Cube prediction model metallogenic prediction method based on three-dimensional modeling

Abstract

The invention belongs to the technical field of mining exploration, and particularly discloses a three-dimensional modeling-based cube prediction model metallogenic prediction method. The method comprises the following steps: collecting geological, geophysical, geochemical and / or remote sensing data of a mining area; cleaning, normalizing and standardizing the data; constructing a three-dimensional geologic model by adopting GIS or CAD; dividing the model into a plurality of cubes, and dynamically adjusting the sizes and shapes of the cubes by adopting a self-adaptive grid division method; geologic features in the cube are analyzed and recognized; a deep learning model is trained by adopting a convolutional neural network based on the recognition features, then data features are automatically extracted, and ore body distribution is predicted; deploying an unmanned aerial vehicle and a sensor network to dynamically acquire data in real time, and feeding back the data to the three-dimensional model for updating; through comparison with an on-site exploration result, the accuracy and reliability of the model are evaluated. The method has the characteristics of high prediction accuracy and timeliness, high adaptability and high exploration efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of mining exploration, and in particular to a cube prediction model mineralization prediction method based on three-dimensional modeling, which has high prediction accuracy and timeliness, strong adaptability and high exploration efficiency. Background Art

[0002] The cube prediction model of three-dimensional modeling is a geological model constructed by combining geology, geological resources and geological engineering with computer technology. It aims to accurately represent the three-dimensional space of the mining area. It is mainly used to identify and predict the location, shape and scale of the ore body, especially the prediction of deep ore bodies and hidden mines. Therefore, it is widely used in mining exploration, geological disaster prediction and prevention, engineering geological survey and design. By dividing the study area into a series of cubes (blocks) of equal size, and discretizing and quantifying the geological, structural, rock mass, ore body and other information in these cubes, researchers can analyze data in a more detailed way, thereby improving the accuracy and effectiveness of exploration; and the three-dimensional cube model also integrates a variety of data including geological, geophysical, geochemical and remote sensing data, so that explorers can visualize the potential distribution of ore bodies in a unified three-dimensional space, thereby identifying possible ore body structures and characteristics. By analyzing the geological characteristics of each cube, explorers can also identify the indicative characteristics of the ore body. For example, by combining geological structure, ore type and other relevant data, the model can help find high-potential mining areas and guide subsequent drilling and sampling work. However, the existing cube prediction model based on three-dimensional modeling still has the following inconveniences in the mineral prospecting process.

[0003] First of all, when using the traditional three-dimensional modeling cube prediction model for ore body prediction, a fixed grid division technology is usually used. This division technology ignores the changes in geological characteristics and data density, resulting in low accuracy in identifying ore body characteristics. Fixed grid division cannot flexibly cope with complex geological structures, which can easily lead to unreliable prediction results, thereby affecting subsequent resource exploration and development efficiency. Improving the accuracy of ore body prediction is an urgent need in the current mining exploration field.

[0004] Secondly, existing ore prediction models mostly rely on traditional machine learning algorithms, which usually require manual feature extraction and show obvious limitations when faced with complex and nonlinear data. This makes it difficult for traditional models to adapt to new or unknown mining conditions, resulting in a significant reduction in prediction results. In order to maintain good prediction performance in a changing geological environment, it is particularly important to develop an automated model with higher adaptability.

[0005] Finally, traditional mining monitoring methods often rely on regular field exploration and data collection, which results in low monitoring frequency, insufficient feedback mechanism, and data delays that prevent the model from quickly responding to changes in the geological environment, resulting in the prediction results not reflecting the actual situation, thus affecting the effectiveness of decision-making. With the rapid changes in geological conditions, there is an urgent need for a solution that can achieve dynamic real-time monitoring and model updates to improve response speed and accuracy during the exploration process. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides a cube prediction model mineralization prediction method based on three-dimensional modeling, which has high prediction accuracy and timeliness, strong adaptability and high exploration efficiency.

[0007] The present invention is implemented in the following way: including data collection, data preprocessing, three-dimensional geological model construction, cube grid division, ore body characteristic analysis, ore body distribution prediction, dynamic real-time monitoring and model updating, model verification and evaluation steps, the specific contents are: A. Data collection: Collect geological, geophysical, geochemical and / or remote sensing data of the target mining area; B. Data preprocessing: cleaning, normalization and standardization of the collected data; C. Construction of 3D geological model: Based on the pre-processed data, a 3D geological model of the target mining area is constructed using GIS or CAD technology; D. Cube meshing: Divide the aforementioned three-dimensional geological model into multiple cube units, and use an adaptive meshing method to dynamically adjust the size and shape of the cube units according to geological characteristics and data density; E. Ore body feature analysis: Analyze and identify the geological features within the aforementioned cubic unit, and use multi-source data fusion technology to improve the accuracy of feature identification; F. Ore body distribution prediction: Based on the features identified by the aforementioned cubic units, a deep learning model is trained using a convolutional neural network. The deep learning model is then used to automatically extract data features and predict ore body distribution. G. Dynamic real-time monitoring and model updating: Deploy drones and sensor networks for dynamic real-time monitoring, obtain field data in real time, and feed the acquired data back to the 3D model for updating; H. Model verification and evaluation: Evaluate the accuracy and reliability of the model by comparing it with the field exploration results.

[0008] Furthermore, in step A, the geophysical data of the target mining area is collected by acquiring underground data using drilling, gravity, electromagnetic and / or seismic methods, and the remote sensing data of the target mining area is collected by identifying and analyzing surface features in combination with remote sensing technology.

[0009] Furthermore, in step B, data cleaning is to eliminate outliers and noise in the collected values ​​by using statistical methods, and normalization and standardization processing is to use a standardized formula to standardize different data to form a unified data format and range. The standardized formula is: , Where: Z is the standardized value, X is the sample value, m is the sample mean, s is the sample standard deviation.

[0010] Furthermore, in the step C, constructing a three-dimensional geological model includes the following sub-steps: C1. Data input: import the collected data after preprocessing into GIS or CAD system; C2. Model drawing: Generate a three-dimensional geological model of the mining area through three-dimensional modeling tools.

[0011] Furthermore, in the step D, the adaptive grid division method specifically includes the following sub-steps: D1. Wavelet transform: Use wavelet transform to analyze the local features of data and extract feature information at different scales; D2. Machine learning optimization: Combine the random forest algorithm to optimize the division of cube units, build and train the random forest model, and then use the random forest model to evaluate the fitness of different division schemes to select the optimal solution; The random forest model is established by the following steps: D21, data preparation: extract feature data from cube cells, construct feature matrix X and target variable Y, and then divide 70-80% of the data in the data set into training set and the remaining data into test set; D22, construction and training: In the process of building each tree in the forest, data samples are randomly extracted, and features are randomly selected at each node for splitting. The number of trees in the forest is set to 80 to 100, and then the training set is used to train the random forest model. Each tree is trained independently to generate the final forest model. D23, Model evaluation: Use accuracy, recall and / or F1-score indicators to evaluate the performance of the final forest model on the test set for the fitness of different partitioning schemes.

[0012] Furthermore, in the step D23, evaluating the performance of the forest model includes the following sub-steps: D231, Set thresholds: Set one or more thresholds in advance. If the accuracy, recall and / or F1-score of a solution is lower than the set threshold, the solution is considered unacceptable. D232, Priority sorting: Prioritize each solution based on its accuracy, recall rate and / or F1-score, and select the solution with the best fitness index; D233. Cross-validation: Repeat the above steps on different validation sets to ensure that the selected scheme has good generalization ability and consistency.

[0013] Furthermore, the wavelet transform in step D1 is used for ore body characteristic analysis and prediction in step E and step F, specifically including: D11, the wavelet transform in step D1 In step E, in the ore body feature analysis stage, the feature information extracted by wavelet transform is used as an important basis for identifying and analyzing geological features in the cubic unit, and the geological features in the cubic unit are more accurately identified and analyzed; D12. In the wavelet transform in step D1, in step F, the feature information extracted in step D1 is used as the input feature of CNN to train a more accurate deep learning model for ore body distribution prediction.

[0014] Furthermore, in the step D2, the random forest algorithm is combined to optimize the division of the cube unit, a random forest model is established and trained, and then the random forest model is used to evaluate the fitness of different division schemes to select the optimal scheme. The specific division optimization steps are as follows: D2A: Based on the fitness evaluation results, the current cube grid division scheme is preliminarily evaluated. The evaluation indicators include accuracy, recall rate, and F1-score to reflect the model's ability to predict mineralization potential. By comparing the evaluation results under different division schemes, the optimization direction and goal are determined; D2B: Based on the preliminary evaluation results, if the model performs poorly on smaller or larger cube units, adjust the size of the grid; if the model overfits on smaller cube units, increase the grid size appropriately; if the model underfits on larger cube units, reduce the grid size appropriately; D2C: Adjust the shape of the grid according to the spatial distribution of geological features and mineralization laws; D2D: According to the fitness evaluation results, the features are ranked in importance, and the features that contribute most to mineralization prediction are selected for extraction and use; D2E: During the optimization of the partitioning scheme, the hyperparameters are adjusted according to the fitness evaluation results. If the model performs well on the training set but poorly on the test set, the number of trees or the maximum depth is appropriately reduced; D2F: After adjusting the grid size, shape, feature selection, and hyperparameters, the random forest model is retrained and fitness evaluation is performed. By comparing the evaluation results before and after optimization, it can be determined whether the partitioning scheme optimization is effective. If the optimization effect is not obvious or the evaluation results deteriorate, the optimization strategy is readjusted.

[0015] Furthermore, in the step D21, constructing the feature matrix X and the target variable Y includes the following sub-steps: D211: Select the features that have an impact on the prediction of ore body distribution, including geological structure type, lithology, mineralization intensity, topography and geomorphology; D212: For each cube unit, extract its specific value or category on the above selected features; these values ​​or categories will constitute each row (or column) of the feature matrix X; D213: Organize the extracted features into a matrix in the order of cube units, where each row (or column) of the matrix represents the feature vector of a cube unit; D214: Target definition: In mineralization prediction, the target variable Y represents the presence or absence of an ore body or a certain property of the ore body; D215: Label each cubic unit to determine whether it contains an ore body or the specific attributes of the ore body; D216: Organize the labeling results into vector form according to the order of cube units (corresponding to the feature matrix X) to form the target variable Y.

[0016] Furthermore, in the step E, the specific steps are as follows: E1: Logarithmic transformation and power transformation of data according to the nature and distribution of features; E2: Perform descriptive statistical analysis on the feature data and calculate its statistics including mean and variance; at the same time, calculate the correlation coefficient between the feature data to obtain the degree of linear relationship between the features; E4: Use geological interpretation methods to interpret the features geologically; E5: Use Bayesian data fusion method to fuse data from different sources and assign appropriate weights to each type of data based on its characteristics and reliability; E6: Analyze and interpret the fused data to reveal the internal connections and patterns among different data; E7: Based on the fused data and geological feature analysis results, the support vector machine algorithm is used to identify the ore body characteristics; E8: Further analysis and processing of the identified features, including feature selection and feature dimensionality reduction; E9: Use neural network data processing methods to deeply mine and extract features, extracting hidden features and patterns; E10: Output the identified and enhanced features as feature vectors and feature maps for use in subsequent ore body distribution prediction models.

[0017] Furthermore, in the F step, a deep learning model is trained using a convolutional neural network, and then the deep learning model is applied to automatically extract data features; wherein the input layer of the convolutional neural network receives the preprocessed cube unit data, and the output layer applies the convolution operation to extract features, and the output layer convolution formula is: , Where: f is the input feature map, g is the convolution kernel, x and y is the coordinate of the output feature map, m and n are the coordinates of the convolution kernel.

[0018] Furthermore, in the step F, the sub-steps of predicting the ore body distribution are as follows: F1: receives the preprocessed cube unit data through the input layer, and the received data is converted into a multidimensional array as input; F2: The convolution layer performs convolution operations on the input data through multiple convolution kernels to extract local features. The convolution operation in this step uses the convolution kernel to slide on the input data for weighted summation and activation. F3: Apply nonlinear activation function ReLU after the convolutional layer to increase the expressiveness of the model; F4: Reduce the dimension of the feature map through pooling operation, reduce the amount of calculation and extract the main features; F5: Flattens the features extracted by the convolutional layer and the pooling layer and inputs them into the fully connected layer for the final classification and regression tasks; F6: Input the preprocessed cube unit data into the trained CNN model and automatically extract data features through the convolution layer; F7: Input the extracted features into the prediction layer of the fully connected layer of the model; F8: In the prediction layer of the fully connected layer, the prediction results of the ore body distribution are obtained by calculation, including the location, scale, and morphological information of the ore body.

[0019] Furthermore, in the G step, the dynamic real-time monitoring and model updating includes the following sub-steps: G1. Drone configuration: Use multiple drones equipped with high-resolution cameras, lidar and / or multispectral sensors to obtain detailed information on the surface and underground of the target mining area; G2. Sensor network deployment: Set up a sensor network in the target mining area to monitor soil moisture, temperature, pH value and / or conductivity in real time; G3. Real-time data collection: drones and sensors transmit data to the central processing system in real time through the Internet of Things technology for preliminary data processing and fusion; G4, Model update: The initially processed and integrated data is fed back in real time, then applied to the 3D geological model through edge computing algorithms, and the model parameters are updated to reflect the latest geological conditions.

[0020] Furthermore, the G3 step is specifically as follows: G301: Configure the parameters of the drone and sensor network transmission equipment according to actual needs, including transmission rate, frequency, and encryption method; G302: Establish communication links between drones, sensor networks, and central processing systems; G303: Start the drone and fly it according to the preset flight route and altitude; G304: The high-resolution camera, lidar, and multispectral sensor on the drone start to work, collecting images, distance, and spectral data on the surface and underground in real time; G305: The collected data is sent to the central processing system in real time through the IoT transmission device; G306: Each sensor in the sensor network collects data according to a preset sampling frequency and time interval, and the collected data is transmitted to a data aggregation node of the sensor network by wireless means; G307: The data aggregation node packages the collected data and sends it to the central processing system in real time through the IoT transmission device; G308: The central processing system receives data from drones and sensor networks in real time; G309: Clean the received data to remove noise, outliers and missing values, and convert the cleaned data into a unified format and timestamp; G310: Spatial alignment and timestamp synchronization of data from different data sources; G311: Use weighted average fusion algorithm to fuse different types of data; G312: Store the fused data in a real-time database in real time.

[0021] Furthermore, in the H step, model verification and evaluation includes the following sub-steps: H1. Select multiple field exploration locations within the target mining area, then conduct drilling and sample collection to obtain actual data; H2. Compare the field exploration data with the model prediction results one by one, record the predicted results and actual results of each sample point, and then compare the predicted results with the actual results to evaluate the accuracy and reliability of the model.

[0022] Furthermore, in the H2 step, the accuracy and reliability of the evaluation model are specifically divided into the following steps: H21: Drilling and sample collection are carried out according to the selected field exploration location to obtain actual data, including geological structure, ore body distribution, and ore grade; H22: Extract the corresponding prediction data from the 3D geological model, including ore body distribution and ore grade prediction results; H23: For each field exploration location, compare the model prediction data with the field exploration data one by one; record the prediction results and actual results of each sample point, including the existence of ore bodies and the difference in ore grades; H24: According to the comparison results, calculate the error indicators, including absolute error, relative error, and mean square error; H25: According to actual needs and application scenarios, set the accuracy threshold, including the error range and the maximum allowable error; if the error of the model prediction result is within the set threshold range, the model is considered to have high accuracy; H26: Count the proportion of sample points that meet the accuracy threshold, that is, the proportion of accurately predicted sample points to the total sample points. This proportion reflects the overall accuracy of the model. H27: Analyze the distribution of error indicators, including the mean, standard deviation, maximum and minimum values ​​of the error; H28: By repeating the comparison and evaluation process on different validation sets, the generalization ability and consistency of the model are evaluated; if the model can show high accuracy and reliability on different validation sets, the model is considered to have good generalization ability; H29: Based on the comparison and evaluation results, write a detailed evaluation report, including data preparation, comparison process, accuracy assessment, and reliability assessment.

[0023] Beneficial effects of the present invention: 1. The present invention uses an adaptive cube grid division method to dynamically adjust the size and shape of the cube according to the changes in geological characteristics and data density. This flexible grid division strategy can more effectively capture complex geological structures, thereby significantly improving the recognition accuracy of ore body characteristics, and further improving the overall reliability of ore body distribution prediction, which can not only reduce resource waste, but also optimize exploration efficiency; 2. The present invention combines the convolutional neural network model with multi-source data analysis to automatically extract data features, which can not only adapt to the complex geological conditions of different mining areas, but also avoid the limitations of traditional machine learning algorithms that require manual feature extraction; and this automated feature learning capability improves the model's adaptability to new data, so that the model can still maintain efficient prediction performance in unknown or changing environments; 3. Through dynamic real-time monitoring and model updating, the present invention can quickly feed back the real-time acquired data to the three-dimensional model, so that the three-dimensional model can continuously reflect the latest geological conditions. This feature not only improves the timeliness of the prediction, but also provides support for real-time decision-making, enhancing the flexibility and efficiency of ore exploration.

[0024] In summary, the present invention has the characteristics of high prediction accuracy and timeliness, strong adaptability and high exploration efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the present invention; In the figure: A-data collection, B-data preprocessing, C-3D geological model construction, D-cube grid division, E-ore body characteristic analysis, F-ore body distribution prediction, G-dynamic real-time monitoring and model updating, H-model verification and evaluation. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] like Figure 1 As shown, the present invention includes data collection, data preprocessing, three-dimensional geological model construction, cube grid division, ore body characteristic analysis, ore body distribution prediction, dynamic real-time monitoring and model updating, model verification and evaluation steps, and the specific contents are: A. Data collection: Collect geological, geophysical, geochemical and / or remote sensing data of the target mining area; B. Data preprocessing: cleaning, normalization and standardization of the collected data; C. Construction of 3D geological model: Based on the pre-processed data, a 3D geological model of the target mining area is constructed using GIS or CAD technology; D. Cube meshing: Divide the aforementioned three-dimensional geological model into multiple cube units, and use an adaptive meshing method to dynamically adjust the size and shape of the cube units according to geological characteristics and data density; E. Ore body feature analysis: Analyze and identify the geological features within the aforementioned cubic unit, and use multi-source data fusion technology to improve the accuracy of feature identification; F. Ore body distribution prediction: Based on the features identified by the aforementioned cubic units, a deep learning model is trained using a convolutional neural network. The deep learning model is then used to automatically extract data features and predict ore body distribution. G. Dynamic real-time monitoring and model updating: Deploy drones and sensor networks for dynamic real-time monitoring, obtain field data in real time, and feed the acquired data back to the 3D model for updating; H. Model verification and evaluation: Evaluate the accuracy and reliability of the model by comparing it with the field exploration results.

[0028] In step A, the geophysical data of the target mining area is collected by obtaining underground data using drilling, gravity, electromagnetic and / or seismic methods, and the remote sensing data of the target mining area is collected by identifying and analyzing surface features in combination with remote sensing technology.

[0029] In step B, data cleaning is to use statistical methods to eliminate outliers and noise in the collected values, and normalization and standardization processing is to use a standardized formula to standardize different data to form a unified data format and range. The standardized formula is: , Where: Z is the standardized value, X is the sample value, m is the sample mean, s is the sample standard deviation.

[0030] In the step C, constructing a three-dimensional geological model includes the following sub-steps: C1. Data input: import the collected data after preprocessing into GIS or CAD system; C2. Model drawing: Generate a three-dimensional geological model of the mining area through three-dimensional modeling tools.

[0031] In the step D, the adaptive grid division method specifically includes the following sub-steps: D1. Wavelet transform: Use wavelet transform to analyze the local features of data and extract feature information at different scales; D2. Machine learning optimization: Combine the random forest algorithm to optimize the division of cube units, build and train the random forest model, and then use the random forest model to evaluate the fitness of different division schemes to select the optimal solution; The random forest model is established by the following steps: D21, data preparation: extract feature data from cube cells, construct feature matrix X and target variable Y, and then divide 70-80% of the data in the data set into training set and the remaining data into test set; D22, construction and training: In the process of building each tree in the forest, data samples are randomly extracted, and features are randomly selected at each node for splitting. The number of trees in the forest is set to 80 to 100, and then the training set is used to train the random forest model. Each tree is trained independently to generate the final forest model. D23, Model evaluation: Use accuracy, recall and / or F1-score indicators to evaluate the performance of the final forest model on the test set for the fitness of different partitioning schemes.

[0032] In the step D23, evaluating the performance of the forest model includes the following sub-steps: D231, Set thresholds: Set one or more thresholds in advance. If the accuracy, recall and / or F1-score of a solution is lower than the set threshold, the solution is considered unacceptable. D232, Priority sorting: Prioritize each solution based on its accuracy, recall rate and / or F1-score, and select the solution with the best fitness index; D233. Cross-validation: Repeat the above steps on different validation sets to ensure that the selected scheme has good generalization ability and consistency.

[0033] The wavelet transform in step D1 extracts feature information at different scales for optimizing grid division. The feature information is used as input feature of the random forest algorithm in step D2. The feature information is used by the random forest model to evaluate the fitness of different cubic unit divisions, thereby selecting the optimal grid division scheme. The feature information includes local anomalies and change trends of the data.

[0034] The wavelet transform in step D1 is used in step E and step F for ore body feature analysis and prediction, specifically including: D11, the wavelet transform in step D1 In step E, in the ore body feature analysis stage, the feature information extracted by wavelet transform is used as an important basis for identifying and analyzing geological features in the cubic unit, and the geological features in the cubic unit are more accurately identified and analyzed; D12. In the wavelet transform in step D1, in step F, the feature information extracted in step D1 is used as the input feature of CNN to train a more accurate deep learning model for ore body distribution prediction.

[0035] In the step D2, the random forest algorithm is combined to optimize the division of the cube unit, and a random forest model is established and trained. Then, the random forest model is used to evaluate the fitness of different division schemes to select the optimal scheme. The specific division optimization steps are as follows: D2A: Based on the fitness evaluation results, the current cube grid division scheme is preliminarily evaluated. The evaluation indicators include accuracy, recall rate, and F1-score to reflect the model's ability to predict mineralization potential. By comparing the evaluation results under different division schemes, the optimization direction and goal are determined; D2B: Based on the preliminary evaluation results, if the model performs poorly on smaller or larger cube units, adjust the size of the grid; if the model overfits on smaller cube units, increase the grid size appropriately; if the model underfits on larger cube units, reduce the grid size appropriately; D2C: Adjust the shape of the grid according to the spatial distribution of geological features and mineralization laws; D2D: According to the fitness evaluation results, the features are ranked in importance, and the features that contribute most to mineralization prediction are selected for extraction and use; D2E: During the optimization of the partitioning scheme, the hyperparameters are adjusted according to the fitness evaluation results. If the model performs well on the training set but poorly on the test set, the number of trees or the maximum depth is appropriately reduced; D2F: After adjusting the grid size, shape, feature selection, and hyperparameters, the random forest model is retrained and fitness evaluation is performed. By comparing the evaluation results before and after optimization, it can be determined whether the partitioning scheme optimization is effective. If the optimization effect is not obvious or the evaluation results deteriorate, the optimization strategy is readjusted.

[0036] In the step D21, the relationship between extracting feature data from the cube unit and the step D1 is as follows: In step D1, wavelet transform is used to analyze local features of data and extract feature information at different scales to reveal implicit patterns and structures in the data, providing a basis for subsequent analysis and modeling. In step D21, feature data is extracted from cube units, which is a further refinement on the basis of step D1, and these feature information are associated with specific cube units. The feature information extracted in step D1 includes geological structure, lithology distribution, and mineralized zones. These features exist in the form of spatial distribution in the three-dimensional geological model. In step D21, by matching these features with cube units, a feature description of each cube unit is formed, so that each cube unit contains a set of specific geological features, which are used as input for subsequent machine learning models.

[0037] In the step D21, constructing the feature matrix X and the target variable Y includes the following sub-steps: D211: Select the features that have an impact on the prediction of ore body distribution, including geological structure type, lithology, mineralization intensity, topography and geomorphology; D212: For each cube unit, extract its specific value or category on the above selected features; these values ​​or categories will constitute each row (or column) of the feature matrix X; D213: Organize the extracted features into a matrix in the order of cube units, where each row (or column) of the matrix represents the feature vector of a cube unit; D214: Target definition: In mineralization prediction, the target variable Y represents the presence or absence of an ore body or a certain property of the ore body; D215: Label each cubic unit to determine whether it contains an ore body or the specific attributes of the ore body; D216: Organize the labeling results into vector form according to the order of cube units (corresponding to the feature matrix X) to form the target variable Y.

[0038] In the step E, the specific steps are as follows: E1: Logarithmic transformation and power transformation of data according to the nature and distribution of features; E2: Perform descriptive statistical analysis on the feature data and calculate its statistics including mean and variance; at the same time, calculate the correlation coefficient between the feature data to obtain the degree of linear relationship between the features; E4: Use geological interpretation methods to interpret the features geologically; E5: Use Bayesian data fusion method to fuse data from different sources and assign appropriate weights to each type of data based on its characteristics and reliability; E6: Analyze and interpret the fused data to reveal the internal connections and patterns among different data; E7: Based on the fused data and geological feature analysis results, the support vector machine algorithm is used to identify the ore body characteristics; E8: Further analysis and processing of the identified features, including feature selection and feature dimensionality reduction; E9: Use neural network data processing methods to deeply mine and extract features, extracting hidden features and patterns; E10: Output the identified and enhanced features as feature vectors and feature maps for use in subsequent ore body distribution prediction models.

[0039] In the step F, a deep learning model is trained using a convolutional neural network, and then the deep learning model is applied to automatically extract data features; wherein the input layer of the convolutional neural network receives the preprocessed cube unit data, and the output layer applies the convolution operation to extract features, and the output layer convolution formula is: , Where: f is the input feature map, g is the convolution kernel, x and y is the coordinate of the output feature map, m and n are the coordinates of the convolution kernel.

[0040] In the F step, the sub-steps for predicting the ore body distribution are as follows: F1: receives the preprocessed cube unit data through the input layer, and the received data is converted into a multidimensional array as input; F2: The convolution layer performs convolution operations on the input data through multiple convolution kernels to extract local features. The convolution operation in this step uses the convolution kernel to slide on the input data for weighted summation and activation. F3: Apply nonlinear activation function ReLU after the convolutional layer to increase the expressiveness of the model; F4: Reduce the dimension of the feature map through pooling operation, reduce the amount of calculation and extract the main features; F5: Flattens the features extracted by the convolutional layer and the pooling layer and inputs them into the fully connected layer for the final classification and regression tasks; F6: Input the preprocessed cube unit data into the trained CNN model and automatically extract data features through the convolution layer; F7: Input the extracted features into the prediction layer of the fully connected layer of the model; F8: In the prediction layer of the fully connected layer, the prediction results of the ore body distribution are obtained by calculation, including the location, scale, and morphological information of the ore body.

[0041] In the G step, dynamic real-time monitoring and model updating includes the following sub-steps: G1. Drone configuration: Use multiple drones equipped with high-resolution cameras, lidar and / or multispectral sensors to obtain detailed information on the surface and underground of the target mining area; G2. Sensor network deployment: Set up a sensor network in the target mining area to monitor soil moisture, temperature, pH value and / or conductivity in real time; G3. Real-time data collection: drones and sensors transmit data to the central processing system in real time through the Internet of Things technology for preliminary data processing and fusion; G4, Model update: The initially processed and integrated data is fed back in real time, then applied to the 3D geological model through edge computing algorithms, and the model parameters are updated to reflect the latest geological conditions.

[0042] The specific process of step G3 is as follows: G301: Configure the parameters of the drone and sensor network transmission equipment according to actual needs, including transmission rate, frequency, and encryption method; G302: Establish communication links between drones, sensor networks, and central processing systems; G303: Start the drone and fly it according to the preset flight route and altitude; G304: The high-resolution camera, lidar, and multispectral sensor on the drone start to work, collecting images, distance, and spectral data on the surface and underground in real time; G305: The collected data is sent to the central processing system in real time through the IoT transmission device; G306: Each sensor in the sensor network collects data according to a preset sampling frequency and time interval, and the collected data is transmitted to a data aggregation node of the sensor network by wireless means; G307: The data aggregation node packages the collected data and sends it to the central processing system in real time through the IoT transmission device; G308: The central processing system receives data from drones and sensor networks in real time; G309: Clean the received data to remove noise, outliers and missing values, and convert the cleaned data into a unified format and timestamp; G310: Spatial alignment and timestamp synchronization of data from different data sources; G311: Use weighted average fusion algorithm to fuse different types of data; G312: Store the fused data in a real-time database in real time.

[0043] In the H step, model verification and evaluation includes the following sub-steps: H1. Select multiple field exploration locations within the target mining area, then conduct drilling and sample collection to obtain actual data; H2. Compare the field exploration data with the model prediction results one by one, record the predicted results and actual results of each sample point, and then compare the predicted results with the actual results to evaluate the accuracy and reliability of the model.

[0044] In the H2 step, the specific process of evaluating the accuracy and reliability of the model is as follows: H21: Drilling and sample collection are carried out according to the selected field exploration location to obtain actual data, including geological structure, ore body distribution, and ore grade; H22: Extract the corresponding prediction data from the 3D geological model, including ore body distribution and ore grade prediction results; H23: For each field exploration location, compare the model prediction data with the field exploration data one by one; record the prediction results and actual results of each sample point, including the existence of ore bodies and the difference in ore grades; H24: According to the comparison results, calculate the error indicators, including absolute error, relative error, and mean square error; H25: According to actual needs and application scenarios, set the accuracy threshold, including the error range and the maximum allowable error; if the error of the model prediction result is within the set threshold range, the model is considered to have high accuracy; H26: Count the proportion of sample points that meet the accuracy threshold, that is, the proportion of accurately predicted sample points to the total sample points. This proportion reflects the overall accuracy of the model. H27: Analyze the distribution of error indicators, including the mean, standard deviation, maximum and minimum values ​​of the error; H28: By repeating the comparison and evaluation process on different validation sets, the generalization ability and consistency of the model are evaluated; if the model can show high accuracy and reliability on different validation sets, the model is considered to have good generalization ability; H29: Based on the comparison and evaluation results, write a detailed evaluation report, including data preparation, comparison process, accuracy assessment, and reliability assessment.

[0045] Example 1: The present invention's 3D-based cube prediction model mineralization prediction method S100: Collect geological, geophysical, geochemical and / or remote sensing data of the target mining area. The geophysical data of the target mining area is collected by drilling, gravity, electromagnetic and / or seismic methods to obtain underground data, and the remote sensing data of the target mining area is collected by combining remote sensing technology to identify and analyze surface features.

[0046] S200: Clean, normalize and standardize the collected data. Data cleaning is to use statistical methods to eliminate abnormal values ​​and noise in the collected values, and normalization and standardization are to use a standardization formula to standardize different data to form a unified data format and range. The standardization formula is: , Where: Z is the standardized value, X is the sample value, μ is the sample mean, and σ is the sample standard deviation.

[0047] S300: Based on the pre-processed data, a 3D geological model of the target mining area is constructed using GIS or CAD technology. The construction of the 3D geological model includes the following steps: S310: Import the pre-processed collected data into a GIS or CAD system.

[0048] S320: Generate a three-dimensional geological model of the mining area using a three-dimensional modeling tool.

[0049] S400: Divide the three-dimensional geological model into a plurality of cubic units, and use an adaptive meshing method to dynamically adjust the size and shape of the cubic units according to geological characteristics and data density. The adaptive meshing method specifically includes the following sub-steps: S410: Analyze local features of data using wavelet transform to extract feature information at different scales.

[0050] S420: Optimize the division of cube units by combining the random forest algorithm, establish and train the random forest model, and then use the random forest model to evaluate the fitness of different division schemes to select the optimal scheme.

[0051] The random forest model is established by the following steps: S421, extracting characteristic data from the cube unit, constructing characteristic matrix X and target variable Y, dividing 70-80% of the data in the data set into a training set and the remaining data into a test set. The characteristic data includes geological characteristics and physical characteristics, the geological characteristics include lithology and porosity, and the physical characteristics include density and magnetism.

[0052] In the above step S421, constructing the feature matrix X and the target variable Y includes the following sub-steps: S4211: Select the features that have an impact on the prediction of ore body distribution, including geological structure type, lithology, mineralization intensity, and topography.

[0053] S4212: For each cubic unit, extract its specific value or category on the above selected features; these values ​​or categories will constitute each row or column of the feature matrix X.

[0054] S4213: Organize the extracted features into a matrix form according to the order of the cube units, and each row (or column) of the matrix represents a feature vector of a cube unit.

[0055] S4214: Target definition: In mineralization prediction, the target variable Y represents the existence or non-existence of an ore body or some property of the ore body.

[0056] S4215: Label each cubic unit to determine whether it contains an ore body or specific attributes of the ore body.

[0057] S4216: Organize the labeling results into a vector form according to the order of the cube units (corresponding to the feature matrix X) to form the target variable Y.

[0058] S422. During the construction of each tree in the forest, data samples are randomly extracted, and features are randomly selected at each node for splitting. The number of trees in the forest is set to 80 to 100. The training set is then used to train the random forest model. Each tree is trained independently to generate the final forest model.

[0059] S423, using accuracy, recall and / or F1-score indicators, the final forest model is used to evaluate the fitness of different partitioning schemes on the test set.

[0060] Among them, evaluating the performance of the forest model includes the following sub-steps: S4231: Set one or more thresholds in advance, for example, set a minimum accuracy standard. If the accuracy, recall and / or F1-score of a solution is lower than the set threshold, the solution is considered unacceptable.

[0061] S4232: Prioritize each solution according to its accuracy, recall rate and / or F1-score, and select the solution with the best fitness index.

[0062] S4233: Repeat the above steps on different validation sets to ensure that the selected solution has good generalization ability and consistency.

[0063] It should be noted that the wavelet transform in step S410 extracts feature information at different scales for optimizing grid division. The feature information is used as input feature of the random forest algorithm in step S420. The feature information is used by the random forest model to evaluate the fitness of different cubic unit divisions, thereby selecting the optimal grid division scheme. The feature information includes local anomalies and change trends of the data.

[0064] S500: Analyze and identify the geological features in the aforementioned cubic unit, and use multi-source data fusion technology to improve the accuracy of feature recognition. The specific steps are as follows: S505: Perform logarithmic transformation and power transformation on the data according to the nature and distribution of the features.

[0065] S510: Perform descriptive statistical analysis on the feature data to calculate the statistics including the mean and variance; and calculate the correlation coefficient between the feature data to obtain the degree of linear relationship between the features.

[0066] S515: Use geological interpretation methods to interpret features geologically.

[0067] S520: Use the Bayesian data fusion method to fuse data from different sources and assign appropriate weights to each type of data based on the characteristics and reliability of the data.

[0068] S525: Analyze and interpret the fused data to reveal the intrinsic connections and patterns between different data.

[0069] S530: Based on the fused data and geological feature analysis results, the support vector machine algorithm is used to identify the ore body characteristics.

[0070] S535: Further analyzing and processing the identified features, including feature selection and feature dimensionality reduction.

[0071] S540: Use neural network data processing methods to deeply mine and extract features to extract hidden features and patterns.

[0072] S545: Output the identified and enhanced features as feature vectors and feature maps for use in subsequent ore body distribution prediction models.

[0073] S600: Based on the features identified by the aforementioned cubic units, a deep learning model is trained using a convolutional neural network, and then the deep learning model is used to automatically extract data features and predict the distribution of the ore body. Specifically, a deep learning model is trained using a convolutional neural network, and then the deep learning model is used to automatically extract data features; wherein the input layer of the convolutional neural network receives preprocessed cubic unit data, and the output layer extracts features by applying a convolution operation, and the output layer convolution formula is: , Where: f is the input feature map, g is the convolution kernel, x and y is the coordinate of the output feature map, m and n are the coordinates of the convolution kernel.

[0074] Among them, in step S600, the sub-steps of predicting the ore body distribution are as follows: S610: receiving the preprocessed cube unit data through the input layer, and the received data is converted into a multidimensional array as input.

[0075] S620: The convolution layer performs a convolution operation on the input data through multiple convolution kernels to extract local features. The convolution operation in this step uses the convolution kernel to slide on the input data to perform weighted summation and activation.

[0076] S630: Apply the nonlinear activation function ReLU after the convolutional layer to increase the expressive power of the model.

[0077] S640: Reduce the dimension of the feature map through pooling operation, reduce the amount of calculation and extract the main features.

[0078] S650: Flatten the features extracted by the convolutional layer and the pooling layer and input them into the fully connected layer for the final classification and regression tasks.

[0079] S660: Input the preprocessed cube unit data into the trained CNN model and automatically extract data features through the convolution layer.

[0080] S670: Input the extracted features into the prediction layer of the fully connected layer of the model.

[0081] S680: In the prediction layer of the fully connected layer, the prediction results of the ore body distribution are obtained by calculation, including the location, scale, and morphological information of the ore body.

[0082] S700: Deploy drones and sensor networks for dynamic real-time monitoring, obtain field data in real time, and feed the acquired data back to the 3D model for updating. Specifically, the following steps are included: S710: Use multiple drones equipped with high-resolution cameras, LiDAR and / or multispectral sensors to obtain detailed information on the surface and underground of target mining areas.

[0083] S720: Set up a sensor network in the target mining area to monitor soil moisture, temperature, pH value and / or conductivity in real time.

[0084] S730: Drones and sensors transmit data to the central processing system in real time through the Internet of Things technology for preliminary data processing and fusion.

[0085] S740: The preliminarily processed and fused data is fed back in real time and then applied to the 3D geological model through edge computing algorithms, and the model parameters are updated to reflect the latest geological conditions.

[0086] The specific process of step S730 is as follows: S7301: Configure the parameters of the drone and sensor network transmission equipment according to actual needs, including transmission rate, frequency, and encryption method.

[0087] S7302: Establish communication links between drones, sensor networks, and central processing systems.

[0088] S7303: Start the drone and fly it according to the preset flight route and altitude.

[0089] S7304: The high-resolution camera, lidar and multispectral sensor on the drone start working to collect images, distance and spectral data of the surface and underground in real time.

[0090] S7305: The collected data is sent to the central processing system in real time through the Internet of Things transmission device.

[0091] S7306: Each sensor in the sensor network collects data according to a preset sampling frequency and time interval, and the collected data is transmitted to a data aggregation node of the sensor network via wireless means.

[0092] S7307: The data aggregation node packages the collected data and sends it to the central processing system in real time through the Internet of Things transmission device.

[0093] S7308: The central processing system receives data from drones and sensor networks in real time.

[0094] S7309: Clean the received data to remove noise, outliers and missing values, and convert the cleaned data into a unified format and timestamp.

[0095] S7310: Spatial alignment and timestamp synchronization of data from different data sources.

[0096] S7311: Use weighted average fusion algorithm to fuse different types of data.

[0097] S7312: Store the fused data in a real-time database in real time.

[0098] S800: Evaluate the accuracy and reliability of the model by comparing it with the field exploration results. This includes the following sub-steps: S810: Select multiple field exploration locations within the target mining area, and then conduct drilling and sample collection to obtain actual data.

[0099] S820: Compare the field exploration data with the model prediction results one by one, record the prediction results and actual results of each sample point, and then compare the prediction results with the actual results to evaluate the accuracy and reliability of the model.

[0100] The specific process of evaluating the accuracy and reliability of the model in step S820 is as follows: S821: Drilling and sample collection are carried out according to the selected field exploration location to obtain actual data, including geological structure, ore body distribution, and ore grade.

[0101] S822: Extract corresponding prediction data from the three-dimensional geological model, including ore body distribution and ore grade prediction results.

[0102] S823: For each field exploration location, the model prediction data is compared with the field exploration data one by one; the prediction results and actual results of each sample point are recorded, including the existence of the ore body and the difference in ore grade.

[0103] S824: Calculate error indicators based on the comparison results, including absolute error, relative error, and mean square error.

[0104] S825: According to actual needs and application scenarios, set accuracy thresholds, including error range and maximum allowable error; if the error of the model prediction result is within the set threshold range, the model is considered to have high accuracy.

[0105] S826: Count the proportion of sample points that meet the accuracy threshold, that is, the proportion of accurately predicted sample points to the total sample points. This proportion reflects the overall accuracy of the model.

[0106] S827: Analyze the distribution of error indicators, including the mean, standard deviation, maximum and minimum values ​​of the error.

[0107] S828: Evaluate the generalization ability and consistency of the model by repeating the comparison and evaluation process on different validation sets; if the model can show high accuracy and reliability on different validation sets, it is considered that the model has good generalization ability.

[0108] S829: Based on the comparison and evaluation results, write a detailed evaluation report, including data preparation, comparison process, accuracy assessment, and reliability assessment.

[0109] Example 2: Model-driven mesh optimization S100, Data Collection: Collect multi-source data, including geological, geophysical and geochemical information.

[0110] S200, data preprocessing: performing outlier elimination and data standardization.

[0111] S300, 3D geological model construction: Establish a preliminary 3D geological model.

[0112] S400, Cube Meshing: Perform preliminary meshing, use finite element analysis to simulate stress distribution under different geological conditions, optimize mesh shape, and adjust meshing strategy in combination with machine learning feedback mechanism.

[0113] S500, Ore body characteristic analysis and prediction: Apply deep learning models to predict ore bodies.

[0114] S600, Dynamic real-time monitoring and model updating: Implement real-time monitoring and feed back updates to the model.

[0115] S700, Model Validation and Evaluation: Evaluate model performance by comparing with field data.

[0116] Summary: This example emphasizes the model-driven mesh optimization strategy that combines finite element analysis and machine learning to ensure the accuracy and adaptability of meshing under complex geological conditions.

[0117] Example 3: Graph Neural Network Combination S100, data collection: collecting data and remote sensing data of the target mining area.

[0118] S200, data preprocessing: data cleaning and standardization.

[0119] S300, three-dimensional geological model construction: generate a three-dimensional model of the mining area.

[0120] S400, cube meshing: perform adaptive meshing and extract local features.

[0121] S500, Ore body feature analysis: The geological body and its features are represented as a graph structure, with nodes as feature information and edges as relationships between features. Graph neural networks (GNNs) are used to capture the unstructured features of the data.

[0122] S600, Ore body distribution prediction: Combine convolutional neural network and graph neural network to predict ore body distribution.

[0123] S700, dynamic real-time monitoring and model updating: real-time monitoring data is fed back to the model.

[0124] S800, Model verification and evaluation: Verify the effectiveness of the model through field exploration data.

[0125] Summary: By introducing graph neural networks, this embodiment can more effectively capture the relationship between unstructured features in the data and improve the accuracy of ore body prediction.

[0126] Example 4: Adaptive Feedback Control System S100, Data Collection: Collect mining area data using a variety of geophysical survey techniques.

[0127] S200, data preprocessing: cleaning and standardization.

[0128] S300, constructing a three-dimensional geological model: constructing a three-dimensional model of the target mining area.

[0129] S400, Cube Meshing: Perform adaptive meshing.

[0130] S500, Ore body characteristic analysis and prediction: Analyze the ore body characteristics and implement deep learning models to predict ore body distribution.

[0131] S600, Dynamic real-time monitoring and model updating: Deploy drones and sensors to achieve real-time data collection and monitoring, real-time decision-making algorithms, and adjust exploration strategies based on data feedback.

[0132] S700, Model Validation and Evaluation: Evaluate the accuracy and reliability of the model by comparing it with field survey results.

[0133] Summary: This embodiment uses an adaptive feedback control system to dynamically adjust the exploration strategy and model parameters according to real-time monitoring data, ensuring that the model always remains in the best state, thereby improving the flexibility and efficiency of exploration.

[0134] Summary of Examples: The above Examples 1 to 4 are designed around adaptive cube meshing, model-driven mesh optimization, graph neural network combination, and adaptive feedback control system. Each example demonstrates different technical means and processes, emphasizing how to combine advanced algorithms and technologies to improve the accuracy and efficiency of prospecting. These innovative methods not only enrich the overall prospecting strategy, but also provide practical solutions for practical applications, ensuring effectiveness and reliability under complex geological conditions.

[0135] Test experiment 1 Purpose: The effects of different implementation examples in ore body prediction were compared, and the accuracy, recall rate and F1-score of each method were verified.

[0136] Experimental steps: S100. Select the test area: select a mining area with known ore body distribution to determine the actual location of the ore body.

[0137] S200, Data Collection: Collect various data of the target area, including geological, geophysical, geochemical and remote sensing data.

[0138] Embodiment implementation: Example 1: Basic solution Example 2: Model-driven mesh optimization Example 3: Graph Neural Network Combination Example 4: Adaptive Feedback Control System S300, model training and prediction: applying corresponding methods to each embodiment to predict ore body distribution.

[0139] S400, result recording: recording the actual ore body position and the predicted ore body position of each embodiment, and calculating the performance index.

[0140] S500, data analysis: analyzing the performance of each embodiment and comparing the results.

[0141] Performance indicators: accuracy, recall and F1-score; The experimental data are shown in Table 1.

[0142] Table 1 Experimental data of test experiment 1 Result analysis: 1. Accuracy: Example 1 performed best, reaching 72.92%.

[0143] 2. Recall rate: The recall rate of Example 1 is also the highest, which is 70%.

[0144] 3. F1-score: The F1-score of Example 1 is 0.7145, indicating that it performs best in terms of comprehensive consideration of precision and recall.

[0145] Summary: Through this test experiment, we can intuitively see the performance differences of different implementation examples in ore body prediction, providing a basis for subsequent improvement and optimization.

[0146] Comparative experiment 1: Comparison between traditional meshing method and adaptive meshing method Existing technical solutions: Fixed grid division: The entire mining area is divided into cube grids of uniform size, ignoring the differences in geological characteristics and data density, and the dynamic adjustment of local characteristics is not considered during data processing.

[0147] Technical solution of the present invention: Adaptive meshing: Dynamically adjust the size and shape of the cube based on geological characteristics and data density.

[0148] Experimental steps: S100, Data Collection: Collect geological, geophysical, geochemical and remote sensing data of the target mining area.

[0149] S200, meshing: using fixed meshing and adaptive meshing methods respectively.

[0150] S300, Model Training and Prediction: Apply deep learning models to each method for ore prediction.

[0151] S400. Result verification: The prediction effects of the two methods are compared as shown in Table 2.

[0152] Table 2 Comparison of the prediction results of Experiment 1 Summary: Adaptive grid division technology effectively improves the ability to capture complex geological features and optimizes the prediction results. The limitation of the fixed grid division method is that it cannot adapt to changes in geological characteristics, resulting in limited prediction results.

[0153] Comparative Experiment 2: Comparison between traditional machine learning and deep learning models Existing technical solutions: Traditional methods: Classic machine learning models such as support vector machines (SVM) or random forests, features need to be extracted manually, and the accuracy of the model depends on the quality of feature selection.

[0154] The technical solution of the present invention is to use a convolutional neural network (CNN) to automatically extract and predict features.

[0155] Experimental steps: S100. Data collection: Use the same dataset.

[0156] S200, Feature Analysis: Ore body prediction using traditional machine learning methods and deep learning models.

[0157] S300. Result verification: Comparison of the prediction effects of the two models is shown in Table 3.

[0158] Table 3 Comparison of the prediction results of Experiment 2 Summary: Deep learning models show higher accuracy and stronger feature capture capabilities than traditional machine learning methods, especially when dealing with nonlinear and high-dimensional data. Traditional machine learning relies heavily on feature selection, while deep learning overcomes this limitation through hierarchical feature learning.

[0159] Comparative experiment 3: Comparison between traditional monitoring and dynamic monitoring Existing technical solutions: regular monitoring, delayed data feedback, untimely model updates, and no consideration of the impact of real-time changes on ore body predictions.

[0160] The technical solution of the present invention is to use drones and sensor networks to perform real-time monitoring, obtain data in real time and update the model.

[0161] Experimental steps: S100, data collection: select the same mining area for comparison.

[0162] S200, monitoring plan: use traditional periodic monitoring and dynamic real-time monitoring respectively.

[0163] S300, model update: the model is updated after each monitoring.

[0164] S400. Result verification: The prediction effects of the two monitoring methods are compared as shown in Table 4.

[0165] Table 4 Comparison of the prediction results of Experiment 3 Summary: Dynamic monitoring technology significantly improves the adaptability and flexibility of the model, enabling it to respond more quickly to changes in the geological environment. The limitation of traditional monitoring methods is that their data feedback is delayed and they fail to effectively capture rapidly changing geological characteristics.

[0166] Comprehensive summary of comparative experiments: Through three comparative experiments, it can be clearly seen that the mineralization prediction method of the cube prediction model based on three-dimensional modeling of the present invention has advantages in various aspects: First, the use of adaptive meshing improves the accuracy of ore body prediction and adapts to different geological conditions.

[0167] Secondly, the deep learning model demonstrates stronger automatic feature extraction capabilities, which improves the prediction effect and significantly improves the accuracy compared to traditional machine learning methods.

[0168] Finally, the dynamic monitoring and real-time feedback mechanism can update the model in a timely manner, significantly improving the timeliness and accuracy of the prediction, and proving its effectiveness under complex geological conditions.

[0169] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A cubic prediction model mineralization prediction method based on three-dimensional modeling, characterized in that It includes data collection, data preprocessing, 3D geological model construction, cube grid division, ore body characteristic analysis, ore body distribution prediction, dynamic real-time monitoring and model updating, model verification and evaluation steps. The specific contents are as follows: A. Data collection: Collect geological, geophysical, geochemical and / or remote sensing data of the target mining area; B. Data preprocessing: cleaning, normalization and standardization of the collected data; C. Construction of 3D geological model: Based on the pre-processed data, a 3D geological model of the target mining area is constructed using GIS or CAD technology; D. Cube meshing: Divide the aforementioned three-dimensional geological model into multiple cube units, and use an adaptive meshing method to dynamically adjust the size and shape of the cube units according to geological characteristics and data density; E. Ore body feature analysis: Analyze and identify the geological features within the aforementioned cubic unit, and use multi-source data fusion technology to improve the accuracy of feature identification; F. Ore body distribution prediction: Based on the features identified by the aforementioned cubic units, a deep learning model is trained using a convolutional neural network. The deep learning model is then used to automatically extract data features and predict ore body distribution. G. Dynamic real-time monitoring and model updating: Deploy drones and sensor networks for dynamic real-time monitoring, obtain field data in real time, and feed the acquired data back to the 3D model for updating; H. Model verification and evaluation: Evaluate the accuracy and reliability of the model by comparing it with the field exploration results.

2. The cubic prediction model mineralization prediction method based on three-dimensional modeling according to claim 1 is characterized in that: In step B, data cleaning is to use statistical methods to eliminate outliers and noise in the collected values, and normalization and standardization processing is to use a standardized formula to standardize different data to form a unified data format and range. The standardized formula is: ， Where: Z is the standardized value, X is the sample value, μ is the sample mean, σ is the sample standard deviation.

3. The cubic prediction model mineralization prediction method based on three-dimensional modeling according to claim 1 is characterized in that: In the step C, constructing a three-dimensional geological model includes the following sub-steps: C1. Data input: import the collected data after preprocessing into GIS or CAD system; C2. Model drawing: Generate a three-dimensional geological model of the mining area through three-dimensional modeling tools.

4. The ore-forming prediction method based on the cube prediction model of three-dimensional modeling according to claim 1 is characterized in that: In the step D, the adaptive grid division method specifically includes the following sub-steps: D1. Wavelet transform: Use wavelet transform to analyze the local features of data and extract feature information at different scales; D2. Machine learning optimization: Combine the random forest algorithm to optimize the division of cube units, build and train the random forest model, and then use the random forest model to evaluate the fitness of different division schemes to select the optimal solution; The random forest model is established by the following steps: D21, data preparation: extract feature data from cube cells, construct feature matrix X and target variable Y, and then divide 70-80% of the data in the data set into training set and the remaining data into test set; D22, construction and training: In the process of building each tree in the forest, data samples are randomly extracted, and features are randomly selected at each node for splitting. The number of trees in the forest is set to 80 to 100, and then the training set is used to train the random forest model. Each tree is trained independently to generate the final forest model. D23, Model evaluation: Use accuracy, recall and / or F1-score indicators to evaluate the performance of the final forest model on the test set for the fitness of different partitioning schemes.

5. The cube prediction model mineralization prediction method based on three-dimensional modeling according to claim 3 is characterized in that: In the step D23, evaluating the performance of the forest model includes the following sub-steps: D231, Set thresholds: Set one or more thresholds in advance. If the accuracy, recall and / or F1-score of a solution is lower than the set threshold, the solution is considered unacceptable. D232, Priority sorting: Prioritize each solution based on its accuracy, recall rate and / or F1-score, and select the solution with the best fitness index; D233. Cross-validation: Repeat the above steps on different validation sets to ensure that the selected scheme has good generalization ability and consistency.

6. The ore-forming prediction method based on the cube prediction model of three-dimensional modeling according to claim 1 is characterized by: In the step E, the specific steps are as follows: E1: Logarithmic transformation and power transformation of data according to the nature and distribution of features; E2: Perform descriptive statistical analysis on the feature data and calculate its statistics including mean and variance; at the same time, calculate the correlation coefficient between the feature data to obtain the degree of linear relationship between the features; E4: Use geological interpretation methods to interpret the features geologically; E5: Use the Bayesian data fusion method to fuse data from different sources and assign appropriate weights to each type of data based on its characteristics and reliability; E6: Analyze and interpret the fused data to reveal the internal connections and patterns among different data; E7: Based on the fused data and geological feature analysis results, the support vector machine algorithm is used to identify the ore body characteristics; E8: Further analysis and processing of the identified features, including feature selection and feature dimensionality reduction; E9: Use neural network data processing methods to deeply mine and extract features, extracting hidden features and patterns; E10: Output the identified and enhanced features as feature vectors and feature maps for use in subsequent ore body distribution prediction models.

7. The method for mineralization prediction based on the cube prediction model of three-dimensional modeling according to claim 1 is characterized in that: In the step F, a deep learning model is trained using a convolutional neural network, and then the deep learning model is applied to automatically extract data features; wherein the input layer of the convolutional neural network receives the preprocessed cube unit data, and the output layer applies the convolution operation to extract features, and the output layer convolution formula is: ， Where: f is the input feature map, g is the convolution kernel, x and y is the coordinate of the output feature map, m and n are the coordinates of the convolution kernel.

8. The method for mineralization prediction based on the cube prediction model of three-dimensional modeling according to claim 7 is characterized by: In the F step, the sub-steps for predicting the ore body distribution are as follows: F1: receives the preprocessed cube unit data through the input layer, and the received data is converted into a multidimensional array as input; F2: The convolution layer performs convolution operations on the input data through multiple convolution kernels to extract local features; The convolution operation in this step uses the convolution kernel to slide over the input data to perform weighted summation and activation; F3: Apply nonlinear activation function ReLU after the convolutional layer to increase the expressiveness of the model; F4: Reduce the dimension of the feature map through pooling operation, reduce the amount of calculation and extract the main features; F5: Flattens the features extracted by the convolutional layer and the pooling layer and inputs them into the fully connected layer for the final classification and regression tasks; F6: Input the preprocessed cube unit data into the trained CNN model and automatically extract data features through the convolution layer; F7: Input the extracted features into the prediction layer of the fully connected layer of the model; F8: In the prediction layer of the fully connected layer, the prediction results of the ore body distribution are obtained by calculation, including the location, scale, and morphological information of the ore body.

9. The method for mineralization prediction based on the cube prediction model of three-dimensional modeling according to claim 1, characterized in that: In the G step, dynamic real-time monitoring and model updating includes the following sub-steps: G1. Drone configuration: Use multiple drones equipped with high-resolution cameras, lidar and / or multispectral sensors to obtain detailed information on the surface and underground of the target mining area; G2. Sensor network deployment: Set up a sensor network in the target mining area to monitor soil moisture, temperature, pH value and / or conductivity in real time; G3. Real-time data collection: drones and sensors transmit data to the central processing system in real time through the Internet of Things technology for preliminary data processing and fusion; G4, Model update: The initially processed and integrated data is fed back in real time, then applied to the 3D geological model through edge computing algorithms, and the model parameters are updated to reflect the latest geological conditions.

10. The method for mineralization prediction based on a cube prediction model of three-dimensional modeling according to any one of claims 1 to 9, characterized in that: In the H step, model verification and evaluation includes the following sub-steps: H1. Select multiple field exploration locations within the target mining area, then conduct drilling and sample collection to obtain actual data; H2. Compare the field exploration data with the model prediction results one by one, record the predicted results and actual results of each sample point, and then compare the predicted results with the actual results to evaluate the accuracy and reliability of the model.

Images