Space-attribute heterogeneous feature-based drainage basin pollution traceability hierarchical network construction method and system

Abstract

The invention relates to a spatial-attribute heterogeneous feature-based drainage basin pollution traceability hierarchical network construction method and system, and belongs to the technical field of environmental science and computer crossing. According to the method, heterogeneous environmental factors such as population, GDP, rainfall, land utilization and water quality are divided into a plurality of layers through combination of attribute correlation analysis and spatial layering modeling, an intra-layer connection relation under spatial adjacency and attribute driving is constructed, an inter-layer interaction rule is designed, a three-dimensional heterogeneous layering graph is formed, and the three-dimensional heterogeneous layering graph is constructed. Therefore, the subsequent high-fidelity modeling of the pollution diffusion path and the accurate quantification of the traceability contribution rate are realized. The method not only can accurately capture a space path and key influence factors of pollution diffusion, but also can realize upstream backtracking identification of downstream pollution points and pollution contribution rate estimation of non-monitoring areas, has the advantages of high adaptability and universality and the like, is suitable for pollutant traceability modeling and environment decision support of multiple types of drainage basins, and has wide application prospects. And the method has good engineering popularization and practical application values.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of environmental science and computer science, and relates to a method and system for constructing a hierarchical network for tracing watershed pollution sources based on spatial-attribute heterogeneous characteristics. Background Technology

[0002] With the increasing severity of pollution problems in watershed ecosystems, how to efficiently and accurately identify pollution sources, understand pollutant migration pathways, and assess the contribution rate of different regions has become a core scientific issue in current watershed environmental governance. In existing research frameworks, pollutant migration processes typically rely on hydrological models (such as SWAT and MIKE) or statistical learning methods (such as regression analysis and spatial interpolation) for modeling. While these methods possess certain simulation and fitting capabilities, they generally suffer from three limitations: 1) Poor data structure adaptability: Traditional methods are unable to fully integrate multi-source environmental data (such as population density, land use, meteorological factors, socio-economic indicators, etc.) that are heterogeneous in origin and different in scale, resulting in insufficient ability of the model to characterize the actual pollution driving factors. 2) Weak spatial structure representation: Watershed systems have obvious spatial topological structure characteristics. Pollutant migration follows complex spatial diffusion and topographic guidance paths. Existing methods often use grids or administrative boundaries as units, failing to depict the real connection logic between nodes. 3) Insufficient description of causal mechanisms and temporal dynamics: Pollution migration is often accompanied by multi-factor coupling effects (such as the linkage effect of rainfall-topography-land use), and has time accumulation and lag feedback mechanisms. Traditional static modeling or single-layer structure methods are difficult to capture these complex feedback paths.

[0003] In recent years, complex network theory has attracted attention in ecosystem and environmental system modeling due to its advantages in multidimensional entity and relationship modeling, structural propagation process analysis, and system stability assessment. However, most existing applications are limited to single-layer networks (such as pollution diffusion maps and ecological interaction maps), failing to systematically express the structural hierarchies, attribute couplings, and cross-layer information flow mechanisms among multi-source factors, thus limiting its application potential in source tracing modeling, impact path identification, and structural optimization control.

[0004] Therefore, there is an urgent need to propose a hierarchical heterogeneous network modeling method that integrates spatial adjacency structure and multi-attribute driving features, and possesses a cross-layer information propagation mechanism, in order to achieve high-fidelity modeling of pollution diffusion paths and accurate quantification of source contribution rates. This invention is proposed to address the above problems, possessing multiple advantages such as strong data fusion capabilities, reasonable structural modeling, and support for cross-layer and temporal propagation modeling, providing new ideas and methodological support for watershed environmental modeling and management decision-making. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method and system for constructing a hierarchical network for tracing pollution sources in watersheds based on spatial-attribute heterogeneous characteristics. This method combines attribute correlation analysis with spatial hierarchical modeling to divide heterogeneous environmental factors such as population, GDP, rainfall, land use, and water quality into multiple layers, and constructs spatial adjacency and attribute-driven intra-layer connectivity relationships, designs inter-layer interaction rules, and forms a three-dimensional heterogeneous hierarchical map, thereby achieving high-fidelity modeling of pollution diffusion paths and accurate quantification of source tracing contribution rates.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for constructing a hierarchical network for tracing watershed pollution sources based on spatial-attribute heterogeneous features, the method specifically includes the following steps: S1. Conduct statistical and correlation analysis on multi-source environmental factors in the watershed, classify population density, GDP, land use, rainfall, hydro-topographic factors, etc. into different categories, and complete the layer division; S2. Using spatial grid units as nodes, construct node attribute vectors for the socio-economic layer, land use layer, and hydro-meteorological layer; S3. Within each layer, establish intra-layer connections based on spatial adjacency and attribute similarity. The socio-economic layer and land use layer are connected using cosine similarity, while the hydro-meteorological layer uses digital elevation model to extract slope and flow direction to generate directed weighted edges. S4. Establish cross-layer connections between different layers through spatial mapping and vertical mapping principles to achieve interaction between attributes and spatial information; S5. Design intra-layer and cross-layer propagation mechanisms to aggregate and update node features and obtain global feature representations of nodes; S6. Introduce a time-dynamic update function to update the state of nodes at different time steps in order to characterize the propagation and evolution of pollutants in the watershed.

[0007] Furthermore, in step S1, the layers are divided into different numbers of layers according to different needs; the layer division includes, but is not limited to, 3 layers: socio-economic layer: node attributes include population density and GDP; land use layer: node attributes include farmland, forest and impermeable surface ratio; hydro-meteorological layer: node attributes include rainfall, slope and flow direction.

[0008] Furthermore, in step S2, the specific steps include: in the three-layer graph structure, constructing the corresponding attribute features of the grid as nodes; the connections between nodes include two types: intra-layer edges: reflecting the similarity or flow relationship between adjacent grids in the same layer; cross-layer edges: representing the attribute coupling of the same grid in different layers; Environmental factors are divided into a socio-economic layer (Socio), a land-use layer (LU), and a hydro-meteorological layer (Hydro). Each layer uses a spatial grid as nodes and explicit attribute vectors as feature representations, forming a multi-layered heterogeneous network structure. (1) No. A layered network, consisting of a binary representation of a set of nodes and a set of edges. This refers to the hierarchical types in a hierarchical network, including the socio-economic layer (Socio), the land use layer (LU), and the hydro-meteorological layer (Hydro). Presentation layer A set of nodes; each node corresponds to a 1 km × 1 km grid cell; Presentation layer The set of edges between internal nodes represents the attribute similarity or flow relationship within the layer; The node feature representation is unified as follows: (2) Presentation layer The eigenvector of node i in the middle. Dimensions representing node features; Indicates that node i is in layer Regarding the value of the k-th feature, The transpose operation converts the feature list into a column vector.

[0009] Furthermore, in step S3, the intra-layer connections between the socio-economic layer and the land use layer are established by calculating the cosine similarity of attributes between nodes and combining it with spatial adjacency; the intra-layer connections of the hydro-meteorological layer are constructed based on slope and flow direction extracted from DEM data, using slope as the edge weight; specifically including: The intra-layer connections between the socio-economic layer and the land use layer are constructed based on spatial adjacency and attribute similarity, with attribute similarity measured by cosine similarity. (3) The cosine similarity function represents the similarity between two attribute vectors. Represents nodes in the socio-economic layer eigenvectors, Represents land use layer nodes eigenvectors, Represents the dot product of two vectors. vector Euclidean norm, vector The Euclidean norm; The final rule for constructing inner joins is expressed as follows: (4) This represents the cross-layer connection between node i in the socio-economic layer and node j in the land use layer; a value of 1 indicates a connection, and 0 indicates no connection. This represents the spatial distance between node i and node j. This represents a distance threshold, controlling the spatial adjacency range of cross-layer edges. This represents the cosine similarity between the feature vectors of the two entities. This represents a similarity threshold; only nodes that exceed this threshold are considered to have sufficient attribute relevance. The hydro-meteorological layer extracts the clear flow direction based on the DEM and constructs a directed weighted network, with slope and flow direction controlling the weights:

[0010] Furthermore, in step S4, cross-layer connections establish vertical mapping edges through nodes in different layers of the same spatial grid to support information flow between multiple layers; specifically including: Cross-layer connections clarify the cross-layer coupling relationships between nodes through vertical mapping and spatial location interaction principles, enhancing information flow and synergy between layers: (7).

[0011] Furthermore, in step S5, the cross-layer propagation mechanism uses a cross-layer weight matrix to control information transfer between different layers, achieving cross-layer fusion of contamination states; the node feature update uses a graph convolution operator to weight and aggregate the features of adjacent nodes, and then reconstructs the features through a nonlinear activation function, specifically including: Information propagation paths are constructed using cross-layer weight matrices, and the inter-layer propagation formula is as follows: (8) .

[0012] Furthermore, in step S6, the time-based dynamic update function is used to describe the change in node contamination status over time, and its form is:

[0013] With the introduction of the time dimension, node features are dynamically updated over time to capture the evolutionary characteristics during the pollution diffusion process. The node state update formula is as follows: .

[0014] Furthermore, the method also includes a network structure performance evaluation step, which includes the calculation and analysis of network connectivity, path characteristics, clustering indices, community structure, and centrality characteristics.

[0015] This invention also provides a hierarchical network construction system for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics.

[0016] The beneficial effects of this invention are as follows: The proposed method for constructing a hierarchical heterogeneous network for watershed pollution source tracing can systematically integrate multi-source heterogeneous information from multiple dimensions such as socio-economic, land use, and hydro-meteorological data within a watershed. This allows for structured modeling and dynamic simulation of pollution diffusion processes within a unified graph framework. Through intra-layer connectivity driven by spatial adjacency and attribute correlation, and a cross-layer connectivity mechanism built on spatial mapping principles, this method effectively characterizes the structural coupling and synergistic driving effects between different types of environmental factors. Furthermore, the introduction of a graph neural propagation mechanism and a time-dynamic update function enables the model to continuously express the evolution path of pollution states and learn propagation patterns. This method not only accurately captures the spatial paths and key influencing factors of pollution diffusion but also enables upstream backtracking identification of downstream pollution points and estimation of pollution contribution rates in non-monitoring areas. It possesses advantages such as reasonable structural expression, sufficient information fusion, accurate propagation modeling, and strong adaptability and versatility. It is applicable to pollutant source tracing modeling and environmental decision support in various types of watersheds and has significant engineering promotion and practical application value.

[0017] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a distribution map of sampling points in the Yanjin River basin, as shown in this embodiment of the invention. Figure 2 This is the attribute correlation matrix (population, GDP, rainfall, land use) in this invention. Figure 3 This is a diagram of the hierarchical heterogeneous network structure in this invention; Figure 4 This is a schematic diagram of the hierarchical connection of nodes in this invention. Detailed Implementation

[0019] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0020] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0021] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0022] In this embodiment, to better explain the hierarchical network construction method of the present invention, the Yanjin River Basin is used as an example to illustrate the specific implementation of the present invention. This embodiment constructs a hierarchical network model with heterogeneous attributes, temporal evolution, and pollution propagation perception capabilities through multi-source environmental data-driven approaches, spatial structure modeling, and cross-layer information fusion.

[0023] I. Attribute Correlation Analysis and Layer Division First, statistical analysis was conducted on the multi-source attribute factors of the watershed, including population density, GDP, annual rainfall, land use (farmland, forest, impervious surfaces), and slope and flow direction. Based on this, the Pearson correlation coefficient was used to calculate the correlation between attributes. The analysis revealed a moderate positive correlation between population and GDP (r≈0.46), a strong negative correlation between farmland and forest (r≈-0.96), while rainfall showed a weak correlation with other factors (r<0.15), indicating that it has an independent hydrological driving mechanism. These relationships form the basis for subsequent hierarchical modeling. Figure 1This is a distribution map of sampling points in the Yanjin River basin, as shown in this embodiment of the invention. Figure 2 This is the attribute correlation matrix (population, GDP, rainfall, land use) in this invention.

[0024] Based on this, the attributes are divided into three types of layers, as shown in Table 1: Table 1 Layer Design and Node Attribute Definition

[0025] II. Construction of Hierarchical Heterogeneous Network Structure In the three-layer graph structure, each 1km² grid is used as a node, and its corresponding attribute features are constructed. The connections between nodes include two types: Inner edge: Reflects the similarity or flow relationship between adjacent grids in the same layer; Cross-layer edge: indicates the coupling of attributes of the same mesh in different layers.

[0026] The modeling details for each layer are as follows: the socio-economic layer and the land use layer are connected by spatial adjacency and attribute cosine similarity; the hydro-meteorological layer uses DEM data to extract flow direction and slope to construct a directed weighted network; nodes between different layers establish cross-layer edges through a vertical mapping mechanism for information exchange.

[0027] Furthermore, the introduction of graph convolution propagation mechanism and temporal state update function enables the model to describe the dynamic process of pollution state evolution over time.

[0028] III. Intra-layer and Cross-layer Propagation Mechanisms To support the effective transfer of information within and between layers, this invention proposes several propagation mechanisms: Intra-layer propagation mechanism: The standard graph convolution operator is used to perform weighted aggregation of adjacent nodes to achieve spatial proximity perception and homogeneous attribute fusion.

[0029] Cross-layer interaction mechanism: By defining a cross-layer propagation weight matrix, information transformation and feature reconstruction of pollution status between different layers are realized.

[0030] Node fusion mechanism: The features of each node in the three layers are aggregated to obtain a fused global representation vector, which is used for subsequent pollution identification and tracking.

[0031] Time-based dynamic update mechanism: Introducing time variables to update the node state at each time step, thus characterizing the spread and evolution of pollutants at different times.

[0032] Figure 3 This is a diagram of the hierarchical heterogeneous network structure in this invention. Figure 4 This is a schematic diagram of the hierarchical connection of nodes in this invention. In this embodiment, the core components of each layer of the hierarchical network structure are as follows: (1) Hierarchical heterogeneous network construction: In order to effectively characterize the spatial diffusion and attribute-driven features of the watershed environmental system, this invention divides environmental factors into socio-economic factors. The network consists of three layers: Socio (solar), LU (land use), and Hydro (hydrometeorological). Each layer uses a spatial grid as nodes and explicit attribute vectors as feature representations, forming a multi-layered heterogeneous network structure. (1) The node feature representation is unified as follows: (2) (2) Intra-layer connectivity strategy: The intra-layer connectivity between the socio-economic layer and the land use layer is constructed based on spatial adjacency and attribute similarity, with attribute similarity based on the remaining... String similarity is used for measurement: (3) The final rule for constructing inner joins is expressed as follows: (4) (3) The spatial guidance network design of the hydro-meteorological layer is designed to reflect the natural migration direction of pollutants. The hydro-meteorological layer extracts the clear water flow direction based on the DEM and constructs a directional addition network. Weighted network, with weights controlled by slope and flow direction:

[0033] (4) Intra-layer heterogeneous convolution mechanism .

[0034] (5) Cross-layer connection mechanism: Cross-layer connection clarifies the cross-layer coupling relationship between nodes through vertical mapping and spatial location interaction principles, and enhances the information flow and synergy between layers: (7) (6) Inter-layer interactive propagation mechanism To enhance the feature interaction capabilities between layers, an information propagation path is constructed using a cross-layer weight matrix. The inter-layer propagation formula is as follows: .

[0035] (7) Node total feature update mechanism The updated feature representations of nodes from each layer are merged to form the final node output representation. The update mechanism is as follows: .

[0036] (8) The time-based dynamic update mechanism introduces a time dimension, and the node characteristics are dynamically updated over time to capture the evolutionary characteristics during the pollution diffusion process. State update formula: .

[0037] The above links are interconnected to form a complete spatial-attribute heterogeneous hierarchical network structure, providing a unified expression structure for graph neural networks and enhancing the spatiotemporal perception capability of pollution migration modeling.

[0038] IV. Network Structure Performance Evaluation In this embodiment, to verify the rationality of the constructed network, the three-layer network structure is evaluated separately: 4.1 Land Use Layer Network Structure Assessment Table 2 Network Structure and Connectivity

[0039] Table 3 Path characteristics and propagation efficiency

[0040] Table 4 Aggregation and Spatial Structure

[0041] Table 5 Community Structure and Centrality Characteristics

[0042] Tables 2–5 comprehensively illustrate that the layer has a balanced structure, good propagation efficiency, and suitable topological properties, providing a foundation for pollution path modeling.

[0043] 4.2 Socioeconomic Layer Network Optimization and Evaluation To improve the connectivity and reliability of the original network, the following optimization mechanisms are introduced. The optimized network performance is significantly improved: Table 6. Network Structure Improvements

[0044] Table 7 Comparison of network structure performance before and after improvement

[0045] Tables 6 and 7 show that the optimized network structure is more robust and has stronger connectivity, making it suitable for graph neural network learning.

[0046] 4.3 Assessment of the hydrometeorological network The hydrological layer focuses on simulating the path of pollutants propagating downstream with water flow, and its topological characteristics are as follows: Table 8 Network Structure and Connectivity

[0047] Table 9 Path characteristics and propagation efficiency

[0048] Table 10 Clustering and Spatial Structure

[0049] Table 11 Community Structure and Centrality Characteristics

[0050] Tables 8–11 show that the hydrological layer map has a reasonable propagation structure, which is suitable for pollution trajectory modeling and control area identification.

[0051] In summary, the technical solution provided by this invention has multiple advantages, such as strong data fusion capability, reasonable structural modeling, and support for cross-layer and temporal propagation modeling, providing new ideas and methodological support for watershed environment modeling and management decision-making.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a hierarchical network for tracing watershed pollution sources based on spatial-attribute heterogeneous characteristics, characterized in that, The method specifically includes the following steps: S1. Conduct statistical and correlation analysis on multi-source environmental factors in the watershed, classify population density, GDP, land use, rainfall, and hydro-topographic factors into different categories, and complete the layer division; S2. Using spatial grid units as nodes, construct node attribute vectors for the socio-economic layer, land use layer, and hydro-meteorological layer; S3. Within each layer, establish intra-layer connections based on spatial adjacency and attribute similarity. The socio-economic layer and land use layer are connected using cosine similarity, while the hydro-meteorological layer uses digital elevation model to extract slope and flow direction to generate directed weighted edges. S4. Establish cross-layer connections between different layers through spatial mapping and vertical mapping principles to achieve interaction between attributes and spatial information; S5. Design intra-layer and cross-layer propagation mechanisms to aggregate and update node features and obtain global feature representations of nodes; S6. Introduce a time-dynamic update function to update the state of nodes at different time steps in order to characterize the propagation and evolution of pollutants in the watershed.

2. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 1, characterized in that, In step S1, the layers are divided into different numbers of layers according to different needs; the layer division includes, but is not limited to, 3 layers: socio-economic layer: node attributes include population density and GDP; land use layer: node attributes include the proportion of farmland, forest and impermeable surface; hydro-meteorological layer: node attributes include rainfall, slope and flow direction.

3. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 2, characterized in that, Step S2 specifically includes: in the three-layer graph structure, constructing the corresponding attribute features of the grid as nodes; the connections between nodes include two types: intra-layer edges: reflecting the similarity or flow relationship between adjacent grids in the same layer; cross-layer edges: representing the attribute coupling of the same grid in different layers; Environmental factors are divided into a socio-economic layer (Socio), a land-use layer (LU), and a hydro-meteorological layer (Hydro). Each layer uses a spatial grid as nodes and explicit attribute vectors as feature representations, forming a multi-layered heterogeneous network structure. （1） No. A layered network, consisting of a binary representation of a set of nodes and a set of edges. This refers to the hierarchical types in a hierarchical network, including the socio-economic layer (Socio), the land use layer (LU), and the hydro-meteorological layer (Hydro). Presentation layer A set of nodes; each node corresponds to a 1 km × 1 km grid cell; Presentation layer The set of edges between internal nodes represents the attribute similarity or flow relationship within the layer; The node feature representation is unified as follows: （2） Presentation layer The eigenvector of node i in the middle. Dimensions representing node features; Indicates that node i is in layer Regarding the value of the k-th feature, The transpose operation converts the feature list into a column vector.

4. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 3, characterized in that, In step S3, the intra-layer connection between the socio-economic layer and the land use layer is established by calculating the cosine similarity of attributes between nodes and combining spatial adjacency; the intra-layer connection of the hydro-meteorological layer is based on the slope and water flow direction extracted from DEM data, constructing directed weighted edges, and using the slope as the edge weight. Specifically, it includes: The intra-layer connections between the socio-economic layer and the land use layer are constructed based on spatial adjacency and attribute similarity, with attribute similarity measured by cosine similarity. （3） The cosine similarity function represents the similarity between two attribute vectors. Represents nodes in the socio-economic layer eigenvectors, Represents land use layer nodes eigenvectors, Represents the dot product of two vectors. vector Euclidean norm, vector The Euclidean norm; The final rule for constructing inner joins is expressed as follows: （4） This represents the cross-layer connection between node i in the socio-economic layer and node j in the land use layer; a value of 1 indicates a connection, and 0 indicates no connection. This represents the spatial distance between node i and node j. This represents a distance threshold, controlling the spatial adjacency range of cross-layer edges. This represents the cosine similarity between the feature vectors of the two entities. This represents a similarity threshold; only nodes that exceed this threshold are considered to have sufficient attribute relevance. The hydro-meteorological layer extracts the clear flow direction based on the DEM and constructs a directed weighted network, with slope and flow direction controlling the weights: 。 5. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 4, characterized in that, In step S4, cross-layer connections establish vertical mapping edges through nodes in different layers of the same spatial grid to support information flow between multiple layers; specifically including: Cross-layer connections clarify the cross-layer coupling relationships between nodes through vertical mapping and spatial location interaction principles, enhancing information flow and synergy between layers: （7）。 6. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 5, characterized in that, In step S5, the cross-layer propagation mechanism uses a cross-layer weight matrix to control information transfer between different layers, achieving cross-layer fusion of contamination states; the node feature update uses a graph convolution operator to weight and aggregate the features of adjacent nodes, and then reconstructs the features through a nonlinear activation function, specifically including: Information propagation paths are constructed using cross-layer weight matrices, and the inter-layer propagation formula is as follows: （8） 。 7. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 6, characterized in that, In step S6, the time-based dynamic update function is used to describe the change in node contamination status over time, and its form is: With the introduction of the time dimension, node features are dynamically updated over time to capture the evolutionary characteristics during the pollution diffusion process. The node state update formula is as follows: 。 8. The method for constructing a hierarchical network for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics according to claim 7, characterized in that, The method also includes a network structure performance evaluation step, which includes the calculation and analysis of network connectivity, path characteristics, clustering indices, community structure, and centrality characteristics.

9. A hierarchical network construction system for watershed pollution source tracing based on spatial-attribute heterogeneous characteristics, characterized in that, The system employs the method described in any one of claims 1 to 8.

Images