A method for predicting mineral resource potential target areas by fusing multi-source data and machine learning

Abstract

The present application relates to the technical field of information technology and geological mineral resources exploration technology. A mineral resource potential target area prediction method fusing multi-source data and machine learning is provided, comprising the following steps: obtaining sample data and constructing a mineralization prediction data set; preprocessing the prediction data set, dividing it into a training set and a test set, and performing data enhancement on the training set through synthetic minority oversampling to obtain an enhanced training set; constructing an initial mineralization prediction model, training the initial mineralization prediction model through the enhanced training set to obtain a mineralization prediction model, and optimizing the parameters of the mineralization prediction model to obtain a mineralization optimized prediction model; performing interpretability processing on the mineralization optimized prediction model; predicting the mineralization potential of the mineral resource target area, and evaluating the mineralization potential prediction result in combination with a visually interpretable image. The present application solves the problem that the traditional mineral resource exploration method is not only time-consuming and labor-intensive, but also difficult to accurately predict the distribution and potential of mineral resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology and geological mineral resources exploration technology, and in particular to a mineral resource potential target area prediction method combining multi-source data and machine learning. BACKGROUND

[0002] Mineral resources, as an important material basis for national economic and social development, their exploration and development are of great significance to the industrialization and modernization process of the country. However, with the increasing depletion of easily identifiable mineral resources on the surface and shallow parts, the exploration of mineral resources in deep and complex geological environments becomes increasingly difficult.

[0003] Traditional mineral resource exploration methods mainly rely on the experience of geologists and geological exploration data, which not only consumes time and effort, but also often fails to accurately predict the distribution and potential of mineral resources. SUMMARY

[0004] The present application aims to provide a mineral resource potential target area prediction method combining multi-source data and machine learning, which aims to solve the problem that traditional mineral resource exploration methods mainly rely on the experience of geologists and geological exploration data, which not only consumes time and effort, but also fails to accurately predict the distribution and potential of mineral resources.

[0005] The present application is achieved by the following technical solutions:

[0006] A mineral resource potential target area prediction method combining multi-source data and machine learning, comprising the steps of:

[0007] Obtaining sample data with ore-forming favorable factors, and constructing an ore-forming prediction data set;

[0008] Preprocessing the prediction data set, dividing the preprocessed ore-forming prediction data set into a training set and a test set, and performing data augmentation on the training set through synthetic minority oversampling to obtain an augmented training set;

[0009] Based on a convolutional neural network model, an initial ore-forming prediction model is constructed, the initial ore-forming prediction model is trained through the augmented training set to obtain an ore-forming prediction model, and the parameters of the ore-forming prediction model are optimized to obtain an ore-forming optimized prediction model;

[0010] The ore-forming optimized prediction model is subjected to interpretability processing, and the abstract feature map of the ore-forming optimized prediction model is converted into a visual interpretation image;

[0011] The ore-forming optimized prediction model is used to predict the mineral resource potential of the target area, and the visual interpretation image is used to evaluate the ore-forming potential prediction results.

[0012] Optionally, the specific process of constructing the ore-forming prediction data set by acquiring sample data with ore-forming favorable factors is as follows:

[0013] The sample data with ore-forming favorable factors is collected through multiple sources, the collected sample data is subjected to multi-source data fusion processing, the sample data subjected to multi-source data fusion processing is labeled with ore spots and non-ore spots, and an ore-forming prediction data set is obtained.

[0014] Optionally, the specific process of preprocessing the prediction data set and dividing the preprocessed ore-forming prediction data set into a training set and a test set is as follows:

[0015] The ore-forming prediction data set is subjected to data cleaning and normalization processing, samples are randomly selected from the data set subjected to data cleaning and normalization processing, the data set is divided into a training set and a test set at a ratio of 7:3, and the ratio of ore spot samples to non-ore spot samples in the training set and the test set is 1:1.

[0016] Optionally, the specific process of enhancing the training set by synthesizing a minority oversampling to obtain an enhanced training set is as follows:

[0017] The minority class samples in the training set are subjected to oversampling processing by using a synthetic minority oversampling technique based on a small sample learning strategy to generate synthetic samples, and an enhanced training set is obtained in combination with the training set.

[0018] Optionally, the specific process of constructing an initial ore-forming prediction model based on a convolutional neural network model is as follows:

[0019] The convolutional neural network model is initialized, and the structure of the convolutional neural network model is set, including an input layer, a plurality of convolutional layers, a pooling layer, a fully connected layer, and an output layer.

[0020] The ore-forming prediction data set is used as input data of the input layer, a plurality of convolutional kernels in the convolutional layer are used to extract ore-forming favorable factors of the input data, each convolutional kernel generates a feature map, the pooling layer is used to downsample the feature map output by the convolutional layer, the fully connected layer is used to convert the feature map output by the pooling layer into a one-dimensional vector and perform nonlinear transformation, and the output layer is used to convert the output of the fully connected layer into an ore-forming probability or a classification result by using an activation function.

[0021] According to the structure of the initialized convolutional neural network model, an initial ore-forming prediction model is constructed using a deep learning framework, and the hyperparameters of the initial ore-forming prediction model are set.

[0022] Optionally, the specific process of training the initial ore-forming prediction model by using the enhanced training set to obtain an ore-forming prediction model is as follows:

[0023] The enhanced training set is input into the initial mineralization prediction model, and the error between the prediction output of the initial mineralization prediction model and the actual label is calculated by the forward propagation algorithm.

[0024] The backpropagation algorithm is used to calculate the gradient layer by layer based on the error and update the weights and bias parameters of the initial mineralization prediction model.

[0025] Repeat the forward and backward propagation process until the preset number of iterations is reached or the model's predictive performance stabilizes.

[0026] Record the weights and bias parameters of the initial mineralization prediction model during each iteration of training, and obtain the mineralization prediction model.

[0027] Optionally, the specific process of optimizing the parameters of the mineralization prediction model to obtain the optimized mineralization prediction model is as follows:

[0028] During training, a manual grid search method is used to optimize the model's hyperparameters. By iterating through all hyperparameter combinations, the performance of each hyperparameter combination is evaluated, and the hyperparameter combination that makes the model perform best on both the training and test sets is selected.

[0029] The feature ranking in the initial mineralization prediction model is optimized by iteratively adjusting the feature order to maximize prediction accuracy and minimize model loss on the test set, thereby obtaining the optimal feature ranking.

[0030] The mineralization prediction model is configured by optimizing hyperparameters and the optimal feature ranking, and then tested and validated using a test set to obtain the optimized mineralization prediction model.

[0031] Optionally, the specific process of performing interpretability processing on the mineralization optimization prediction model, converting the abstract feature map of the mineralization optimization prediction model into a visually interpretable image, is as follows:

[0032] By using the deconvolution method, the weights of the convolutional filters learned in the mineralization optimization prediction model are projected back into the input space to generate a visual feature map, showing how each convolutional filter responds to mineralization favorable factors.

[0033] Calculate the gradient of the convolutional neural network output relative to the input, identify the input features that have the greatest impact on the network's prediction, and overlay the calculated gradient onto the visualized feature map to highlight the areas that contribute the most to the prediction of mineral potential.

[0034] By using interpretable tools, the abstract feature maps of the mineralization optimization prediction model are converted into visually interpretable images, and the extent to which the mineralization potential prediction results of the mineralization optimization prediction model are affected by the enhanced training set is analyzed and demonstrated.

[0035] Optionally, the specific process of predicting the mineralization potential of a mineral resource target area using a mineralization optimization prediction model and evaluating the mineralization potential prediction results in conjunction with visually interpretable images is as follows:

[0036] Mineral resource target area data is input into the mineralization optimization prediction model. The forward propagation algorithm of the mineralization optimization prediction model is used to calculate the mineralization probability and classification results of each point in the mineral resource target area, and generate a mineralization probability and classification distribution map.

[0037] Based on the mineralization probability and classification distribution map, combined with geological background information and mineral exploration experience, different mineralization potential areas are divided, and a mineral resource potential evaluation map is drawn.

[0038] Visually interpretable images generated by interpretable tools are used to analyze the decision-making basis of the metallogenic optimization prediction model in the process of predicting metallogenic potential. The principle and process of the metallogenic optimization prediction model in identifying potential metallogenic areas based on favorable metallogenic factors are shown. The metallogenic potential prediction results are compared with geological theories and mineral exploration experience, and the reliability of the metallogenic potential prediction results is evaluated.

[0039] Based on the reliability assessment of the comprehensive mineral resource potential evaluation map and the metallogenic potential prediction results, priority exploration areas for mineral resource target areas are determined.

[0040] Optionally, the stability and generalization ability of the mineralization optimization prediction model can be verified through cross-validation, and the performance of the mineralization optimization prediction model can be comprehensively evaluated by using the receiver operating characteristic curve and the corresponding area under the curve as evaluation indicators.

[0041] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0042] Improving prediction accuracy: By integrating multi-source data, including geological, geophysical, geochemical, and remote sensing data, a more comprehensive and accurate mineralization prediction dataset can be constructed, capturing more information related to favorable mineralization factors, thereby improving the accuracy and reliability of mineral resource potential target area prediction.

[0043] Addressing the problem of data imbalance: In mineral resource exploration, the number of samples for certain favorable mineralization factors may be small, leading to data imbalance. This invention effectively solves this problem by using synthetic minority oversampling technology to enhance the dataset. This not only improves the generalization ability of the model but also ensures that the model will not be biased due to data imbalance during the prediction process.

[0044] Optimize model performance: By training with an enhanced training set obtained from synthetic minority oversampling and then optimizing the parameters, an optimized mineralization prediction model was obtained, which can further improve the model's prediction performance and make it more adaptable and robust when facing complex geological environments and various mineralization conditions.

[0045] Enhanced model interpretability: The mineralization optimization prediction model was processed for interpretability by converting the model's abstract feature map into a visually interpretable image. This not only makes the model's prediction results more intuitive and easy to understand, but also provides geologists with more information about mineralization mechanisms and the distribution of favorable factors, helping them to understand the prediction results more deeply and carry out subsequent geological exploration work.

[0046] Improving exploration efficiency: The method provided by this invention can quickly and accurately determine the distribution area of ​​potential mineral resources, thereby greatly improving the efficiency of mineral resource exploration, helping to reduce exploration costs, shorten the exploration cycle, and provide stronger support for the development and utilization of mineral resources. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the mineral resource potential target area prediction method that integrates multi-source data and machine learning according to an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0049] Example 1

[0050] Reference Figure 1 A method for predicting mineral resource potential target areas by integrating multi-source data and machine learning includes the following steps:

[0051] Step 1: Obtain sample data with favorable mineralization factors and construct a mineralization prediction dataset.

[0052] In this embodiment, the specific process of obtaining sample data with favorable mineralization factors and constructing a mineralization prediction dataset is as follows:

[0053] Sample data with favorable mineralization factors are collected from multiple sources. The collected sample data is then fused to resolve issues such as inconsistent data formats and mismatched coordinate systems. Mineralization point and non-mineralization point labels are set on the fused sample data to obtain a mineralization prediction dataset.

[0054] In this embodiment, sample data with favorable mineralization factors are collected from multiple sources, including geological survey reports, remote sensing images, geochemical analysis data, and other mineralization-related sample data. It is ensured that the collected sample data contains multiple favorable mineralization factors, such as geological features (e.g., strata, structures, magmatic activity), remote sensing alteration features (e.g., vegetation cover, soil moisture, surface temperature), and geochemical features (e.g., elemental content, isotope ratios). Multi-source data fusion integrates data from different sources into a unified dataset. Based on known geological and mineral information or field exploration results, mineral-bearing and non-mineral-bearing labels are assigned to the fused sample data; a mineral-bearing label indicates that mineral resources are known to exist at the sample point, while a non-mineral-bearing label indicates that mineral resources are not known to exist at the sample point or that mineral resources are scarce. The labeled sample data is then organized into a mineralization prediction dataset, including the coordinates, feature values, and label information of the sample points. The mineralization prediction dataset contains sufficient sample points to cover various geological and geochemical features within the target area, thereby ensuring the accuracy of the model's predictions.

[0055] Step 2: Preprocess the prediction dataset. Divide the preprocessed mineralization prediction dataset into a training set and a test set. Then, perform data augmentation on the training set by synthesizing minority oversampling to obtain an augmented training set.

[0056] In this embodiment, the specific process of preprocessing the prediction dataset and dividing the preprocessed mineralization prediction dataset into training and testing sets is as follows:

[0057] The mineralization prediction dataset was cleaned and normalized. Samples were randomly selected from the cleaned and normalized dataset and divided into training and test sets in a 7:3 ratio. The ratio of mineralized samples to non-mineralized samples in both the training and test sets was 1:1.

[0058] In this embodiment, data cleaning includes removing duplicates, handling missing values, and denoising. Removing duplicates involves checking for identical sample points in the mineralization prediction dataset and deleting these duplicates. Handling missing values ​​involves selecting appropriate strategies for samples containing missing values ​​based on the number and nature of the missing values. For example, if there are few missing values, interpolation methods (such as linear interpolation, mean interpolation, etc.) or values ​​based on neighboring points can be used to fill in the missing values; if there are many missing values, these samples need to be deleted directly. Denoising can be achieved by applying methods such as moving averages and median filtering, using filtering techniques or statistical methods to remove noise from the data and improve data quality. Normalization methods include min-max scaling and Z-score normalization. Based on expertise in geology, geochemistry, and remote sensing, features that have a significant impact on mineralization prediction are selected and retained; features are then transformed or combined as necessary to extract more useful information. For example, the content of multiple geochemical elements can be converted into ratios or indices.

[0059] In this embodiment, the specific process of augmenting the training set by synthesizing minority oversampling to obtain the augmented training set is as follows:

[0060] A synthetic minority oversampling technique based on a few-shot learning strategy is adopted to oversample the minority class samples in the training set to generate synthetic samples, which are then combined with the training set to obtain an enhanced training set.

[0061] In this embodiment, minority class samples, namely mineral deposit samples (assumed to be positive class), are identified in the training set. Since mineral deposits are usually far fewer than non-mineral deposits in mineral exploration, mineral deposit samples are considered minority class samples. For each minority class sample, its k nearest neighbors are found in the feature space using a distance metric (such as Euclidean distance, Manhattan distance, or cosine similarity). The nearest neighbors can be samples of the same class (i.e., other mineral deposit samples). For each minority class sample and its k nearest neighbors, a neighbor is randomly selected, and a new sample point is randomly generated on the line connecting this neighbor and the original minority class sample. This new sample point is the synthetic sample, which combines the features of the original minority class sample and the selected neighbor.

[0062] The specific generation process can be described as follows:

[0063] For each minority class sample x i Find its k nearest neighbors Randomly select a neighbor In x i and Generate a new sample point x between them new x newThe expression for is shown in equation (1):

[0064]

[0065] Here, `rand(0,1)` represents a random number generated within the interval [0,1]. The above generation process is repeated until the desired number of synthetic samples is reached. The number of synthetic samples can be determined based on the actual situation. The generated synthetic samples are then merged with the samples in the original training set to obtain the augmented training set. The augmented training set contains more minority class samples, which helps improve the model's classification performance on the minority class. When generating synthetic samples, a certain degree of randomness can be introduced to increase sample diversity. In practical applications, the parameters of synthetic minority oversampling (such as the k-value, the number of synthetic samples, etc.) need to be adjusted according to the specific characteristics of the dataset and the model's performance.

[0066] Step 3: Construct an initial mineralization prediction model based on a convolutional neural network model. Train the initial mineralization prediction model using an enhanced training set to obtain a mineralization prediction model. Optimize the parameters of the mineralization prediction model to obtain an optimized mineralization prediction model.

[0067] In this embodiment, the specific process of constructing the initial mineralization prediction model based on the convolutional neural network model is as follows:

[0068] The convolutional neural network (CNN) model is initialized, setting its structure, including an input layer, several convolutional layers, pooling layers, fully connected layers, and an output layer. The input layer receives a preprocessed mineralization prediction dataset in two-dimensional or three-dimensional matrix format, containing multi-source information such as geological features, remote sensing alteration features, and geochemical features. The convolutional layers use multiple convolutional kernels (also called filters), each responsible for extracting specific features from the input data (such as geological structures, fault systems, and remote sensing anomalies). Each convolutional layer contains multiple convolutional operations, generating a feature map. The pooling layers downsample the feature maps output by the convolutional layers to reduce data dimensionality and computational cost while preserving important features. Pooling methods include max pooling and average pooling. The fully connected layers transform the feature maps output by the pooling layers into one-dimensional vectors and perform non-linear transformations (such as the ReLU activation function) to extract higher-level features. The output layer uses an activation function (such as softmax) to convert the output of the fully connected layers into mineralization probabilities or classification results (mineralized or non-mineralized points).

[0069] The mineralization prediction dataset is used as input data for the input layer. Several convolutional kernels in the convolutional layer extract mineralization-favorable factors from the input data, with each kernel generating a feature map. Pooling layers downsample the feature maps output from the convolutional layers. Fully connected layers transform the feature maps output from the pooling layers into one-dimensional vectors and perform non-linear transformations. The output layer uses an activation function to convert the output of the fully connected layer into a mineralization probability or classification result. Based on the structure of the initialized convolutional neural network model, an initial mineralization prediction model is constructed using a deep learning framework, and its hyperparameters are set. The deep learning framework can be TensorFlow, PyTorch, or other similar frameworks. The initial mineralization prediction model is constructed based on the determined model structure and hyperparameters. During the construction process, the shape of the input data, the configuration of the convolutional layers, the pooling layers, the fully connected layers, and the output layer need to be specified.

[0070] In this embodiment, the specific process of training the initial mineralization prediction model by enhancing the training set to obtain the mineralization prediction model is as follows:

[0071] Set the initial parameters of the convolutional neural network model, including learning rate, weight decay, batch size, etc.; initialize the weights and biases of the convolutional kernels using random initialization methods, such as Gaussian or uniform distribution; define the number of iterations or termination conditions for the training loop, such as reaching a preset accuracy or loss no longer significantly decreasing; in each iteration, randomly sample a batch of data from the augmented training set for training; input the augmented training set batch into the initial mineralization prediction model, and calculate the error between the predicted output of the initial mineralization prediction model and the actual label using the forward propagation algorithm; progressively extract features and generate predicted outputs through structures such as convolutional layers, pooling layers, and fully connected layers; calculate the error between the predicted output and the actual label using the cross-entropy loss function (or other loss functions suitable for classification tasks).

[0072] Based on the gradient of the loss function, the backpropagation algorithm is used to calculate the gradient of the parameters of each layer. Optimization algorithms (such as Adam, SGD, etc.) are then applied to update the weights and bias parameters of the initial mineralization prediction model in order to reduce the loss.

[0073] Repeat the forward and backward propagation process until the preset number of iterations is reached or the model's predictive performance stabilizes; record the weights and bias parameters of the initial mineralization prediction model during each iteration of training, and obtain the mineralization prediction model.

[0074] In this embodiment, the specific process of optimizing the parameters of the mineralization prediction model to obtain the optimized mineralization prediction model is as follows:

[0075] During training, a manual grid search method is used to optimize the model's hyperparameters. By iterating through all hyperparameter combinations, the performance of each combination is evaluated, and the combination that best performs on both the training and test sets is selected. The search range for hyperparameters is defined for the convolutional neural network model. Hyperparameters include, but are not limited to, learning rate, dropout rate (to prevent overfitting), kernel size, number of filters within a convolutional layer, batch size, and number of training iterations. The manual grid search method systematically iterates through all possible hyperparameter combinations. For each combination, the model is reinitialized and trained on the entire augmented training dataset. During training, the performance of each hyperparameter combination on the training and test sets, such as accuracy and loss, is monitored and recorded. Based on the performance evaluation results, the hyperparameter combination that best performs on both the training and validation sets is selected.

[0076] The feature order in the initial mineralization prediction model is optimized by iteratively adjusting the feature order to maximize prediction accuracy and minimize model loss on the test set, thus obtaining the optimal feature order. The mineralization prediction model is then configured using the optimized hyperparameters and the optimal feature order, and tested and validated using the test set to obtain the optimized mineralization prediction model.

[0077] Step 4: Perform interpretability processing on the mineralization optimization prediction model to convert the abstract feature map of the mineralization optimization prediction model into a visually interpretable image.

[0078] In this embodiment, the specific process of performing interpretability processing on the mineralization optimization prediction model, converting the abstract feature map of the mineralization optimization prediction model into a visually interpretable image, is as follows:

[0079] By using deconvolution, the weights of the convolutional filters learned in the mineralization optimization prediction model are projected back into the input space to generate a visual feature map, demonstrating how each convolutional filter responds to mineralization-favorable factors. The trained mineralization optimization prediction model is loaded, the convolutional layers to be visualized are selected, and deconvolution operations are applied to map the outputs of the convolutional layers back to the spatial dimension of the input data. The resulting feature map is then visualized, such as adjusting color mapping and contrast, to more clearly display the features.

[0080] The gradient of the convolutional neural network output relative to the input is calculated to identify the input features that have the greatest impact on the network's predictions. The calculated gradient is then overlaid on a visualized feature map to highlight the regions that contribute the most to the prediction of mineral potential. Using the automatic differentiation function provided by the deep learning framework, the gradient of the model output relative to the input data is calculated. The gradient is then normalized to more intuitively display its distribution. The gradient is then overlaid on a visualized feature map to highlight the regions that contribute the most to the prediction of mineral potential.

[0081] Interpretable tools are used to convert the abstract feature maps of the mineralization optimization prediction model into visually interpretable images. This analysis demonstrates the extent to which the predicted mineralization potential of the model is influenced by the augmented training set. Suitable interpretable tools for convolutional neural networks are selected, such as Grad-CAM (Gradient-weighted Class Activation Mapping) and Guided Backpropagation. Grad-CAM generates a visual image reflecting the model's areas of interest by calculating a weighted sum of gradients and feature maps. Guided Backpropagation combines backpropagation with specific rules to generate more refined feature visualizations. Based on the selected tools, the model's input and output formats are adjusted. The interpretable tools are run to generate visually interpretable images. Post-processing of the generated images, such as adjusting color and contrast, is performed to better showcase the feature regions of interest to the model.

[0082] Step 5: Predict the mineralization potential of the target area using the mineralization optimization prediction model, and evaluate the prediction results by combining visually interpretable images.

[0083] In this embodiment, the specific process of predicting the mineralization potential of a mineral resource target area using a mineralization optimization prediction model and evaluating the mineralization potential prediction results in conjunction with visually interpretable images is as follows:

[0084] Mineral resource target area data is input into the mineralization optimization prediction model. The forward propagation algorithm of the mineralization optimization prediction model is used to calculate the mineralization probability and classification results of each point in the mineral resource target area, and generate a mineralization probability and classification distribution map.

[0085] Based on the mineralization probability and classification distribution map, combined with geological background information and mineral exploration experience, different mineralization potential areas can be divided into high-potential areas, medium-potential areas, low-potential areas, and background areas, and a mineral resource potential evaluation map can be drawn. Geological background information includes geological background information such as strata, structure, and magmatic activity; mineral exploration experience includes the distribution of known mineral occurrences and metallogenic models.

[0086] Visually interpretable images generated by interpretable tools are used to analyze the decision-making basis of the metallogenic optimization prediction model in the process of predicting metallogenic potential. The principle and process of the metallogenic optimization prediction model in identifying potential metallogenic areas based on favorable metallogenic factors are demonstrated. The metallogenic potential prediction results are compared with geological theories and mineral exploration experience, and the reliability of the metallogenic potential prediction results is evaluated. The reliability evaluation includes the accuracy, consistency and interpretability of the metallogenic potential prediction results.

[0087] Based on a comprehensive assessment of the reliability of the mineral resource potential evaluation map and the metallogenic potential prediction results, priority exploration areas for mineral resource target areas are determined. A detailed mineral resource potential prediction report is prepared, including data preparation, prediction process, and result evaluation; all relevant data, model parameters, prediction results, and evaluation indicators are recorded.

[0088] Example 2

[0089] Based on Example 1, in this example, the stability and generalization ability of the mineralization optimization prediction model are verified by cross-validation, and the performance of the mineralization optimization prediction model is comprehensively evaluated by using the receiver operating characteristic curve and the corresponding area under the curve as evaluation indicators.

[0090] In this embodiment, the configured model is validated on an independent test set to evaluate its generalization ability. Performance metrics such as accuracy, recall, and F1 score are calculated on the test set, and the receiver operating characteristic (ROC) curve and its corresponding area under the curve are plotted. To further verify the model's stability and generalization ability, a k-fold cross-validation method can be used. The dataset is divided into k subsets. Each time, k-1 subsets are used for training, and the remaining subset is used for validation. This process is repeated k times, selecting a different subset as the validation set each time. Finally, the average performance metrics from the k validations are calculated. The optimized and validated model is saved as the final mineralization optimization prediction model. A detailed performance evaluation report is generated, including metrics such as model accuracy, recall, F1 score, ROC curve, and AUC value, as well as the results of cross-validation.

[0091] Based on the model's predictions (probability of mineralization) and actual labels (mineralized or non-mineralized locations) on the validation set, calculate the True Positive Rate (TPR) and False Positive Rate (FPR) at different thresholds. Plot the Receiver Operating Characteristic (ROC) curve with FPR on the x-axis and TPR on the y-axis. The area under the curve (AUC) can be calculated using methods such as numerical integration or the trapezoidal rule. The AUC value ranges from 0 to 1, with a higher value indicating better model performance. Compare the average performance metrics of the models under different hyperparameter combinations, especially the AUC value; select the hyperparameter combination that best performs on both the training and validation sets as the final model configuration; select the optimal mineralization prediction model based on the performance evaluation results; ensure that the selected model has high precision, recall, F1 score, and a large AUC value on the validation set. The final model is validated using an independent test set to ensure its generalization ability; performance metrics such as accuracy, recall, F1 score, ROC curve, and AUC value are calculated on the test set; the final model is applied to predict the mineralization potential of mineral resource target areas; and the decision-making basis and prediction results of the model are analyzed in conjunction with visually interpretable images.

[0092] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting mineral resource potential target areas by integrating multi-source data and machine learning, characterized in that, Including the following steps: Obtain sample data with favorable mineralization factors and construct a mineralization prediction dataset; The prediction dataset is preprocessed, and the preprocessed mineralization prediction dataset is divided into a training set and a test set. The training set is augmented by synthesizing a minority oversampling to obtain an augmented training set. An initial mineralization prediction model is constructed based on a convolutional neural network model. The initial mineralization prediction model is trained by an enhanced training set to obtain a mineralization prediction model. The parameters of the mineralization prediction model are then optimized to obtain an optimized mineralization prediction model. Interpretability processing is performed on the mineralization optimization prediction model to convert the abstract feature map of the mineralization optimization prediction model into a visually interpretable image; The specific process of performing interpretability processing on the mineralization optimization prediction model, converting the abstract feature map of the mineralization optimization prediction model into a visually interpretable image, is as follows: By using the deconvolution method, the weights of the convolutional filters learned in the mineralization optimization prediction model are projected back into the input space to generate a visual feature map, showing how each convolutional filter responds to mineralization favorable factors. Calculate the gradient of the convolutional neural network output relative to the input, and identify the input features that have the greatest impact on the network's predictions; The calculated gradients are superimposed on the visualized feature map to highlight the areas that contribute the most to the prediction of mineralization potential. The abstract feature map of the mineralization optimization prediction model is converted into a visually interpretable image using interpretable tools, and the extent to which the mineralization potential prediction results of the mineralization optimization prediction model are affected by the enhanced training set is analyzed and shown. The mineralization potential of mineral resource target areas is predicted by using a mineralization optimization prediction model, and the prediction results are evaluated by combining visually interpretable images. The specific process of predicting the mineralization potential of a mineral resource target area using a mineralization optimization prediction model, and evaluating the prediction results in conjunction with visually interpretable images, is as follows: Mineral resource target area data is input into the mineralization optimization prediction model. The forward propagation algorithm of the mineralization optimization prediction model is used to calculate the mineralization probability and classification results of each point in the mineral resource target area, and generate a mineralization probability and classification distribution map. Based on the mineralization probability and classification distribution map, combined with geological background information and mineral exploration experience, different mineralization potential areas are divided, and a mineral resource potential evaluation map is drawn. Visually interpretable images generated by interpretable tools are used to analyze the decision-making basis of the metallogenic optimization prediction model in the process of predicting metallogenic potential. The principle and process of the metallogenic optimization prediction model in identifying potential metallogenic areas based on favorable metallogenic factors are shown. The metallogenic potential prediction results are compared with geological theories and mineral exploration experience, and the reliability of the metallogenic potential prediction results is evaluated. Based on the reliability assessment of the comprehensive mineral resource potential evaluation map and the metallogenic potential prediction results, priority exploration areas for mineral resource target areas are determined.

2. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of obtaining sample data with favorable mineralization factors and constructing a mineralization prediction dataset is as follows: Sample data with favorable mineralization factors are collected from multiple sources. The collected sample data is then subjected to multi-source data fusion processing. Mineralization point and non-mineralization point labels are set on the sample data after multi-source data fusion processing to obtain a mineralization prediction dataset.

3. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of preprocessing the prediction dataset and dividing the preprocessed mineralization prediction dataset into training and testing sets is as follows: The mineralization prediction dataset was cleaned and normalized. Samples were randomly selected from the cleaned and normalized dataset and divided into training and test sets in a 7:3 ratio. The ratio of mineralized samples to non-mineralized samples in both the training and test sets was 1:

1.

4. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of augmenting the training set by synthesizing a minority oversampling to obtain an augmented training set is as follows: A synthetic minority oversampling technique based on a few-shot learning strategy is adopted to oversample the minority class samples in the training set to generate synthetic samples, which are then combined with the training set to obtain an enhanced training set.

5. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of constructing the initial mineralization prediction model based on the convolutional neural network model is as follows: Initialize the convolutional neural network model and set its structure, including the input layer, several convolutional layers, pooling layers, fully connected layers, and the output layer. The mineralization prediction dataset is used as the input data for the input layer. Several convolutional kernels in the convolutional layer extract the mineralization favorable factors from the input data, and each convolutional kernel generates a feature map. The pooling layer downsamples the feature map output by the convolutional layer. The fully connected layer transforms the feature map output by the pooling layer into a one-dimensional vector and performs a non-linear transformation. The output layer uses an activation function to convert the output of the fully connected layer into a mineralization probability or classification result. Based on the structure of the initialized convolutional neural network model, an initial mineralization prediction model is constructed using a deep learning framework, and the hyperparameters of the initial mineralization prediction model are set.

6. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of training the initial mineralization prediction model by enhancing the training set to obtain the mineralization prediction model is as follows: The enhanced training set is input into the initial mineralization prediction model, and the error between the prediction output of the initial mineralization prediction model and the actual label is calculated by the forward propagation algorithm. The backpropagation algorithm is used to calculate the gradient layer by layer based on the error and update the weights and bias parameters of the initial mineralization prediction model. Repeat the forward and backward propagation process until the preset number of iterations is reached or the model's predictive performance stabilizes. Record the weights and bias parameters of the initial mineralization prediction model during each iteration of training, and obtain the mineralization prediction model.

7. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The specific process of optimizing the parameters of the mineralization prediction model to obtain the optimized mineralization prediction model is as follows: During training, a manual grid search method is used to optimize the model's hyperparameters. By iterating through all hyperparameter combinations, the performance of each hyperparameter combination is evaluated, and the hyperparameter combination that makes the model perform best on both the training and test sets is selected. The feature ranking in the initial mineralization prediction model is optimized by iteratively adjusting the feature order to maximize prediction accuracy and minimize model loss on the test set, thereby obtaining the optimal feature ranking. The mineralization prediction model is configured by optimizing hyperparameters and the optimal feature ranking, and then tested and validated using a test set to obtain the optimized mineralization prediction model.

8. The method for predicting mineral resource potential target areas by integrating multi-source data and machine learning as described in claim 1, characterized in that, The stability and generalization ability of the mineralization optimization prediction model were verified by cross-validation. The performance of the mineralization optimization prediction model was comprehensively evaluated by using the receiver operating characteristic curve and the corresponding area under the curve as evaluation indicators.

Images