Deep lithology identification method, apparatus and device, and storage medium

Abstract

The invention belongs to the technical field of geophysical exploration, and provides a deep lithology identification method and device, equipment and a storage medium. The method comprises the steps of obtaining a data set; wherein the data set comprises a feature set and a lithology label; performing locality preserving projection on the data set to obtain a preprocessed data set; constructing a machine learning model, training the machine learning model by using the preprocessed data set to obtain an identification model for deep lithology identification, and in the training process, introducing distance constraints and geological constraints in sequence; wherein the distance constraint is that clustering labels are distributed by minimizing the distance between a data point and a clustering center closest to the data point, and an initial clustering center is selected to optimize distribution; and the geological constraint takes a lithologic label provided by the preprocessed data set as a constraint condition. According to the method, the local neighborhood relation between the data points is considered, so that the prediction accuracy is remarkably improved under the condition that the label data is limited and unbalanced.

Description

Technical Field

[0001] The present application relates to the field of geophysical exploration technology, and in particular to a deep lithology identification method, device, equipment and storage medium. Background Art

[0002] Integrated geophysical surveys, including gravity, magnetic, electromagnetic, and seismic methods, are key means of revealing the underground structure and material properties of mining areas. However, integrating these diverse measured data to improve the accuracy and reliability of underground structure characterization still faces major challenges. In recent years, with the development of machine learning technology, the fusion of multi-source geophysical data has been greatly improved, significantly improving the interpretation accuracy and reliability in applications such as mineral resource prediction, geological mapping, and rock classification.

[0003] Although image fusion techniques can reduce interpretation errors and computational requirements in multi-source data fusion applications, they are limited by the lack of drill hole lithology labels as constraints in lithology identification. Traditional machine learning techniques (such as support vector machines, back propagation neural networks, and random forests) have demonstrated significant results in assessing the probability of mineralization in three-dimensional mineral exploration, but these methods are usually limited to dividing areas into mineral-bearing and non-mineral-bearing areas, ignoring the important role of lithology. In addition, these techniques are highly dependent on labeled data, and their performance is significantly restricted when labeled data is scarce and diverse.

[0004] Subsurface lithology identification is crucial to understanding the formation, type, and distribution of mineral deposits. However, the scarcity and imbalance of drilling data labels have hampered progress in this field. To overcome these limitations, recent studies have introduced innovative machine learning strategies to improve the accuracy of lithology identification by dealing with the problems of insufficient labeled data and class imbalance. For example, the ALSLINB method proposed by Ren et al. combines active learning and semi-supervised learning to expand the label set of nonlinear drilling data, but has difficulties in generalizing to unknown data types. Lv et al. subsequently proposed SsL-VGMM, which handles limited and unbalanced data and adapts to new data types through a variational Gaussian mixture model. Despite this, these methods still face the problem of uneven spatial distribution of drilling data when applied in practical scenarios. Summary of the invention

[0005] The present application provides a deep lithology identification method, device, equipment and storage medium, which significantly improves the prediction accuracy when the label data is limited and unbalanced by considering the local neighborhood relationship between data points.

[0006] In a first aspect, the present application provides a deep lithology identification method, comprising: Acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds to a lithology label in a one-to-one spatial position; Performing local preserving projection on the data set to obtain a preprocessed data set; A machine learning model is constructed and trained using the preprocessed data set to obtain an identification model for deep lithology identification. During the training process, distance constraints and geological constraints are introduced in sequence; wherein the distance constraint is to assign cluster labels by minimizing the distance between a data point and the cluster center closest to the data point, and an initial cluster center is selected to optimize the assignment; and the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

[0007] Further, the data set is subjected to local preservation projection to obtain a preprocessed data set, including: Constructing a weight matrix; wherein the weight matrix is ​​defined by a Gaussian kernel function, and the weight matrix is ​​used to capture the local adjacency relationship between each data point in the data set; Finding a projection matrix such that the data still maintains the original local structure in the low-dimensional space determined by the projection matrix is ​​used as the optimization goal, and an objective function is established; A projection matrix is ​​determined based on the objective function, and the data set is locally projected using the projection matrix to obtain a preprocessed data set.

[0008] Furthermore, the objective function is expressed as: ; In the formula, represents a matrix of numbers, represents a diagonal matrix, represents the identity matrix, A represents the projection matrix, x i and x j Indicates the data set i and j data points, W ij Represents data points x i and data points x j The weight matrix of T Represents matrix transpose.

[0009] Furthermore, the distance constraint includes: Randomly select a sample point from the preprocessed data set as the first initial cluster center ; Calculate the shortest distance between each sample point and the current cluster center; Calculate the probability of each sample point being selected as the next cluster center. The calculation formula is: ; In the formula, Represents the probability of each sample point being selected as the next cluster center, x represents the sample point, X 1 means preprocessing the dataset, Indicates the shortest distance between each sample point and the current cluster center; Select the sample point corresponding to the maximum probability value or the maximum probability distribution as the next cluster center; The cluster centers and cluster assignments of sample points are iteratively updated based on the k-means algorithm until the cluster assignments no longer change. The category to which the known sample points have the highest probability of belonging is calculated by comparing the posterior probabilities of each labeled sample point in each cluster and selecting the category with the highest probability.

[0010] Furthermore, the geological constraints include: Based on the distance constraint, determine and identify all the labeled data points that have different labels from the current labeled data points; Apply a penalty function to the probability of the labeled data point in the current cluster: ; In the formula, Represents data points Belongs to distribution degree of responsibility, represents a collection of data point indices, The labels of the data points in the distribution The labels of the data points with the highest probability are different. represents the penalty factor, otherwise express .

[0011] Furthermore, the method further comprises: The accuracy, confusion matrix and F1 score are used as evaluation indicators to evaluate the lithology classification and prediction performance of the recognition model for deep lithology identification; wherein the calculation formulas for the accuracy and F1 score are as follows: ; In the formula, Accuracy Indicates the accuracy, and Respectively represent the number of samples classified or predicted accurately and the total number of samples, Precision Indicates the accuracy, Recall represents the recall rate, TPrepresents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted positive examples, FN Represents the number of incorrectly predicted positive examples.

[0012] In a second aspect, the present application provides a deep lithology identification device, the device comprising: A data acquisition module is configured to acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds to a lithology label in a one-to-one spatial position; A data processing module is configured to perform local preserving projection on the data set to obtain a preprocessed data set; The model training module is configured to build a machine learning model, train the machine learning model using the preprocessed data set, and obtain an identification model for deep lithology identification. During the training process, distance constraints and geological constraints are introduced in sequence; wherein the distance constraint is to assign cluster labels by minimizing the distance between a data point and the cluster center closest to the data point, and an initial cluster center is selected to optimize the assignment; and the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

[0013] In a third aspect, the present application provides an electronic device, comprising: a processor and a memory; the memory is used to store instructions; the processor is used to execute the instructions in the memory, so that the electronic device executes the method described in the first aspect.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect.

[0015] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0016] The deep lithology identification method, device, equipment and storage medium provided by the present application have the following beneficial effects: This application is based on multi-source geophysical data to achieve high-precision identification and positioning of deep lithology in complex underground structures; in the case of scarce label data and imbalanced categories, through application verification in the Bayan Obo deposit, it can accurately identify key lithologies such as slate, dolomite, mineralized dolomite and sandstone, and accurately locate mineralized dolomite within a depth range of 200 to 1600 meters, which is highly consistent with drilling data. This application provides a new solution for deep mineral exploration in complex geological environments, and lays the foundation for future research on the mineralization mechanism and lithology distribution pattern of specific mining areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] Figure 1 A flowchart of a deep lithology identification method is shown as an exemplary embodiment; Figure 2 A structural diagram of a deep lithology identification system is shown as an exemplary embodiment; Figure 3 A schematic diagram of a data set shown in an exemplary embodiment, wherein: (a) resistivity profile; (b) velocity profile; (c) density profile; (d) magnetic susceptibility profile; (e) drilling data; Figure 4 FIG. 1 is a feature preprocessing result diagram showing resistivity characteristics as an example of an exemplary embodiment, wherein: (a) original data distribution; (b) data distribution after Gaussian transformation; (c) normalized data distribution; Figure 5 The evaluation result diagram of different experiments on the field data set is shown as an exemplary embodiment, wherein: (a) the confusion matrix of the training set of Experiment 1; (b) the F1 score of the training set of Experiment 1; (c) the confusion matrix of the test set of Experiment 1; (d) the F1 score of the test set of Experiment 1; (e) the confusion matrix of the training set of Experiment 2; (f) the F1 score of the training set of Experiment 2; (g) the confusion matrix of the test set of Experiment 2; (h) the F1 score of the test set of Experiment 2; (i) the confusion matrix of the training set of Experiment 3; (j) the F1 score of the training set of Experiment 3; (k) the confusion matrix of the test set of Experiment 3; (l) the F1 score of the test set of Experiment 3; (m) the confusion matrix of the training set of Experiment 4; (n) the F1 score of the training set of Experiment 4; (o) the confusion matrix of the test set of Experiment 4; (p) the F1 score of the test set of Experiment 4; Figure 6 A lithology prediction cross section diagram is shown for an exemplary embodiment; Figure 7 The figure is a structural diagram of a deep lithology identification device showing an exemplary embodiment.

[0019] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0020] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0021] It should be noted that the links involved in this application and the relevant information in the links, platform-related information, etc. (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data that are understood and authorized by the relevant users or fully authorized by all parties, and the collection, use, processing, transmission, provision, disclosure and application of the relevant data comply with the laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and comply with the principles of legality, legitimacy and necessity.

[0022] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0023] The embodiment of the present application provides a prior evidence collection method. It should be noted that the deep lithology identification method provided in the present application can be implemented on any electronic device with data processing capabilities, or it can be a deep lithology identification system. It should be noted that the deep lithology identification system can be deployed separately on an electronic device in any environment (for example: deployed separately on an edge server in an edge environment), or it can be deployed entirely in a cloud environment, or it can be deployed in a distributed manner in different environments.

[0024] For example, the deep lithology identification system can be logically divided into multiple parts, each with different functions. The parts in the deep lithology identification system can be deployed in any two or three of the electronic device (located on the user side, such as the client), the edge environment, and the cloud environment. The edge environment is an environment that includes a collection of edge electronic devices that are close to the electronic device. The edge electronic devices include: edge servers, edge stations with computing power, etc. The various parts of the deep lithology identification system deployed in different environments or devices work together to realize the functions of the data processing platform.

[0025] It should be understood that this application does not restrictively divide which parts of the deep lithology identification system are deployed in what specific environment. In actual application, it can be adaptively deployed according to the computing power of electronic devices, the resource availability of edge environments and cloud environments, or specific application requirements.

[0026] Figure 1 is a flowchart of a deep lithology identification method shown in an exemplary embodiment. Figure 2 As shown, the method includes steps S1 to S3, which are described in detail as follows: S1: Acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds one-to-one with a lithology label in terms of spatial position.

[0027] In this embodiment, the purpose of step S1 is to establish a data set. For example, data obtained by comprehensive geophysical exploration including gravity, magnetic method, electromagnetic method and seismic method can be used to construct a multi-source geophysical data set according to spatial correspondence.

[0028] Specifically, the dataset can be represented as , which consists of two parts: feature set and label. The feature and label have a one-to-one correspondence in spatial position. Indicates the first, second, ..., nth feature set. In view of the continuity of geological structure, it is particularly important to maintain the local neighbor information of the data. Therefore, when constructing the data set matrix, this embodiment not only considers the traditional geological features, but also takes into account the spatial location information. (horizontal position coordinates) and (depth position coordinates) are important features incorporated into the model. The feature set can be expressed as , which is composed of different geophysical data, among which , Represents physical parameters, such as resistivity, density, Represents the number of samples. The lithology label refers to the lithology category, which is provided by the borehole geological data in this embodiment. Samples with known lithology are called labeled data or labeled data, and samples with unknown lithology are called unlabeled data.

[0029] S2: Perform local preservation projection on the dataset to obtain a preprocessed dataset.

[0030] In this embodiment, the purpose of step S2 is to process the data set constructed in step S1 to obtain a preprocessed data set. Data processing includes various transformations and feature dimensionality reduction methods, using a combination of various quasi-Gaussian transformations to make the data conform to a "quasi-Gaussian" distribution to be more suitable for a Gaussian mixture model; using the local preserving projection (LPP) method to reduce the dimensionality of multi-source geophysical data, maintaining the spatial proximity of the data, so as to improve the model's ability to recognize complex geological structures.

[0031] In an exemplary embodiment, a Locality Preserving Projections (LPP) technique is used, which specifically achieves dimensionality reduction by preserving the local structure of the data. The processed data set not only has fewer features but also retains key spatial structure properties.

[0032] In the LPP method, the key step is to construct a weight matrix , which captures the local adjacency between data points. This weight matrix is ​​usually defined by a Gaussian kernel function, where two data points close to each other and will be given a larger weight. Then, the goal of LPP is to find a projection matrix so that the data still maintains its original local structure in the new low-dimensional space. The optimization objective function of this process is: ; In the formula, represents a matrix of numbers, represents a diagonal matrix, represents the identity matrix, A represents the projection matrix, x i and x j Indicates the data set i and j data points, W ij Represents data points x i and data points x j The weight matrix of T Represents matrix transpose.

[0033] S3: Build a machine learning model and train it with the preprocessed data set to obtain an identification model for deep lithology identification. During the training process, introduce distance constraints and geological constraints in turn; the distance constraint assigns cluster labels by minimizing the distance between the data point and the cluster center closest to the data point, and selects the initial cluster center to optimize the assignment; the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

[0034] In this embodiment, the purpose of step S3 is to train the constructed machine learning model through the preprocessed data obtained in step S2 to obtain an identification model that can realize deep lithology identification. Specifically, in the face of the scarcity of lithology label data and the imbalance of various lithology labels in the real data of the mining area, a double-constrained semi-supervised learning method (TCS-VGMM) is proposed by improving the VGMM algorithm, which includes two constraints: distance constraint (KMeanspp) and geological constraint (penalty function).

[0035] In an exemplary embodiment, the distance constraint is to assign cluster labels by minimizing the distance between the data point and its nearest cluster center through kmeanspp, and select the initial cluster center to optimize this assignment. The basic principle is: randomly (uniformly distributed) select a sample point from the preprocessed data set as the first initial cluster center Then calculate the shortest distance between each sample and the current cluster center, using Represents; then calculate the probability of each sample point being selected as the next cluster center , and finally select the sample point corresponding to the maximum probability value (or probability distribution) as the next cluster center: ; In the formula, Represents the probability of each sample point being selected as the next cluster center, x represents the sample point, X 1 means preprocessing the dataset, Indicates the shortest distance between each sample point and the current cluster center.

[0036] Then, the cluster centers and cluster assignments of data points are iteratively updated through the standard k-means algorithm until the cluster assignments no longer change. The most likely category of a known data point is then calculated by comparing the posterior probabilities of each labeled data point in each cluster and selecting the category with the highest probability.

[0037] In this embodiment, the clustering and classification performance can be enhanced through the interaction of distance constraints and label constraints. The distance constraint optimizes the mean through the KMeans++ algorithm , covariance and blend weights The goal is to minimize the distance between the data point and the cluster center. The process iteratively updates the responsibility matrix ( ) and optimize the Gaussian distribution parameters in the M step to improve the clustering structure by maximizing the likelihood function. Label constraints work in the E step to prevent conflicts between Gaussian components and improve the optimization effect. After convergence, data points are assigned to Gaussian components based on the maximum responsibility value and classified according to the training labels. Unconstrained Gaussian components are marked as undefined, which helps data prediction. The label constraint is expressed as: ; in, represents the total number of data points, represents the number of cluster centers, represents the jth cluster center, x i Indicates i data points, J Represents a label constraint.

[0038] On this basis, geological constraints based on penalty functions are further imposed, and the provided lithology labels are used as constraints to guide the KMeanspp algorithm to achieve more appropriate data partitioning. Its basic principle is: for each class, samples with the same label (lithology) can be assigned to the same or different Gaussian components; while samples with different labels cannot be assigned to the same Gaussian components. That is, all labeled data points with different labels from the current labeled data points will be identified. A penalty function is applied to the probability of these data points on the current cluster: ; In the formula, Represents data points Belongs to distribution degree of responsibility, is a collection of indices of data points whose labels are related to the The labels of the data points with the highest probability in each cluster are different. is the penalty factor. , otherwise express This is equivalent to avoiding assigning these data points to this cluster. This step is the key to implementing the constraints, which directly uses the known label information to influence the cluster assignment process.

[0039] Finally, the adjusted posterior probability (R) is normalized to ensure that the sum of the cluster probabilities for each data point is 1. This ensures that the adjusted R can be correctly interpreted as a probability distribution.

[0040] In an exemplary embodiment, the accuracy rate ( Accuracy), confusion matrix and F1 score are three commonly used model performance indicators to evaluate the lithology classification and prediction performance of the trained model. Accuracy is the ratio of the number of correctly predicted instances to the total number of instances. The confusion matrix is ​​a matrix that specifically describes the performance of the model and is used to evaluate the accuracy of the classification model. For the confusion matrix (taking the binary classification problem as an example), it usually includes four parts: True Positive ( TP ), False Positive ( FP ), True Negative ( TN ), False Negative( FN ), which represent the number of correctly predicted positive examples, the number of incorrectly predicted positive examples (actually negative examples), the number of correctly predicted negative examples, and the number of incorrectly predicted negative examples (actually positive examples. The F1 score is a comprehensive consideration of the precision rate ( Precision ) and recall ( Recall ) are measured. Their calculation formulas are: ; In the formula, Accuracy Indicates the accuracy, and Respectively represent the number of samples classified or predicted accurately and the total number of samples, Precision Indicates the accuracy, Recall represents the recall rate, TP represents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted positive examples, FN Represents the number of incorrectly predicted positive examples.

[0041] The higher the precision, the more accurate the model is in predicting the positive class and the fewer false positives it has. The higher the recall, the better the model is at identifying actual positive examples and the fewer positive examples it misses. The F1 score ranges from 0 to 1, with 1 representing the best possibility and 0 representing the worst performance. The higher the value, the better the model performs in considering both precision and comprehensiveness (i.e., recall).

[0042] The following embodiments of this application will take the measured data of the Bayan Obo mining area as an example to verify the deep lithology identification method proposed in the above embodiments.

[0043] Figure 2This is a structural diagram of a deep lithology identification system shown in an exemplary embodiment, and the deep lithology identification system is used to execute the deep lithology identification method described above. The deep lithology identification system includes a first module 201, a second module 202 and a third module 203, which are used to respectively execute steps S1 to S3 in the deep lithology identification method process, which are described in detail as follows.

[0044] The first module 201 is used to generate a data set based on multi-source geoscientific data. The generated data set is as follows: Figure 3 shown.

[0045] The second module 202 is used to perform data processing, including various transformations and feature dimensionality reduction.

[0046] When performing data set preprocessing, this embodiment takes resistivity characteristics as an example to demonstrate how to convert data into a form that approximates Gaussian distribution. Figure 4 (a) is the distribution characteristics of resistivity data. Figure 4 (b) shows the features after Box-Cox transformation. It is found that the data already belongs to Gaussian distribution. Then, the feature is standardized to obtain Figure 4 (c) After completing the Gaussian transformation of all features, the local preserving projection algorithm (LPP) is used to reduce the dimensionality of data features including position coordinates, resistivity, magnetic susceptibility, density, and velocity. Here, reduction to 3 dimensions is used as an example.

[0047] The third module 203 is used to implement DCSM model (machine learning model) training and evaluation.

[0048] Combination Figure 1 As shown in the flowchart, before inputting data into the TCS-VGMM model, the features are first normalized and dimensionally reduced. Since the theoretical synthetic data already conforms to the Gaussian distribution, the "quasi-Gaussian" transformation is not used as a preprocessing method, and the processed data is used for the training of the TCS-VGMM model. Based on a large number of experiments, the initial parameters of all models in this embodiment are and They are set to 0.1 respectively, the model is iterated 1000 times, and the tolerance of variational inference is set to 1e-8.

[0049] This example applies DCSM to the Bayan Obo deposit to evaluate its effectiveness in a complex geological environment. Four experiments were conducted using different experiments to evaluate field data sets. The four experiments are: Experiment 1: Select drilling holes ①⑥ as training sets, and ②③④⑤ as test sets.

[0050] Experiment 2: Select drilling holes ①②⑥ as training sets, and ③④⑤ as test sets.

[0051] Experiment 3: Select drilling holes ①②⑤⑥ as training sets, and ③④ as test sets.

[0052] Experiment 4: Select drilling holes ①②④⑤⑥ as training sets and ③ as test sets.

[0053] The results of the four experiments are as follows Figure 5 As shown in the figure, these experiments are based on field data, using labels of non-training boreholes as the test dataset. The small figures in each row represent the results of a single experiment, showing the confusion matrix and F1 score of the training and test sets. Classes 1 to 4 on the x-axis represent slate, dolomite, mineralized dolomite, and sandstone, respectively. The experiments show that adding labeled data significantly improves model performance. In the training set, the accuracy increased from 95.06% to 97.71%, while the F1 score increased from 0.9391 to 0.9511. The confusion matrix shows that due to data imbalance, eight sandstone samples were misclassified as slate because the number of slate samples was several times more. Data augmentation may help to solve this problem in future studies. In the test set, adding more labeled data increased the accuracy from 48.54% to 82.69% and the F1 score from 0.4310 to 0.8494, indicating that increasing the number of boreholes can improve prediction accuracy in complex geological environments.

[0054] Therefore, this example takes the lithology distribution predicted by the model in Experiment 4 as the final result of the discussion. Figure 6 As shown in the predicted cross-section, ⑦ represents the newly studied drill hole (KY14-02-01).

[0055] Based on the latest field observations and drilling data, this embodiment systematically verifies and analyzes the reliability of the prediction results of the DCSM model. In the geological map, the study section shows conglomerate (including granite outcrops), slate, dolomite and gneissic tonalite from south to north. The prediction results of the DCSM model accurately reflect the spatial distribution characteristics of this lithology, identifying differentiated new lithologies at both ends of the section, and predicting slate and dolomite in the middle, which is highly consistent with the geological mapping data. In particular, the scientific research deep well KY14-02-01 (drilling depth of about 2,000 meters) 140 meters south-west of the section provides key verification: drilling reveals that 0-0.7 kilometers are mainly slate, 0.7-1.5 kilometers are ore-bearing dolomite, 1.5-1.8 kilometers are interbedded slate and dolomite, and below 1.8 kilometers are mainly slate. This distribution sequence is highly consistent with the model prediction results.

[0056] The DCSM model shows that the ore-bearing dolomite beneath the main and eastern mining areas transforms into slate at a depth of about 1.8 kilometers, which is consistent with the deep drilling data. Combined with the regional tectonic evolution background, it is speculated that during the mineralization period of the Bayan Obo area about 1.3 billion years ago, there may have been a deep carbonate magma channel between the main and eastern mines, but it was destroyed during the later multi-stage tectonic transformation process. This understanding provides a new perspective for understanding the mineralization process in the region.

[0057] Figure 7 is a structural diagram of a deep rock identification device shown in an exemplary embodiment. The present application embodiment provides a deep rock identification device, such as Figure 7 As shown, the deep lithology identification device includes: The data acquisition module 701 is configured to acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds to a lithology label in a one-to-one spatial position; A data processing module 702 is configured to perform local preserving projection on the data set to obtain a preprocessed data set; The model training module 703 is configured to construct a machine learning model, train the machine learning model using the preprocessed data set, and obtain an identification model for deep lithology identification. During the training process, distance constraints and geological constraints are introduced in sequence; wherein, the distance constraint is to assign cluster labels by minimizing the distance between a data point and the cluster center closest to the data point, and select an initial cluster center to optimize the assignment; and the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

[0058] In some embodiments, the data acquisition module is further configured to: Constructing a weight matrix; wherein the weight matrix is ​​defined by a Gaussian kernel function, and the weight matrix is ​​used to capture the local adjacency relationship between each data point in the data set; Finding a projection matrix such that the data still maintains the original local structure in the low-dimensional space determined by the projection matrix is ​​used as the optimization goal, and an objective function is established; A projection matrix is ​​determined based on the objective function, and the data set is locally projected using the projection matrix to obtain a preprocessed data set.

[0059] In some embodiments, the objective function is expressed as: ; In the formula, represents a matrix of numbers, represents a diagonal matrix, represents the identity matrix, A represents the projection matrix, xi and x j Indicates the data set i and j data points, W ij Represents data points x i and data points x j The weight matrix of T Represents matrix transpose.

[0060] In some embodiments, the distance constraint includes: Randomly select a sample point from the preprocessed data set as the first initial cluster center .

[0061] Calculate the shortest distance between each sample point and the current cluster center; Calculate the probability of each sample point being selected as the next cluster center. The calculation formula is: ; In the formula, Represents the probability of each sample point being selected as the next cluster center, x represents the sample point, X 1 means preprocessing the dataset, Indicates the shortest distance between each sample point and the current cluster center; Select the sample point corresponding to the maximum probability value or the maximum probability distribution as the next cluster center; The cluster centers and cluster assignments of sample points are iteratively updated based on the k-means algorithm until the cluster assignments no longer change. The category to which the known sample points have the highest probability of belonging is calculated by comparing the posterior probabilities of each labeled sample point in each cluster and selecting the category with the highest probability.

[0062] In some embodiments, the geological constraints include: Based on the distance constraint, determine and identify all the labeled data points that have different labels from the current labeled data points; Apply a penalty function to the probability of the labeled data point in the current cluster: ; In the formula, Represents data points Belongs to distribution degree of responsibility, represents a collection of data point indices, The labels of the data points in the distribution The labels of the data points with the highest probability are different. represents the penalty factor, otherwise express .

[0063] In some embodiments, the model training module is further configured to: The accuracy, confusion matrix and F1 score are used as evaluation indicators to evaluate the lithology classification and prediction performance of the recognition model for deep lithology identification; wherein the calculation formulas for the accuracy and F1 score are as follows: ; In the formula, Accuracy Indicates the accuracy, and Respectively represent the number of samples classified or predicted accurately and the total number of samples, Precision Indicates the accuracy, Recall represents the recall rate, TP represents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted positive examples, FN Represents the number of incorrectly predicted positive examples.

[0064] The deep lithology identification device provided in this embodiment can be used to execute the above-mentioned deep lithology identification method. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0065] An embodiment of the present application provides an electronic device, which may include: a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus, the memory is used to store computer execution instructions, and the processor is used to call the computer execution instructions in the memory to execute the deep lithology identification method shown in any of the above method embodiments.

[0066] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the present application may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.

[0067] The present application provides a computer-readable storage medium, on which computer-executable instructions are stored; when the computer-executable instructions are executed by a processor, they are used to implement a deep lithology identification method as in any of the above-mentioned embodiments.

[0068] An embodiment of the present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned deep lithology identification method is implemented.

[0069] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0070] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A deep lithology identification method, characterized in that: include: Acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds to a lithology label in a one-to-one spatial position; Performing local preserving projection on the data set to obtain a preprocessed data set; A machine learning model is constructed and trained using the preprocessed data set to obtain an identification model for deep lithology identification. During the training process, distance constraints and geological constraints are introduced in sequence; wherein the distance constraint is to assign cluster labels by minimizing the distance between a data point and the cluster center closest to the data point, and an initial cluster center is selected to optimize the assignment; and the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

2. The method according to claim 1, characterized in that Performing local preservation projection on the data set to obtain a preprocessed data set, including: Constructing a weight matrix; wherein the weight matrix is ​​defined by a Gaussian kernel function, and the weight matrix is ​​used to capture the local adjacency relationship between each data point in the data set; Finding a projection matrix such that the data still maintains the original local structure in the low-dimensional space determined by the projection matrix is ​​used as the optimization goal, and an objective function is established; A projection matrix is ​​determined based on the objective function, and the data set is locally projected using the projection matrix to obtain a preprocessed data set.

3. The method according to claim 1, characterized in that The objective function is expressed as: ； In the formula, represents a matrix of numbers, represents a diagonal matrix, represents the identity matrix, A represents the projection matrix, x i and x j Indicates the data set i and j data points, W ij Represents data points x i and data points x j The weight matrix of T Represents matrix transpose.

4. The method according to claim 1, characterized in that: The distance constraints include: Randomly select a sample point from the preprocessed data set as the first initial cluster center ; Calculate the shortest distance between each sample point and the current cluster center; Calculate the probability of each sample point being selected as the next cluster center. The calculation formula is: ； In the formula, Represents the probability of each sample point being selected as the next cluster center, x represents the sample point, X 1 means preprocessing the dataset, Indicates the shortest distance between each sample point and the current cluster center; Select the sample point corresponding to the maximum probability value or the maximum probability distribution as the next cluster center; The cluster centers and cluster assignments of sample points are iteratively updated based on the k-means algorithm until the cluster assignments no longer change. The category to which the known sample points have the highest probability of belonging is calculated by comparing the posterior probabilities of each labeled sample point in each cluster and selecting the category with the highest probability.

5. The method according to claim 1, characterized in that The geological constraints include: Based on the distance constraint, determine and identify all the labeled data points that have different labels from the current labeled data points; Apply a penalty function to the probability of the labeled data point in the current cluster: ； In the formula, Represents data points Belongs to distribution degree of responsibility, represents a collection of data point indices, The labels of the data points in the distribution The labels of the data points with the highest probability are different. represents the penalty factor, otherwise express .

6. The method according to claim 1, characterized in that The method further comprises: The accuracy, confusion matrix and F1 score are used as evaluation indicators to evaluate the lithology classification and prediction performance of the recognition model for deep lithology identification; wherein the calculation formulas for the accuracy and F1 score are as follows: ； In the formula, Accuracy Indicates the accuracy, and Respectively represent the number of samples classified or predicted accurately and the total number of samples, Precision Indicates the accuracy, Recall represents the recall rate, TP Represents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted positive examples, FN Represents the number of incorrectly predicted positive examples.

7. A deep lithology identification device, characterized in that: The device comprises: A data acquisition module is configured to acquire a data set; wherein the data set includes a feature set and a lithology label, and each feature included in the feature set corresponds to a lithology label in a one-to-one spatial position; A data processing module is configured to perform local preserving projection on the data set to obtain a preprocessed data set; The model training module is configured to build a machine learning model, train the machine learning model using the preprocessed data set, and obtain an identification model for deep lithology identification. During the training process, distance constraints and geological constraints are introduced in sequence; wherein the distance constraint is to assign cluster labels by minimizing the distance between a data point and the cluster center closest to the data point, and an initial cluster center is selected to optimize the assignment; and the geological constraint uses the lithology label provided by the preprocessed data set as a constraint condition.

8. An electronic device, characterized in that: The electronic device comprises: a processor and a memory; the memory is used to store instructions; the processor is used to execute the instructions in the memory, so that the electronic device executes the method as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

10. A computer program product, characterized in that The method comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.

Images