Intelligent traceability method and system for micro-plastic emission in offshore neighborhood

Abstract

The invention relates to the technical field of micro-plastic traceability, in particular to an intelligent traceability method and system for micro-plastic emission in an offshore temporary region. The method comprises the following steps: acquiring environmental dynamic element information and micro-plastic observation information which influence micro-plastic transportation and distribution in a target offshore neighborhood, and constructing a multi-source data set; according to the multi-source data set, training an artificial intelligence model by taking at least one offshore specific physical mechanism represented by the environmental dynamics elements and pollutant spatial and temporal distribution characteristics represented by the micro-plastic observation information as constraint conditions by taking the offshore specific physical mechanism to obtain an intelligent traceability model; and inputting the multi-source data set into a pre-constructed intelligent traceability model, and outputting a traceability result calculated by the intelligent traceability model. According to the method, data availability differences of different offshore environments are flexibly dealt with, a traceability conclusion with high interpretability is obtained, and the spatial position of an emission source is accurately identified.

Description

Technical Field

[0001] This application relates to the field of microplastic traceability technology, and in particular to an intelligent traceability method and system for microplastic emissions in near-shore areas. Background Technology

[0002] Microplastic pollution has become a global coastal environmental problem. Most existing source tracing methods have the following limitations: Relying on offline sampling and laboratory analysis is time-consuming and labor-intensive, and cannot achieve rapid response and real-time monitoring.

[0003] The models are oversimplified: Most numerical or statistical models ignore the high-frequency and complex dynamic processes in the nearshore waters, resulting in rough source tracing results and high uncertainty.

[0004] The problem of "scenario generalization": Existing methods fail to closely fit the unique characteristics of strong land-sea interaction and dense human activities in nearshore areas, and the scenarios they target lack specificity and accuracy.

[0005] Therefore, there is an urgent need for an innovative solution that can deeply integrate the unique physical mechanisms of nearshore waters to achieve accurate and efficient traceability. Summary of the Invention

[0006] This application provides an intelligent source tracing method and system for microplastic emissions in nearshore areas to solve the above-mentioned problems.

[0007] In a first aspect, this application provides an intelligent source tracing method for microplastic emissions in nearshore areas. The method includes: acquiring environmental dynamics information and microplastic observation information affecting microplastic transport and distribution within a target nearshore area, and constructing a multi-source dataset; based on the multi-source dataset, training an artificial intelligence model with at least one nearshore-specific physical mechanism characterized by the environmental dynamics and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information, using the nearshore-specific physical mechanism as a constraint, to obtain an intelligent source tracing model, enabling the intelligent source tracing model to learn the microplastic transport laws driven by the physical mechanism; inputting the multi-source dataset into the pre-constructed intelligent source tracing model, and outputting the source tracing result calculated by the intelligent source tracing model, wherein the source tracing result includes at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources.

[0008] Through the above technical solutions, and by deeply integrating physical mechanism constraints with artificial intelligence models, three core advantages are achieved: First, improved source tracing accuracy, the model reduces overfitting to sparse data under the guidance of physical laws, and more accurately identifies the spatial location of emission sources; second, enhanced interpretability of results, the quantitative output of source tracing probability maps, transport paths and contribution rates provides intuitive basis for management decisions, and the correlation of physical evidence improves the transparency and verifiability of conclusions; third, strong adaptability, the fusion of multi-source data and alternative implementation methods enable it to flexibly cope with the differences in data availability in different nearshore environments.

[0009] Optionally, the environmental dynamics information includes periodic hydrological cycle information, water body characteristic gradient information, and wave energy information. The construction of the periodic hydrological cycle information includes: analyzing the reciprocating pumping effect and net transport direction of pollutants caused by tidal asymmetry based on tidal harmonic constant and tidal field data, and constructing the periodic hydrological cycle information. The construction of the water body characteristic gradient information includes: obtaining the spatial distribution of salinity and temperature through satellite remote sensing or on-site measurements, thereby identifying the spatial range and frontal structure of estuarine plumes, and constructing the water body characteristic gradient information. The spatial range of the estuarine plume is a fan-shaped region extending from the estuary to the open sea. The acquisition of wave energy information includes: obtaining effective wave height and wave period data through wave observation stations or wave models, and constructing the wave energy information for assessing the potential intensity of sediment resuspension.

[0010] Optionally, the microplastic observation information includes a spatiotemporal sequence of concentration and a chemical composition fingerprint. The spatiotemporal sequence of concentration is obtained by deploying an automatic monitoring buoy network or conducting periodic cruise surveys to capture the dynamic changes in microplastic distribution. The chemical composition fingerprint is obtained by performing spectral analysis on samples to obtain their main polymer composition, which is used to establish the chemical correlation between pollutants at the observation point and potential emission sources. The chemical correlation is then introduced as a priori weight into the source contribution rate calculation of the intelligent source tracing model.

[0011] Optionally, the nearshore-specific physical mechanism includes at least one of the following: tidal pumping effect, plume aggregation effect, and wave resuspension effect. The training of the artificial intelligence model using the nearshore-specific physical mechanism as a constraint includes: the tidal pumping effect refers to the phenomenon where the net transport of pollutants is not zero during the tidal cycle due to tidal asymmetry; the plume aggregation effect refers to the accumulation of microplastics near the salinity front due to differences in water density and circulation structure; the wave resuspension effect refers to the process where historical microplastics deposited in sediment are resuspended and re-enter the water body under the influence of wave energy; for the tidal pumping effect, the residual flow field derived based on the harmonic constant is added as a physical constraint term to the loss function, forcing the model's output transport law to be consistent with the net transport direction; for the plume aggregation effect, salinity gradient data is converted into a spatial attention weight matrix, causing the model to prioritize calculating the pollutant source probability within the plume's influence range; for the wave resuspension effect, an effective wave height threshold is set to activate the resuspension source contribution calculation module, and a positive correlation is established between the contribution weight and wave energy.

[0012] Optionally, the artificial intelligence model adopts a physical information neural network architecture or a spatiotemporal graph neural network architecture. When adopting the physical information neural network architecture, the residual of the control equation for nearshore pollutant transport is used as a physical constraint loss term, which, together with the data fitting loss term, constitutes a multi-objective loss function for optimization. When adopting the spatiotemporal graph neural network architecture, the monitoring stations are constructed as graph nodes, and the environmental dynamics elements and the microplastic observation information are used as node and edge features, respectively, while simultaneously modeling spatial correlation and temporal dynamics. Both architectures optimize data fitting and consistency with physical laws through multi-task learning, ensuring that the model learns from observation data while adhering to physical constraints.

[0013] Optionally, the specific implementation of the physical information neural network architecture includes: establishing a nearshore pollutant transport control equation, including an advection diffusion equation and source-sink terms, wherein the velocity field in the nearshore pollutant transport control equation is driven by the environmental dynamics elements; using the residual of the control equation as a physical constraint loss term, which together with the data fitting loss term constitutes a multi-objective loss function; and using a gradient descent algorithm to simultaneously minimize the data fitting error and the physical equation residual, so that the trained network parameters both reproduce the observed data and satisfy physical laws.

[0014] Optionally, the source tracing results include a dynamic source probability map, a dominant transport path, and a quantitative analysis of source contribution rates. The dynamic source probability map dynamically displays the probability of each potential emission source in the form of a heat map and is updated based on real-time input data. The dominant transport path is simulated using a particle tracking algorithm based on an inverse probability field, used to visualize the most likely trajectory of microplastics from the source point to the observation point. The quantitative analysis of source contribution rates is used to calculate and output the relative percentage contribution of each potential emission source to the pollution concentration at the observation point.

[0015] Optionally, the generation of the dominant transport path includes: based on a trained intelligent source tracing model, inverting to obtain the forward probability field of pollutants reaching the observation point from all potential source points, or the reverse probability field of tracing back from the observation point to the source point; releasing a large number of virtual particles in the reverse probability field and determining their motion direction according to the probability field gradient, and implementing a random walk simulation; aggregating the motion trajectories of all virtual particles, extracting the path with the highest frequency of occurrence through kernel density estimation as the dominant transport path, and calculating the statistical confidence of the dominant transport path.

[0016] Optionally, the method further includes a physical mechanism analysis and interpretability output step for the source tracing results: extracting and analyzing the key environmental dynamics features that the intelligent source tracing model relies on during the decision-making process; associating and matching the determination of high-probability source tracing areas with the spatiotemporal activity characteristics of specific physical mechanisms to generate an association report; outputting readable source tracing conclusions, clearly pointing out the physical evidence supporting the source tracing conclusions, including the temporal matching of tidal phase and concentration peak, the spatial coupling of plume structure and high-probability areas, and the synchronicity of wave events and the surge in resuspension source contributions.

[0017] Secondly, this application provides an intelligent source tracing system for microplastic emissions in nearshore areas. The system includes: a data construction module for acquiring environmental dynamics information and microplastic observation information affecting microplastic transport and distribution within the target nearshore area, and constructing a multi-source dataset; a model training module for training an artificial intelligence model based on the multi-source dataset, using at least one nearshore-specific physical mechanism represented by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants represented by the microplastic observation information, with the nearshore-specific physical mechanism as a constraint, to obtain an intelligent source tracing model, enabling the intelligent source tracing model to learn the microplastic transport laws driven by the physical mechanism; and a source tracing output module for inputting the multi-source dataset into the pre-constructed intelligent source tracing model and outputting the source tracing results calculated by the intelligent source tracing model, wherein the source tracing results include at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application; Figure 2 A flowchart illustrating an intelligent source tracing method for microplastic emissions in nearshore areas, provided as an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an intelligent traceability system for microplastic emissions in nearshore areas, provided as an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0022] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0023] Microplastic pollution has become a global coastal environmental problem. Existing source tracing methods mostly suffer from the following limitations: reliance on offline sampling and laboratory analysis, which is time-consuming and labor-intensive, and cannot achieve rapid response and real-time monitoring. Overly simplistic models: most numerical or statistical models ignore the high-frequency and complex dynamic processes in coastal areas, resulting in crude source tracing results and high uncertainty. "Scenario generalization" problem: existing methods fail to closely align with the unique characteristics of strong land-sea interactions and intensive human activities in coastal areas; their models generalize to a limited range of scenarios, lacking specificity and accuracy. Therefore, there is an urgent need for an innovative solution that can deeply integrate the unique physical mechanisms of coastal areas to achieve accurate and efficient source tracing.

[0024] Based on this, this application provides an intelligent source tracing method and system for microplastic emissions in nearshore areas. Through the deep integration of physical mechanism constraints and artificial intelligence models, it achieves three core advantages: First, it improves the accuracy of source tracing. Under the guidance of physical laws, the model reduces overfitting to sparse data and more accurately identifies the spatial location of emission sources. Second, it enhances the interpretability of results. The quantitative output of source tracing probability maps, transport paths, and contribution rates provides intuitive basis for management decisions, and the correlation of physical evidence improves the transparency and verifiability of conclusions. Third, it has strong adaptability. The fusion of multi-source data and alternative implementation methods enable it to flexibly cope with the differences in data availability in different nearshore environments.

[0025] Figure 1 This application provides an illustration of an application scenario. In the process of tracing microplastic emissions in nearshore areas, the method provided in this application can be used to flexibly address the differences in data availability in different nearshore environments, obtain highly interpretable tracing conclusions, and accurately identify the spatial location of emission sources.

[0026] Specifically, the method provided in this application is applied to any server. The server interacts with a set of multi-source sensors, and obtains environmental dynamics information and microplastic observation information provided by the multi-source sensor set through the server to construct a multi-source dataset. Based on the multi-source dataset, at least one nearshore-specific physical mechanism characterized by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information are used as constraints to train an artificial intelligence model to obtain an intelligent source tracing model. The intelligent source tracing model learns the microplastic transport laws driven by the physical mechanism. The multi-source dataset is input into the pre-constructed intelligent source tracing model, and the source tracing results calculated by the intelligent source tracing model are output. The source tracing results are provided to the corresponding supervisors. The source tracing results include at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources. By deeply integrating physical constraints with artificial intelligence models, three core advantages have been achieved: First, improved source tracing accuracy. Guided by physical laws, the model reduces overfitting to sparse data and more accurately identifies the spatial location of emission sources. Second, enhanced interpretability of results. The quantitative output of source tracing probability maps, transport paths, and contribution rates provides intuitive basis for management decisions, and the correlation of physical evidence improves the transparency and verifiability of conclusions. Third, strong adaptability. The fusion of multi-source data and alternative implementation methods enable it to flexibly cope with the differences in data availability in different nearshore environments.

[0027] For specific implementation details, please refer to the following examples.

[0028] Figure 2 This is a flowchart illustrating an embodiment of XXX provided in this application. The method of this embodiment can be applied to the server in the above scenario. For example... Figure 2 As shown, the method includes: S201. Obtain environmental dynamics information and microplastic observation information that affect the transport and distribution of microplastics in the nearshore area of ​​the target, and construct a multi-source dataset; S202. Based on the multi-source dataset, the nearshore-specific physical mechanism characterized by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information are used as constraints to train the artificial intelligence model to obtain an intelligent source tracing model, so that the intelligent source tracing model learns the microplastic transport law driven by the physical mechanism. S203. Input the multi-source dataset into the pre-built intelligent source tracing model and output the source tracing result calculated by the intelligent source tracing model. The source tracing result includes at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources.

[0029] Technical Background and Working Principle: Nearshore areas serve as crucial channels for land-based microplastic input into the ocean, presenting complex challenges for pollution source tracing. Microplastic transport is strongly regulated by environmental dynamics factors such as tidal cycles, salinity gradients, and wave energy, making it difficult for traditional statistical models to capture the nonlinear processes driven by these physical mechanisms. Furthermore, microplastic observation data is typically sparse and exhibits spatiotemporal heterogeneity, leading to significant uncertainty in source tracing results. This embodiment embeds physical mechanisms characterized by environmental dynamics factors (such as tidal pumping, plume accumulation, and wave resuspension) as constraints into an artificial intelligence model. This forces the model to follow physical laws while learning the distribution of observational data, thereby simulating the transport behavior of microplastics in real hydrological environments and enabling reliable inference of emission source locations and contribution levels from multi-source data.

[0030] Technical Solution and Component Functions: This solution first collects environmental dynamics information (including periodic hydrological cycle information based on tidal harmonic constant and tidal field data analysis, water body characteristic gradient information such as salinity and temperature distribution obtained through satellite remote sensing or field measurements to identify the range of estuarine plumes, and wave energy information obtained through wave observation stations or wave models to assess sediment resuspension potential) and microplastic observation information (including concentration spatiotemporal sequences obtained through automatic monitoring buoy networks or cruise surveys, and chemical composition fingerprints obtained through spectral analysis to establish source association prior weights) in a unified format, constructing a multi-source dataset; subsequently, in the model training module, an intelligent source tracing model is constructed based on the Physical Information Neural Network (PINN) or Spatiotemporal Graph Neural Network (ST-GNN) architecture, which will be used to train near-shore waters. Ocean-specific physical mechanisms (such as the residual current field corresponding to tidal pumping effect, the salinity gradient attention weight corresponding to plume aggregation effect, and the wave height threshold activation corresponding to wave resuspension effect) are embedded as constraints through loss functions or attention mechanisms. Through multi-task learning, the model simultaneously optimizes data fit (such as concentration prediction error) and physical consistency (such as control equation residuals), enabling the model to learn the microplastic transport laws driven by physical mechanisms. Finally, in the source inference module, the multi-source dataset is input into the trained model, and the output is a dynamic source inference probability map (indicating the spatial location and contribution of potential emission sources in the form of a heat map), the dominant transport path (simulated by inverse probability field particle tracking), and a quantitative analysis of source contribution rate (normalized calculation of the relative contribution percentage of each source). The interpretability output module is used to associate physical evidence (such as matching tidal phase with concentration peak) to enhance the credibility of the results.

[0031] Beneficial effects: This embodiment achieves three core advantages through the deep integration of physical mechanism constraints and artificial intelligence models: First, it improves the accuracy of source tracing. Under the guidance of physical laws, the model reduces overfitting to sparse data and more accurately identifies the spatial location of emission sources. Second, it enhances the interpretability of results. The quantitative output of source tracing probability maps, transport paths, and contribution rates provides intuitive basis for management decisions, and the correlation of physical evidence improves the transparency and verifiability of conclusions. Third, it is highly adaptable. The fusion of multi-source data and alternative implementation methods enable it to flexibly cope with the differences in data availability in different nearshore environments.

[0032] Alternative or modified implementation methods: At the data acquisition level, environmental dynamics information can be obtained by using on-site flow profilers or high-frequency ground wave radar instead of satellite remote sensing to acquire salinity gradients, or by using numerical wave prediction systems (such as the SWAN model) instead of wave observation stations; microplastic observation information can be obtained by using filter sampling combined with mass spectrometry analysis instead of spectral fingerprints, or by using UAV remote sensing for surface concentration estimation. At the model training level, physical constraints can be implemented by using the Lagrange multiplier method, adaptive weight adjustment, or a penalty term based on reinforcement learning instead of a fixed loss function term; the artificial intelligence model architecture can be replaced by a spatiotemporal convolutional network (ST-CNN) or a graph attention network (GAT) to optimize computational efficiency. At the output level, the source probability map can be modified into a contour map or a 3D volume rendering form; the generation of the dominant transport path can use a deterministic shortest path algorithm (such as Dijkstra) instead of random walks; and the calculation of source contribution rate can be optimized based on Bayesian inference or principal component analysis.

[0033] In some embodiments, environmental dynamics information includes periodic hydrological cycle information, water body characteristic gradient information, and wave energy information. The construction of periodic hydrological cycle information includes: based on tidal harmonic constant and tidal field data, analyzing the reciprocating pumping effect and net transport direction of pollutants caused by tidal asymmetry, and constructing periodic hydrological cycle information. The construction of water body characteristic gradient information includes: obtaining the spatial distribution of salinity and temperature through satellite remote sensing or field measurements, and identifying the spatial range and frontal structure of estuarine plumes, and constructing water body characteristic gradient information. The spatial range of estuarine plumes is a fan-shaped area extending from the estuary to the open sea. The acquisition of wave energy information includes: obtaining effective wave height and wave period data through wave observation stations or wave models, and constructing wave energy information for assessing the potential intensity of sediment resuspension.

[0034] Technical Background and Working Principle: The transport pathways of microplastics in nearshore environments are highly dependent on the coupling effects of multiple environmental dynamic factors: tidal asymmetry causes a net transport shift of pollutants during the tidal cycle through a reciprocating pumping effect; estuarine plumes, due to salinity and temperature gradients, form density-driven circulation structures that induce microplastic aggregation in frontal regions; and wave energy triggers the resuspension of historically deposited microplastics through substrate shear stress, thereby altering the concentration distribution in the water column. Traditional methods often treat these factors in isolation, making it difficult to capture their synergistic effects. This embodiment systematically acquires periodic hydrological cycles, water characteristic gradients, and wave energy information to construct a physically consistent environmental dynamic dataset. This provides a mechanism-driven constraint basis for subsequent intelligent models, ensuring that the source tracing process accurately reflects the dominant influence of actual nearshore physical processes.

[0035] Technical Solution and Component Functions: This solution constructs three types of environmental dynamic element information through a multimodal data acquisition and analysis module: Periodic hydrological cycle process information is obtained by inputting tidal harmonic constants (such as the M2 and S2 tidal amplitudes and phases acquired from tide gauges or satellite altimeters) and high-resolution tidal field data (such as data generated through field measurements using numerical models HYCOM or ADCP) into a tidal asymmetry analytical algorithm to calculate the residual flow vector field and quantify the net transport direction and flux caused by the reciprocating pumping effect; Water body characteristic gradient information is obtained by acquiring spatially continuous surface salinity and temperature data through satellite remote sensing (such as salinity and temperature products from MODIS or Sentinel-3) or field measurements (such as CTD profilers). Gradient calculation and front detection algorithms are used to identify the spatial range (fan-shaped expansion structure) of estuarine plumes and their frontal positions, thereby extracting the spatial boundary parameters of plume aggregation effects; Wave energy information is obtained through wave observation stations (such as buoys or shore-based radar) or wave models (such as SWAN or WAVEWATCH). III) Obtain effective wave height and wave period time series data, and calculate the potential intensity threshold for sediment resuspension using the wave shear stress formula, combined with sediment type information. These three types of information, after spatiotemporal alignment and standardization, together constitute an environmental dynamics element dataset, which serves as the input source for physical constraints in the mechanism embedding during the training phase.

[0036] Beneficial effects: This embodiment achieves a systematic and quantitative characterization of environmental dynamics elements, bringing three core advantages: First, by simultaneously integrating tidal pumping, plume aggregation, and wave resuspension effects, it comprehensively covers the key physical driving factors of nearshore microplastic transport, reducing model bias caused by missing physical processes; second, high-resolution data (such as remotely sensed salinity gradients and model residual current fields) provide explicit spatial constraints, enhancing the reliability of source tracing results in complex terrain areas; and third, the modular design allows for flexible selection of input sources based on data availability, improving the applicability of the method in different geographical scenarios.

[0037] Alternative or modified implementation methods: At the data acquisition level, information on periodic hydrological cycles can be obtained by using data-assimilated tidal forecasting models (such as NAO.99b or FES2014) instead of measured harmonic constants, or by supplementing tidal field data with surface flow fields retrieved from high-frequency ground wave radar; water characteristic gradient information can be obtained through seabed sensor arrays (such as moored CTD chains) or UAV-borne thermal infrared sensors to acquire higher-frequency vertical gradients, replacing the large-scale coverage of satellite remote sensing; wave energy information can be obtained by using parameterized models based on wind speed data (such as the Empirical Wave Height formula) or wave parameters retrieved from shore-based video monitoring systems, replacing traditional wave observation stations. At the data processing level, tidal asymmetry analysis can be modified into simplified calculations based on tidal ellipse parameters; plume range identification can be achieved by automatically segmenting salinity fronts using machine learning classifiers (such as random forests); wave resuspension assessment can be performed by introducing sediment transport models (such as SEDTRANS) to couple dynamic changes in the seabed.

[0038] In some embodiments, microplastic observation information includes a spatiotemporal sequence of concentration and a chemical composition fingerprint. The spatiotemporal sequence of concentration is obtained by deploying an automated monitoring buoy network or periodic cruise surveys to capture dynamic changes in microplastic distribution. The chemical composition fingerprint is obtained by performing spectral analysis on samples to obtain their main polymer composition, which is used to establish the chemical correlation between pollutants at the observation point and potential emission sources, and the chemical correlation is introduced as a priori weight into the source contribution rate calculation of the intelligent source tracing model.

[0039] Technical Background and Working Principle: The distribution of microplastics in nearshore environments exhibits highly dynamic characteristics, making accurate source tracing difficult solely based on concentration data. While spatiotemporal concentration sequences can reflect the migration patterns of pollutants under the influence of tides and currents, their spatial representativeness is limited by the density of monitoring points. Chemical fingerprints, on the other hand, establish a chemical correlation between the "pollution source and the environmental medium" through polymer composition characteristics, providing intrinsic evidence for source tracing. This embodiment constructs a multidimensional observation dataset by simultaneously acquiring concentration dynamics and chemical fingerprint information: concentration sequences capture external transport processes, while chemical fingerprints reveal the source attributes of substances. The combination of these two forms a complementary verification mechanism of "spatiotemporal distribution + compositional characteristics," significantly improving the physicochemical consistency of source tracing inference.

[0040] Technical Solution and Component Functions: This solution constructs a microplastics observation information system through a three-dimensional monitoring network: the spatiotemporal concentration sequence is obtained by deploying an automatic monitoring buoy network (integrating laser diffraction particle counters or fluorescence spectroscopy sensors to achieve hourly continuous monitoring) and periodic cruise surveys (using standard plankton net stratified sampling to construct a three-dimensional concentration field). After quality control and spatiotemporal interpolation, a continuous concentration distribution field is generated to capture the transport dynamics at the tidal cycle scale. The chemical composition fingerprint is obtained by performing micro Fourier transform infrared spectroscopy (μ-FTIR) or Raman spectroscopy analysis on the collected samples to obtain the characteristic peak intensity ratios of major polymers such as polyethylene (PE) and polypropylene (PP). A fingerprint database is constructed through principal component analysis (PCA), and the chemical correlation between the observation point and potential emission sources (such as sewage treatment plants and fishery activity areas) is established through similarity calculation. Finally, the chemical correlation is quantified into a priori weight matrix and injected into the source contribution rate calculation module of the intelligent source tracing model through a Bayesian framework, so that the model further strengthens the chemical evidence-driven source tracing decision on the basis of physical constraints.

[0041] Beneficial effects: This embodiment achieves three major technological breakthroughs: First, by coordinating the observation of concentration and composition, it overcomes the limitations of single-dimensional data, enabling the source tracing results to simultaneously satisfy transport kinetics and the law of conservation of matter; Second, the prior weights provided by chemical fingerprints significantly improve the contribution rate analysis capability in multi-source mixed scenarios, especially suitable for complex catchment areas such as estuaries; Third, the combination of automatic monitoring and laboratory analysis realizes full-chain data support from on-site perception to model decision-making.

[0042] Alternative or modified implementation methods: Concentration monitoring can utilize remote sensing inversion technology (such as floating debris index based on Sentinel-2 imagery) to expand spatial coverage, or use autonomous underwater vehicles (AUVs) equipped with sensors to achieve intensive observation of key areas; chemical analysis can be modified to use pyrolysis-gas chromatography / mass spectrometry (Py-GC / MS) to detect trace samples, or use fluorescence labeling technology to track emissions from specific sources; prior weight construction can use machine learning classifiers (such as support vector machines) to replace principal component analysis, or introduce Monte Carlo simulation to quantify the uncertainty of fingerprint matching; at the data fusion level, chemical fingerprints can be further combined with stable isotope characteristics to form a multi-index source tracing system.

[0043] In some embodiments, nearshore-specific physical mechanisms include at least one of tidal pumping effect, plume aggregation effect, and wave resuspension effect. Tidal pumping effect refers to the phenomenon that the net transport of pollutants is not zero during the tidal cycle due to tidal asymmetry. Plume aggregation effect refers to the phenomenon of microplastic aggregation near salinity fronts due to differences in water density and circulation structure. Wave resuspension effect refers to the process in which historical microplastics deposited in bottom sediments are resuspended into the water body under the action of wave energy. For tidal pumping effect, the residual flow field derived based on harmonic constant is added as a physical constraint term to the loss function to force the transport law output by the model to be consistent with the net transport direction. For plume aggregation effect, salinity gradient data is converted into a spatial attention weight matrix so that the model prioritizes calculating the pollutant source probability within the influence range of plumes. For wave resuspension effect, an effective wave height threshold is set to activate the resuspension source contribution calculation module, and a positive correlation between contribution weight and wave energy is established.

[0044] Technical Background and Working Principle: The transport of microplastics in nearshore environments is controlled by multiple physical mechanisms: tidal pumping generates net residual flow through nonlinear tidal dynamics, leading to directional transport of pollutants in apparent reciprocating motion; plume aggregation is formed by secondary circulation driven by density gradients, creating a converging flow field near salinity fronts; and wave resuspension disrupts the sediment-pollutant balance through seabed shear forces, causing historically deposited microplastics to re-enter the water. Traditional data-driven models struggle to incorporate these complex physical processes, while pure numerical simulations are limited by parameterization errors. This embodiment transforms the physical mechanisms into computable constraints, maintaining the flexibility of the artificial intelligence model while ensuring its output strictly adheres to nearshore hydrodynamic laws.

[0045] Technical Solution and Component Functions: This solution achieves constraint integration of three major effects through physical mechanism quantification and embedded modules: For the tidal pumping effect, based on tidal harmonic analysis, the Euler residual current is calculated from the amplitude and phase difference of major tidal constituents such as M2 and S2. The residual current vector field is added as a physical constraint term to the model loss function. By minimizing the difference between the model output concentration field and the residual current transport direction, the net transport pattern caused by tidal asymmetry is forced to be learned. For the plume aggregation effect, the horizontal gradient modulus is calculated using salinity data from satellite remote sensing or field measurements. It is then transformed into a spatial attention weight matrix in the [0,1] interval using the sigmoid function. This matrix is ​​then multiplied by the Hadamard product of the source probability distribution output by the model. The calculation automatically strengthens the source weighting of the salinity front region during model training. For the wave resuspension effect, the bottom shear stress τb=0.5ρfwUb² (where fw is the friction coefficient and Ub is the near-bottom wave trajectory velocity) is calculated based on wave observation data. The critical shear stress threshold τcr=0.2-0.5N / m² (adjusted according to the bottom sediment type) is set. When τb>τcr, the resuspension source contribution calculation module is activated, and a positive correlation between the effective wave height Hs and the resuspension contribution weight is established through linear transformation: Wresus=α·(Hs-Hthr) (where α is the proportionality coefficient and Hthr is the threshold wave height). This dynamic weight is injected into the source contribution rate calculation process.

[0046] Beneficial effects: This embodiment achieves three major breakthroughs through the computable transformation of physical mechanisms: First, it transforms qualitative physical concepts into quantitative constraints, enabling the model to maintain physical rationality even in data-scarce regions; second, the introduction of spatial attention weights significantly improves the model's source resolution in key regions such as plumes; and third, the dynamic threshold mechanism ensures the spatiotemporal adaptive calculation of wave resuspension contributions, effectively capturing the impact of sudden resuspension events.

[0047] Alternative or modified implementation methods: Tidal pumping constraints can be modified to be based on the concept of Lagrange residual flow, using particle tracking to calculate the net displacement field as the constraint target; plume attention weights can use feature extraction based on convolutional neural networks to replace gradient calculation, or introduce temperature gradients to jointly construct a multi-parameter attention mechanism; wave resuspension thresholds can be set as a dynamic function τcr=f(D50) that varies with sediment grain size, or machine learning classifiers (such as support vector machines) can be used to predict the resuspension state based on multiple parameters (wave height, period, sediment); the mathematical expression of the constraint terms can use the Lagrange multiplier method to construct equality constraints, or use adaptive weights to balance the contribution ratio of physical constraints and data fitting.

[0048] In some embodiments, when a physical information neural network architecture is adopted, the residuals of the control equations for nearshore pollutant transport are used as physical constraint loss terms, which together with the data fitting loss terms constitute a multi-objective loss function for optimization. When a spatiotemporal graph neural network architecture is adopted, monitoring stations are constructed as graph nodes, and environmental dynamics elements and microplastic observation information are used as node and edge features, respectively, while simultaneously modeling spatial correlation and temporal dynamics. Both architectures optimize data fitting and consistency with physical laws through multi-task learning, ensuring that the model learns from observation data and complies with physical constraints.

[0049] Technical Background and Working Principle: Traditional data-driven models in nearshore microplastic tracing often neglect the constraints of physical mechanisms, leading to significant deviations between predicted results and actual conditions. Physical Information Neural Networks (PINNs) enhance generalization ability and interpretability by embedding the residuals of the pollutant transport control equations into the loss function, forcing the model output to conform to physical laws. Spatiotemporal Graph Neural Networks (ST-GNNs) utilize graph structures to model the spatial correlations between monitoring stations and combine time series data to dynamically capture the evolutionary characteristics of microplastic transport. Both architectures employ multi-task learning mechanisms to maintain physical consistency while fitting observational data, solving the overfitting problem of pure data-driven models under sparse data.

[0050] Technical Solution and Component Functions: When using the Physical Information Neural Network (PINN) architecture, the control equations for nearshore pollutant transport (such as the advection-diffusion equation and source-sink terms) are first established, where the velocity field is driven by environmental dynamic factors (such as tides, salinity gradients, and wave energy). The residuals of the control equations are used as physical constraint loss terms, which, together with data fitting loss terms (such as mean squared error), constitute a multi-objective loss function. These two terms are simultaneously minimized using a gradient descent algorithm (such as Adam), ensuring that the trained network parameters both reproduce the observed data and satisfy physical laws. When using the Spatiotemporal Graph Neural Network (ST-GNN) architecture, monitoring stations are constructed as graph nodes, environmental dynamic factors (such as tidal residual currents and salinity gradients) are used as node features, and microplastic observation information (such as spatiotemporal concentration sequences) is used as edge features. Graph convolutional layers are used to capture spatial dependencies, and time series modules (such as recurrent neural networks) are combined to model dynamic changes. During training, a multi-task loss function is used to simultaneously optimize data reconstruction errors and physical constraint terms (such as mechanism matching loss based on attention weights). Both architectures support modular extensions. For example, PINN can integrate automatic differentiation to calculate residuals, while ST-GNN can introduce a graph attention mechanism to enhance the learning of key nodes.

[0051] Beneficial effects: By adopting the PINN or ST-GNN architecture, robust predictions were achieved in data-sparse scenarios, significantly improving the physical rationality and interpretability of the source tracing results; the multi-task learning mechanism ensured that the model not only learned the statistical characteristics of the observed data, but also complied with the nearshore-specific physical constraints, reducing the risk of overfitting and providing a reliable basis for decision support.

[0052] Alternative or modified implementation methods: Physical information neural networks can use deep residual networks (ResNet) or convolutional neural networks (CNN) instead of basic fully connected networks to enhance feature extraction capabilities; spatiotemporal graph neural networks can be replaced by spatiotemporal convolutional networks (STCN) or Transformer-based sequence models to optimize long-term time-series dependency modeling; the weight allocation of multi-task loss functions can use adaptive adjustment strategies (such as dynamic weighting based on validation set performance) instead of fixed proportions; the implementation of physical constraint terms can be modified to use the Lagrange multiplier method or adversarial training to balance data fitting and physical consistency.

[0053] In some embodiments, a nearshore pollutant transport control equation is established, including an advection diffusion equation and source-sink terms. The velocity field in the nearshore pollutant transport control equation is driven by environmental dynamics factors. The residual of the control equation is used as a physical constraint loss term, which together with the data fitting loss term constitutes a multi-objective loss function. The gradient descent algorithm is used to minimize both the data fitting error and the physical equation residual, so that the trained network parameters can both reproduce the observed data and satisfy physical laws.

[0054] Technical background and working principle The transport of microplastics in nearshore environments follows fundamental physical laws such as the conservation of mass and momentum. Traditional numerical simulation methods require precise boundary conditions and parameterization schemes, resulting in high computational costs and sensitivity to model parameters. Physical information neural networks (PINNs) transform the forward modeling problem into a constrained optimization problem by directly embedding the residuals of the governing equations into the neural network training process. This approach retains the flexibility of data-driven methods while ensuring that the output results consistently conform to physical laws. This method is particularly suitable for tracing pollutant sources in complex nearshore environments, where the coupling effects of advection, diffusion, and source-sink processes significantly influence the spatiotemporal distribution characteristics of microplastics.

[0055] Technical Solution and Component Functions: First, the governing equations for nearshore pollutant transport are established, including advection-diffusion equations and source-sink terms. The velocity field in the advection term is driven by environmental dynamic factors (tidal residual current, density flow induced by salinity gradient, etc.), the diffusion term considers turbulent diffusion effects, and the source-sink terms include emission sources and sedimentation processes. The residuals of the governing equations are used as physical constraint loss terms, which, together with the data fitting loss term based on microplastic observation concentrations, constitute a multi-objective loss function. The residuals of the partial differential equations are calculated using automatic differentiation techniques, and gradient descent algorithms (such as Adam or L-BFGS) are used to simultaneously minimize the data fitting error and the physical equation residuals. During training, the network parameters are iteratively updated until the physical residuals and data errors reach a preset convergence threshold. The final network accurately fits the observed data and satisfies the physical constraints of the governing equations throughout the computational domain.

[0056] Beneficial effects: By directly embedding the physical control equations into the neural network training, it is ensured that the model output satisfies the physical laws at any spatial location and time point, which significantly improves the prediction reliability in sparse data regions; the application of automatic differentiation technology avoids the discretization error in traditional numerical methods and realizes the accurate implementation of physical constraints; the optimization strategy of multi-objective loss function balances the requirements of data fitting accuracy and physical consistency, and enhances the generalization ability of the model.

[0057] Alternative or modified implementation methods: The governing equations can incorporate a sedimentation term parameterized by microplastic sedimentation velocity, or a degradation term considering the influence of bioattachment; the calculation of the physical constraint loss term can use a weak or variational form instead of the strong form to reduce the requirement for solution smoothness; the weight allocation of the multi-objective loss function can adopt an adaptive adjustment strategy, dynamically balancing the proportion of data terms and physical terms according to the training phase; the optimization algorithm can be replaced by the quasi-Newton method or the conjugate gradient method to adapt to training data of different scales; the network architecture can adopt Fourier neural networks or deep residual networks to enhance the ability to capture high-frequency features.

[0058] In some embodiments, the source tracing results include a dynamic source probability map, a dominant transport path, and a quantitative analysis of source contribution rates. The dynamic source probability map dynamically displays the probability of each potential emission source in the form of a heatmap and is updated based on real-time input data. The dominant transport path is simulated by a particle tracking algorithm based on an inverse probability field and is used to visualize the most likely trajectory of microplastics from the source point to the observation point. The quantitative analysis of source contribution rates is used to calculate and output the relative percentage contribution of each potential emission source to the pollution concentration at the observation point.

[0059] Technical Background and Working Principle: Nearshore microplastic source tracing requires not only identifying the location of potential emission sources but also understanding the transport pathways of pollutants and the relative contribution of each source. Dynamic source tracing probability maps visually display the emission probability at different spatial locations using heatmaps; their dynamic characteristics reflect the impact of changing environmental conditions on the tracing results. Dominant transport pathways, based on gradient information in the probability field, simulate the most probable trajectory of pollutants through particle tracking, revealing key transport channels for microplastics from the source area to the observation area. Quantitative source contribution analysis transforms probability values ​​into specific contribution percentages through normalization, providing a quantitative basis for pollution liability determination and control priority assessment. The organic combination of these three components enables a full-chain source tracing analysis, from spatial location and pathway identification to liability quantification.

[0060] Technical Solution and Component Functions: The dynamic source tracing probability map is achieved through spatial interpolation and heatmap rendering of the probability field output by the intelligent source tracing model. Color gradients represent the source tracing probability intensity of different regions, and the map is dynamically updated based on real-time input environmental dynamics data. The generation of dominant transport paths is based on an inverse probability field. A large number of virtual particles are released into this field, and random walk simulations are performed based on local probability gradients. After aggregating all particle trajectories, kernel density estimation is used to extract the most frequently occurring connected paths. Quantitative source contribution rate analysis calculates the relative percentage contribution of each source to a specific observation point by normalizing the probability weights of each potential emission source with the concentration data at the observation point. The contribution rate calculation also considers the spatial overlap effect and temporal variation characteristics of different sources. The three components share the same probability field data foundation but use different post-processing algorithms to extract targeted information.

[0061] Beneficial effects: By providing dynamically updated probability maps, visualized transport paths, and quantified contribution rate analysis, a multi-dimensional traceability result display system is formed, which greatly improves the interpretability and practicality of the results; the dynamic update mechanism enables the traceability results to reflect changes in environmental conditions in a timely manner, enhancing the system's real-time response capability; path visualization helps to understand complex near-shore transport processes, while contribution rate quantification provides a clear scientific basis for management decisions.

[0062] Alternative or modified implementation methods: Dynamic source probabilities can be replaced by contour maps, 3D isosurfaces, or animation sequences instead of static heatmaps; dominant transport path generation can be replaced by deterministic shortest path algorithms, streamline tracing methods, or path planning based on deep reinforcement learning instead of random walk simulations; source contribution rate calculation can be replaced by Bayesian inference, principal component analysis regression, or contribution allocation methods based on Shapley values ​​instead of simple normalization calculations; results display can be integrated into WebGIS platforms or mobile applications, supporting interactive queries and multi-dimensional data filtering.

[0063] In some embodiments, the generation of the dominant transport path includes: based on a trained intelligent source tracing model, inverting the forward probability field of pollutants arriving at the observation point from all potential source points, or the reverse probability field of tracing back from the observation point to the source point; releasing a large number of virtual particles in the reverse probability field and determining their motion direction according to the probability field gradient, and implementing a random walk simulation; aggregating the motion trajectories of all virtual particles, extracting the path with the highest frequency of occurrence through kernel density estimation as the dominant transport path, and calculating the statistical confidence of the dominant transport path.

[0064] Technical Background and Working Principle: The transport paths of microplastics in nearshore environments are comprehensively influenced by physical mechanisms such as tides, currents, and waves, resulting in highly nonlinear and spatiotemporally variable migration trajectories. The generation of dominant transport paths aims to visualize the most probable trajectories of microplastics from potential emission sources to observation points, thereby revealing the main migration channels and key impact areas of pollutants. The inverse probability field, based on the inversion calculation of the intelligent source tracing model, reflects the inverse probabilistic influence of the observation point on the potential source; its gradient field implicitly contains the transport direction driven by physical mechanisms. Virtual particle random walk simulations are implemented in this field to statistically capture the possible movement patterns of microplastics in complex environments, while kernel density estimation extracts high-frequency paths from a large number of trajectories and quantifies the reliability of the paths through confidence level calculations, ensuring consistency between the results and real physical processes.

[0065] Technical Solution and Component Functions: This implementation first utilizes a trained intelligent source tracing model to invert and obtain the reverse probability field (or forward probability field) for tracing pollutants from the observation point back to the source point. This probability field is generated by numerical differentiation of the source tracing results output by the model to produce a gradient field. Subsequently, a large number of virtual particles are released in the gradient field. Each particle's motion direction is determined by the local probability field gradient, and a random walk simulation is performed to generate multiple possible trajectories. The gradient direction guides the particles to move towards high-probability regions, simulating the transport process of microplastics. After aggregating all particle trajectories, the kernel density estimation method is used to smooth the spatial distribution of the trajectories. The most frequently occurring connected path is extracted as the dominant transport path, and a statistical confidence level (such as percentage or p-value) is calculated based on the frequency of the path's occurrence in the simulation to quantify the path's reliability. The intelligent source tracing model is responsible for providing the probability field data, the virtual particle simulation module executes the gradient-driven random walk, and the trajectory aggregation and kernel density estimation module processes the trajectory data and outputs the path results and confidence index.

[0066] Beneficial effects: By generating dominant transport paths, the migration trajectory of microplastics from the source to the observation point can be intuitively displayed, which helps to identify key transport channels and pollution hotspots, and improves the spatial interpretability and decision support capabilities of the source tracing results; the calculation of statistical confidence provides a quantitative indicator of path reliability, enhances the verifiability and scientific validity of the results, and facilitates environmental managers to formulate targeted control measures.

[0067] Alternative or modified implementation methods: The number of virtual particles released can be dynamically adjusted based on computational resources, for example, ranging from thousands to millions of particles; the random walk algorithm can use the Metropolis-Hastings method or Langevin dynamics instead of basic gradient following to introduce randomness and enhance path diversity; kernel density estimation can be replaced by a grid-based probability accumulation method or clustering analysis techniques (such as DBSCAN) for path extraction; the inversion of the probability field can also use a forward probability field instead of a backward probability field, or a combination of both for bidirectional path optimization; in addition, gradient calculation can use the Sobel operator or Gaussian gradient instead of numerical differentiation to improve stability.

[0068] In some embodiments, the method further includes a physical mechanism analysis and interpretability output step for the source tracing results: extracting and analyzing the key environmental dynamics features that the intelligent source tracing model relies on in the decision-making process; associating and matching the determination of high-probability source tracing areas with the spatiotemporal activity characteristics of specific physical mechanisms to generate an association report; outputting readable source tracing conclusions, clearly pointing out the physical evidence supporting the source tracing conclusions, including the temporal matching of tidal phase and concentration peak, the spatial coupling of plume structure and high-probability areas, and the synchronicity of wave events and the surge in resuspension source contributions.

[0069] Technical Background and Working Principle: While intelligent source tracing models can output high-precision tracing results during the prediction process, their decision-making logic is often considered a "black box," reducing the acceptability and verifiability of the results. Physical mechanism analysis aims to reveal the correspondence between the model's internal decisions and external physical processes. By analyzing the model's sensitivity to input features (such as tides, salinity gradients, and wave energy), it identifies the key physical factors driving the tracing results. Its core principle lies in the fact that the model learns the implicit correlation between environmental dynamics and microplastic transport during training. These correlations can be extracted through gradient analysis or attention mechanisms, and high-probability tracing areas are matched with the spatiotemporal activity characteristics of specific physical mechanisms (such as tidal pumping, plume aggregation, and wave resuspension). This establishes a transparent mapping from data-driven prediction to physical interpretation, enhancing the scientific rigor and credibility of the tracing conclusions.

[0070] Technical Solution and Component Functions: This implementation achieves physical mechanism analysis and interpretable output through the following components: First, the feature extraction module uses gradient backpropagation or attention weight analysis techniques to quantify the dependence of the intelligent source tracing model on various environmental dynamic elements (tidal phase, salinity gradient amplitude, significant wave height) and identify key decision-making features. Second, the spatiotemporal matching module performs correlation analysis between the high-probability source tracing area output by the model and the spatiotemporal activity data of the physical mechanism, including calculating the temporal correlation between tidal phase and concentration peak, assessing the spatial overlap between the plume front position and the high-probability area, and detecting the synchronicity between wave event time and the surge in resuspension source contribution. Finally, the report generation module automatically compiles a structured correlation report based on the matching results, clearly indicating the physical evidence supporting the source tracing conclusion, and outputting the conclusion in readable text form, such as "The high source tracing probability in region A is mainly driven by the ebb tide dominance and the salinity front aggregation effect." All components work collaboratively to achieve a complete analytical chain from model features to physical evidence.

[0071] Beneficial effects: Through physical mechanism analysis and interpretable output, the transparency and verifiability of source tracing results are significantly improved, enabling environmental managers to understand the physical basis of model decisions and enhancing their trust in source tracing conclusions; the automatic generation of structured reports saves manual analysis costs and improves analysis efficiency; clear physical evidence provides direct scientific basis for pollution source control measures and supports precise environmental governance.

[0072] Alternative or modified implementation methods: Feature extraction can use SHAP value analysis, integrated gradients, or hierarchical attention mechanisms to replace basic gradient analysis; spatiotemporal matching can use cross-correlation analysis, dynamic time warping, or spatiotemporal clustering algorithms for correlation quantification; report generation can support multiple output formats such as XML, JSON, or PDF, and can integrate natural language generation technology to automatically generate descriptive text; the matching scope of physical evidence can be extended to additional environmental parameters such as wind speed and direction, and runoff to enhance the comprehensiveness of the analysis.

[0073] Figure 3 This application provides a schematic diagram of the structure of an intelligent traceability system for microplastic emissions in nearshore areas, as shown in one embodiment. Figure 3 As shown, the intelligent traceability system 300 for microplastic emissions in nearshore areas in this embodiment includes: a data construction module 301, a model training module 302, and a traceability output module 303.

[0074] Data construction module 301 is used to acquire environmental dynamics information and microplastic observation information that affect the transport and distribution of microplastics in the nearshore area of ​​the target, and to construct a multi-source dataset; The model training module 302 is used to train the artificial intelligence model based on the multi-source dataset, using the nearshore specific physical mechanism characterized by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information as constraints, to obtain an intelligent source tracing model, so that the intelligent source tracing model learns the microplastic transport law driven by the physical mechanism. The source tracing output module 303 is used to input the multi-source dataset into a pre-built intelligent source tracing model and output the source tracing results calculated by the intelligent source tracing model. The source tracing results include at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources.

[0075] Optionally, in the data construction module 301, the environmental dynamics element information includes periodic hydrological cycle process information, water body characteristic gradient information, and wave energy information; The construction of the periodic hydrological cycle process information includes: based on tidal harmonic constant and tidal field data, analyzing the reciprocating pumping effect and net transport direction of pollutants caused by tidal asymmetry, and constructing the periodic hydrological cycle process information; The construction of the water body characteristic gradient information includes: obtaining the spatial distribution of salinity and temperature through satellite remote sensing or on-site measurement, identifying the spatial range and frontal structure of the estuarine plume, and constructing the water body characteristic gradient information. The spatial range of the estuarine plume is a fan-shaped region extending from the estuary to the open sea. The acquisition of wave energy information includes: obtaining effective wave height and wave period data through wave observation stations or wave models, constructing the wave energy information, and using it to assess the potential intensity of sediment resuspension.

[0076] Optionally, in the data construction module 301, the microplastic observation information includes a concentration spatiotemporal sequence and a chemical composition fingerprint; The concentration spatiotemporal sequence is obtained by deploying an automatic monitoring buoy network or periodic cruise surveys to capture dynamic changes in microplastic distribution; The chemical composition fingerprint is obtained by performing spectral analysis on the sample to obtain its main polymer composition, which is used to establish the chemical correlation between pollutants at the observation point and potential emission sources. The chemical correlation is then introduced as a priori weight into the source contribution rate calculation of the intelligent source tracing model.

[0077] Optionally, when training the artificial intelligence model using the near-shore specific physical mechanism as a constraint, the model training module 302 is specifically used for: The tidal pumping effect refers to the phenomenon that the net transport of pollutants is not zero during the ebb and flow of the tides due to tidal asymmetry. The feather aggregation effect refers to the phenomenon of microplastic aggregation near a salinity front due to differences in water density and circulation structure. The wave resuspension effect refers to the process by which historical microplastics deposited in bottom sediment are resuspended into the water body under the action of wave energy. For the tidal pumping effect, the residual flow field derived based on the harmonic constant is added as a physical constraint term to the loss function, forcing the transport law output by the model to be consistent with the net transport direction; For the plume aggregation effect, the salinity gradient data is transformed into a spatial attention weight matrix, so that the model prioritizes calculating the pollutant source probability within the plume's influence range; For the wave resuspension effect, an effective wave height threshold is set to activate the resuspension source contribution calculation module, and a positive correlation is established between the contribution weight and the wave energy.

[0078] Optionally, in the model training module 302, the artificial intelligence model adopts a physical information neural network architecture or a spatiotemporal graph neural network architecture; When the physical information neural network architecture is adopted, the residual of the control equation for nearshore pollutant transport is used as the physical constraint loss term, which together with the data fitting loss term constitutes a multi-objective loss function for optimization. When the spatiotemporal graph neural network architecture is adopted, the monitoring stations are constructed as graph nodes, and the environmental dynamics elements and the microplastic observation information are respectively used as node and edge features, while simultaneously modeling spatial correlation and temporal dynamics. Both architectures optimize data fit and consistency with physical laws through multi-task learning, ensuring that the model learns from the observed data while adhering to physical constraints.

[0079] Optionally, the specific implementation of the physical information neural network architecture in the model training module 302 is specifically used for: A nearshore pollutant transport control equation is established, including an advection diffusion equation and source-sink terms, wherein the velocity field in the nearshore pollutant transport control equation is driven by the environmental dynamics elements. The residuals of the control equations are used as physical constraint loss terms, which together with the data fitting loss terms constitute a multi-objective loss function. By minimizing both the data fitting error and the physical equation residuals using the gradient descent algorithm, the trained network parameters can both reproduce the observed data and satisfy the physical laws.

[0080] Optionally, in the source tracing output module 303, the source tracing results include a dynamic source tracing probability map, dominant transport path, and quantitative analysis of source contribution rate; The dynamic source tracing probability map dynamically displays the probability of each potential emission source in the form of a heat map and is updated based on real-time input data; The dominant transport path is simulated using a particle tracking algorithm based on an inverse probability field, which is used to visualize the most likely trajectory of microplastics from the source point to the observation point. The quantitative analysis of source contribution rate is used to calculate and output the relative percentage contribution of each potential emission source to the pollution concentration at the observation point.

[0081] Optionally, when generating the dominant transport path, the traceability output module 303 is specifically used for: Based on the trained intelligent source tracing model, the forward probability field of pollutants reaching the observation point from all potential source points is obtained, or the reverse probability field is traced back from the observation point to the source point. A large number of virtual particles are released in the inverse probability field, and their motion direction is determined according to the probability field gradient to implement a random walk simulation. The motion trajectories of all virtual particles are aggregated, and the most frequently occurring path is extracted as the dominant transport path through kernel density estimation. The statistical confidence of the dominant transport path is then calculated.

[0082] Optionally, the system 300 further includes a source tracing conclusion extension module 304, specifically used for: Extract and analyze the key environmental dynamics features that the intelligent tracing model relies on during the decision-making process; The determination of high-probability tracing areas is correlated and matched with the spatiotemporal activity characteristics of specific physical mechanisms to generate a correlation report; Output readable source attribution conclusions, clearly indicating the physical evidence supporting these conclusions, including the temporal matching of tidal phases and concentration peaks, the spatial coupling of plume structures and high-probability regions, and the synchronicity of wave events and surges in resuspension source contributions.

[0083] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A smart source tracing method for microplastic emissions in nearshore areas, characterized in that, include: To acquire environmental dynamics information and microplastic observation information affecting microplastic transport and distribution in the nearshore area of ​​the target, and to construct a multi-source dataset; Based on the multi-source dataset, at least one nearshore-specific physical mechanism characterized by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information are used as constraints to train the artificial intelligence model, thereby obtaining an intelligent source tracing model. The intelligent source tracing model learns the microplastic transport law driven by the physical mechanism. The multi-source dataset is input into a pre-built intelligent source tracing model, and the source tracing results calculated by the intelligent source tracing model are output. The source tracing results include at least a source tracing probability map for identifying the spatial location and contribution of one or more potential emission sources.

2. The method according to claim 1, characterized in that, The environmental dynamics information includes information on periodic hydrological cycles, water body characteristic gradients, and wave energy. The construction of the periodic hydrological cycle process information includes: based on tidal harmonic constant and tidal field data, analyzing the reciprocating pumping effect and net transport direction of pollutants caused by tidal asymmetry, and constructing the periodic hydrological cycle process information; The construction of the water body characteristic gradient information includes: obtaining the spatial distribution of salinity and temperature through satellite remote sensing or on-site measurement, identifying the spatial range and frontal structure of the estuarine plume, and constructing the water body characteristic gradient information. The spatial range of the estuarine plume is a fan-shaped region extending from the estuary to the open sea. The acquisition of wave energy information includes: obtaining effective wave height and wave period data through wave observation stations or wave models, constructing the wave energy information, and using it to assess the potential intensity of sediment resuspension.

3. The method according to claim 2, characterized in that, The microplastic observation information includes a spatiotemporal sequence of concentration and a chemical composition fingerprint; The concentration spatiotemporal sequence is obtained by deploying an automatic monitoring buoy network or periodic cruise surveys to capture dynamic changes in microplastic distribution; The chemical composition fingerprint is obtained by performing spectral analysis on the sample to obtain its main polymer composition, which is used to establish the chemical correlation between pollutants at the observation point and potential emission sources. The chemical correlation is then introduced as a priori weight into the source contribution rate calculation of the intelligent source tracing model.

4. The method according to claim 2, characterized in that, The nearshore-specific physical mechanisms include at least one of tidal pumping effect, plume aggregation effect, and wave resuspension effect. The training of the artificial intelligence model using these nearshore-specific physical mechanisms as constraints includes: The tidal pumping effect refers to the phenomenon that the net transport of pollutants is not zero during the ebb and flow of the tides due to tidal asymmetry. The feather aggregation effect refers to the phenomenon of microplastic aggregation near a salinity front due to differences in water density and circulation structure. The wave resuspension effect refers to the process by which historical microplastics deposited in bottom sediment are resuspended into the water body under the action of wave energy. For the tidal pumping effect, the residual flow field derived based on the harmonic constant is added as a physical constraint term to the loss function, forcing the transport law output by the model to be consistent with the net transport direction; For the plume aggregation effect, the salinity gradient data is transformed into a spatial attention weight matrix, so that the model prioritizes calculating the pollutant source probability within the plume's influence range; For the wave resuspension effect, an effective wave height threshold is set to activate the resuspension source contribution calculation module, and a positive correlation is established between the contribution weight and the wave energy.

5. The method according to claim 3, characterized in that, The artificial intelligence model adopts a physical information neural network architecture or a spatiotemporal graph neural network architecture; When the physical information neural network architecture is adopted, the residual of the control equation for nearshore pollutant transport is used as the physical constraint loss term, which together with the data fitting loss term constitutes a multi-objective loss function for optimization. When the spatiotemporal graph neural network architecture is adopted, the monitoring stations are constructed as graph nodes, and the environmental dynamics elements and the microplastic observation information are respectively used as node and edge features, while simultaneously modeling spatial correlation and temporal dynamics. Both architectures optimize data fit and consistency with physical laws through multi-task learning, ensuring that the model learns from the observed data while adhering to physical constraints.

6. The method according to claim 5, characterized in that, The specific implementation of the physical information neural network architecture includes: A nearshore pollutant transport control equation is established, including an advection diffusion equation and source-sink terms, wherein the velocity field in the nearshore pollutant transport control equation is driven by the environmental dynamics elements. The residuals of the control equations are used as physical constraint loss terms, which together with the data fitting loss terms constitute a multi-objective loss function. By minimizing both the data fitting error and the physical equation residuals using the gradient descent algorithm, the trained network parameters can both reproduce the observed data and satisfy the physical laws.

7. The method according to claim 6, characterized in that, The source tracing results include dynamic source tracing probability maps, dominant transport paths, and quantitative analysis of source contribution rates. The dynamic source tracing probability map dynamically displays the probability of each potential emission source in the form of a heat map and is updated based on real-time input data; The dominant transport path is simulated using a particle tracking algorithm based on an inverse probability field, which is used to visualize the most likely trajectory of microplastics from the source point to the observation point. The quantitative analysis of source contribution rate is used to calculate and output the relative percentage contribution of each potential emission source to the pollution concentration at the observation point.

8. The method according to claim 7, characterized in that, The generation of the dominant transport path includes: Based on the trained intelligent source tracing model, the forward probability field of pollutants reaching the observation point from all potential source points is obtained, or the reverse probability field is traced back from the observation point to the source point. A large number of virtual particles are released in the inverse probability field, and their motion direction is determined according to the probability field gradient to implement a random walk simulation. The motion trajectories of all virtual particles are aggregated, and the most frequently occurring path is extracted as the dominant transport path through kernel density estimation. The statistical confidence of the dominant transport path is then calculated.

9. The method according to claim 8, characterized in that, The method also includes steps for analyzing the physical mechanism of the tracing results and outputting interpretability: Extract and analyze the key environmental dynamics features that the intelligent tracing model relies on during the decision-making process; The determination of high-probability tracing areas is correlated and matched with the spatiotemporal activity characteristics of specific physical mechanisms to generate a correlation report; Output readable source attribution conclusions, clearly indicating the physical evidence supporting these conclusions, including the temporal matching of tidal phases and concentration peaks, the spatial coupling of plume structures and high-probability regions, and the synchronicity of wave events and surges in resuspension source contributions.

10. A smart traceability system for microplastic emissions in nearshore areas, characterized in that, The method applied to any one of claims 1-9 includes: The data construction module is used to acquire environmental dynamics information and microplastic observation information that affect the transport and distribution of microplastics in the nearshore area of ​​the target, and to construct a multi-source dataset; The model training module is used to train the artificial intelligence model based on the multi-source dataset, using the nearshore specific physical mechanism characterized by the environmental dynamics elements and the spatiotemporal distribution characteristics of pollutants characterized by the microplastic observation information as constraints, to obtain an intelligent source tracing model, so that the intelligent source tracing model learns the microplastic transport law driven by the physical mechanism. The source tracing output module is used to input the multi-source dataset into a pre-built intelligent source tracing model and output the source tracing results calculated by the intelligent source tracing model. The source tracing results include at least a source tracing probability map for identifying the spatial location and contribution level of one or more potential emission sources.

Images