is obtained by plotting the true positive rate on the y-axis and the false positive rate on the x-axis, the area under the curve AUC is used as a measure of the performance of the classifier; the AUC value ranges from 0 to 1, and the AUC value of a reliable model is greater than 0.5, that is, a model with an AUC value greater than 0.5 is considered a reliable model, and the closer to 1, the more reliable the model; Step six discriminates the deposit type The content of each trace element in the sphalerite to be determined is obtained, and the reliable machine learning model obtained in step five is used to predict the deposit type, that is, the content set of each trace element in the sphalerite to be determined is obtained, a machine learning model is established by using a random forest and a gradient boosting algorithm, a classification matrix of trace elements in the sphalerite to be determined is obtained, and the deposit type is determined according to the classification matrix.
2. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: In step two, the k-nearest neighbor method is used for interpolation, and the center logarithmic ratio transformation is used to make the data conform to the normal distribution.
3. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: During data preprocessing, The elements Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu with less than 40% missing values of sphalerite trace element content are selected; The missing values refer to the content lower than the detection limit of the test instrument and part of the research on individual elements without test work; For the missing values of sphalerite trace elements Ag, As, Cd, Co, Ga, Ge, Sb, Pb, Fe, Mn, In, Sn and Cu in the database, the nearest neighbor method based on Euclidean distance without changing the covariance of the data set is used in the data processing software XLSTAT to interpolate. In order to make the sphalerite trace elements conform to the normal distribution, the center logarithmic ratio transformation is performed on the sphalerite trace elements in ioGAS.
4. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: In step three, 200-400 samples are extracted from each deposit type of the established sphalerite trace element database for training the random forest and gradient boosting classifier.
5. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: In step three, the training and test classification matrix is used to evaluate the accuracy of model training, each column represents the predicted deposit type, and the total number of each column represents the number of data predicted as the deposit type; each row represents the true deposit type of the data, and the total number of data in each row represents the number of data instances of the deposit type.
6. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: During model training, the existing sphalerite trace element training set is used for training.
7. The method for accurately identifying deposit types based on trace elements of sphalerite according to claim 1, characterized in that: The key parameters of the machine learning model are: n_estimators=5000, max_depth=3, min_samples_split=6.