Unseen ore sample selection method and system, storage medium and electronic equipment

Abstract

The invention discloses an unseen ore sample selection method and system, a storage medium and an electronic device, and relates to the technical field of metallogenic prediction, and the method comprises the steps: carrying out the anomaly detection of collected unseen ore samples through employing a depth support vector data description algorithm, and obtaining an anomaly score; setting an abnormal score threshold value, and selecting the unseen ore samples of which the abnormal scores are lower than the threshold value as a pre-selection set of the unseen ore samples; using a K-means clustering algorithm to divide the pre-selection set of the samples without the mine into a plurality of clusters; and through a cluster center undersampling technology, retaining a cluster center sample of each cluster as a final selection set of unseen samples. According to the method, the reliability of the unseen ore sample can be improved, and meanwhile, the loss of a large amount of information caused by random undersampling is relieved, so that the selected unseen ore sample is representative.

Description

Technical Field

[0001] The present invention relates to the technical field of mineralization prediction, and in particular to a method, system, storage medium and electronic equipment for selecting unseen mineral samples. Background Art

[0002] In the field of mineral resource exploration, accurate prediction of potential mineralization areas is of great significance for optimizing exploration efficiency and reducing exploration costs. Traditional mineralization prediction methods mostly rely on samples with ore-seeking (i.e. samples verified by drilling), while samples without ore-seeking (i.e. samples not verified by drilling) are often selected by random undersampling. However, random undersampling has problems of inaccurate sample selection and data imbalance, which may lead to erroneous data in the sample set without ore-seeking, affecting the performance of the prediction model. How to select reliable and representative samples in the area without ore-seeking and reduce data redundancy has become an important research topic in the field of mineralization prediction. Summary of the invention

[0003] The purpose of the present invention is to solve the problems of unreliable data and data redundancy when selecting samples without mineral deposits, and to propose a method for selecting samples without mineral deposits, comprising the following steps:

[0004] S1. Collect samples without mineral deposits;

[0005] S2, using the deep support vector data description algorithm to perform anomaly detection on samples without ore, and obtain anomaly scores;

[0006] S3, setting an abnormality score threshold, and selecting unseen samples with abnormality scores lower than the threshold as a pre-selected set of unseen samples;

[0007] S4, using K-means clustering algorithm, divide the pre-selected set of samples without mineralization into multiple clusters;

[0008] S5. Through the cluster center under-sampling technology, the cluster center samples of each cluster are retained as the final selection set of samples without mineralization.

[0009] Furthermore, the loss function in the training phase of the deep support vector data description algorithm is:

[0010]

[0011] Among them, loss is the loss function of the deep support vector data description algorithm training stage, R represents the radius of the sphere, ν∈(0,1] represents the weight parameter for controlling the proportion of outliers, and n represents the number of samples without mineralization. is a neural network that supports deep vector data description algorithms, x i represents the i-th sample without mineralization, represents the connection weights of the deep support vector data description algorithm, L represents the number of layers of the neural network, W l represents the weight of the lth layer, c represents the center of the hypersphere, ||·|| F represents the Frobenius norm, and λ>0 represents the weight parameter of the regularization term.

[0012] Furthermore, the abnormality score calculation formula is:

[0013]

[0014] Among them, S(x i ) represents x i The anomaly score, x i represents the i-th unseen sample, φ(x i ; W * ) indicates that the parameter of the deep support vector data description algorithm is x i and W * The neural network, W * represents the optimal connection weight obtained by the trained deep support vector data description algorithm, and c represents the center of the hypersphere.

[0015] Furthermore, an abnormal score threshold is set according to the normal distribution of the abnormal score.

[0016] Furthermore, the pre-selected set of unseen samples is divided into multiple clusters. First, K cluster centers are initialized, and the distance from each unseen sample in the pre-selected set of unseen samples to each cluster center is calculated:

[0017] d(x i ,μ k )=‖x i -μ k ‖ 2 ,

[0018] Among them, d(x i ,μ k ) represents the i-th unseen sample x in the pre-selected set of unseen samples i The distance to the kth cluster center;

[0019] According to the distance, each sample is assigned to the nearest cluster center, the cluster center of each cluster is updated, and the iteration is repeated until the positions of the cluster centers converge to obtain the cluster clusters.

[0020] The present invention also provides a system for selecting unseen ore samples, comprising:

[0021] Sample collection module, used to collect samples where no minerals have been found;

[0022] Anomaly detection module, used to perform anomaly detection on samples without mineralization using deep support vector data description algorithm to obtain anomaly scores;

[0023] A sample selection module is used to set an anomaly score threshold and select unseen samples with anomaly scores lower than the threshold as a pre-selected set of unseen samples;

[0024] A sample clustering module is used to divide the pre-selected set of unseen samples into multiple clusters using the K-means clustering algorithm;

[0025] The sample determination module is used to retain the cluster center samples of each cluster as the final selection set of samples without mineralization through the cluster center under-sampling technology, and use the final selection set of samples without mineralization for mineralization prediction.

[0026] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for selecting samples without mineral deposits is implemented.

[0027] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned method for selecting unseen samples.

[0028] The beneficial effects brought by the technical solution provided by the present invention are:

[0029] The present invention firstly adopts a deep support vector data description algorithm to perform anomaly detection on samples not seen by the mine, removes samples with high anomaly scores, retains samples with low anomaly scores for clustering, ensures the reliability of samples in the pre-selected set of samples not seen by the mine, and retains the cluster center sample of each cluster cluster as the final sample not seen by the mine through the cluster center undersampling technology, and removes other samples in the cluster cluster, so as to reduce redundant samples and ensure the representativeness of samples not seen by the mine. The present invention can improve the reliability of samples not seen by the mine, and at the same time alleviate the large amount of information loss caused by random undersampling, so as to ensure that the selected samples not seen by the mine are representative. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of a method for selecting a sample without a known ore according to an embodiment of the present invention;

[0031] Figure 2 It is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0032] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0033] Example 1: A flow chart of the method for selecting unseen ore samples according to an embodiment of the present invention is as follows Figure 1 , specifically including the following steps:

[0034] S1. Collect samples without mineralization. Samples without mineralization come from areas that have not been verified by drilling, and samples that are very close to samples with mineralization are eliminated to ensure that there is a clear spatial gap between them to reduce the possibility of misselection.

[0035] S2. Use the deep support vector data description algorithm to perform anomaly detection on samples without mineralization and obtain anomaly scores.

[0036] The Deep Support Vector Data Description (DeepSVDD) algorithm maps sample data to a high-dimensional feature space through a neural network and constructs a hypersphere in this space. The center c is the center of the normal data, and the radius R is the minimum sphere surrounding these normal samples. The goal of DeepSVDD is to minimize the distance from the normal sample (i.e., the sample without minerals) to the center c and maximize the radius R of the sphere so that it can surround all normal samples.

[0037] In a preferred embodiment of the present invention, the loss function in the training phase of the deep support vector data description algorithm is:

[0038]

[0039] Among them, loss is the loss function of the deep support vector data description algorithm training stage, R represents the radius of the sphere, ν∈(0,1] represents the weight parameter for controlling the proportion of outliers, n represents the number of samples without mineralization, is a neural network that supports deep vector data description algorithms, x i represents the i-th sample without mineralization, represents the connection weights of the deep support vector data description algorithm, L represents the number of layers of the neural network. Indicates The weight of the layer, c represents the center of the hypersphere, ||·|| F represents the Frobenius norm, and λ>0 represents the weight parameter of the regularization term.

[0040] For each unseen sample x i , the DeepSVDD model determines whether it is an abnormal sample by calculating the distance between the sample and the center of the sphere c. After the DeepSVDD training is completed, the model assigns an abnormal score S(x i ), the anomaly score formula is:

[0041]

[0042] Among them, S(x i ) represents x i The anomaly score, x i represents the i-th unseen sample, φ(x i ; W * ) indicates that the parameter of the deep support vector data description algorithm is x i and W * The neural network, W * represents the optimal connection weight obtained by the trained deep support vector data description algorithm, and c represents the center of the hypersphere.

[0043] S3. Set an abnormality score threshold, and select unseen samples with abnormality scores lower than the threshold as a pre-selected set of unseen samples.

[0044] According to the normal distribution of abnormal scores, the abnormal score threshold σ is set. i )<σ, indicating that sample x i If the difference between the sample and the ore sample is large, then x i As normal data, otherwise it is abnormal data, where the normal data set is used as a pre-selected set of samples that have not been found.

[0045] S4. Use the K-means clustering algorithm to divide the pre-selected set of unseen samples into multiple clusters.

[0046] In a preferred embodiment of the present invention, K cluster centers are first initialized, and the distance from each unseen sample in the pre-selected set of unseen samples to each cluster center is calculated:

[0047] d(x i ,μ k )=||x i -μ k || 2 ,

[0048] Among them, d(x i ,μ k ) represents the i-th unseen sample x in the pre-selected set of unseen samples i The distance to the kth cluster center, that is, the centroid position;

[0049] According to the distance from each unseen sample in the pre-selected set of unseen samples to each cluster center, each sample is assigned to the nearest cluster center, the cluster center of each cluster is updated, and the iteration is repeated until the cluster center position converges and the centroid no longer changes significantly, thus obtaining multiple cluster clusters of the pre-selected set of unseen samples.

[0050] S5. Through the cluster center under-sampling technology, the cluster center samples of each cluster are retained as the final selection set of samples without mineralization.

[0051] For each cluster C k , only retain its centroid μ k , as the representative sample of the cluster, and the rest of the samples are removed. This can reduce redundant samples and ensure that the information in each cluster is preserved through the centroid.

[0052] Embodiment 2: The present invention also provides a system for selecting unseen mineral samples, comprising:

[0053] Sample collection module, used to collect samples where no minerals have been found;

[0054] Anomaly detection module, used to perform anomaly detection on samples without mineralization using deep support vector data description algorithm to obtain anomaly scores;

[0055] A sample selection module is used to set an anomaly score threshold and select unseen samples with anomaly scores lower than the threshold as a pre-selected set of unseen samples;

[0056] A sample clustering module is used to divide the pre-selected set of unseen samples into multiple clusters using the K-means clustering algorithm;

[0057] The sample determination module is used to retain the cluster center samples of each cluster as the final selection set of samples without mineralization through the cluster center under-sampling technology, and use the final selection set of samples without mineralization for mineralization prediction.

[0058] Embodiment 3: In an exemplary embodiment, a computer-readable storage medium is included, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for selecting samples without mineral deposits is implemented.

[0059] Example 4: Please refer to Figure 2 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0060] Wherein, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned method for selecting samples without mineral deposits.

[0061] The unseen ore sample selection method adopted in the present invention can effectively solve the data imbalance problem in mineral resource exploration and significantly improve the performance of the mineral resource prediction model. It is of great significance in the field of mineralization prediction and has broad application prospects.

[0062] In the field of mineralization prediction, it is necessary to collect samples with and without ore. The samples with ore come from known ore samples verified by drilling data. The characteristics of the samples with ore include geological factors, ore body location, depth and mineral content, etc. The samples without ore come from areas that have not been verified by drilling, and samples that are very close to the samples with ore are eliminated to ensure that there is a clear spatial gap between them to reduce the possibility of misselection.

[0063] The collected actual ore-seeking samples are expanded using a generative adversarial network, and the collected actual ore-seeking samples and the expanded ore-seeking samples are used as positive samples. The collected samples without ore-seeking are subjected to the method of the embodiment of the present invention to obtain negative samples. The number of positive samples and negative samples is kept consistent, that is, the number of positive samples is the same as the number of cluster centers K. The positive samples and negative samples are combined to construct a balanced data set: Among them, D final represents a balanced data set, X mineral is a positive sample, The distribution of geological features among samples in the balanced data set is clear, ensuring that the model will not be biased towards a certain type of sample due to data imbalance during training.

[0064] The random forest classifier is used as the benchmark model to train the constructed balanced dataset. The dataset is divided into a training set and a test set in a ratio of 6:4. The evaluation indicators include AUC (area under the curve) and classification accuracy. The AUC value of the trained model on the test set is 0.89, which is significantly higher than the traditional random undersampling method (AUC value is 0.75).

[0065] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for selecting unseen ore samples, characterized in that: The following steps are involved: S1. Collect samples without mineral deposits; S2, using the deep support vector data description algorithm to perform anomaly detection on samples without ore, and obtain anomaly scores; S3, setting an abnormality score threshold, and selecting unseen samples with abnormality scores lower than the threshold as a pre-selected set of unseen samples; S4, using K-means clustering algorithm, divide the pre-selected set of samples without mineralization into multiple clusters; S5. Through the cluster center under-sampling technology, the cluster center samples of each cluster are retained as the final selection set of samples without mineralization, and the final selection set of samples without mineralization is used for mineralization prediction.

2. The method for selecting unseen ore samples according to claim 1, characterized in that: The loss function in the training phase of the deep support vector data description algorithm is: Among them, loss is the loss function of the deep support vector data description algorithm training stage, R represents the radius of the sphere, ν∈(0,1] represents the weight parameter for controlling the proportion of outliers, and n represents the number of samples without mineralization. is a neural network that supports deep vector data description algorithms, x i represents the i-th sample without mineralization, represents the connection weights of the deep support vector data description algorithm, L represents the number of layers of the neural network, W 1 represents the weight of the ιth layer, c represents the center of the hypersphere, ∥·∥ F represents the Frobenius norm, and λ>0 represents the weight parameter of the regularization term.

3. The method for selecting unseen samples according to claim 1, characterized in that: The abnormality score calculation formula is: Among them, S(x i ) represents x i The anomaly score, x i represents the i-th unseen sample, φ(x i ; W * ) indicates that the parameter of the deep support vector data description algorithm is x i and W * The neural network, W * represents the optimal connection weight obtained by the trained deep support vector data description algorithm, and c represents the center of the hypersphere.

4. The method for selecting unseen samples according to claim 1, characterized in that: The anomaly score threshold is set according to the normal distribution of anomaly scores.

5. The method for selecting unseen samples according to claim 1, characterized in that: The pre-selected set of samples without known ore is divided into multiple clusters. First, K cluster centers are initialized, and the distance from each sample without known ore to each cluster center in the pre-selected set of samples without known ore is calculated: d(x i ,m k )=‖x i -m k ‖ 2 , Among them, d(x i ,μ k ) represents the i-th unseen sample x in the pre-selected set of unseen samples i The distance to the kth cluster center; According to the distance, each sample is assigned to the nearest cluster center, the cluster center of each cluster is updated, and the iteration is repeated until the positions of the cluster centers converge to obtain the cluster clusters.

6. A system for selecting samples of unseen minerals, characterized in that: include: Sample collection module, used to collect samples where no minerals have been found; Anomaly detection module, used to perform anomaly detection on samples without mineralization using deep support vector data description algorithm to obtain anomaly scores; A sample selection module is used to set an anomaly score threshold and select unseen samples with anomaly scores lower than the threshold as a pre-selected set of unseen samples; A sample clustering module is used to divide the pre-selected set of unseen samples into multiple clusters using the K-means clustering algorithm; The sample determination module is used to retain the cluster center samples of each cluster as the final selection set of samples without mineralization through the cluster center under-sampling technology, and use the final selection set of samples without mineralization for mineralization prediction.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

8. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 5.

Images