A natural resource management system and method based on graph neural network optimization

Abstract

The present invention relates to the field of natural resource data processing technology, and in particular to a natural resource management system and method based on graph neural network optimization. The method comprises the following steps: dividing a target plot into soil layers and lateral areas to obtain a land micro-unit node array; inputting the target plot's soil characteristic data, land micro-unit slope mapping data, and rainfall data into a land spatial feature fitting model as node feature vectors of the land micro-unit nodes for spatial feature aggregation to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit nodes; and obtaining the deep soil plastic pollutant concentration through a migration and diffusion model based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of plastic pollutants at each land micro-unit node, and the soil layer depth of each land micro-unit node. The land micro-unit node locations with plastic pollutant content greater than a preset threshold are screened as key treatment areas, and plastic pollutant cleaning decisions are executed for the key treatment areas.

Description

Technical Field

[0001] The present invention relates to the technical field of natural resource data processing, and in particular to a natural resource management system and method based on graph neural network optimization. Background Art

[0002] In traditional land resource management, plastic pollutant cleanup methods mainly rely on manual experience and detection equipment for surface detection. However, these methods can usually only be applied to the cleanup of plastic pollutants in shallow soil. For the distribution of plastic pollutants in deeper soil layers, traditional detection methods are not only time-consuming and labor-intensive, but also difficult to achieve efficient and accurate cleanup. In addition, the distribution of plastic pollutants in the soil is also affected by environmental factors such as rainfall and topography, which causes the sedimentation and migration of plastic pollutants. The plastic pollutant content in the surface soil cannot accurately reflect the plastic pollutant situation in the deep soil of the same area. This limitation makes it difficult to implement land plastic pollutant control based on existing technologies in deep soil areas. Therefore, how to accurately predict the distribution of plastic pollutants in deep soil and provide more accurate decision-making support for land resource and environmental management has become an urgent problem to be solved in the current field of natural resource management. Summary of the Invention

[0003] The present invention provides a natural resource management system and method based on graph neural network optimization to solve the problem that land plastic pollutants in deep soil areas are difficult to effectively control.

[0004] To achieve the above objectives, on the one hand, the present invention provides a natural resource management system based on graph neural network optimization, the system comprising: a land micro-unit data acquisition module, a land spatial feature fitting module, and a plastic pollutant migration assessment module, and the modules are connected in sequence.

[0005] The land micro-unit data acquisition module is used to divide the target plot into soil layers from shallow soil to deep soil at preset intervals to obtain vertical soil layers of the target plot, divide the vertical soil layers into horizontal areas at preset intervals to obtain a land micro-unit node array, obtain soil through a soil sampler arranged at a preset sampling point of the target plot and perform basic soil property detection to obtain soil property data of the target plot including soil density, soil porosity, and soil particle size, measure the slope of different positions of the target plot through a slope meter to obtain land micro-unit slope mapping data, and retrieve the rainfall data of the meteorological station of the target plot; and use the target plot soil property data, land micro-unit slope mapping data, and rainfall data as node feature vectors of the land micro-unit nodes.

[0006] The land spatial feature fitting module is used to input the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node. The land spatial feature fitting model is a mapping relationship between the land micro-unit node characteristics and the vertical migration coefficient of plastic pollutants at the land micro-unit node established based on soil property data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data obtained from historical sampling data as training data and model training through a graph neural network.

[0007] The plastic pollutant migration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of the plastic pollutants of each land micro-unit node and the soil layer depth of each land micro-unit node through a migration and diffusion model, screen the land micro-unit node locations where the soil plastic pollutant concentration is greater than a preset threshold as key treatment areas, and execute plastic pollutant cleaning decisions for the key treatment areas.

[0008] Furthermore, the land spatial feature fitting module includes: a node distance calculation module, a spatial graph network generation module, a connection determination module, a degree matrix generation module, a feature aggregation matrix generation module, and a graph convolution module, and the modules are connected in sequence.

[0009] The node distance calculation module is used to calculate the spatial distance between nodes according to the spatial coordinates of each node of the land micro-unit node array in the target plot spatial graph network through Euclidean distance to obtain the land micro-unit node spatial distance.

[0010] The spatial graph network generation module is used to mark the corresponding two land micro-unit nodes as adjacent nodes and establish node connection edges to obtain the target land parcel spatial graph network when the spatial distance between the land micro-unit nodes is less than a preset association threshold.

[0011] The connection determination module is used to create a node of size 1 according to the number of land micro-unit nodes in the target land space graph network. The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is:

[0012] .

[0013] in is the adjacency matrix No. Rank Elements of a column.

[0014] The degree matrix generation module is used to calculate the land micro-unit degree matrix based on the element nodes in the adjacency matrix. for:

[0015] .

[0016] in is the number of land micro-unit nodes, It is the first Rank Elements of a column.

[0017] The feature aggregation matrix generation module is used to obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and the degree matrix. for:

[0018] .

[0019] in is the land micro-unit feature aggregation matrix.

[0020] The graph convolution module is used to perform graph convolution operations on node feature vectors based on the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit.

[0021] Furthermore, the graph convolution module includes a connected layer aggregation module and an output module.

[0022] The layer aggregation module is used to input the feature vector of the input node of the input layer. Each intermediate layer aggregates the features of the adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output as follows:

[0023] .

[0024] in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer.

[0025] The output module is used to output the vertical migration coefficient of plastic pollutants in land micro-units in the output layer.

[0026] Furthermore, the training process of the land spatial feature fitting module includes:

[0027] According to the historical sampling data of the concentration of plastic pollutants in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is:

[0028] .

[0029] in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants.

[0030] A four-layer graph neural network is set, in which the node feature dimension of the input layer is set to 5, the feature dimension of the middle layer is set to 64, and the feature dimension of the output layer is set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data are used as the input of the training set, and the vertical migration coefficient of plastic pollutants is used as the output of the training set to train the land spatial feature fitting model.

[0031] The mean square error loss function is used as the loss function, and the loss function is:

[0032] .

[0033] in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value.

[0034] The loss function is optimized according to the Adam optimization algorithm, and the weight matrix of each layer is iteratively updated until the loss value stops decreasing after a preset number of consecutive iterations.

[0035] Furthermore, the plastic pollutant migration assessment module includes a surface data acquisition module and a plastic pollutant concentration assessment module that are connected to each other.

[0036] The surface data acquisition module is used to sample and detect the soil of the first layer of land micro-unit nodes in the land micro-unit node array to obtain the soil plastic pollutant concentration of the corresponding land micro-unit nodes, and to obtain the soil layer depth of the land micro-unit nodes according to the preset intervals of the soil layer division from shallow soil to deep soil.

[0037] The plastic pollutant concentration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit node of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant of the land micro-unit node, and the soil layer depth of the land micro-unit node through the migration and diffusion model:

[0038] .

[0039] in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is Yes, etc. Soil layer depth, Yes, etc. Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

[0040] Based on the same inventive concept, the present invention provides a natural resource management method based on graph neural network optimization, the method comprising:

[0041] The target plot is divided into soil layers from shallow soil to deep soil at preset intervals to obtain vertical soil layers of the target plot, and the vertical soil layers are divided into horizontal regions at preset intervals to obtain a land micro-unit node array. Soil is obtained by a soil sampler arranged at a preset sampling point of the target plot, and basic soil properties are tested to obtain soil property data of the target plot, including soil density, soil porosity, and soil particle size. The slope of different positions of the target plot is measured by a slope meter to obtain land micro-unit slope mapping data, and the rainfall data of the meteorological station of the target plot is retrieved; the soil property data of the target plot, the land micro-unit slope mapping data, and the rainfall data are used as the node feature vectors of the land micro-unit nodes.

[0042] The node feature vector of the land micro-unit node is input into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node. The land spatial feature fitting model is a mapping relationship between the land micro-unit node characteristics and the vertical migration coefficient of plastic pollutants at the land micro-unit node established based on soil property data obtained from historical sampling data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data as training data, and model training is performed through a graph neural network.

[0043] Based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of plastic pollutants at each land micro-unit node and the soil layer depth of each land micro-unit node, the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot is obtained by recursive calculation through the migration and diffusion model. The land micro-unit node locations with plastic pollutant content greater than the preset threshold are screened as key treatment areas, and plastic pollutant cleaning decisions are implemented in the key treatment areas.

[0044] Furthermore, the method of inputting the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node includes:

[0045] According to the spatial coordinates of each node of the land micro-unit node array in the target plot spatial graph network, the spatial distance between nodes is calculated by Euclidean distance to obtain the spatial distance of the land micro-unit nodes.

[0046] When the spatial distance between land micro-unit nodes is less than a preset association threshold, the corresponding two land micro-unit nodes are marked as adjacent nodes and node connection edges are established to obtain the target land parcel spatial graph network.

[0047] According to the number of land micro-unit nodes in the target plot spatial graph network, a size of The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is:

[0048] .

[0049] in is the adjacency matrix No. Rank Elements of a column.

[0050] The land micro-unit degree matrix is ​​calculated based on the element nodes in the adjacency matrix for:

[0051] .

[0052] in is the number of land micro-unit nodes, It is the first Rank Elements of a column.

[0053] Obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and degree matrix for:

[0054] .

[0055] in is the land micro-unit feature aggregation matrix.

[0056] The vertical migration coefficient of plastic pollutants in land microunits is obtained by performing graph convolution operation on the node feature vector according to the land microunit feature aggregation matrix.

[0057] Furthermore, the method of performing a graph convolution operation on the node feature vector according to the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit includes:

[0058] The input layer inputs the node feature vector. Each layer in the middle layer aggregates the features of the adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output:

[0059] .

[0060] in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer.

[0061] The output layer outputs the vertical migration coefficient of plastic pollutants in land micro-units.

[0062] Furthermore, the method of using soil property data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data obtained from historical sampling data as training data and performing model training through a graph neural network includes:

[0063] According to the historical sampling data of the concentration of plastic pollutants in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is:

[0064] .

[0065] in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants.

[0066] A four-layer graph neural network is set, in which the node feature dimension of the input layer is set to 5, the feature dimension of the middle layer is set to 64, and the feature dimension of the output layer is set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data are used as the input of the training set, and the vertical migration coefficient of plastic pollutants is used as the output of the training set to train the land spatial feature fitting model.

[0067] The mean square error loss function is used as the loss function, and the loss function is:

[0068] .

[0069] in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value.

[0070] The loss function is optimized according to the Adam optimization algorithm, and the weight matrix of each layer is iteratively updated until the loss value stops decreasing after a preset number of consecutive iterations.

[0071] Furthermore, the method for obtaining the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot by recursively calculating the soil plastic pollutant concentration of the first layer land micro-unit nodes of the target plot, the vertical migration coefficient of the plastic pollutants of each land micro-unit node, and the soil layer depth of each land micro-unit node through a migration and diffusion model includes:

[0072] The soil of the first layer of land micro-unit nodes in the land micro-unit node array is sampled and tested to obtain the soil plastic pollutant concentration of the corresponding land micro-unit node, and the soil layer depth of the land micro-unit node is obtained according to the preset interval of soil layer division from shallow soil to deep soil.

[0073] According to the soil plastic pollutant concentration of the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant at the land micro-unit node, and the soil layer depth of the land micro-unit node, the soil plastic pollutant concentration at the deep land micro-unit node of the target plot is recursively calculated through the migration and diffusion model:

[0074] .

[0075] in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is Yes, etc. Soil layer depth, Yes, etc. Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

[0076] Compared with the existing technology, the beneficial effects of the present invention are: by constructing a land spatial feature fitting model based on a graph neural network, the spatial dependency relationship between soil nodes can be effectively captured, and the migration state of plastic pollutants between different soil layers is further combined with the migration diffusion model to predict deep soil plastic pollutants, thereby realizing the overall assessment of key governance areas of deep soil and providing effective decision-making support for the precise management of land resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0078] Figure 1 This is a schematic diagram of the module composition of a natural resource management system based on graph neural network optimization according to Example 1 of the present invention;

[0079] Figure 2 This is a flowchart of a natural resource management method based on graph neural network optimization according to Example 2 of the present invention. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0081] Example 1: Figure 1As shown, this embodiment provides a natural resource management system based on graph neural network optimization, which includes: a land micro-unit data acquisition module, a land spatial feature fitting module, and a plastic pollutant migration assessment module, and the modules are connected in sequence.

[0082] The land micro-unit data acquisition module is used to divide the target plot from shallow soil to deep soil according to preset intervals to obtain vertical soil layers of the target plot, divide the vertical soil layers into horizontal areas according to preset intervals to obtain land micro-unit node arrays, obtain soil through soil samplers arranged at preset sampling points of the target plot and perform basic soil property detection to obtain soil property data of the target plot including soil density, soil porosity, and soil particle size, measure the slope of different positions of the target plot through a slope meter to obtain land micro-unit slope mapping data, and retrieve rainfall data of the meteorological station of the target plot; the soil property data of the target plot, Land micro-unit slope mapping data and rainfall data serve as node feature vectors for land micro-unit nodes. The soil property data and slope are important factors affecting the migration of water and pollutants, determining soil permeability and the rate of plastic particle migration. Rainfall, soil density, soil porosity, soil particle size, and slope interact to influence the migration of plastic particles in the soil. The target plot's soil property data is sampled and measured layer by layer. These soil property data are measured using established methods, such as the ring knife method for soil density, the immersion method for soil porosity, and the sieving-straw method for soil particle size. Land micro-unit slope mapping data is measured using the center point of each surface grid as the measurement location. The slopes of each vertical layer node are annotated according to the land micro-unit slope mapping data for the same columnar layer node. The rainfall data is summarized and statistically analyzed using a rolling time window method for the previous 90 days' rainfall. For example, the soil layer division is as follows: the target plot is a land with an area of ​​100m×100m and a depth of 10m. The target plot is divided into soil layers from shallow soil to deep soil according to 0.5m to obtain the vertical soil layers of the target plot, with a total of 20 layers. The plot is divided horizontally into 10m×10m grids, with a total of 100 nodes in each layer. The land micro-unit node array consists of 20 soil layers, 100 nodes in each layer, and a total of 2000 nodes.

[0083] The land spatial feature fitting module is used to input the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node. The land spatial feature fitting model is a mapping relationship between the land micro-unit node characteristics and the vertical migration coefficient of plastic pollutants at the land micro-unit node established based on soil property data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data obtained from historical sampling data as training data and model training through a graph neural network.

[0084] The plastic pollutant migration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of the plastic pollutants of each land micro-unit node and the soil layer depth of each land micro-unit node through a migration and diffusion model, screen the land micro-unit node locations where the soil plastic pollutant concentration is greater than a preset threshold as key treatment areas, and execute plastic pollutant cleaning decisions for the key treatment areas.

[0085] It should be noted that the land spatial feature fitting module includes: a node distance calculation module, a spatial graph network generation module, a connection judgment module, a degree matrix generation module, a feature aggregation matrix generation module, and a graph convolution module, and the modules are connected in sequence.

[0086] The node distance calculation module is used to calculate the spatial distance between nodes according to the spatial coordinates of each node of the land micro-unit node array in the target plot spatial graph network through Euclidean distance to obtain the land micro-unit node spatial distance.

[0087] For example: The spatial coordinates of the first node are , the spatial coordinates of the second node are , then the spatial distance between nodes is calculated by Euclidean distance to obtain the spatial distance between land micro-unit nodes: .

[0088] The spatial graph network generation module is used to mark the corresponding two land micro-unit nodes as adjacent nodes and establish node connection edges to obtain the target land parcel spatial graph network when the spatial distance between the land micro-unit nodes is less than a preset association threshold.

[0089] For example, when the association threshold is set to 15, since the land micro-unit node spatial distance between the first node and the second node is 14.14, which is less than 15, the first node and the second node are marked as adjacent nodes and a node connection edge is established.

[0090] The connection determination module is used to create a node of size 1 according to the number of land micro-unit nodes in the target land space graph network. The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is:

[0091] .

[0092] in is the adjacency matrix No. Rank Elements of a column.

[0093] For example, if the target plot has 4 nodes (n=4), the size of the creation is The land micro-unit zero matrix. Assuming that there are connecting edges between node 1 and node 2, and node 2 and node 3, the adjacency matrix is: .

[0094] The degree matrix generation module is used to calculate the land micro-unit degree matrix based on the element nodes in the adjacency matrix. for:

[0095] .

[0096] in is the number of land micro-unit nodes, It is the first Rank Elements of a column.

[0097] For example: , , similarly we can get , , then the degree matrix .

[0098] The feature aggregation matrix generation module is used to obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and the degree matrix. for:

[0099] .

[0100] in is the land micro-unit feature aggregation matrix.

[0101] For example: When , When the land micro-unit characteristic aggregation matrix .

[0102] The graph convolution module is used to perform graph convolution operations on node feature vectors based on the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit.

[0103] It should be noted that the graph convolution module includes a connected layer aggregation module and an output module.

[0104] The layer aggregation module is used to input the feature vector of the input node of the input layer. Each intermediate layer aggregates the features of the adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output as follows:

[0105] .

[0106] in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer.

[0107] The output module is used to output the vertical migration coefficient of plastic pollutants in land micro-units in the output layer.

[0108] Aggregating the node features in the graph through the land micro-unit feature aggregation matrix can effectively collect information from the adjacent nodes of the land micro-unit nodes and generate new feature representations, ensuring that the features of the land micro-unit nodes not only rely on their own attributes, but also incorporate features from their adjacent nodes. Especially when there are unevenly distributed soil features, the normalization effect of the feature aggregation matrix can reduce the impact of node degree differences on the prediction results, promote the balance of information fusion between different nodes, and enable the graph neural network to capture the complex spatial dependencies between nodes in the graph structure, and enable the plastic pollutant migration coefficient of each soil micro-unit node to reflect the environmental conditions of the surrounding soil layers. For example, in the calculation process of soil plastic pollutant migration, the characteristics of the upper soil layer will affect the deeper soil layer. The feature aggregation matrix enables the graph neural network to capture this depth influence and improve the accuracy of subsequent assessment of soil plastic pollutant content in different soil layers.

[0109] It should be noted that the training process of the land spatial feature fitting module includes:

[0110] According to the historical sampling data of the concentration of plastic pollutants in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is:

[0111] .

[0112] in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants.

[0113] For example: when the concentration of plastic pollutants in the upper soil layer in a certain vertical direction in the land micro-unit node array is g / , the concentration of plastic pollutants in the lower soil g / , the depth difference between the upper and lower adjacent layers of soil m, then the vertical migration coefficient of plastic pollutants corresponding to the land micro-unit node is , when the concentration of plastic pollutants in the upper soil layer in a certain vertical direction in the land micro-unit node array is g / , the concentration of plastic pollutants in the lower soil g / , the depth of the upper and lower adjacent layers of soil m, then the vertical migration coefficient of plastic pollutants corresponding to the land micro-unit node is The larger the vertical migration coefficient of plastic pollutants, the more difficult it is for plastic pollutants in the soil to diffuse and migrate, resulting in a larger gradient of soil plastic pollutant concentration differences. The smaller the vertical migration coefficient of plastic pollutants, the easier it is for plastic pollutants in the soil to diffuse and migrate, resulting in a smaller gradient of soil plastic pollutant concentration differences.

[0114] A four-layer graph neural network is set, in which the node feature dimension of the input layer is set to 5, the feature dimension of the middle layer is set to 64, and the feature dimension of the output layer is set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data are used as the input of the training set, and the vertical migration coefficient of plastic pollutants is used as the output of the training set to train the land spatial feature fitting model.

[0115] The mean square error loss function is used as the loss function, and the loss function is:

[0116] .

[0117] in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value.

[0118] The loss function is optimized according to the Adam optimization algorithm, and the weight matrix of each layer is iteratively updated until the loss value stops decreasing after a preset number of consecutive iterations.

[0119] It should be noted that the plastic pollutant migration assessment module includes a surface data acquisition module and a plastic pollutant concentration assessment module that are connected to each other.

[0120] The surface data acquisition module is used to sample and detect the soil of the first layer of land micro-unit nodes in the land micro-unit node array to obtain the soil plastic pollutant concentration of the corresponding land micro-unit nodes, and to obtain the soil layer depth of the land micro-unit nodes according to the preset intervals of the soil layer division from shallow soil to deep soil.

[0121] The plastic pollutant concentration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit node of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant of the land micro-unit node, and the soil layer depth of the land micro-unit node through the migration and diffusion model:

[0122] .

[0123] in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is Yes, etc. Soil layer depth, Yes, etc. Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

[0124] For example: When The horizontal Vertical Soil plastic pollutant concentrations at land micro-unit nodes g / , No. The horizontal Vertical The vertical migration coefficient of plastic pollutants in land micro-unit nodes Each soil layer is divided into layers at intervals of 0.5m, i.e. Soil layer depth , No. Soil layer depth , No. Soil layer depth , then The horizontal Vertical Soil plastic pollutant concentrations at land micro-unit nodes g / , when the The horizontal Vertical The vertical migration coefficient of plastic pollutants in land micro-unit nodes When The horizontal Vertical Soil plastic pollutant concentrations at land micro-unit nodes 0.93g / When the concentration of soil plastic pollutants at the first-layer land micro-unit nodes is measured, the above process can be used to evaluate the concentration of soil plastic pollutants at the corresponding deep-layer land micro-unit nodes, thereby providing a basis for further determining key governance areas.

[0125] Example 2: Based on the same inventive concept, Figure 2 As shown, this embodiment provides a natural resource management method based on graph neural network optimization, the method comprising:

[0126] S1. Divide the target plot into soil layers from shallow soil to deep soil at preset intervals to obtain vertical soil layers of the target plot, divide the vertical soil layers into horizontal regions at preset intervals to obtain a land micro-unit node array, obtain soil through a soil sampler arranged at a preset sampling point of the target plot and perform basic soil property detection to obtain soil property data of the target plot including soil density, soil porosity, and soil particle size, measure the slope at different positions of the target plot through a slope meter to obtain land micro-unit slope mapping data, and retrieve the rainfall data of the meteorological station of the target plot; use the target plot soil property data, land micro-unit slope mapping data, and rainfall data as the node feature vector of the land micro-unit node.

[0127] S2. Input the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node. The land spatial feature fitting model is based on soil property data obtained from historical sampling data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data as training data, and uses graph neural network to perform model training to establish a mapping relationship between land micro-unit node characteristics and the vertical migration coefficient of plastic pollutants at land micro-unit nodes.

[0128] S3. Based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of plastic pollutants at each land micro-unit node and the soil layer depth of each land micro-unit node, the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot is recursively calculated through the migration and diffusion model. The land micro-unit node locations with plastic pollutant content greater than the preset threshold are selected as key treatment areas, and plastic pollutant cleaning decisions are implemented in the key treatment areas.

[0129] It should be noted that the method of inputting the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node includes:

[0130] S21. Calculate the spatial distance between nodes using Euclidean distance according to the spatial coordinates of each node of the land micro-unit node array in the target plot spatial graph network to obtain the spatial distance of the land micro-unit nodes.

[0131] S22. When the spatial distance between the land micro-unit nodes is less than a preset association threshold, the corresponding two land micro-unit nodes are marked as adjacent nodes and node connection edges are established to obtain the target land parcel spatial graph network.

[0132] S23, create a network of land micro-units with a size of The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is:

[0133] .

[0134] in is the adjacency matrix No. Rank Elements of a column.

[0135] S24. Calculate the land micro-unit degree matrix based on the element nodes in the adjacency matrix for:

[0136] .

[0137] in is the number of land micro-unit nodes, It is the first Rank Elements of a column.

[0138] S25. Obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and degree matrix for:

[0139] .

[0140] in is the land micro-unit feature aggregation matrix.

[0141] S26. Perform graph convolution operation on the node feature vector according to the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit.

[0142] It should be noted that the method of performing graph convolution operation on node feature vectors according to the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in land micro-units includes:

[0143] S261, the input layer inputs the node feature vector, and each intermediate layer aggregates the features of adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output:

[0144] .

[0145] in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer.

[0146] S262. The output layer outputs the vertical migration coefficient of plastic pollutants in land micro-units.

[0147] It should be noted that the method of using soil property data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data obtained from historical sampling data as training data and performing model training through a graph neural network includes:

[0148] A1. Based on the historical sampling data of plastic pollutant concentrations in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is:

[0149] .

[0150] in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants.

[0151] A2. Set up a four-layer graph neural network, where the node feature dimension of the input layer is set to 5, the feature dimension of the middle layer is set to 64, and the feature dimension of the output layer is set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data are used as the input of the training set, and the vertical migration coefficient of plastic pollutants is used as the output of the training set to train the land spatial feature fitting model.

[0152] A3. Use the mean square error loss function as the loss function. The loss function is:

[0153] .

[0154] in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value.

[0155] A4. Optimize the loss function using the Adam optimization algorithm, iteratively updating the weight matrix of each layer until the loss value stops decreasing after a preset number of consecutive iterations.

[0156] It should be noted that the method for recursively calculating the soil plastic pollutant concentration at the deep land micro-unit nodes of the target plot based on the soil plastic pollutant concentration at the first layer of the land micro-unit nodes of the target plot, the vertical migration coefficient of the plastic pollutants at each land micro-unit node, and the soil layer depth of each land micro-unit node through a migration and diffusion model includes:

[0157] S31. Sampling and testing the soil of the first layer of land micro-unit nodes in the land micro-unit node array to obtain the soil plastic pollutant concentration of the corresponding land micro-unit nodes, and obtaining the soil layer depth of the land micro-unit nodes according to the preset intervals of soil layer division from shallow soil to deep soil.

[0158] S32. Based on the soil plastic pollutant concentration at the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant at the land micro-unit node, and the soil layer depth of the land micro-unit node, the soil plastic pollutant concentration at the deep land micro-unit node of the target plot is recursively calculated using the migration and diffusion model:

[0159] .

[0160] in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is Yes, etc. Soil layer depth, Yes, etc. Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

[0161] It should be noted that, regarding the method in the above embodiment, the specific manner of performing each operation has been described in detail in Example 1 of the method, and will not be elaborated here.

[0162] Finally, it should be noted that although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A natural resource management system based on graph neural network optimization, characterized in that: The system includes: a land micro-unit data acquisition module, a land spatial feature fitting module, and a plastic pollutant migration assessment module, and the modules are connected in sequence; The land micro-unit data acquisition module is used to divide the target plot of land from shallow soil to deep soil at preset intervals to obtain vertical soil layers of the target plot of land, divide the vertical soil layers into horizontal regions at preset intervals to obtain a land micro-unit node array, obtain soil through soil samplers arranged at preset sampling points of the target plot of land and perform basic soil property tests to obtain soil property data of the target plot of land, including soil density, soil porosity, and soil particle size, measure the slope of different locations of the target plot of land using a slope meter to obtain land micro-unit slope mapping data, and retrieve rainfall data from the meteorological station to which the target plot of land belongs; and use the target plot of land property data, land micro-unit slope mapping data, and rainfall data as node feature vectors of land micro-unit nodes; The land spatial feature fitting module is used to input the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node. The land spatial feature fitting model is a mapping relationship between land micro-unit node features and land micro-unit node plastic pollutant vertical migration coefficients established by model training using soil property data, land micro-unit slope mapping data, adjacent two-layer soil plastic pollutant concentration data, and rainfall data obtained from historical sampling data as training data through a graph neural network; The plastic pollutant migration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit nodes of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of the plastic pollutants at each land micro-unit node, and the soil layer depth of each land micro-unit node through a migration and diffusion model, and select the land micro-unit node locations where the soil plastic pollutant concentration is greater than a preset threshold as key treatment areas, and execute plastic pollutant cleanup decisions for the key treatment areas; The land spatial feature fitting module includes: a node distance calculation module, a spatial graph network generation module, an adjacency determination module, a degree matrix generation module, a feature aggregation matrix generation module, and a graph convolution module, and the modules are connected in sequence; The node distance calculation module is used to calculate the spatial distance between nodes according to the spatial coordinates of each node of the land micro-unit node array in the target land plot spatial graph network by using Euclidean distance to obtain the spatial distance of the land micro-unit nodes; The spatial graph network generation module is configured to mark two corresponding land micro-unit nodes as adjacent nodes and establish node connection edges to obtain a target land parcel spatial graph network when the spatial distance between the land micro-unit nodes is less than a preset association threshold; The adjacency determination module is used to create a node with a size of The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is: ； in is the adjacency matrix No. Rank Elements of the column; The degree matrix generation module is used to calculate the land micro-unit degree matrix based on the element nodes in the adjacency matrix. for: ； in is the number of land micro-unit nodes, It is the first Rank Elements of the column; The feature aggregation matrix generation module is used to obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and the degree matrix. for: ； in is the land micro-unit characteristic aggregation matrix; The graph convolution module is used to perform a graph convolution operation on the node feature vector according to the land micro-unit feature aggregation matrix to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit; The graph convolution module includes a connected layer aggregation module and an output module; The layer aggregation module is used to input the feature vector of the input node of the input layer. Each intermediate layer aggregates the features of the adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output as follows: ； in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer; The output module is used to output the vertical migration coefficient of plastic pollutants in land micro-units in the output layer.

2. A natural resource management system based on graph neural network optimization according to claim 1, characterized in that: The training process of the land space feature fitting module includes: According to the historical sampling data of the concentration of plastic pollutants in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is: ； in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants; A four-layer graph neural network was set up, with the node feature dimension of the input layer set to 5, the feature dimension of the middle layer set to 64, and the feature dimension of the output layer set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data were used as the input of the training set, and the vertical migration coefficient of plastic pollutants was used as the output of the training set to train the land spatial feature fitting model. The mean square error loss function is used as the loss function, and the loss function is: ； in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value; The loss function is optimized according to the Adam optimization algorithm, and the weight matrix of each layer is iteratively updated until the loss value stops decreasing after a preset number of consecutive iterations.

3. The natural resource management system based on graph neural network optimization according to claim 1 is characterized in that: The plastic pollutant migration assessment module includes a surface data acquisition module and a plastic pollutant concentration assessment module connected to each other; The surface data acquisition module is used to sample and detect the soil of the first layer of land micro-unit nodes in the land micro-unit node array to obtain the soil plastic pollutant concentration of the corresponding land micro-unit node, and to obtain the soil layer depth of the land micro-unit node according to the preset interval of soil layer division from shallow soil to deep soil; The plastic pollutant concentration assessment module is used to recursively calculate the soil plastic pollutant concentration of the deep land micro-unit node of the target plot based on the soil plastic pollutant concentration of the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant of the land micro-unit node, and the soil layer depth of the land micro-unit node through the migration and diffusion model: ； in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is It is Soil layer depth, It is Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

4. A natural resource management method based on graph neural network optimization, the method being implemented based on the system according to any one of claims 1 to 3, characterized in that: The method comprises: The target plot is divided into soil layers from shallow to deep soil layers at preset intervals to obtain vertical soil layers of the target plot. The vertical soil layers are divided into horizontal regions at preset intervals to obtain a land micro-unit node array. Soil is sampled by placing a soil sampler at a preset sampling point on the target plot and basic soil property tests are performed to obtain soil property data of the target plot, including soil density, soil porosity, and soil particle size. The slope of different locations on the target plot is measured using a slope meter to obtain land micro-unit slope mapping data. The rainfall data of the meteorological station to which the target plot belongs is retrieved. The target plot soil property data, land micro-unit slope mapping data, and rainfall data are used as node feature vectors of the land micro-unit nodes. The node feature vectors of the land micro-unit nodes are input into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vectors of the land micro-unit nodes of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit nodes. The land spatial feature fitting model is a mapping relationship between land micro-unit node features and vertical migration coefficients of plastic pollutants at land micro-unit nodes established by model training using soil property data obtained from historical sampling data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data; The concentration of soil plastic pollutants at the deep land micro-unit nodes of the target plot is recursively calculated based on the soil plastic pollutant concentration at the first-layer land micro-unit nodes of the target plot, the vertical migration coefficient of plastic pollutants at each land micro-unit node, and the soil layer depth of each land micro-unit node through the migration and diffusion model. The locations of land micro-unit nodes with plastic pollutant content greater than the preset threshold are selected as key treatment areas, and plastic pollutant cleanup decisions are implemented in the key treatment areas; The method of inputting the node feature vector of the land micro-unit node into the land spatial feature fitting model to perform spatial feature aggregation on the node feature vector of the land micro-unit node of the target plot to obtain the vertical migration coefficient of plastic pollutants at the land micro-unit node includes: According to the spatial coordinates of each node of the land micro-unit node array in the target plot spatial graph network, the spatial distance between nodes is calculated by Euclidean distance to obtain the spatial distance of the land micro-unit node; When the spatial distance between land micro-unit nodes is less than the preset association threshold, the corresponding two land micro-unit nodes are marked as adjacent nodes and node connection edges are established to obtain the target land parcel spatial graph network; According to the number of land micro-unit nodes in the target plot spatial graph network, a size of The land micro-unit zero matrix is ​​updated according to the node connection edge to obtain the land micro-unit adjacency matrix , the update rule is: ； in is the adjacency matrix No. Rank Elements of the column; The land micro-unit degree matrix is ​​calculated based on the element nodes in the adjacency matrix for: ； in is the number of land micro-unit nodes, It is the first Rank Elements of the column; Obtain the land micro-unit feature aggregation matrix based on the adjacency matrix and degree matrix for: ； in is the land micro-unit characteristic aggregation matrix; According to the land micro-unit feature aggregation matrix, the node feature vector is subjected to graph convolution operation to obtain the vertical migration coefficient of plastic pollutants in the land micro-unit; The method for obtaining the vertical migration coefficient of plastic pollutants in land micro-units by performing a graph convolution operation on the node feature vector according to the land micro-unit feature aggregation matrix includes: The input layer inputs the node feature vector. Each layer in the middle layer aggregates the features of the adjacent nodes through the land micro-unit feature aggregation matrix and performs nonlinear transformation through the activation function to obtain the layer output: ； in is the activation function, It is The output features of the layer, It is The output features of the layer, It is The weight of the layer; The output layer outputs the vertical migration coefficient of plastic pollutants in land micro-units.

5. The natural resource management method based on graph neural network optimization according to claim 4 is characterized in that: The method of using soil property data, land micro-unit slope mapping data, plastic pollutant concentration data of two adjacent layers of soil, and rainfall data obtained from historical sampling data as training data and performing model training through a graph neural network includes: According to the historical sampling data of the concentration of plastic pollutants in two adjacent layers of soil, the corresponding vertical migration coefficient of plastic pollutants is: ； in is the concentration of plastic pollutants in the upper soil layer, is the concentration of plastic pollutants in the lower soil layer, is the depth difference between adjacent soil layers, is the vertical migration coefficient of plastic pollutants; A four-layer graph neural network was set up, with the node feature dimension of the input layer set to 5, the feature dimension of the middle layer set to 64, and the feature dimension of the output layer set to 1. Soil density, soil porosity, soil particle size, land micro-unit slope mapping data, and rainfall data were used as the input of the training set, and the vertical migration coefficient of plastic pollutants was used as the output of the training set to train the land spatial feature fitting model. The mean square error loss function is used as the loss function, and the loss function is: ； in is the vertical migration coefficient of plastic pollutants predicted by the model, is the actual vertical migration coefficient of plastic pollutants in the training set samples, is the number of training set samples, is the loss value; The loss function is optimized according to the Adam optimization algorithm, and the weight matrix of each layer is iteratively updated until the loss value stops decreasing after a preset number of consecutive iterations.

6. The natural resource management method based on graph neural network optimization according to claim 4 is characterized in that: The method for obtaining the soil plastic pollutant concentration of the deep land micro-unit node of the target plot by recursively calculating the soil plastic pollutant concentration of the first layer land micro-unit node of the target plot, the vertical migration coefficient of the plastic pollutant at each land micro-unit node, and the soil layer depth of each land micro-unit node through a migration and diffusion model includes: The soil of the first layer of land micro-unit nodes in the land micro-unit node array is sampled and tested to obtain the soil plastic pollutant concentration of the corresponding land micro-unit node, and the soil layer depth of the land micro-unit node is obtained according to the preset interval of soil layer division from shallow soil to deep soil; According to the soil plastic pollutant concentration of the first-layer land micro-unit node, the vertical migration coefficient of the plastic pollutant at the land micro-unit node, and the soil layer depth of the land micro-unit node, the soil plastic pollutant concentration at the deep land micro-unit node of the target plot is recursively calculated through the migration and diffusion model: ； in is the number of vertical layers of land micro-unit nodes, It is The horizontal Vertical The soil plastic pollutant concentration of each land micro-unit node, It is The horizontal Vertical The vertical migration coefficient of plastic pollutants at the land micro-unit node is It is Soil layer depth, It is Soil layer depth, is the base of natural logarithms, It is The horizontal Vertical The concentration of soil plastic pollutants at each land micro-unit node.

Images