Micro-plastic pollution tracing method, system and equipment and storage medium

Abstract

The invention relates to a microplastic pollution traceability method, system and device and a storage medium, and is applied to the field of water pollution traceability. Cleaning, standardizing and spatially aligning micro-plastic data and then storing; constructing a micro-plastic pollution source space-time diagram, and performing definition reinforcement on the micro-plastic pollution source space-time diagram to obtain a reinforced micro-plastic pollution source space-time diagram; performing space-time dependence modeling on the enhanced micro-plastic pollution source space-time diagram to obtain a dynamic pollution propagation probability diagram, extracting feature vectors of social economic plastic data and micro-plastic multi-modal feature data in the micro-plastic data, and performing fusion to obtain fused multi-modal features; performing model training by fusing the multi-modal features and the dynamic pollution propagation probability graph to obtain a traceability model; and setting a dynamic threshold value and a dynamic risk level, and performing microplastic pollution tracing and pollution alarm according to the dynamic threshold value, the dynamic risk level and the tracing model. The technical effect of the invention is that the microplastic pollution traceability precision is improved.

Description

Technical Field

[0001] This application relates to the field of water pollution source tracing, and in particular to a method, system, device and storage medium for tracing microplastic pollution sources. Background Technology

[0002] Emerging pollutants are newly discovered pollutants that pose significant risks to the ecological environment and human health. Microplastics are one of the major emerging pollutants, with complex and spatially heterogeneous sources, and are closely related to human activities within the region. Conducting regional-scale source apportionment is fundamental to the risk management of microplastics in water bodies.

[0003] Currently, the source tracing of microplastic pollution in water bodies mainly relies on three types of technical approaches. The first type is analytical methods based on the physicochemical characteristics of microplastics; the second type uses chemical tracer correlation strategies; and the third type is spatial statistics and model simulation.

[0004] Existing technologies for tracing the source of microplastic pollution in water bodies use a single type of data, resulting in insufficient specificity of microplastic characteristic sources and thus affecting the accuracy of microplastic pollution tracing. Summary of the Invention

[0005] To improve the accuracy of microplastic pollution source tracing, this application provides a microplastic pollution source tracing method, system, equipment, and storage medium.

[0006] Firstly, this application provides a method for tracing the source of microplastic pollution, employing the following technical solution: Receive microplastic data, which includes socioeconomic plastic data, microplastic multimodal characteristic data, and spatial heterogeneous network data; The microplastic data is stored after being cleaned, standardized, and spatially aligned. A spatiotemporal map of microplastic pollution sources is constructed, and the spatiotemporal map of microplastic pollution sources is defined and enhanced based on the microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources. Spatiotemporal dependency modeling is performed on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map. Feature vectors of the socioeconomic plastic data and the microplastic multimodal feature data are extracted and fused to obtain fused multimodal features. The source tracing model is obtained by training the model using the fused multimodal features and the dynamic pollution propagation probability map; Set dynamic thresholds and dynamic risk levels, and perform microplastic pollution source tracing and pollution alarm based on the dynamic thresholds, dynamic risk levels, and source tracing model.

[0007] The above technical solution collects multi-source heterogeneous data, including socioeconomic plastics data, microplastic multimodal characteristic data, and spatial heterogeneous network data, as microplastic feature sources. A multi-source heterogeneous data fusion architecture is constructed to integrate these three types of heterogeneous data: socioeconomic flux, microplastic multimodal characteristics, and spatial network data. Without any one of these, the technology cannot achieve a closed-loop analysis of pollution source-migration path-environmental impact. This multi-level, multimodal data fusion and modeling method overcomes the limitations of a single data source, enabling precise location and dynamic tracking of microplastic pollution sources.

[0008] In one specific implementation scheme, the method is based on a socio-economic statistical yearbook, and the receiving of microplastic data includes: The socioeconomic data on plastics were obtained by analyzing the socioeconomic statistical yearbook. Plastic samples were obtained from river and lake basins. The physical characteristics of microplastics were obtained by identifying the sampled plastic samples, and the chemical characteristics of microplastics were obtained by analyzing the sampled plastic samples. The physical and chemical characteristics of microplastics are fused across modes to obtain microplastic fusion characteristics, and microplastic abundance data are calculated based on the microplastic fusion characteristics. By combining the physical characteristics, chemical characteristics, and abundance data of the microplastics, multimodal characteristic data of microplastics is obtained. Receive spatially heterogeneous network data, which includes river networks, ocean current paths, meteorological data, and monitoring station locations.

[0009] The above technical solution collects multi-source heterogeneous data, including socioeconomic plastics data, microplastic multimodal characteristic data, and spatial heterogeneous network data, as microplastic feature sources. A multi-source heterogeneous data fusion architecture is constructed to integrate these three types of heterogeneous data: socioeconomic flux, microplastic multimodal characteristics, and spatial network data. Without any one of these, the technology cannot achieve a closed-loop analysis of pollution source-migration path-environmental impact. This multi-level, multimodal data fusion and modeling method overcomes the limitations of a single data source, enabling precise location and dynamic tracking of microplastic pollution sources.

[0010] Furthermore, when acquiring multimodal microplastic feature data, collaborative computation of microplastic features is performed. This multimodal feature vector calculation improves the accuracy of microplastic source identification, comprehensively considering multiple physical characteristics such as shape, color, size, and polymer type of microplastics, overcoming the limitations of single-feature identification. In particular, the introduction of microplastic abundance data (DFI) quantifies the dispersion of microplastic features, providing more comprehensive feature information for subsequent source tracing analysis.

[0011] In one specific implementation, the method is based on a unified geographic coordinate system, and storing the microplastic data after cleaning, standardizing, and spatially aligning the microplastic data includes: Outliers in the microplastic data are removed, and missing values ​​in the microplastic data are filled using Kriging space interpolation. Units were standardized for the microplastic data, and the socioeconomic plastic data were normalized. The socioeconomic plastics data, the microplastics multimodal feature data, and the spatial heterogeneous network data are mapped to the unified geographic coordinate system to construct a river network spatial topology map; A spatiotemporal database is established, and the socioeconomic plastic data, the microplastic multimodal feature data, and the spatial heterogeneous network data are stored in the spatiotemporal database according to timestamps, spatial locations, and data types.

[0012] Through the above technical solutions, the data repository is the core foundation of the microplastic pollution tracing system, providing a data engine for the construction of microplastic pollution tracing.

[0013] In a specific feasible implementation, the method is based on ecological and environmental monitoring and published data and the STGNN model. The step of constructing a spatiotemporal map of microplastic pollution sources, and defining and enhancing the spatiotemporal map of microplastic pollution sources based on the microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources, includes: Based on the STGNN model, a spatiotemporal graph of microplastic pollution sources is constructed according to the river network spatial topology and the spatial heterogeneous network data. The spatiotemporal graph of microplastic pollution sources includes nodes and edges. The pollution source was determined based on the spatially heterogeneous network data; The node is defined based on the pollution source and the location of the monitoring station; The edges are defined according to the direction of water flow in the river network and the ocean current path; The abundance data, the physical characteristics, and the socioeconomic plastics data are attached to the node as node features; The water flow velocity and wind direction intensity are obtained by analyzing the ecological environment monitoring data. The edges are weighted according to the water flow velocity and wind direction intensity to obtain a spatiotemporal map of enhanced microplastic pollution sources.

[0014] In one specific implementation scheme, the step of performing spatiotemporal dependency modeling on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map includes: The river network topology was aggregated using GraphSAGE to enhance the spatiotemporal map of microplastic pollution sources. A dynamic pollution propagation probability map is obtained by learning the temporal changes of the abundance data using a temporal convolution method.

[0015] Through the above technical solutions, the spatial heterogeneity network construction module integrates various spatial data (such as river networks, ocean current paths, and land use types) to construct a comprehensive spatial network map of microplastic pollution sources. This method considers the heterogeneity of environmental factors and can more accurately reflect the propagation characteristics of microplastics in different spatial environments. A spatiotemporal graph neural network (STGNN) source tracing engine is constructed. From a spatial dimension, GraphSAGE is used to aggregate river network topology relationships; from a temporal dimension, the TCN module captures the impact of sudden changes in human activities before and after the pandemic. Without this, it would be impossible to dynamically quantify the contribution rate of pollution sources, achieve real-time simulation of pollution migration paths, and improve source tracing accuracy.

[0016] In one specific implementation, the method is based on a multimodal encoder, and the step of extracting feature vectors from the socioeconomic plastics data and the microplastics multimodal feature data and fusing them to obtain fused multimodal features includes: The physical features, chemical features, and socioeconomic plastic data are extracted to obtain a multimodal feature vector; Receive the preset feature weight information of the multimodal feature vector; The first multimodal feature vector is obtained by concatenating the multimodal feature vectors according to the preset feature weight information; The fused multimodal features are obtained by using the multimodal encoder to reduce the dimensionality of the first multimodal feature vector.

[0017] The above technical solution introduces a cross-modal attention fusion mechanism and uses chemical modality weight allocation and socio-economic modality dynamic calibration to resolve multi-source data conflicts.

[0018] In a specific implementation scheme, the step of training the model using the fused multimodal features and the dynamic pollution propagation probability map to obtain the source tracing model includes: The model is trained using the fused multimodal features and the dynamic pollution propagation probability map. The loss function for model training includes a regression loss function and a classification loss function, calculated as follows: ; in, Here, N is the regression loss function, N is the number of monitoring points, and MSE / MAE are the weighting coefficients. To predict abundance values ​​for the model, To monitor abundance values ​​in the field; ; in, Let C be the classification loss function, and C be the pollution source category. For category weights, According to the category of pollution source, This represents the class probabilities predicted by the model.

[0019] Through the above technical solution, a Spatiotemporal Graph Neural Network (STGNN) source tracing engine is constructed. From a spatial dimension, GraphSAGE is used to aggregate river network topology relationships; from a temporal dimension, a TCN module is used to capture the impact of sudden changes in human activities before and after the pandemic. Without this, it would be impossible to dynamically quantify the contribution rate of pollution sources and achieve real-time simulation of pollution migration paths. Simultaneously, chemical mode weight allocation and socio-economic mode dynamic calibration are employed to resolve multi-source data conflicts. Compared to traditional models, the source tracing and positioning accuracy of this invention can be controlled at the hundred-meter level.

[0020] Secondly, this application provides a microplastic pollution tracing system, which adopts the following technical solution: The system includes: A microplastics data receiving module is used to receive microplastics data, which includes socioeconomic plastics data, microplastics multimodal characteristic data, and spatial heterogeneity network data. A microplastic data processing and storage module is used to clean, standardize, and spatially align the microplastic data before storing it. The spatiotemporal map generation module is used to construct a spatiotemporal map of microplastic pollution sources. Based on the microplastic data, the spatiotemporal map of microplastic pollution sources is defined and enhanced to obtain an enhanced spatiotemporal map of microplastic pollution sources. The source tracing model preparation module is used to perform spatiotemporal dependency modeling on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map, extract feature vectors from the socioeconomic plastic data and the microplastic multimodal feature data, and fuse them to obtain fused multimodal features; The source tracing model training module is used to train the model using the fused multimodal features and the dynamic pollution propagation probability map to obtain the source tracing model; The traceability module is used to set dynamic thresholds and dynamic risk levels, and to perform microplastic pollution traceability and pollution alarm based on the dynamic thresholds, dynamic risk levels and traceability models.

[0021] Thirdly, this application provides a computer device that adopts the following technical solution: it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and executed as described above in the microplastic pollution tracing method.

[0022] Fourthly, this application provides a computer-readable storage medium, which employs the following technical solution: storing a computer program that can be loaded by a processor and executed as described above for tracing the source of microplastic pollution.

[0023] In summary, this application has the following beneficial technical effects: (1) This application collects multi-source heterogeneous data, including socioeconomic plastics data, microplastics multimodal feature data, and spatial heterogeneous network data, as microplastics feature sources. It constructs a multi-source heterogeneous data fusion architecture, integrating the three types of heterogeneous data: socioeconomic flux, microplastics multimodal features, and spatial network data. Without any one of these, the technology cannot achieve closed-loop analysis of pollution source-migration path-environmental impact. This multi-level, multimodal data fusion and modeling method can overcome the limitations of a single data source and achieve accurate location and dynamic tracking of microplastic pollution sources.

[0024] (2) When acquiring multimodal feature data of microplastics, multimodal feature co-computation of microplastics is performed. The accuracy of microplastic source identification is improved by calculating multimodal feature vectors. It comprehensively considers multiple physical features of microplastics such as shape, color, size and polymer type, and overcomes the limitations of single feature identification. In particular, the introduction of microplastic abundance data (diversity index DFI) can quantify the dispersion of microplastic features and provide more comprehensive feature information for subsequent source tracing analysis.

[0025] (3) The spatial heterogeneity network construction module integrates various spatial data (such as river networks, ocean current paths, land use types, etc.) to construct a comprehensive spatial network map of microplastic pollution sources. This method considers the heterogeneity of environmental factors and can more accurately reflect the propagation characteristics of microplastics in different spatial environments. A spatiotemporal graph neural network (STGNN) source tracing engine is constructed. From the spatial dimension, GraphSAGE is used to aggregate river network topology relationships; from the temporal dimension, the TCN module is used to capture the impact of sudden changes in human activities before and after the epidemic. Without this, it is impossible to dynamically quantify the contribution rate of pollution sources, realize real-time simulation of pollution migration paths, and improve the accuracy of source tracing.

[0026] (4) Introduce a cross-modal attention fusion mechanism and use chemical modality weight allocation and socio-economic modality dynamic calibration to resolve multi-source data conflicts. Attached Figure Description

[0027] Figure 1 This is a flowchart of a microplastic pollution tracing method in an embodiment of this application.

[0028] Figure 2 This is the system architecture of this application.

[0029] Figure 3 This is the pollution propagation probability map output by STGNN.

[0030] Figure 4 This is a structural block diagram of a microplastic pollution tracing method according to an embodiment of this application.

[0031] Figure reference numerals: 401, Microplastics data receiving module; 402, Microplastics data processing and storage module; 403, Spatiotemporal diagram generation module; 404, Traceability model preparation module; 405, Traceability model training module; 406, Traceability module. Detailed Implementation

[0032] The following is in conjunction with the appendix Figures 1-4 This application will be described in further detail.

[0033] This application discloses a method for tracing the source of microplastic pollution, which is used to improve the accuracy of tracing the source of microplastic pollution.

[0034] Emerging pollutants are newly discovered pollutants that pose significant risks to the ecological environment and human health. Microplastics are one of the major emerging pollutants, with complex and spatially heterogeneous sources, and are closely related to human activities within the region. Conducting regional-scale source apportionment is fundamental to the risk management of microplastics in water bodies. Currently, there is a lack of source apportionment studies for typical microplastics, and existing pollution source identification relies heavily on the professional knowledge and practical experience of relevant personnel, which is highly subjective and hinders the in-depth development of microplastic risk management.

[0035] To address the above technical bottlenecks, it is essential to improve the source apportionment technology system for microplastics in water bodies and effectively carry out source tracing work. Only in this way can we clarify our thinking, take targeted measures, and further develop a technology system suitable for the treatment of microplastics in water bodies, thereby achieving the goal of "control".

[0036] Currently, the source tracing of microplastic pollution in water bodies mainly relies on three types of technical approaches.

[0037] The first category is analytical methods based on the physicochemical characteristics of microplastics: by identifying the polymer type, morphology (size, shape, color), surface aging traces (such as cracks, biofilm adhesion), and additive components (such as plasticizers, flame retardants) of the particles, the possible source category (such as clothing fibers, packaging fragments, industrial raw materials) can be inferred. This method relies on techniques such as Fourier transform infrared spectroscopy (FTIR), Raman spectroscopy, and scanning electron microscopy (SEM).

[0038] The second approach employs a chemical tracer association strategy: detecting source markers that coexist with microplastics, such as fluorescent whitening agents in detergents, drug residues or artificial sweeteners in wastewater, to establish spatial correlations between pollutants and domestic sewage / industrial emissions.

[0039] The third category is spatial statistics and model simulation: using Geographic Information Systems (GIS) to integrate pollution distribution data with the spatial location of potential sources (such as wastewater treatment plants, docks, and farmland), and calculating source contribution rates through models such as Principal Component Analysis (PCA) and Positive Matrix Factorization (PMF). Some studies attempt to combine microbial community characteristics (such as plastic ring biomarkers) to assist in source tracing.

[0040] However, existing technologies still face significant bottlenecks that restrict the accuracy of source tracing and its widespread application. These bottlenecks are threefold: First, insufficient specificity of characteristic sources, as environmental degradation of microplastics leads to blurred physicochemical characteristics (e.g., polyethylene fragments cannot distinguish the source of agricultural film / packaging bags); second, lack of spatial dynamic correlation, as traditional models (PCA / PMF) neglect the co-evolution of pollution source spatial distribution density and microplastic communities; and finally, deficiencies in the analysis of mixed pollution, lacking a standard source component database, and model parameters relying on subjective experience.

[0041] The aforementioned shortcomings directly lead to two limitations in current source tracing research: first, high-precision characteristic analysis at the laboratory scale cannot be extrapolated to watershed management (e.g., only determining that "fibers originate from the washing process," but failing to pinpoint specific discharge areas); second, the source contribution rates output by spatial models lack credibility due to their detachment from the true biochemical properties of microplastic communities (e.g., attributing pollution in a certain area to "industrial activities" without supporting characteristic evidence). Especially for large, shallow lakes, with their strong water system connectivity and significant pollution source polymorphism (intertwined with agriculture, aquaculture, urban clusters, and shipping), traditional methods struggle to achieve precise spatial location of pollution sources. Therefore, this invention integrates microplastic biogeochemical characteristics, socioeconomic data, and spatial network modeling to develop an intelligent source tracing system suitable for precise source apportionment of microplastic pollution in water bodies.

[0042] Therefore, this application proposes a method for tracing the source of microplastic pollution, which improves the accuracy of tracing the source of microplastic pollution.

[0043] like Figure 1 As shown, the method includes: S10, Receive microplastic data, which includes socioeconomic plastic data, multimodal microplastic characteristic data, and spatial heterogeneity network data.

[0044] Specifically, such as Figure 2 As shown, this application adopts a three-layer progressive system architecture to realize microplastic pollution source tracing analysis: The first layer, the data acquisition layer, integrates multi-source heterogeneous data, including socio-economic flux data, microplastic multimodal characteristic data, and spatial heterogeneous network data.

[0045] The first step is to collect microplastic data from three sources: socioeconomic plastic data, microplastic multimodal characteristic data, and spatial heterogeneous network data.

[0046] S20, after cleaning, standardizing and spatially aligning the microplastic data, stores the microplastic data.

[0047] Specifically, microplastic data is stored after being processed by cleaning (e.g., removing outliers and filling in missing values), standardization (e.g., unifying units), and spatial alignment (e.g., mapping data to a unified geographic coordinate system).

[0048] S30. Construct a spatiotemporal map of microplastic pollution sources. Based on microplastic data, define and enhance the spatiotemporal map of microplastic pollution sources to obtain an enhanced spatiotemporal map of microplastic pollution sources.

[0049] Specifically, such as Figure 2 As shown, the second-layer intelligent computing engine layer models the spatiotemporal dynamic relationship through a spatiotemporal graph neural network (STGNN) to obtain a spatiotemporal graph of microplastic pollution sources. Based on the microplastic data, the spatiotemporal graph of microplastic pollution sources is defined and enhanced to obtain an enhanced spatiotemporal graph of microplastic pollution sources.

[0050] S40, a dynamic pollution propagation probability map is obtained by performing spatiotemporal dependency modeling on the enhanced spatiotemporal map of microplastic pollution sources, extracting feature vectors from socioeconomic plastic data and microplastic multimodal feature data and fusing them to obtain fused multimodal features.

[0051] Specifically, spatiotemporal dependency modeling of the enhanced microplastic pollution source spatiotemporal map yields a dynamic pollution propagation probability map, such as... Figure 2 As shown, the second layer uses a cross-modal feature fusion module to perform feature fusion on socio-economic plastics data and microplastics multimodal feature data to obtain fused multimodal features, thereby achieving deep correlation of multi-source data.

[0052] S50 uses a model training method that integrates multimodal features and dynamic pollution propagation probability maps to obtain a source tracing model.

[0053] Specifically, the source tracing model is a multimodal fusion system with STGNN as its core. It uses multimodal features and dynamic pollution propagation probability maps as inputs to train the model and obtain a trained source tracing model.

[0054] S60 sets dynamic thresholds and dynamic risk levels, and performs microplastic pollution source tracing and pollution alarm based on the dynamic thresholds, dynamic risk levels, and source tracing models.

[0055] Specifically, dynamic threshold setting: An adaptive quantile adjustment mechanism is introduced to set contribution rate thresholds based on historical data quantiles (e.g., contribution rate of a source > historical 95th quantile → orange alert). Dynamic risk level classification: Contribution rate + propagation speed → generating red / orange / yellow three-level warning signals.

[0056] Microplastic pollution source tracing and pollution alarms are conducted based on dynamic thresholds, dynamic risk levels, and source tracing models. For example... Figure 2 As shown, the third-layer visualization decision-making layer outputs a dynamic source tracing heat map that intuitively displays the pollution distribution, and simultaneously generates a source contribution rate early warning system to support risk management decisions.

[0057] The dynamic source tracing heatmap generation includes: 1. Spatial rendering: Based on the pollution propagation probability output by STGNN, the heatmap gradient (dark red = high pollution) is dynamically displayed on the map.

[0058] 2. Time axis control: Supports sliding the time axis to view historical / real-time / predictive pollution distribution (such as diffusion simulation for the next 72 hours).

[0059] Contribution Rate Early Warning System: 1. Visualization panel: Heat map, contribution rate chart, and early warning information are synchronized and linked, supporting filtering by region / time / pollution source type; real-time output of early warning list sorted by high-risk sources.

[0060] 2. Decision support function: Displays hotspots of pollution source contribution rates and relevant responsible units or departments, allows for telephone contact to suggest investigations, and organizes rectification after confirming the cause of pollution incidents.

[0061] This study collects multi-source heterogeneous data, including socioeconomic plastics data, microplastic multimodal characteristic data, and spatial heterogeneous network data, as microplastic feature sources. A multi-source heterogeneous data fusion architecture is constructed to integrate these three types of heterogeneous data: socioeconomic flux, microplastic multimodal characteristics, and spatial network data. Without any one of these, the technology cannot achieve a closed-loop analysis of pollution source-migration path-environmental impact. This multi-level, multimodal data fusion and modeling method overcomes the limitations of a single data source, enabling precise location and dynamic tracking of microplastic pollution sources.

[0062] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the method is based on socioeconomic statistical yearbooks, and the step of receiving microplastic data can be specifically performed as follows: First, analyze the socio-economic statistical yearbook to obtain socio-economic plastic data. Specifically, the statistics bureau obtains socio-economic statistical yearbook data, statistical data on industry, agriculture, logistics, etc., and calculates the amount of plastic loss in each industry based on (such as industrial output value, express delivery volume, tire loss rate), and quantifies the amount of plastic loss in each city in river and lake basins (such as express packaging loss rate = annual express delivery volume × unit packaging loss coefficient). Then, plastic samples were collected from river and lake basins. Specifically, plastic samples were collected from sampling points located in river and lake basins. Then, the physical characteristics of microplastics are obtained by identifying the sampled plastic samples, and the chemical characteristics of microplastics are obtained by analyzing the sampled plastic samples. Specifically, the physical characteristics of microplastics include physical characteristics such as shape, color, size, and polymer type. The polymer type is identified by micro-FTIR spectroscopy, and the shape, color, and size of microplastics are measured by measuring the sampled plastic samples. In the calculation module, the source of microplastics is determined by analyzing the polymer type (PE / PP / PET ratio). Then, the physical and chemical characteristics of microplastics are fused across modalities to obtain microplastic fused characteristics. Microplastic abundance data is then calculated based on these fused characteristics. Specifically, the microplastic abundance data is represented by the diversity index (DFI). The specific steps are as follows: Step 1: Feature Embedding → Extracting high-dimensional feature vectors using CNN: Input feature matrix:

[0063] CNN Feature Extraction:

[0064] illustrate: • Original feature vector (physical measurement value of microscopic infrared spectroscopy) • A: Feature extraction function of convolutional neural network • : High-dimensional embedded feature vector (k is the embedding dimension) Step 2: Cross-modal fusion → Attention mechanism weighting: Global features:

[0065] Attention score:

[0066] Normalized weights:

[0067] Optimization notes: 1. Enhance representation capabilities by adding a nonlinear activation function σ (such as ReLU).

[0068] 2. Explicitly define global features The characteristic mean is denoted as .

[0069] 3. Introduce a bias term b to improve the model's flexibility.

[0070] 4. Use the vector concatenation operator [⋅∣∣⋅] to replace the fuzzy symbol "∣∣".

[0071] Step 3: Calculate the Diversity Index (DFI):

[0072] Information entropy calculation:

[0073] Parameter definition: : Feature mode set; K: Feature V i Number of categories after discretization (requires binning preprocessing); C k The set of samples within the k-th bin interval; N: Total number of sampling points in the watershed.

[0074] Then, by combining the physical characteristics, chemical characteristics, and abundance data of microplastics, multimodal characteristic data of microplastics are obtained; specifically, ; Finally, spatially heterogeneous network data is received, which includes river networks, ocean current paths, meteorological data, and monitoring station locations. Specifically, GIS tools (Gaode Map API + POIKit crawler) are used to crawl POI data such as river networks, ocean current paths, and land use types, and integrate meteorological data (wind direction, precipitation) and the topological relationships of monitoring station locations.

[0075] This study collects multi-source heterogeneous data, including socioeconomic plastics data, microplastic multimodal characteristic data, and spatial heterogeneous network data, as microplastic feature sources. A multi-source heterogeneous data fusion architecture is constructed to integrate these three types of heterogeneous data: socioeconomic flux, microplastic multimodal characteristics, and spatial network data. Without any one of these, the technology cannot achieve a closed-loop analysis of pollution source-migration path-environmental impact. This multi-level, multimodal data fusion and modeling method overcomes the limitations of a single data source, enabling precise location and dynamic tracking of microplastic pollution sources.

[0076] Furthermore, when acquiring multimodal microplastic feature data, collaborative computation of microplastic features is performed. This multimodal feature vector calculation improves the accuracy of microplastic source identification, comprehensively considering multiple physical characteristics such as shape, color, size, and polymer type of microplastics, overcoming the limitations of single-feature identification. In particular, the introduction of microplastic abundance data (DFI) quantifies the dispersion of microplastic features, providing more comprehensive feature information for subsequent source tracing analysis.

[0077] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the method is based on a unified geographic coordinate system. The step of cleaning, standardizing, and spatially aligning the microplastic data before storing it can be specifically performed as follows: First, outliers in the microplastic data are removed, and missing values ​​are filled using Kriging interpolation. Specifically, data cleaning includes removing outliers (such as abnormal abundance caused by extreme weather) and filling missing values ​​(using Kriging interpolation). Then, the units of the microplastic data were standardized and the socioeconomic plastic data were normalized. Specifically, data standardization includes unifying units (such as converting microplastic abundance to "particles / m³") and normalizing socioeconomic indicators. Next, the socioeconomic plastics data, microplastics multimodal characteristic data, and spatial heterogeneity network data are mapped to a unified geographic coordinate system to construct a river network spatial topology map. Specifically, spatial alignment of the data includes mapping the data to a unified geographic coordinate system (such as WGS-84) to construct a spatial topology network. Finally, a spatiotemporal database is established, which stores socioeconomic plastic data, microplastic multimodal characteristic data, and spatially heterogeneous network data in a layered manner according to timestamp, spatial location, and data type. Specifically, a spatiotemporal database (such as PostGIS) is established and stored in a layered manner according to timestamp, spatial location, and data type.

[0078] The data repository is the core foundation of the microplastic pollution tracing system, providing a data engine for the construction of microplastic pollution tracing.

[0079] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the method constructs a spatiotemporal map of microplastic pollution sources based on published ecological and environmental monitoring data and the STGNN model. The step of defining and enhancing the spatiotemporal map of microplastic pollution sources based on microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources can be specifically performed as follows: Based on the STGNN model, a spatiotemporal graph of microplastic pollution sources is constructed from the river network spatial topology map and spatially heterogeneous network data. The spatiotemporal graph of microplastic pollution sources includes nodes and edges. Specifically, STGNN is a deep learning model that combines graph neural networks and time series learning methods to capture the spatial and temporal dependencies in data. Based on the STGNN model, a spatiotemporal graph of microplastic pollution sources is constructed from the river network spatial topology map and spatially heterogeneous network data. The spatiotemporal graph of microplastic pollution sources includes nodes and edges. Pollution sources can be identified based on spatial heterogeneous network data. Specifically, spatial heterogeneous network data includes capturing POI data such as river networks, ocean current paths, and land use types, which can help identify pollution sources (e.g., factories). Nodes are defined based on the location of pollution sources and monitoring stations, and edges are defined based on the direction of water flow in the river network and ocean current paths. Specifically, the graph structure is defined as follows: node = monitoring point / area, edge = spatial connection (such as river flow direction, ocean current path) or similarity (similarity of land use type). Abundance data, physical characteristics, and socioeconomic data on microplastics are appended as node features to the nodes. Water flow velocity and wind direction intensity are obtained by analyzing publicly available ecological and environmental monitoring data. Weights are then applied to the edges based on water flow velocity and wind direction intensity to generate a spatiotemporal map of enhanced microplastic pollution sources. Specifically, the spatiotemporal attributes are: Node Features = (Microplastic Abundance, Polymer Type, Socioeconomic Indicators), Edge Weights = Dynamic Interaction Intensity (e.g., Water Flow Velocity, Wind Direction Intensity). Water flow velocity and wind direction intensity are derived from publicly available ecological and environmental monitoring data.

[0080] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the step of performing spatiotemporal dependency modeling on the enhanced spatiotemporal map of microplastic pollution sources to obtain a dynamic pollution propagation probability map can be specifically performed as follows: The river network topology of the spatiotemporal map of microplastic pollution sources is enhanced by using GraphSAGE aggregation. Specifically, GraphSAGE is used as the core spatial aggregator. By using GraphSAGE as the core spatial aggregator, multi-layer spatiotemporal convolution is used to capture cross-node propagation patterns and capture spatial correlations between nodes (such as the impact of upstream pollution on downstream). A dynamic pollution propagation probability map is obtained by learning the temporal changes in abundance data using temporal convolution methods. Specifically, temporal convolution (TCN) is used to learn the temporal changes in microplastic abundance (such as a surge in abundance during the rainy season). The final output is a dynamic pollution propagation probability map (predicting future diffusion paths). Figure 3 This is the pollution propagation probability map output by STGNN.

[0081] The spatial heterogeneity network construction module integrates various spatial data (such as river networks, ocean current paths, and land use types) to construct a comprehensive spatial network map of microplastic pollution sources. This method considers the heterogeneity of environmental factors and can more accurately reflect the propagation characteristics of microplastics in different spatial environments. A spatiotemporal graph neural network (STGNN) source tracing engine is constructed. From a spatial dimension, GraphSAGE is used to aggregate river network topology relationships; from a temporal dimension, the TCN module captures the impact of sudden changes in human activities before and after the pandemic. Without this engine, it is impossible to dynamically quantify the contribution rate of pollution sources, achieve real-time simulation of pollution migration paths, and improve source tracing accuracy.

[0082] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the method is based on a multimodal encoder. The step of extracting feature vectors from socioeconomic plastic data and microplastic multimodal feature data and fusing them to obtain fused multimodal features can be specifically performed as follows: Multimodal feature vectors are obtained by extracting physical features, chemical features, and socioeconomic plastic data. Specifically, the embedding vectors of microphysical features, chemical features, and socioeconomic features are extracted respectively, which are the multimodal feature vectors. The system receives preset feature weight information from multimodal feature vectors, concatenates the multimodal feature vectors according to the preset feature weight information to obtain the first multimodal feature vector, and uses a multimodal encoder to reduce the dimensionality of the first multimodal feature vector to obtain the fused multimodal features. Specifically, it calculates the weight of each modality (e.g., industrial areas focus on socio-economic data, and estuaries focus on hydrological data), concatenates the multimodal features, and then uses an autoencoder to reduce the dimensionality, retaining key information to obtain the fused multimodal features.

[0083] A cross-modal attention fusion mechanism is introduced, and chemical modality weight allocation and socio-economic modality dynamic calibration are used to resolve multi-source data conflicts.

[0084] In one embodiment, to improve the accuracy of microplastic pollution source tracing, the step of training a source tracing model using a fusion of multimodal features and a dynamic pollution propagation probability map can be specifically performed as follows: The model is trained using a fusion of multimodal features and a dynamic pollution propagation probability map. The loss function for model training includes a regression loss function and a classification loss function, calculated as follows: ; in, Here, N is the regression loss function, N is the number of monitoring points, and MSE / MAE are the weighting coefficients. To predict abundance values ​​for the model, To monitor abundance values ​​in the field; ; in, Let C be the classification loss function, and C be the pollution source category. For category weights, According to the category of pollution source, This represents the class probabilities predicted by the model.

[0085] Specifically, the input to the source tracing model training is a fusion of multimodal features and a dynamic pollution propagation probability map, and the output is the pollution source contribution rate (e.g., an industrial zone contributing 30% and a wastewater treatment plant contributing 45%). The loss function for model training mainly includes a combination of regression loss (predicted abundance vs. actual abundance) and classification loss (source type identification), calculated as follows: ; in, Here, N is the regression loss function, N is the number of monitoring points, and MSE / MAE are the weighting coefficients. To predict abundance values ​​for the model, To monitor abundance values ​​in the field; ; in, Let C be the classification loss function, and C be the pollution source category. For category weights, According to the category of pollution source, This represents the class probabilities predicted by the model.

[0086] A spatiotemporal graph neural network (STGNN) source tracing engine is constructed. Spatially, it utilizes GraphSAGE to aggregate river network topology relationships; temporally, it uses a TCN module to capture the impact of sudden changes in human activities before and after the pandemic. Without this engine, it is impossible to dynamically quantify the contribution rate of pollution sources and achieve real-time simulation of pollution migration paths. Simultaneously, chemical mode weight allocation and dynamic calibration of socio-economic modes are employed to resolve multi-source data conflicts. Compared to traditional models, the source tracing and positioning accuracy of this invention can be controlled within the hundreds of meters.

[0087] Based on the above method, this application also discloses a microplastic pollution tracing system. For example... Figure 4 The system includes the following modules: The microplastic data receiving module 401 is used to receive microplastic data, which includes socioeconomic plastic data, microplastic multimodal characteristic data, and spatial heterogeneity network data. Microplastic data processing and storage module 402 is used to store microplastic data after cleaning, standardizing and spatially aligning it. The spatiotemporal map generation module 403 is used to construct a spatiotemporal map of microplastic pollution sources. Based on microplastic data, the spatiotemporal map of microplastic pollution sources is defined and enhanced to obtain an enhanced spatiotemporal map of microplastic pollution sources. The source tracing model preparation module 404 is used to perform spatiotemporal dependency modeling on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map, extract feature vectors from socioeconomic plastic data and microplastic multimodal feature data and fuse them to obtain fused multimodal features; The source tracing model training module 405 is used to train the model using fused multimodal features and dynamic pollution propagation probability maps to obtain the source tracing model; The traceability module 406 is used to set dynamic thresholds and dynamic risk levels, and to perform microplastic pollution traceability and pollution alarm based on the dynamic thresholds, dynamic risk levels and traceability models.

[0088] In one embodiment, the microplastic data receiving module 401 is specifically used to parse socioeconomic statistical yearbooks to obtain socioeconomic plastic data; sample plastic samples from river and lake basins; identify the physical characteristics of microplastics from the sampled plastic samples and analyze the sampled plastic samples to obtain microplastic chemical characteristics; fuse the physical and chemical characteristics of microplastics across modes to obtain microplastic fusion characteristics, and calculate microplastic abundance data based on the microplastic fusion characteristics; combine the physical, chemical, and abundance data of microplastics to obtain multimodal microplastic characteristic data; and receive spatial heterogeneous network data, which includes river networks, ocean current paths, meteorological data, and monitoring station locations.

[0089] In one embodiment, the microplastic data processing and storage module 402 is specifically used to remove outliers from the microplastic data, fill in missing values ​​in the microplastic data using Kriging spatial interpolation, unify the units of the microplastic data, and normalize the socioeconomic plastic data; map the socioeconomic plastic data, microplastic multimodal feature data, and spatial heterogeneous network data to a unified geographic coordinate system to construct a river network spatial topology map; and establish a spatiotemporal database, storing the socioeconomic plastic data, microplastic multimodal feature data, and spatial heterogeneous network data in the spatiotemporal database according to timestamps, spatial locations, and data types.

[0090] In one embodiment, the spatiotemporal graph generation module 403 is specifically used to construct a spatiotemporal graph of microplastic pollution sources based on the STGNN model, using the river network spatial topology map and spatial heterogeneous network data. The spatiotemporal graph of microplastic pollution sources includes nodes and edges. The module determines pollution sources based on the spatial heterogeneous network data. Nodes are defined based on the pollution sources and monitoring station locations. Edges are defined based on the water flow direction of the river network and ocean current paths. Abundance data, physical characteristics, and socioeconomic plastic data are added to the nodes as node features. The module analyzes the ecological environment monitoring data to obtain water flow velocity and wind direction intensity, and applies weights to the edges based on the water flow velocity and wind direction intensity to obtain a reinforced spatiotemporal graph of microplastic pollution sources.

[0091] In one embodiment, the source tracing model preparation module 404 is specifically used to aggregate and enhance the river network topology of the spatiotemporal map of microplastic pollution sources using GraphSAGE; and to learn the temporal changes of abundance data using a temporal convolution method to obtain a dynamic pollution propagation probability map.

[0092] In one embodiment, the traceability model preparation module 404 is specifically used to extract physical features, chemical features, and socioeconomic plastic data to obtain a multimodal feature vector; receive preset feature weight information of the multimodal feature vector; concatenate the multimodal feature vector according to the preset feature weight information to obtain a first multimodal feature vector; and use a multimodal encoder to reduce the dimensionality of the first multimodal feature vector to obtain a fused multimodal feature.

[0093] In one embodiment, the source tracing model training module 405 is specifically used to train the model using fused multimodal features and a dynamic pollution propagation probability map. The loss function for model training includes a regression loss function and a classification loss function, calculated as follows: ; in, Here, N is the regression loss function, N is the number of monitoring points, and MSE / MAE are the weighting coefficients. To predict abundance values ​​for the model, To monitor abundance values ​​in the field; ; in, Let C be the classification loss function, and C be the pollution source category. For category weights, According to the category of pollution source, This represents the class probabilities predicted by the model.

[0094] This application also discloses a computer device.

[0095] Specifically, the computer device includes a memory and a processor, with the memory storing a computer program that can be loaded by the processor and executed as described above for tracing the source of microplastic pollution.

[0096] This application also discloses a computer-readable storage medium.

[0097] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed as described above in a microplastic pollution tracing method. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for tracing the source of microplastic pollution, characterized in that, The method includes: Receive microplastic data, which includes socioeconomic plastic data, microplastic multimodal characteristic data, and spatial heterogeneous network data; The microplastic data is stored after being cleaned, standardized, and spatially aligned. A spatiotemporal map of microplastic pollution sources is constructed, and the spatiotemporal map of microplastic pollution sources is defined and enhanced based on the microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources. Spatiotemporal dependency modeling is performed on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map. Feature vectors of the socioeconomic plastic data and the microplastic multimodal feature data are extracted and fused to obtain fused multimodal features. The source tracing model is obtained by training the model using the fused multimodal features and the dynamic pollution propagation probability map; Set dynamic thresholds and dynamic risk levels, and perform microplastic pollution source tracing and pollution alarm based on the dynamic thresholds, dynamic risk levels, and source tracing model.

2. The method according to claim 1, characterized in that, The method is based on a socio-economic statistical yearbook, and the received microplastic data includes: The socioeconomic data on plastics were obtained by analyzing the socioeconomic statistical yearbook. Plastic samples were obtained from river and lake basins. The physical characteristics of microplastics were obtained by identifying the sampled plastic samples, and the chemical characteristics of microplastics were obtained by analyzing the sampled plastic samples. The physical and chemical characteristics of microplastics are fused across modes to obtain microplastic fusion characteristics, and microplastic abundance data are calculated based on the microplastic fusion characteristics. By combining the physical characteristics, chemical characteristics, and abundance data of the microplastics, multimodal characteristic data of microplastics is obtained. Receive spatially heterogeneous network data, which includes river networks, ocean current paths, meteorological data, and monitoring station locations.

3. The method according to claim 2, characterized in that, The method is based on a unified geographic coordinate system, and the process of cleaning, standardizing, and spatially aligning the microplastic data before storing the microplastic data includes: Outliers in the microplastic data are removed, and missing values ​​in the microplastic data are filled using Kriging space interpolation. Units were standardized for the microplastic data, and the socioeconomic plastic data were normalized. The socioeconomic plastics data, the microplastics multimodal feature data, and the spatial heterogeneous network data are mapped to the unified geographic coordinate system to construct a river network spatial topology map; A spatiotemporal database is established, and the socioeconomic plastic data, the microplastic multimodal feature data, and the spatial heterogeneous network data are stored in the spatiotemporal database according to timestamps, spatial locations, and data types.

4. The method according to claim 3, characterized in that, The method is based on ecological and environmental monitoring and published data and the STGNN model. The step of constructing a spatiotemporal map of microplastic pollution sources, and defining and enhancing the spatiotemporal map of microplastic pollution sources based on the microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources, includes: Based on the STGNN model, a spatiotemporal graph of microplastic pollution sources is constructed according to the river network spatial topology and the spatial heterogeneous network data. The spatiotemporal graph of microplastic pollution sources includes nodes and edges. The pollution source was determined based on the spatially heterogeneous network data; The node is defined based on the pollution source and the location of the monitoring station; The edges are defined according to the direction of water flow in the river network and the ocean current path; The abundance data, the physical characteristics, and the socioeconomic plastics data are attached to the node as node features; The water flow velocity and wind direction intensity are obtained by analyzing the ecological environment monitoring data. The edges are weighted according to the water flow velocity and wind direction intensity to obtain a spatiotemporal map of enhanced microplastic pollution sources.

5. The method according to claim 4, characterized in that, The process of performing spatiotemporal dependency modeling on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map includes: The river network topology was aggregated using GraphSAGE to enhance the spatiotemporal map of microplastic pollution sources. A dynamic pollution propagation probability map is obtained by learning the temporal changes of the abundance data using a temporal convolution method.

6. The method according to claim 5, characterized in that, The method is based on a multimodal encoder, and the step of extracting feature vectors from the socioeconomic plastics data and the microplastics multimodal feature data and fusing them to obtain fused multimodal features includes: The physical features, chemical features, and socioeconomic plastic data are extracted to obtain a multimodal feature vector; Receive the preset feature weight information of the multimodal feature vector; The first multimodal feature vector is obtained by concatenating the multimodal feature vectors according to the preset feature weight information; The fused multimodal features are obtained by using the multimodal encoder to reduce the dimensionality of the first multimodal feature vector.

7. The method according to claim 6, characterized in that, The process of training the source tracing model using the fused multimodal features and the dynamic pollution propagation probability map includes: The model is trained using the fused multimodal features and the dynamic pollution propagation probability map. The loss function for model training includes a regression loss function and a classification loss function, calculated as follows: ； in, Here, N is the regression loss function, N is the number of monitoring points, and MSE / MAE are the weighting coefficients. To predict abundance values ​​for the model, To monitor abundance values ​​in the field; ； in, Let C be the classification loss function, and C be the pollution source category. For category weights, According to the category of pollution source, This represents the class probabilities predicted by the model.

8. A microplastic pollution tracing system, characterized in that, The system includes: The microplastic data receiving module (401) is used to receive microplastic data, which includes socioeconomic plastic data, microplastic multimodal characteristic data, and spatial heterogeneity network data. A microplastic data processing and storage module (402) is used to clean, standardize, and spatially align the microplastic data before storing the microplastic data. The spatiotemporal map generation module (403) is used to construct a spatiotemporal map of microplastic pollution sources and to define and enhance the spatiotemporal map of microplastic pollution sources based on the microplastic data to obtain an enhanced spatiotemporal map of microplastic pollution sources. The source tracing model preparation module (404) is used to perform spatiotemporal dependency modeling on the enhanced microplastic pollution source spatiotemporal map to obtain a dynamic pollution propagation probability map, extract the feature vectors of the socioeconomic plastic data and the microplastic multimodal feature data and fuse them to obtain fused multimodal features; The source tracing model training module (405) is used to train the model using the fused multimodal features and the dynamic pollution propagation probability map to obtain the source tracing model; The traceability module (406) is used to set dynamic thresholds and dynamic risk levels, and to perform microplastic pollution traceability and pollution alarm based on the dynamic thresholds, the dynamic risk levels and the traceability model.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed according to any one of claims 1 to 7.

Images