A method and system for source apportionment of heavy metals in soil

Abstract

This invention discloses a method and system for source analysis of heavy metals in soil. The method includes acquiring soil data, extracting graph structure features from the soil data using a graph convolutional neural network model, inputting the graph structure features into a self-organizing competitive learning neural network model to obtain the optimal matching unit between neuron weights and soil data, calculating a planar visualization map of single-element components of heavy metals in the soil data using the two-dimensional visualization of the self-organizing competitive learning neural network model based on the planar visualization map of single-element components, and determining the source of heavy metals in the soil data based on the single-element component planar visualization map. By extracting the graph structure features of the soil data, the self-organizing competitive method can fully learn the correlation information between soil data, improving the efficiency and accuracy of analyzing the source of heavy metals in soil data, and overcoming the drawback of strong subjectivity in heavy metal source analysis.

Description

Technical Field

[0001] This invention relates to the technical field of soil heavy metal pollution prevention and control, and in particular to a method and system for source analysis of soil heavy metals. Background Technology

[0002] Soil is the material foundation upon which human beings depend for survival, and maintaining healthy soil conditions is an important strategy for formulating sustainable soil development. However, with rapid economic development and industrial progress, the health of soil is constantly deteriorating, especially with heavy metal pollution, which poses a potential threat to ecosystems and human health. Therefore, accurately identifying the sources of heavy metals in soil can effectively improve the efficiency of heavy metal pollution prevention and control.

[0003] Heavy metals entering the soil can come from various sources, including agricultural pollution, such as fertilizers, irrigation, pesticides, and herbicides, which introduce heavy metal elements into the soil. Natural and anthropogenic influences also contribute to this, such as the control of elements like Mn, Cr, and Ni in the soil by the parent material. Transportation and industrial sources are also major sources of heavy metals in the soil, such as emissions from the electronics industry, industrial emissions, and vehicle exhaust. Atmospheric deposition is also an important source of heavy metals in the soil.

[0004] Due to differences in geographical location and environmental factors, there are certain correlations among heavy metal elements in soil. These correlations result in graph-structured relationships between samples. Furthermore, current methods for analyzing the sources of heavy metals suffer from the drawback of being highly subjective. Summary of the Invention

[0005] This invention aims to at least address the technical problems existing in the prior art. To this end, this invention proposes a method and system for source apportionment of heavy metals in soil, which can improve the efficiency and accuracy of analyzing the sources of heavy metals in soil data, and overcome the drawback of strong subjectivity in the apportionment of heavy metal sources.

[0006] In a first aspect, the present invention provides a method for source analysis of heavy metals in soil, comprising the following steps:

[0007] Obtain soil data;

[0008] The graph structure features of the soil data were extracted using a graph convolutional neural network model.

[0009] The graph structure features are input into a self-organizing competitive learning neural network model to obtain the best matching unit between the neuron weight values ​​and the soil data.

[0010] Based on the optimal matching unit and the neuron weight values, and using the two-dimensional visualization of the self-organizing competitive learning neural network model, a planar visualization of the single-element components of heavy metals in the soil data is calculated.

[0011] Based on the single-element component planar visualization, the source of heavy metals in the soil data is determined.

[0012] According to embodiments of the present invention, at least the following technical effects are achieved:

[0013] This method acquires soil data, extracts the graph structure features of the soil data using a graph convolutional neural network model, inputs the graph structure features into a self-organizing competitive learning neural network model, obtains the best matching unit between the neuron weight values ​​and the soil data, and uses the two-dimensional visualization of the self-organizing competitive learning neural network model to calculate the planar visualization map of single-element components of heavy metals in the soil data. Based on the planar visualization map of single-element components, the source of heavy metals in the soil data is determined. By extracting the graph structure features of the soil data, the self-organizing competitive method can fully learn the correlation information between soil data, improving the efficiency and accuracy of analyzing the source of heavy metals in soil data, and overcoming the drawback of strong subjectivity in the analysis of heavy metal source.

[0014] According to some embodiments of the present invention, the step of extracting the graph structure features of the soil data using a graph convolutional neural network model includes:

[0015] The initial graph structure features of the soil data are calculated as follows: And pre-set the initial weight values ​​w for each layer of the graph convolutional neural network model. i (0) Where N is the number of soil data points, and l is the number of layers in the graph convolutional neural network model. The input to the graph convolutional neural network model is the k-th metal element of the i-th soil data.

[0016] The similarity matrix of the initial graph structure features is constructed based on the fully connected Gaussian kernel distance, and the adjacency matrix and degree matrix are constructed based on the similarity matrix;

[0017] The graph structure features of the l-th layer of the soil data are calculated based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The calculation formula for the graph structure features of the l-th layer of the soil data is as follows:

[0018]

[0019]

[0020]

[0021]

[0022] in, Let A be the first matrix, A be the adjacency matrix, and I be the identity matrix. Let D be the second matrix, D be the degree matrix, and δ be the nonlinear activation function. w represents the graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model. i (l-1) The element weight value is the i-th soil data in the (l-1)-th layer of the graph convolutional neural network model.

[0023] According to some embodiments of the present invention, the step of inputting the graph structure features into a self-organizing competitive learning neural network model to obtain the optimal matching unit between the neuron weight values ​​and the soil data includes:

[0024] Step S1: Initialize the neuron weight value of each neuron to obtain the first neuron weight value of each neuron;

[0025] Step S2: Standardize the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer to obtain standardized neuron weight values ​​and standardized graph structure features. The calculation formula for standardizing the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer is as follows:

[0026]

[0027]

[0028] in, W represents the standardized graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model. j Let be the weight value of the j-th neuron. Let be the standardized neuron weight value of the j-th neuron, and M be the number of neurons;

[0029] Step S3: Calculate the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features. The calculation formula for calculating the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features is as follows:

[0030]

[0031]

[0032] Where g is the best matching unit for the i-th soil data, d gLet d be the minimum distance from the i-th soil data point to all neurons. j Let be the distance from the i-th soil data point to the j-th neuron;

[0033] Step S4: Calculate the neuron neighborhood radius based on the preset initial neuron neighborhood radius, wherein the calculation formula for calculating the neuron neighborhood radius based on the preset initial neuron neighborhood radius is:

[0034]

[0035] Where, N g (t) represents the neighborhood radius of the neuron in the t-th iteration, T is the preset maximum number of iterations, and N g (0) represents the initial radius of the neuron's neighborhood;

[0036] Step S5: Update the neuron weight value of each neuron according to the neighborhood radius of the neuron and the distance from the soil data to each neuron, and obtain the updated neuron weight value of the i-th soil data in the t-th iteration;

[0037] Step S6: If i equals N, the neuron weight value for the t-th iteration is obtained. If i is less than N, the weight value of the (i+1)-th soil data is calculated based on the updated neuron weight value of the i-th soil data. The distance from the (i+1)-th soil data to each neuron and the optimal matching unit of the (i+1)-th soil data are calculated based on the graph structure features of the (i+1)-th soil data in the l-th layer. The neuron weight value of each neuron is updated based on the neighborhood radius of the neuron and the distance from the (i+1)-th soil data to each neuron, resulting in the updated neuron weight value for the (i+1)-th soil data in the t-th iteration. This process is repeated until i equals N, yielding the neuron weight value for the t-th iteration. The formula for calculating the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data is as follows:

[0038]

[0039] Where β(t) is the pre-set learning rate for the t-th iteration;

[0040] Step S7: Calculate the loss function value for the t-th iteration based on the best matching unit and the second best matching unit. If t equals the maximum number of iterations, obtain the neuron weight values ​​for the soil data. If t is less than the maximum number of iterations, then t = t + 1. Update the graph structure features of the soil data in the graph convolutional neural network model based on the loss function value for the t-th iteration. Obtain the neuron weight values ​​for the (t+1)-th iteration based on the updated graph structure features. Continue this process until t equals the maximum number of iterations to obtain the neuron weight values ​​for the soil data. The formula for calculating the loss function value for the t-th iteration based on the best matching unit and the second best matching unit is:

[0041]

[0042]

[0043] Among them, g i For the i-th soil data, This is the second best matching unit for the i-th soil data.

[0044] According to some embodiments of the present invention, updating the neuron weight value of each neuron based on the neighborhood radius of the neuron and the distance from the soil data to each neuron, to obtain the updated neuron weight value of the i-th soil data in the t-th iteration, includes:

[0045] Compare the distance from the soil data to each neuron with the size of the neuron's neighborhood radius: update the neuron weight value of the neuron whose distance from the soil data to the neuron is less than or equal to the neuron's neighborhood radius to the neuron weight value of the i-th soil data;

[0046] The neuron weights of neurons whose distance from the soil data to the neuron is greater than the neuron's neighborhood radius remain unchanged.

[0047] According to some embodiments of the present invention, the method for source apportionment of heavy metals in soil further includes:

[0048] Obtain the latitude and longitude of the soil data;

[0049] The spatial distribution map of heavy metals in the soil data is calculated based on the latitude and longitude of the soil elements, the best matching unit of the soil elements, and the single-element component planar visualization map.

[0050] The spatial distribution characteristics of heavy metals in the soil data are obtained based on the spatial distribution map.

[0051] A second aspect of the present invention provides a source apportionment system for heavy metals in soil, the soil heavy metal source apportionment system comprising:

[0052] The data acquisition module is used to acquire soil data;

[0053] The graph structure feature extraction module is used to extract the graph structure features of the soil data through a graph convolutional neural network model.

[0054] The optimal matching unit and neuron weight value calculation module are used to input the graph structure features into the self-organizing competitive learning neural network model to obtain the optimal matching unit between the neuron weight values ​​and the soil data.

[0055] The single-element component planar visualization acquisition module is used to calculate the single-element component planar visualization of heavy metals in the soil data based on the best matching unit and the neuron weight value, and using the two-dimensional visualization of the self-organizing competitive learning neural network model.

[0056] The source correlation analysis module is used to determine the source of heavy metals in the soil data based on the single-element component planar visualization.

[0057] This system acquires soil data and extracts its graph structure features using a graph convolutional neural network model. These features are then input into a self-organizing competitive learning neural network model to obtain the optimal matching unit between neuron weights and soil data. Based on this optimal matching unit and neuron weights, and utilizing the two-dimensional visualization of the self-organizing competitive learning neural network model, a planar visualization map of single-element components of heavy metals in the soil data is calculated. Based on this planar visualization map, the source of heavy metals in the soil data is determined. By extracting the graph structure features of the soil data, the self-organizing competitive method can fully learn the correlation information between soil data, improving the efficiency and accuracy of analyzing the source of heavy metals in soil data and overcoming the drawback of strong subjectivity in heavy metal source analysis.

[0058] According to some embodiments of the present invention, the graph structure feature extraction module further includes:

[0059] The data initialization module calculates the initial graph structure features of the soil data as follows: And pre-set the initial weight values ​​w for each layer of the graph convolutional neural network model. i (0) Where N is the number of soil data points, and l is the number of layers in the graph convolutional neural network model. The input to the graph convolutional neural network model is the k-th metal element of the i-th soil data.

[0060] The matrix construction module is used to construct a similarity matrix of the initial graph structure features based on the fully connected Gaussian kernel distance, and to construct an adjacency matrix and a degree matrix based on the similarity matrix;

[0061] The graph structure feature calculation module for soil data is used to calculate the graph structure features of the l-th layer of the soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The calculation formula for calculating the graph structure features of the l-th layer of the soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features is as follows:

[0062]

[0063]

[0064]

[0065]

[0066] in, Let A be the first matrix, A be the adjacency matrix, and I be the identity matrix. Let D be the second matrix, D be the degree matrix, and δ be the nonlinear activation function. w represents the graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model. i (l-1) The element weight value is the i-th soil data in the (l-1)-th layer of the graph convolutional neural network model.

[0067] According to some embodiments of the present invention, the optimal matching unit and the neuron weight value calculation module further include:

[0068] The neuron weight initialization module is used to initialize the neuron weight value of each neuron, and obtain the first neuron weight value of each neuron;

[0069] The data standardization module is used to standardize the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer to obtain standardized neuron weight values ​​and standardized graph structure features. The calculation formula for standardizing the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer is as follows:

[0070]

[0071]

[0072] in, W represents the standardized graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model.j Let be the weight value of the j-th neuron. Let be the standardized neuron weight value of the j-th neuron, and M be the number of neurons;

[0073] The neuron distance calculation module is used to calculate the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features. The calculation formula for calculating the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features is as follows:

[0074]

[0075]

[0076] Where g is the best matching unit for the i-th soil data, d g Let d be the minimum distance from the i-th soil data point to all neurons. j Let be the distance from the i-th soil data point to the j-th neuron;

[0077] The neuron neighborhood radius calculation module is used to calculate the neuron neighborhood radius based on a preset initial neuron neighborhood radius. The calculation formula for calculating the neuron neighborhood radius based on the preset initial neuron neighborhood radius is as follows:

[0078]

[0079] Where, N g (t) represents the neighborhood radius of the neuron in the t-th iteration, T is the preset maximum number of iterations, and N g (0) represents the initial radius of the neuron's neighborhood;

[0080] The neuron weight value update module is used to update the neuron weight value of each neuron according to the neighborhood radius of the neuron and the distance from the soil data to each neuron, so as to obtain the updated neuron weight value of the i-th soil data in the t-th iteration;

[0081] The soil data neuron weight value iteration module is used to obtain the neuron weight value of the t-th iteration if i equals N, and to calculate the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data if i is less than N. It also calculates the distance from the (i+1)-th soil data to each neuron and the optimal matching unit of the (i+1)-th soil data based on the graph structure features of the (i+1)-th soil data in the l-th layer, and updates the neuron weight value of each neuron based on the neighborhood radius of the neuron and the distance from the (i+1)-th soil data to each neuron, obtaining the updated neuron weight value of the (i+1)-th soil data in the t-th iteration. This process is repeated until i equals N, obtaining the neuron weight value in the t-th iteration. The calculation formula for obtaining the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data is as follows:

[0082]

[0083] Where β(t) is the pre-set learning rate for the t-th iteration;

[0084] The neuron weight value iteration module is used to calculate the loss function value of the t-th iteration based on the best matching unit and the second best matching unit in the t-th iteration. If t equals the maximum number of iterations, the neuron weight value of the soil data is obtained. If t is less than the maximum number of iterations, then t = t + 1. The graph structure features of the soil data in the graph convolutional neural network model are updated based on the loss function value of the t-th iteration. The neuron weight value of the (t+1)-th iteration is obtained based on the updated graph structure features, and so on, until t equals the maximum number of iterations, to obtain the neuron weight value of the soil data. The calculation formula for calculating the loss function value of the t-th iteration based on the best matching unit and the second best matching unit in the t-th iteration is as follows:

[0085]

[0086]

[0087] Among them, g i For the i-th soil data, This is the second best matching unit for the i-th soil data.

[0088] A third aspect of the present invention provides an electronic device for source analysis of heavy metals in soil, comprising at least one control processor and a memory for communicative connection to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the above-described source analysis method for heavy metals in soil.

[0089] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described method for source analysis of heavy metals in soil.

[0090] It should be noted that the beneficial effects of the second to fourth aspects of the present invention compared with the prior art are the same as the beneficial effects of the above-described soil heavy metal source analysis system compared with the prior art, and will not be described in detail here.

[0091] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0092] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0093] Figure 1 This is a flowchart of a method for source analysis of heavy metals in soil according to an embodiment of the present invention;

[0094] Figure 2 This is a schematic diagram of the overall process of a method for source analysis of heavy metals in soil according to an embodiment of the present invention.

[0095] Figure 3 This is a planar visualization diagram of a single-element component according to an embodiment of the present invention;

[0096] Figure 4 This is a spatial distribution map of heavy metals in soil data according to an embodiment of the present invention;

[0097] Figure 5 This is a flowchart of a soil heavy metal source analysis system according to an embodiment of the present invention. Detailed Implementation

[0098] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0099] In the description of this invention, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of technical features indicated, or implicitly indicating the order of the technical features indicated.

[0100] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the drawings and are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0101] In the description of this invention, it should be noted that, unless otherwise explicitly defined, terms such as "setting," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0102] Due to differences in geographical location and environmental factors, there are certain correlations among heavy metal elements in soil. These correlations result in graph-structured relationships between samples. Furthermore, current methods for analyzing the sources of heavy metals suffer from the drawback of being highly subjective.

[0103] To address the aforementioned technical deficiencies, referring to... Figure 1 and Figure 2 The present invention also provides a method for source analysis of heavy metals in soil, comprising:

[0104] Step S101: Obtain soil data;

[0105] Step S102: Extract graph structure features from soil data using a graph convolutional neural network model;

[0106] Step S103: Input the graph structure features into the self-organizing competitive learning neural network model to obtain the best matching unit between the neuron weight values ​​and the soil data;

[0107] Step S104: Based on the best matching unit and neuron weight values, and using the two-dimensional visualization of the self-organizing competitive learning neural network model, calculate the planar visualization of single-element components of heavy metals in soil data.

[0108] Step S105: Based on the single-element component planar visualization, determine the source of heavy metals in the soil data.

[0109] This method acquires soil data, extracts the graph structure features of the soil data using a graph convolutional neural network model, inputs the graph structure features into a self-organizing competitive learning neural network model, obtains the best matching unit between the neuron weight values ​​and the soil data, and uses the two-dimensional visualization of the self-organizing competitive learning neural network model to calculate the planar visualization map of single-element components of heavy metals in the soil data. Based on the planar visualization map of single-element components, the source of heavy metals in the soil data is determined. By extracting the graph structure features of the soil data, the self-organizing competitive method can fully learn the correlation information between soil data, improving the efficiency and accuracy of analyzing the source of heavy metals in soil data, and overcoming the drawback of strong subjectivity in the analysis of heavy metal source.

[0110] In some embodiments, step S102 may include, but is not limited to, steps S201 to S203:

[0111] Step S201: Calculate the initial graph structure features of the soil data. And pre-set the initial weight values ​​w for each layer of the graph convolutional neural network model. i (0) Where N is the number of soil data points, and l is the number of layers in the graph convolutional neural network model. The input graph is the k-th metal element of the i-th soil data in the convolutional neural network model.

[0112] Step S202: Construct a similarity matrix of the initial graph structure features based on the fully connected Gaussian kernel distance, and construct an adjacency matrix and a degree matrix based on the similarity matrix;

[0113] Step S203: Calculate the graph structure features of the l-th layer of the soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The calculation formula for the graph structure features of the l-th layer of the soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features is as follows:

[0114]

[0115]

[0116]

[0117]

[0118] in, Let A be the first matrix, A be the adjacency matrix, and I be the identity matrix. Let D be the second matrix, D be the degree matrix, and δ be the nonlinear activation function. For the graph structure features of the i-th soil data in the l-th layer of the graph convolutional neural network model, w i (l-1) Let represent the element weights of the i-th soil data in the (l-1)-th layer of the graph convolutional neural network model.

[0119] In some embodiments, step S103 may include, but is not limited to, steps S301 to S307:

[0120] Step S301: Initialize the neuron weight value of each neuron to obtain the first neuron weight value of each neuron;

[0121] Step S302: Standardize the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer to obtain standardized neuron weight values ​​and standardized graph structure features. The calculation formula for standardizing the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer is as follows:

[0122]

[0123]

[0124] in, W represents the standardized graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model. j Let be the weight value of the j-th neuron. Let be the standardized neuron weight value of the j-th neuron, and M be the number of neurons;

[0125] Step S303: Calculate the distance from soil data to each neuron and the optimal matching unit of soil data based on the standardized neuron weight values ​​and standardized graph structure features. The formula for calculating the distance from soil data to each neuron and the optimal matching unit of soil data based on the standardized neuron weight values ​​and standardized graph structure features is as follows:

[0126]

[0127]

[0128] Where g is the best matching unit for the i-th soil data, d g Let d be the minimum distance from the i-th soil data point to all neurons. j Let be the distance from the i-th soil data point to the j-th neuron;

[0129] Step S304: Calculate the neuron neighborhood radius based on the preset initial neuron neighborhood radius. The formula for calculating the neuron neighborhood radius based on the preset initial neuron neighborhood radius is as follows:

[0130]

[0131] Where, N g (t) represents the neighborhood radius of the neuron in the t-th iteration, T is the preset maximum number of iterations, and N g (0) represents the initial radius of the neuron's neighborhood;

[0132] Step S305: Update the neuron weight value of each neuron according to the neighborhood radius of the neuron and the distance from the soil data to each neuron, and obtain the updated neuron weight value of the i-th soil data in the t-th iteration;

[0133] Step S306: If i equals N, then obtain the neuron weight value for the t-th iteration. If i is less than N, then calculate the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data. Calculate the distance from the (i+1)-th soil data to each neuron and the optimal matching unit of the (i+1)-th soil data based on the graph structure features of the (i+1)-th soil data in layer l. Update the neuron weight value of each neuron based on the neighborhood radius of the neuron and the distance from the (i+1)-th soil data to each neuron, obtaining the updated neuron weight value for the (i+1)-th soil data in the t-th iteration. Continue this process until i equals N, obtaining the neuron weight value for the t-th iteration. The formula for calculating the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data is:

[0134]

[0135] Where β(t) is the pre-set learning rate for the t-th iteration;

[0136] Step S307: Calculate the loss function value for the t-th iteration based on the best and second-best matching units. If t equals the maximum number of iterations, obtain the neuron weight values ​​for the soil data. If t is less than the maximum number of iterations, then t = t + 1. Update the graph structure features of the soil data in the graph convolutional neural network model based on the loss function value for the t-th iteration. Obtain the neuron weight values ​​for the (t+1)-th iteration based on the updated graph structure features. Continue this process until t equals the maximum number of iterations to obtain the neuron weight values ​​for the soil data. The formula for calculating the loss function value for the t-th iteration based on the best and second-best matching units is as follows:

[0137]

[0138]

[0139] Among them, g i For the i-th soil data, For the second best matching unit of the i-th soil data, refer to Figure 4 Each hexagon represents a neuron, and each neuron has 6 adjacent neurons in the output layer.

[0140] In some embodiments, step S305 may include, but is not limited to, steps S401 to S402:

[0141] Step S401: Compare the distance from the soil data to each neuron with the size of the neuron's neighborhood radius: Update the neuron weight value of the neurons whose distance from the soil data to the neuron is less than or equal to the neuron's neighborhood radius to the neuron weight value of the i-th soil data.

[0142] Step S402: The neuron weights of neurons whose distance from soil data to neurons is greater than the neuron's neighborhood radius remain unchanged.

[0143] In some embodiments, the source apportionment method for heavy metals in soil may include, but is not limited to, steps S501 to S503:

[0144] Step S501: Obtain the latitude and longitude of the soil data;

[0145] Step S502: Calculate the spatial distribution map of heavy metals in the soil data based on the latitude and longitude of soil elements, the best matching unit of soil elements, and the single-element component planar visualization map.

[0146] Step S503: Obtain the spatial distribution characteristics of heavy metals in soil data based on the spatial distribution map.

[0147] To facilitate understanding by those skilled in the art, the following set of experimental data is provided:

[0148] Reference Figure 3 , Figure 3 Each hexagon represents a neuron, and the intensity of the neuron's color indicates its weight. Elements with the same color distribution can be clustered into a single heavy metal source. The color distribution pattern reveals three main sources of heavy metals in the soil data: Source 1 consists of As, Cd, Cu, and Ni; Source 2 consists of Pb and Cr; and Source 3 consists of Hg and Zn.

[0149] Reference Figure 4 (Sample numbers: Jiangmen City Pengjiang District (JP), Jiangmen City Xinhui District (JX), Jiangmen City Jianghai District (JJ), Jiangmen City Taishan City (JT), Jiangmen City Heshan City (JH), Jiangmen City Kaiping City (JK), Jiangmen City Enping City (JE))

[0150] from Figure 3 It can be seen that the color distributions of As, Cd, Cu, and Ni are similar, originating from the same source. Since the first source portion of the component planar diagrams for As, Cd, Cu, and Ni has a darker color distribution, that is... Figure 4 The sample corresponding to the first source has a greater impact on the above four elements. Pb and Cr, however, have a greater impact. Figure 3 The data shows similar color distributions. The first and second source regions in the component planar plots of these two elements have darker color distributions, so Pb and Cr can be explained as being influenced by a mixture of source one and source two. Similarly, the source analysis of Zn and Hg is the same as that of As, Cd, Cu, and Ni.

[0151] As, Cd, Cu, and Ni originate from the same source. Figure 4 The sample distribution shows that the majority of the samples are from Taishan City, Jiangmen, indicating that the influence is primarily from Taishan. According to the survey, Taishan City is the most populous and has the largest land area among the three districts and four cities of Jiangmen. Taishan's economy is mainly based on agriculture, forestry, animal husbandry, and fisheries. Analysis ultimately identified Source 1 as a source of human activity, primarily agricultural activity.

[0152] Source 2 is dominated by Pb and Cr. It can be observed that samples (JX-1, JX-3, JP-8, JT-12, JK-11, JH-2) have a significant impact on the component planar graph of neuron weights, and Source 1 (Taishan) also has a significant impact on this element group. To our knowledge, the main contributors to Jiangmen City's GDP from transportation are Pengjiang District, Kaiping City, and Xinhui District, followed by Enping, Heshan, Jianghai, and Taishan. Pengjiang District, as the center of Jiangmen, has the most developed transportation industry. Based on the analysis, Source 2 was ultimately determined to be both anthropogenic and transportation-related.

[0153] Source 3 is primarily composed of Zn and Hg, and their component planar diagrams correspond. Figure 4 The third source component, with 50% of the sample points in Xinhui District falling within this source, followed by Heshan City and Pengjiang District. Based on analysis and summary, source 3 can be identified as an industrial source.

[0154] Additionally, refer to Figure 5 One embodiment of the present invention provides a source apportionment system for heavy metals in soil, including a data acquisition module 1100, a graph structure feature extraction module 1200, an optimal matching unit and neuron weight value calculation module 1300, a single-element component planar visualization graph acquisition module 1400, and a source correlation analysis module 1500, wherein:

[0155] The data acquisition module 1100 is used to acquire soil data;

[0156] The graph structure feature extraction module 1200 is used to extract graph structure features from soil data through a graph convolutional neural network model.

[0157] The optimal matching unit and neuron weight value calculation module 1300 is used to input graph structure features into the self-organizing competitive learning neural network model to obtain the optimal matching unit between neuron weight values ​​and soil data.

[0158] The single-element component planar visualization acquisition module 1400 is used to calculate the single-element component planar visualization of heavy metals in soil data based on the best matching unit and neuron weight values, and by utilizing the two-dimensional visualization of a self-organizing competitive learning neural network model.

[0159] The Source Correlation Analysis Module 1500 is used to determine the source of heavy metals in soil data based on a single-element component planar visualization.

[0160] This system acquires soil data and extracts its graph structure features using a graph convolutional neural network model. These features are then input into a self-organizing competitive learning neural network model to obtain the optimal matching unit between neuron weights and soil data. Based on this optimal matching unit and neuron weights, and utilizing the two-dimensional visualization of the self-organizing competitive learning neural network model, a planar visualization map of single-element components of heavy metals in the soil data is calculated. Based on this planar visualization map, the source of heavy metals in the soil data is determined. By extracting the graph structure features of the soil data, the self-organizing competitive method can fully learn the correlation information between soil data, improving the efficiency and accuracy of analyzing the source of heavy metals in soil data and overcoming the drawback of strong subjectivity in heavy metal source analysis.

[0161] In some embodiments, the graph structure feature extraction module further includes:

[0162] The data initialization module calculates the initial graph structure features of the soil data. And pre-set the initial weight values ​​w for each layer of the graph convolutional neural network model. i (0) Where N is the number of soil data points, and l is the number of layers in the graph convolutional neural network model. The input graph is the k-th metal element of the i-th soil data in the convolutional neural network model.

[0163] The matrix construction module is used to construct a similarity matrix of the initial graph structure features based on the fully connected Gaussian kernel distance, and to construct an adjacency matrix and a degree matrix based on the similarity matrix;

[0164] The graph structure feature calculation module for soil data is used to calculate the graph structure features of the l-th layer of soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The calculation formula for the graph structure features of the l-th layer of soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features is as follows:

[0165]

[0166]

[0167]

[0168]

[0169] in, Let A be the first matrix, A be the adjacency matrix, and I be the identity matrix. Let D be the second matrix, D be the degree matrix, and δ be the nonlinear activation function. For the graph structure features of the i-th soil data in the l-th layer of the graph convolutional neural network model, w i (l-1) Let represent the element weights of the i-th soil data in the (l-1)-th layer of the graph convolutional neural network model.

[0170] In some embodiments, the optimal matching unit and neuron weight value calculation module further includes:

[0171] The neuron weight initialization module is used to initialize the neuron weight value of each neuron, and obtain the first neuron weight value of each neuron;

[0172] The data standardization module is used to standardize the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer, obtaining standardized neuron weight values ​​and standardized graph structure features. The calculation formula for standardizing the weight values ​​of the first neuron and the graph structure features of the i-th soil data in the l-th layer is as follows:

[0173]

[0174]

[0175] in, W represents the standardized graph structure feature of the i-th soil data in the l-th layer of the graph convolutional neural network model. j Let be the weight value of the j-th neuron. Let be the standardized neuron weight value of the j-th neuron, and M be the number of neurons;

[0176] The neuron distance calculation module is used to calculate the distance from soil data to each neuron and the best matching unit of soil data based on the standardized neuron weight values ​​and standardized graph structure features. The calculation formula for the distance from soil data to each neuron and the best matching unit of soil data based on the standardized neuron weight values ​​and standardized graph structure features is as follows:

[0177]

[0178]

[0179] Where g is the best matching unit for the i-th soil data, d g Let d be the minimum distance from the i-th soil data point to all neurons. j Let be the distance from the i-th soil data point to the j-th neuron;

[0180] The neuron neighborhood radius calculation module is used to calculate the neuron neighborhood radius based on a preset initial neuron neighborhood radius. The calculation formula for the neuron neighborhood radius based on the preset initial neuron neighborhood radius is as follows:

[0181]

[0182] Where, N g (t) represents the neighborhood radius of the neuron in the t-th iteration, T is the preset maximum number of iterations, and N g (0) represents the initial radius of the neuron's neighborhood;

[0183] The neuron weight update module is used to update the neuron weight value of each neuron according to the neighborhood radius of the neuron and the distance from the soil data to each neuron, so as to obtain the updated neuron weight value of the i-th soil data in the t-th iteration.

[0184] The soil data neuron weight value iteration module is used to obtain the neuron weight value of the t-th iteration if i equals N, and to calculate the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data if i is less than N. It then calculates the distance from the (i+1)-th soil data to each neuron and the optimal matching unit based on the graph structure features of the (i+1)-th soil data in layer l. Finally, it updates the neuron weight value of each neuron based on the neighborhood radius and the distance from the (i+1)-th soil data to each neuron, obtaining the updated neuron weight value of the (i+1)-th soil data in the t-th iteration. This process continues until i equals N, yielding the neuron weight value for the t-th iteration. The formula for calculating the weight value of the (i+1)-th soil data based on the updated neuron weight value of the i-th soil data is as follows:

[0185]

[0186] Where β(t) is the pre-set learning rate for the t-th iteration;

[0187] The neuron weight value iteration module is used to calculate the loss function value of the t-th iteration based on the best and second-best matching units in the t-th iteration. If t equals the maximum number of iterations, the neuron weight values ​​for the soil data are obtained; if t is less than the maximum number of iterations, then t = t + 1. The graph structure features of the soil data in the graph convolutional neural network model are updated based on the loss function value of the t-th iteration. The neuron weight values ​​for the (t+1)-th iteration are obtained based on the updated graph structure features, and so on, until t equals the maximum number of iterations, at which point the neuron weight values ​​for the soil data are obtained. The formula for calculating the loss function value of the t-th iteration based on the best and second-best matching units in the t-th iteration is as follows:

[0188]

[0189]

[0190] Among them, g i For the i-th soil data, This is the second best matching unit for the i-th soil data.

[0191] It should be noted that this system embodiment is based on the same inventive concept as the above system embodiment. Therefore, the relevant content of the above method embodiment is also applicable to this system embodiment, and will not be repeated here.

[0192] This application also provides an electronic device for source analysis of heavy metals in soil, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the source analysis method for heavy metals in soil as described above.

[0193] The processor and memory can be connected via a bus or other means.

[0194] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0195] The non-transient software program and instructions required to implement the soil heavy metal source analysis method of the above embodiments are stored in memory. When executed by the processor, the soil heavy metal source analysis method of the above embodiments is executed, for example, the method described above is executed. Figure 1 The method steps S101 to S105.

[0196] This application also provides a computer-readable storage medium storing computer-executable instructions for performing: the soil heavy metal source analysis method described above.

[0197] The computer-readable storage medium stores computer-executable instructions that are executed by a processor or controller, for example, by a processor in the above-described electronic device embodiment, causing the processor to perform the soil heavy metal source analysis method described above, for example, performing the above-described... Figure 1 The method steps S101 to S105.

[0198] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program units, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program units, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0199] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for source apportionment of heavy metals in soil, characterized in that, The source apportionment methods for heavy metals in the soil include: Obtain soil data; The graph structure features of the soil data were extracted using a graph convolutional neural network model. The graph structure features are input into a self-organizing competitive learning neural network model to obtain the optimal matching unit between the neuron weights and the soil data, specifically: Step S1: Initialize the neuron weight value of each neuron to obtain the first neuron weight value of each neuron; Step S2: Combine the weight values ​​of the first neuron with the weight values ​​of the second neuron. The first layer The graph structure features of each soil data point are standardized to obtain standardized neuron weight values ​​and standardized graph structure features. Specifically, the first neuron weight value is then compared with the first... The first layer The formula for calculating data standardization based on the graph structure characteristics of soil data is as follows: in, For the graph convolutional neural network model, the first... The first layer Standardized graph structure characteristics of soil data For the first The weight values ​​of each neuron, For the first The standardized neuron weights of each neuron. The number of neurons; Step S3: Calculate the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features. The calculation formula for calculating the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features is as follows: in, For the first The best matching unit for each soil data For the first The minimum distance among all neurons from each soil data point. For the first Soil data to the first The distance between neurons; Step S4: Calculate the neuron neighborhood radius based on the preset initial neuron neighborhood radius, wherein the calculation formula for calculating the neuron neighborhood radius based on the preset initial neuron neighborhood radius is: in, For the first The radius of the neuron's neighborhood in the next iteration. The preset maximum number of iterations, The initial radius of the neuron's neighborhood; Step S5: Update the neuron weight value of each neuron according to the neighborhood radius of the neuron and the distance from the soil data to each neuron, to obtain the first... The iteration of the ... The neuron weight values ​​updated based on soil data; Step S6, if the above If it equals N, then we get the first... The neuron weight values ​​in the next iteration, if the If it is less than N, then according to the first... The neuron weights after updating the soil data were calculated to obtain the first... The weight value of the soil data, according to the first... The weight value of the soil data and the first soil data are related to the first The first layer The graph structure features of the soil data were calculated to obtain the first... The distance from each soil data point to each neuron and the distance from the first soil data point to each neuron are related to the distance from the first soil data point to each neuron. The best matching unit for each soil data, and based on the neighborhood radius of the neuron and the first... The distance from each soil data point to each neuron is used to update the neuron weight value of each neuron, resulting in the first... The iteration of the ... The neuron weights are updated based on the soil data, and so on, until... Equal to N, obtain the first The neuron weight values ​​in the nth iteration, wherein the weights are based on the nth iteration. The neuron weights after updating the soil data were calculated to obtain the first... The formula for calculating the weight value of each soil data point is: in, The learning rate for the t-th iteration is set in advance; Step S7, according to the first The best matching unit and the second best matching unit in the second iteration are calculated to obtain the... The loss function value of the next iteration, if the If the number of iterations equals the maximum number of iterations, then the neuron weight values ​​of the soil data are obtained. If it is less than the maximum number of iterations, then According to the first The loss function value of the iteration updates the graph structure features of the soil data in the graph convolutional neural network model, and the updated graph structure features are used to obtain the first iteration. The neuron weights for each iteration are then calculated, and so on, until... The maximum number of iterations is equal to the neuron weight values ​​of the soil data, where the weight values ​​are obtained based on the first iteration. The optimal matching unit and the second-best matching unit of the iteration are calculated to obtain the... The formula for calculating the loss function value of the next iteration is: in, For the first The best matching unit for each soil data For the first The second best matching unit for soil data; Based on the optimal matching unit and the neuron weight values, and using the two-dimensional visualization of the self-organizing competitive learning neural network model, a planar visualization of the single-element components of heavy metals in the soil data is calculated. Based on the single-element component planar visualization, the source of heavy metals in the soil data is determined.

2. The method for source apportionment of heavy metals in soil according to claim 1, characterized in that, The extraction of graph structure features from the soil data using a graph convolutional neural network model includes: The initial graph structure features of the soil data are calculated as follows: And pre-set the initial weight values ​​for each layer of the graph convolutional neural network model. ,in, The number of soil data points. The number of layers in the graph convolutional neural network model. For the input of the graph convolutional neural network model, the first... The first soil data One metallic element; The similarity matrix of the initial graph structure features is constructed based on the fully connected Gaussian kernel distance, and the adjacency matrix and degree matrix are constructed based on the similarity matrix; The soil data is calculated based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The graph structure features of the layer, wherein the soil data is calculated based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features to obtain the first layer of the graph structure features. The formula for calculating the graph structure features of a layer is: in, For the first matrix, It is an adjacency matrix. It is the identity matrix. For the second matrix, For degree matrix, It is a non-linear activation function. For the graph convolutional neural network model, the first... The first layer The graphical structural characteristics of soil data For the graph convolutional neural network model, the first... The first layer The element weight values ​​of each soil data point.

3. The method for source analysis of heavy metals in soil according to claim 2, characterized in that, The neuron weight value of each neuron is updated based on the neighborhood radius of the neuron and the distance from the soil data to each neuron, to obtain the first... The iteration of the ... The updated neuron weights based on the soil data include: Compare the distance from the soil data to each neuron with the size of the neuron's neighborhood radius: update the neuron weight value of neurons whose distance from the soil data to the neuron is less than or equal to the neuron's neighborhood radius to the nth neuron. Neuron weight values ​​for each soil data point; The neuron weights of neurons whose distance from the soil data to the neuron is greater than the neuron's neighborhood radius remain unchanged.

4. The method for source analysis of heavy metals in soil according to claim 3, characterized in that, The source apportionment method for heavy metals in soil also includes: Obtain the latitude and longitude of the soil data; The spatial distribution map of heavy metals in the soil data is calculated based on the latitude and longitude of the soil elements, the best matching unit of the soil elements, and the single-element component planar visualization map. The spatial distribution characteristics of heavy metals in the soil data are obtained based on the spatial distribution map.

5. A source apportionment system for heavy metals in soil, characterized in that, The soil heavy metal source apportionment system includes: The data acquisition module is used to acquire soil data; The graph structure feature extraction module is used to extract the graph structure features of the soil data through a graph convolutional neural network model. The optimal matching unit and neuron weight calculation module is used to input the graph structure features into the self-organizing competitive learning neural network model to obtain the optimal matching unit between the neuron weight values ​​and the soil data. The optimal matching unit and neuron weight calculation module further includes: The neuron weight initialization module is used to initialize the neuron weight value of each neuron, and obtain the first neuron weight value of each neuron; The data standardization module is used to compare the weight values ​​of the first neuron with those of the second neuron. The first layer The graph structure features of each soil data point are standardized to obtain standardized neuron weight values ​​and standardized graph structure features. Specifically, the first neuron weight value is then compared with the first... The first layer The formula for calculating data standardization based on the graph structure characteristics of soil data is as follows: in, For the graph convolutional neural network model, the first... The first layer Standardized graph structure characteristics of soil data For the first The weight values ​​of each neuron, For the first The standardized neuron weights of each neuron. The number of neurons; The neuron distance calculation module is used to calculate the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features. The calculation formula for calculating the distance from the soil data to each neuron and the optimal matching unit of the soil data based on the standardized neuron weight values ​​and the standardized graph structure features is as follows: in, For the first The best matching unit for each soil data For the first The minimum distance among all neurons from each soil data point. For the first Soil data to the first The distance between neurons; The neuron neighborhood radius calculation module is used to calculate the neuron neighborhood radius based on a preset initial neuron neighborhood radius. The calculation formula for calculating the neuron neighborhood radius based on the preset initial neuron neighborhood radius is as follows: in, For the first The radius of the neuron's neighborhood in the next iteration. The preset maximum number of iterations, The initial radius of the neuron's neighborhood; The neuron weight update module is used to update the neuron weight value of each neuron based on the neighborhood radius of the neuron and the distance from the soil data to each neuron, to obtain the weight value of the first neuron. The iteration of the ... The neuron weight values ​​updated based on soil data; The neural network weight value iteration module for soil data is used to... If it equals N, then we get the first... The neuron weight values ​​in the next iteration, if the If it is less than N, then according to the first... The neuron weights after updating the soil data were calculated to obtain the first... The weight value of the soil data, according to the first... The weight value of the soil data and the first soil data are related to the first The first layer The graph structure features of the soil data were calculated to obtain the first... The distance from each soil data point to each neuron and the distance from the first soil data point to each neuron are related to the distance from the first soil data point to each neuron. The best matching unit for each soil data, and based on the neighborhood radius of the neuron and the first... The distance from each soil data point to each neuron is used to update the neuron weight value of each neuron, resulting in the first... The iteration of the ... The neuron weights are updated based on the soil data, and so on, until... Equal to N, obtain the first The neuron weight values ​​in the nth iteration, wherein the weights are based on the nth iteration. The neuron weights after updating the soil data were calculated to obtain the first... The formula for calculating the weight value of each soil data point is: in, The learning rate for the t-th iteration is set in advance; The neuron weight value iteration module is used to iterate based on the first... The best matching unit and the second best matching unit in the second iteration are calculated to obtain the... The loss function value of the next iteration, if the If the number of iterations equals the maximum number of iterations, then the neuron weight values ​​of the soil data are obtained. If it is less than the maximum number of iterations, then According to the first The loss function value of the iteration updates the graph structure features of the soil data in the graph convolutional neural network model, and the updated graph structure features are used to obtain the first iteration. The neuron weights for each iteration are then calculated, and so on, until... The maximum number of iterations is equal to the neuron weight values ​​of the soil data, where the weight values ​​are obtained based on the first iteration. The optimal matching unit and the second-best matching unit of the iteration are calculated to obtain the... The formula for calculating the loss function value of the next iteration is: in, For the first The best matching unit for each soil data For the first The second best matching unit for soil data; The single-element component planar visualization acquisition module is used to calculate the single-element component planar visualization of heavy metals in the soil data based on the best matching unit and the neuron weight value, and using the two-dimensional visualization of the self-organizing competitive learning neural network model. The source correlation analysis module is used to determine the source of heavy metals in the soil data based on the single-element component planar visualization.

6. The soil heavy metal source apportionment system according to claim 5, characterized in that, The graph structure feature extraction module also includes: The data initialization module calculates the initial graph structure features of the soil data as follows: And pre-set the initial weight values ​​for each layer of the graph convolutional neural network model. ,in, The number of soil data points. The number of layers in the graph convolutional neural network model. For the input of the graph convolutional neural network model, the first... The first soil data One metallic element; The matrix construction module is used to construct a similarity matrix of the initial graph structure features based on the fully connected Gaussian kernel distance, and to construct an adjacency matrix and a degree matrix based on the similarity matrix; The graph structure feature calculation module for soil data is used to calculate the first degree feature of the soil data based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features. The graph structure features of the layer, wherein the soil data is calculated based on the adjacency matrix, degree matrix, initial weight values, and initial graph structure features to obtain the first layer of the graph structure features. The formula for calculating the graph structure features of a layer is: in, For the first matrix, It is an adjacency matrix. It is the identity matrix. For the second matrix, For degree matrix, It is a non-linear activation function. For the graph convolutional neural network model, the first... The first layer The graphical structural characteristics of soil data For the graph convolutional neural network model, the first... The first layer The element weight values ​​of each soil data point.

7. A source analysis device for heavy metals in soil, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the source apportionment method for heavy metals in soil as described in any one of claims 1 to 4.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the source analysis method for heavy metals in soil as described in any one of claims 1 to 4.

Images