Geochemical anomaly identification method and system based on causal discovery and deep learning

Abstract

The invention discloses a geochemical anomaly identification method and system based on causal discovery and deep learning, and relates to the technical field of geochemical anomaly identification, and the method comprises the steps: constructing a graph structure with sampling points as nodes based on exploration geochemical data, the node features being element concentration values of the sampling points, and constructing edges based on a geographic proximity relation; a linear non-Gaussian directed acyclic model is adopted to carry out causal analysis on the node features, and a causal weight matrix is generated; optimizing a graph structure constructed based on geochemical data based on a causal weight matrix, and performing causal adjustment on node features; inputting the causal-adjusted graph structure into a graph attention network, and performing weighted aggregation on node information through an attention mechanism; and training the graph attention network model to complete geochemical anomaly recognition. According to the method, the influence of noise data on an identification result is effectively reduced by removing false correlation features and strengthening causal correlation features, and the method has good robustness and interpretability.

Description

Technical Field

[0001] The present invention relates to the technical field of geochemical anomaly identification, and in particular to a geochemical anomaly identification method and system based on causal discovery and deep learning. Background Art

[0002] Deep learning has strong nonlinear fitting and feature extraction capabilities, and can deeply mine complex patterns hidden in geochemical data. It has become a cutting-edge technology for mining geochemical exploration data to identify geochemical patterns related to mineralization. However, deep learning is a data-driven model that often does not fully take into account metallogenic theory and expert knowledge. It only learns its potential patterns from training data and focuses on the correlation between elements and ore deposits without fully considering the causal relationship between them. As a result, it faces a series of urgent problems in terms of interpretability, generalization, and robustness. In addition, the current deep learning models for geochemical anomaly identification are mainly based on a single grid unit or pixel block (image), which is difficult to fully characterize the spatial pattern characteristics of geochemistry and the spatial coupling relationship between ore deposits and geochemical elements. The graph-based deep learning algorithm inputs all nodes and their connection relationships into the network, so it can better capture the relationship between training samples and effectively identify geochemical anomalies. Graph attention network is an important graph deep neural network method, which is widely used in geochemical anomaly identification and extraction. Data noise may occur during the sampling and preprocessing of exploration geochemical data. For example, due to the limitation of instrument analysis accuracy, some geochemical data cannot be effectively detected, resulting in missing values. These data interference problems caused by noise and incomplete data have brought challenges to the application of deep learning in geochemical anomaly identification, especially graph neural networks. When the neighborhood nodes in the graph are disturbed, the classification and prediction accuracy of the model will be affected and reduced. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that data noise affects the accuracy of graph-based deep learning algorithms in geochemical anomaly identification, and propose a geochemical anomaly identification method based on causal discovery and deep learning, comprising the following steps:

[0004] S1. Based on the exploration geochemical data, a graph structure with sampling points as nodes is constructed, where the node feature is the element concentration value of the sampling point, and the edge is constructed based on the geographical proximity relationship;

[0005] S2, using a linear non-Gaussian directed acyclic model to perform causal analysis on node features and generate a causal weight matrix;

[0006] S3, optimize the graph structure built based on geochemical data based on the causal weight matrix, and make causal adjustments to node features;

[0007] S4, input the causally adjusted graph structure into the graph attention network, and perform weighted aggregation of node information through the attention mechanism;

[0008] S5. Train the graph attention network model to complete geochemical anomaly identification.

[0009] Furthermore, in S1, based on the exploration geochemical data, before constructing a graph structure with sampling points as nodes, the exploration geochemical data is standardized.

[0010] Furthermore, the Delaunay triangulation method is used to construct the adjacency matrix.

[0011] Furthermore, the causal weight matrix is ​​generated by the formula Α=Bχ+e, where χ represents the observed variable, B is a lower triangular matrix representing the causal weights between variables, and e is a non-Gaussian noise term.

[0012] Furthermore, a threshold method is used to perform causal adjustment on node features, eliminating node features whose causal weights are lower than the set threshold, and assigning different weights to node features whose causal weights are higher than the set threshold according to the causal weight values.

[0013] Furthermore, the attention mechanism of the graph attention network calculates the attention weight through the following formula:

[0014]

[0015] Among them, α ij represents the attention weights of the i-th node and the j-th node, represents a trainable parameter vector, W represents a trainable weight matrix, represents the feature vector of the i-th node, represents the feature vector of the jth node, ∥ represents the concatenation operation, LeakyRe LU represents the linear activation function, N i Represents the set of neighboring nodes of the i-th node.

[0016] Furthermore, the training data of the graph attention network model consists of positive samples and negative samples. The positive samples are sampling points of known mineral deposits, and the negative samples are random sampling points far away from the positive sample area with a number equivalent to that of positive samples.

[0017] The present invention also provides a geochemical anomaly identification system, comprising:

[0018] A graph structure construction unit is used to construct a graph structure with sampling points as nodes based on exploration geochemical data, where the node feature is the element concentration value of the sampling point, and edges are constructed based on geographical proximity;

[0019] A causal analysis unit, used to perform causal analysis on node features using a linear non-Gaussian directed acyclic model to generate a causal weight matrix;

[0020] A causal adjustment unit is used to optimize the graph structure built based on geochemical data based on the causal weight matrix and perform causal adjustment on node features;

[0021] The node information weighted aggregation unit is used to input the causally adjusted graph structure into the graph attention network and perform weighted aggregation of node information through the attention mechanism;

[0022] The geochemical anomaly identification unit is used to train the graph attention network model to complete geochemical anomaly identification.

[0023] The beneficial effects brought by the technical solution provided by the present invention are:

[0024] The present invention first constructs a graph structure based on exploration geochemical data, introduces a causal discovery algorithm, removes pseudo-correlation features in geochemical data in the graph structure, and strengthens causal correlation features, effectively reducing the impact of noise data on recognition results, improving the expressiveness of the graph structure, and combining the graph attention mechanism to perform weighted aggregation on node information, thereby improving the model's recognition of geochemical anomaly patterns. The present invention not only has good robustness and interpretability, but also has strong adaptability and scalability, providing reliable technical support for geological prospecting and mineral resource evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of a geochemical anomaly identification method based on causal discovery and deep learning according to an embodiment of the present invention;

[0026] Figure 2 is a cause-effect relationship diagram of geochemical element data of a certain region constructed by using a cause-effect analysis method in an embodiment of the present invention;

[0027] Figure 3 It is a causal relationship diagram of geochemical element data screened out from a certain area in an embodiment of the present invention, showing the absolute values ​​of causal weights between elements (lithium is taken as an example in the diagram);

[0028] Figure 4 It is a comparison diagram between the method of an embodiment of the present invention and a graph attention network without introducing causal relationship. DETAILED DESCRIPTION

[0029] 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.

[0030] The flowchart of the geochemical anomaly identification method according to the embodiment of the present invention is as follows: Figure 1, specifically including the following steps:

[0031] S1. Based on the exploration geochemical data, a graph structure with sampling points as nodes is constructed, where the node features are the element concentration values ​​of the sampling points, and edges are constructed based on geographical proximity.

[0032] For the original exploration geochemical data, it is necessary to perform standardization first. Data standardization is a process of converting data of different dimensions or magnitudes into a unified standard for comparison, analysis and weighting. In data processing, the choice of standardization method has an important impact on the analysis results. The following are several commonly used data standardization methods:

[0033] Conventional standardization methods:

[0034] (1) Range standardization: linearly transform the data to the interval [0,1]. This method is simple and easy to implement, but it is sensitive to outliers. If there are extremely large or extremely small values ​​in the data, most of the data may be concentrated in a smaller interval, affecting subsequent analysis.

[0035] (2) Z-score standardization: The data are converted to a standard normal distribution by subtracting the mean and dividing by the standard deviation. This method can eliminate the dimension and magnitude differences of the data and make different indicators comparable, but it requires that the data obey or approximately obey the normal distribution.

[0036] Geochemical Zoning Standardization Method:

[0037] The mineralization and halo elements are standardized according to geochemical zoning. This method can effectively suppress non-mineral anomalies in high-background areas, strengthen mineral-induced anomalies in low-background areas, and highlight prospecting information. It is particularly suitable for extracting prospecting information from a large amount of regional geochemical exploration data.

[0038] Factor partition standardization method after ILR transformation:

[0039] First, the geochemical data is transformed by ILR data to eliminate the closure effect of the data, and then factor analysis is performed. According to the factor partitioning results, the sample data is partitioned and standardized in combination with the EDA method. This method can remove the influence of high background values ​​in the geological body, make the delineated anomalies correspond better to the spatial position of the ore deposit, and is less affected by human factors.

[0040] The Delaunay triangulation method is used to construct the adjacency matrix. For example, if there are four points A, B, C, and D, two triangles ABC and BCD can be constructed respectively at a suitable threshold distance. However, it is necessary to ensure that the circumscribed circle of any triangle does not contain other points. The adjacency relationship is established according to the sides of the triangle, such as AB is connected, BC is connected, CD is connected, and AD is not connected. At this time, 1 indicates that the two points are connected, and 0 indicates that they are not adjacent to construct the corresponding matrix.

[0041] S2. Use the linear non-Gaussian directed acyclic model (LiNGAM) to perform causal analysis on node features and generate a causal weight matrix.

[0042] Specifically, the LiNGAM algorithm assumes that these elements all contain a set of non-Gaussian noise, such as interference caused by errors in the sampling process or errors in measurement, and then they are brought into the LiNGAM causal weight matrix generation formula, which is Α=Bχ+e, where A represents the causal weight matrix, χ is a vector representing the values ​​of all observed variables, B is a lower triangular matrix representing the causal weights between variables, and e is a non-Gaussian noise term. We can then bring each pair of elements into the formula, and determine whether there is an explainable causal relationship between a pair of elements by whether the pair of elements can be represented by the formula, and determine their causal direction and causal weight.

[0043] refer to Figure 2 and Figure 3 , Figure 2 is a causal relationship diagram of geochemical element data of a certain region constructed by the causal analysis method in an embodiment of the present invention, wherein the direction of the arrow indicates the direction of the causal relationship between the data variables, Figure 3 It is a causal relationship diagram of geochemical element data screened from a certain area in an embodiment of the present invention, showing the absolute values ​​of the causal weights between elements. The direction of the arrow also indicates the causal direction, and the thickness of the line indicates the absolute values ​​of the causal weights between a pair of elements. Figure 3 It can be seen that there is a strong causal relationship between K2O and Li, indicating that K2O has an important influence on the lithium mineralization process.

[0044] S3. Optimize the graph structure built based on geochemical data based on the causal weight matrix and make causal adjustments to node features.

[0045] The threshold method is used to perform causal adjustment on node features, and node features with causal weights lower than the set threshold are eliminated. For node features with causal weights higher than the set threshold, variables with higher causal weights are given greater weights according to the causal weight values. The specific threshold can be set according to the distribution characteristics of geochemical data. Causal adjustment weakens the influence of pseudo-correlation features and highlights key variables with strong causal relationships, thereby enhancing the expressive power of key variables in node features. The geochemical map after causal adjustment can more accurately reflect the intrinsic causal relationship between geochemical elements and mineral deposits, eliminate pseudo-correlation features in the data, and improve the expressive power of geochemical maps.

[0046] S4. Input the causally adjusted graph structure into the graph attention network, perform weighted aggregation on the node information through the attention mechanism, and generate an optimized node feature representation.

[0047] The attention mechanism of the graph attention network calculates the attention weight through the following formula:

[0048]

[0049] Among them, α ij represents the attention weights of the i-th node and the j-th node, represents a trainable parameter vector, W represents a trainable weight matrix, represents the feature vector of the i-th node, represents the feature vector of the jth node, ∥ represents the concatenation operation, LeakyRe LU represents the linear activation function, N i Represents the neighboring nodes of the i-th node.

[0050] Through the attention mechanism, the weighted fusion of the central node's information of its neighboring nodes is completed, making the key variables more representative in the aggregated feature expression.

[0051] S5. Train the graph attention network model to complete geochemical anomaly identification.

[0052] The training data of the graph attention network model consists of positive samples and negative samples. The positive samples are sampling points of known mineral deposits, and the negative samples are random sampling points far away from the positive sample area with the same number as the positive samples. Through multiple training iterations and optimization, the graph attention network model can accurately aggregate node feature information, effectively improve the accuracy of geochemical anomaly identification, and enhance anti-disturbance capabilities.

[0053] In an exemplary embodiment, a geochemical anomaly identification system is provided, the system comprising:

[0054] A graph structure construction unit is used to construct a graph structure with sampling points as nodes based on exploration geochemical data, where the node feature is the element concentration value of the sampling point, and edges are constructed based on geographical proximity;

[0055] A causal analysis unit, used to perform causal analysis on node features using a linear non-Gaussian directed acyclic model to generate a causal weight matrix;

[0056] A causal adjustment unit is used to optimize the graph structure built based on geochemical data based on the causal weight matrix and perform causal adjustment on node features;

[0057] The node information weighted aggregation unit is used to input the causally adjusted graph structure into the graph attention network and perform weighted aggregation of node information through the attention mechanism;

[0058] The geochemical anomaly identification unit is used to train the graph attention network model to complete geochemical anomaly identification.

[0059] In order to verify the effectiveness of the method of the present invention, in the embodiment of the present invention, the method of the present invention is compared with the graph attention network without introducing causal relationship. The comparison results are as follows: Figure 4 As shown. Figure 4 It can be seen that the method of the present invention has strong anti-disturbance ability and high accuracy in identifying geochemical anomalies.

[0060] Through the above steps, the present invention realizes the measurement of the causal relationship between geochemical data and mineral deposits and the effective identification and extraction of geochemical spatial patterns. The model not only effectively eliminates pseudo-correlation features and improves the anti-interference ability of data, but also further improves the recognition effect of geochemical anomalies through the optimized graph attention network, providing scientific support for geological prospecting and resource evaluation.

[0061] 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 geochemical anomaly identification method based on causal discovery and deep learning, characterized in that: The following steps are involved: S1. Based on the exploration geochemical data, a graph structure with sampling points as nodes is constructed, where the node feature is the element concentration value of the sampling point, and the edge is constructed based on the geographical proximity relationship; S2, using a linear non-Gaussian directed acyclic model to perform causal analysis on node features and generate a causal weight matrix; S3, optimize the graph structure based on the causal weight matrix and make causal adjustments to node features; S4, input the causally adjusted graph structure into the graph attention network, and perform weighted aggregation of node information through the attention mechanism; S5. Train the graph attention network model to complete geochemical anomaly identification.

2. A geochemical anomaly identification method based on causal discovery and deep learning according to claim 1, characterized in that: In S1, based on the exploration geochemical data, the exploration geochemical data are standardized before constructing a graph structure with sampling points as nodes.

3. A geochemical anomaly identification method based on causal discovery and deep learning according to claim 1, characterized in that: The Delaunay triangulation method is used to construct the adjacency matrix.

4. The geochemical anomaly identification method based on causal discovery and deep learning according to claim 1 is characterized in that: The causal weight matrix is ​​generated by the formula Α=Bχ+e, where A represents the causal weight matrix, χ represents the observed variables, B is a lower triangular matrix representing the causal weights between variables, and e is a non-Gaussian noise term.

5. The geochemical anomaly identification method based on causal discovery and deep learning according to claim 1, characterized in that: The threshold method is used to perform causal adjustment on node features, eliminating node features whose causal weight is lower than the set threshold. For node features whose causal weight is higher than the set threshold, different weights are assigned according to the causal weight value.

6. The method for identifying geochemical anomalies based on causal discovery and deep learning according to claim 1, characterized in that: The attention mechanism of the graph attention network calculates the attention weight through the following formula: Among them, α ij represents the attention weights of the i-th node and the j-th node, represents a trainable parameter vector, W represents a trainable weight matrix, represents the feature vector of the i-th node, represents the feature vector of the jth node, ∥ represents the concatenation operation, LeakyRe LU represents the linear activation function, N i Represents the neighboring nodes of the i-th node.

7. The geochemical anomaly identification method based on causal discovery and deep learning according to claim 1 is characterized in that: The training data of the graph attention network model consists of positive samples and negative samples. The positive samples are sampling points of known mineral deposits, and the negative samples are random sampling points in the non-positive sample area with the same number as the positive samples.

8. A geochemical anomaly identification system based on causal discovery and deep learning, characterized in that: include: A graph structure construction unit is used to construct a graph structure with sampling points as nodes based on exploration geochemical data, where the node feature is the element concentration value of the sampling point, and edges are constructed based on geographical proximity; A causal analysis unit, used to perform causal analysis on node features using a linear non-Gaussian directed acyclic model to generate a causal weight matrix; A causal adjustment unit, which is used to optimize the graph structure based on the causal weight matrix and perform causal adjustment on node features; The node information weighted aggregation unit is used to input the causally adjusted graph structure into the graph attention network and perform weighted aggregation of node information through the attention mechanism; The geochemical anomaly identification unit is used to train the graph attention network model to complete geochemical anomaly identification.

Images