Data integration risk assessment system for multi-source exposure of perfluoroalkyl / polyfluoroalkyl substances

Abstract

The invention relates to the technical field of data integration, and particularly discloses a perfluoro / polyfluoroalkyl substance multi-source exposure data integration risk assessment system, which is characterized in that environmental exposure data of perfluoro / polyfluoroalkyl substances is acquired through a multi-source environmental sensor array, and a PFAS multi-mode exposure feature database is established; carrying out pollution source isotope fingerprint analysis, and obtaining source contribution rate distribution maps of three pollution sources of industrial emission, surface runoff and atmospheric settlement through a nonlinear source analysis algorithm; constructing a three-dimensional geographic information dynamic migration model according to the source contribution rate distribution map, and generating a multi-medium dynamic migration flux matrix; a composite risk assessment model is established based on the multi-medium dynamic migration flux matrix, probability risk assessment is executed in combination with an ecological toxicity threshold database, and a space gridding risk grade map is output; the method not only fills the blank of the prior art in the aspects of multi-medium dynamic modeling and nonlinear source analysis, but also provides powerful technical support for environmental pollution control and ecological risk prevention and control.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data integration, and relates to a data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substances. BACKGROUND

[0002] As a class of environmental pollutants with high persistence, bioaccumulation and potential toxicity, perfluoro / polyfluoroalkyl substances (PFAS) are widely present in industrial emissions, consumer product residues, and natural environmental media (such as water bodies, soil, and atmosphere), forming a typical multi-source, cross-media exposure risk pattern. In this context, it is of great scientific significance and practical necessity to carry out integrated risk assessment of PFAS multi-source exposure data. First, multi-source exposure integration assessment can truly reflect the comprehensive exposure load of the population and the ecosystem in different time, space and environmental media, avoiding the risk of underestimation caused by single medium or single exposure path assessment. Secondly, the migration and exposure process of PFAS has high spatiotemporal dynamics and medium coupling, and it is necessary to integrate various types of sensor data, monitoring records, meteorological parameters and geographic information and other multi-source heterogeneous data to accurately model its environmental behavior and exposure risk. Therefore, from the perspective of data integration, building a unified and multi-dimensional risk assessment system is a key technical path to improve PFAS pollution monitoring capability, develop scientific governance strategies, and protect public health and ecological safety.

[0003] However, the current technology still faces many technical defects and practical application bottlenecks in PFAS multi-source exposure data integration and risk assessment. First, existing monitoring and assessment systems mostly focus on a single medium, such as only focusing on PFAS concentration in drinking water or exposure level of particulate matter adsorbing PFAS in the atmosphere, lacking comprehensive exposure modeling capability across media and paths, resulting in one-sided evaluation results and distorted risk judgments. Secondly, the heterogeneity and spatiotemporal scale inconsistency of multi-source data have not been effectively solved, and existing systems often lack heterogeneous spatiotemporal alignment algorithms specific to PFAS characteristics in the data preprocessing stage, making it difficult to achieve high-quality data fusion and limiting the accuracy and reliability of subsequent models. In addition, traditional risk assessment methods are mostly based on static concentration thresholds or empirical rules, which are difficult to capture the complex migration, transformation and accumulation processes of PFAS in the environment, especially when facing nonlinear, multivariate driven pollution processes, lacking sufficient modeling capability and prediction accuracy.

[0004] The current technology generally lacks the ability to analyze the contribution rate of pollution sources, and cannot effectively identify the relative weight of different source items (such as industrial emissions, surface runoff, and atmospheric deposition) in the overall exposure risk, resulting in difficulty in accurately implementing control measures. In addition, the existing system still mainly uses static charts or text reports in the risk expression method, lacks visual and dynamic risk output mechanism, and is difficult to meet the real-time response needs of environmental supervision. SUMMARY

[0005] In view of the problems existing in the prior art, the present application provides a data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substance, which is used to solve the above technical problems.

[0006] In order to achieve the above-mentioned purpose and other purposes, the technical scheme adopted by the present application is as follows: The present application provides a data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substance, which comprises: A feature database construction module: collecting environmental exposure data of perfluoro / polyfluoroalkyl substance through a multi-source environmental sensor array, performing heterogenous spatio-temporal alignment processing on the collected data, and establishing a PFAS multi-modal exposure feature database; A distribution map acquisition module: based on the PFAS multi-modal exposure feature database, performing pollution source isotope fingerprint analysis, and obtaining source contribution rate distribution maps of three types of pollution sources (industrial emissions, surface runoff, and atmospheric deposition) through a nonlinear source analysis algorithm; A flux matrix generation module: constructing a three-dimensional geographic information dynamic migration model according to the source contribution rate distribution map, combining ground elevation, groundwater flow field, and meteorological parameter data set, performing migration flux evolution calculation of PFAS in soil-water-air three-phase medium, and generating a multi-medium dynamic migration flux matrix; A grade map output module: based on the multi-medium dynamic migration flux matrix, establishing a composite risk assessment model, using a long short-term memory network weighted by an attention mechanism to predict dynamic exposure concentration, combining an ecological toxicity threshold database to perform probability risk assessment, and outputting a spatial gridded risk grade map; An early warning heat map generation module: performing spatio-temporal interpolation optimization on the spatial gridded risk grade map, generating a graded risk early warning heat map, and pushing it to an environmental supervision terminal through a geographic information system service interface.

[0007] The operation logic of the feature database construction module is as follows: The water phase ion selective electrode sensor, the soil solid phase adsorption sensor and the atmospheric gas chromatography sensor are deployed to form a three-dimensional sensing network to synchronously collect original data of water body dissolved PFAS concentration, soil adsorbed PFAS content and atmospheric aerosol PFAS flux; the original data of each sensor is subjected to outlier screening, and the sliding window mean method is used to fill in the missing data section to obtain a standardized time series concentration sequence; The standardized time series concentration sequence is subjected to space-time benchmark alignment, a timestamp marker is marked by a satellite positioning and timing system, and a space coordinate benchmark is unified by a geographic information system to generate a space-time aligned exposure data set; a multi-modal data correlation matrix is established based on the space-time aligned exposure data set, the covariance intensity coefficient between water-soil-air three-phase data is calculated, and the cross-medium correlation characteristics are identified; the internal consistency of the correlation matrix is verified by using the Cronbach coefficient, and the discrete monitoring node data with a coupling degree lower than a set threshold is removed; the verified correlation matrix is subjected to wavelet packet transform filtering processing to eliminate device background noise interference, and a PFAS multi-modal exposure feature database is output.

[0008] A multi-modal data correlation matrix is established based on the space-time aligned exposure data set, the covariance intensity coefficient between water-soil-air three-phase data is calculated, and the cross-medium correlation characteristics are identified; the internal consistency of the correlation matrix is verified by using the Cronbach coefficient, and the discrete monitoring node data with a coupling degree lower than a set threshold is removed, and the specific operation logic is as follows: The water phase concentration, soil adsorption amount and atmospheric flux data in the space-time aligned exposure data set are split into a three-dimensional data matrix according to a monitoring time window, and the matrix dimensions correspond to geographic coordinates, timestamps and medium types respectively; the three-dimensional data matrix is subjected to standardized dimensionless processing, and the water phase concentration unit, soil adsorption amount unit and atmospheric flux unit are uniformly converted into a standard Gini coefficient; a water-soil dual-medium covariance matrix is calculated based on the standardized Gini coefficient, a soil-air dual-medium covariance matrix is calculated synchronously, and two groups of medium covariance intensity distribution maps are obtained; a positive correlation feature section in the covariance intensity distribution map is extracted, a water-soil-air three-phase covariance intensity coefficient is selected according to a preset intensity threshold, and the intensity threshold is dynamically adjusted based on the quartile distance of historical monitoring data; the selected covariance intensity coefficient is subjected to spatial topology analysis, and a multi-modal data correlation matrix containing medium interaction weights is constructed; The three-phase data coupling relationship of each monitoring node is extracted from the multi-modal data correlation matrix, and the internal consistency coefficient of the node is calculated; the internal consistency coefficient is compared with a preset reliability threshold, the reliability threshold is generated by equi-proportional scaling according to the intensity threshold; the monitoring node with an internal consistency coefficient lower than the reliability threshold in three continuous time windows is marked as a discrete node, and a discrete node spatial distribution heat map is generated; the multi-modal data correlation matrix is topologically reconstructed based on the discrete node distribution heat map, and an optimized correlation matrix is generated after removing the abnormal nodes.

[0009] The operation logic of the distribution map acquisition module is: S21: Extracting δ¹³C, δ¹ 8 O isotope ratio and fluorocarbon chain length distribution data from the PFAS multi-modal exposure feature database to generate a multi-dimensional isotope fingerprint feature vector; S22: Constructing an isotope reference fingerprint library of industrial emission sources, surface runoff sources and atmospheric deposition sources, wherein the industrial source fingerprint contains hexafluoropropylene oligomer characteristic peaks, and the surface runoff source fingerprint contains perfluorocarboxylic acid homolog abundance ratio characteristics; S23: Using a hybrid weight algorithm based on Mahalanobis distance improvement to calculate the similarity matching value of the multi-dimensional isotope fingerprint feature vector and the reference fingerprint library; establishing a source contribution rate optimization model by a nonlinear least squares method, and inversely calculating the initial contribution rate of the three types of pollution sources with the concentration data of the monitoring points as the constraint conditions; verifying the initial contribution rate by residual concentration deconvolution, and returning to adjust the mixing weight parameters when the residual error exceeds the set threshold; S24: Generating a spatial interpolation map according to the verified contribution rate data, and outputting the source contribution rate distribution map of the three types of pollution sources, i.e. industrial emissions, surface runoff and atmospheric deposition.

[0010] The operation logic of S23 is: Extracting the industrial emission source characteristic variance matrix, the surface runoff source covariance matrix and the atmospheric deposition source mean vector from the reference fingerprint library as the initial weight parameters; calculating the improved Mahalanobis distance between the generated multi-dimensional isotope fingerprint feature vector and the characteristic matrix of the three types of pollution sources, wherein the improved weight is a linear combination of 0.6 times the source characteristic variance and 0.4 times the monitoring concentration gradient; performing inverse proportional normalization processing on the distance calculation results of the three types of pollution sources to generate three matching values, i.e. industrial emission similarity, surface runoff similarity and atmospheric deposition similarity; establishing a similarity matching matrix containing latitude and longitude coordinates to store the matching value data of each monitoring point and the type of pollution source; Based on the similarity matching matrix, a pollution source contribution degree correlation equation is constructed, and the initial values of the contribution rate weight coefficients of industrial emissions, surface runoff and atmospheric deposition are set to 1 / 3; the obtained real-time monitoring concentration data is used as the boundary condition to establish a concentration constraint equation group; the Levenberg-Marquardt algorithm is applied to iteratively solve the nonlinear equation group to obtain the numerical solution of the initial contribution rate of the three types of pollution sources; the initial contribution rate is subjected to forced normalization processing to make the sum of the contribution rates of the three types of pollution sources constant at 100%; According to the initial contribution rate, the standard deviation residual of the measured concentration and the theoretical concentration of the monitoring points is calculated; the residual is compared with the coupling degree verification threshold, and the parameter adjustment instruction is triggered when the residual exceeds 1.5 times the threshold; the mixing weight proportion is dynamically adjusted according to the residual distribution characteristics; the adjusted weight parameters are used for re-verification and execution until the residual error meets the threshold requirement.

[0011] The steps of the flux matrix generation module are: Based on the source contribution rate distribution map, the range of the three-dimensional modeling space is delineated, and the Kriging interpolation algorithm is used to generate the spatial distribution grid of the initial load of pollutants. The digital elevation model and the groundwater flow field monitoring data are spatially overlaid and analyzed to calculate the terrain slope parameters and the groundwater seepage velocity vector of each grid cell. The rainfall intensity and wind speed data in the meteorological parameters are analyzed, and the atmospheric deposition flux and aerosol diffusion rate are calculated in combination with the covariance intensity coefficient. The three-phase coupled migration equation set is established by the soil pore water convection-adsorption equation, the water body interphase distribution equation, and the atmospheric diffusion equation. The finite volume method is used to spatially discretize the coupled equation set, and the groundwater flow field boundary conditions are set based on the seepage velocity. The preconditioned conjugate gradient method is used to iteratively solve the discretized equation set, and the time-varying distribution data of the flux at each medium interface are output. The soil vertical infiltration flux, water body horizontal migration flux, and atmospheric dry and wet deposition flux data are fused to construct a multi-medium dynamic migration flux matrix containing the time dimension.

[0012] The detailed sub-steps of the flux matrix generation module include: Read the unified geographic coordinate reference data, perform coordinate conversion on the digital elevation model, and generate a standardized terrain elevation matrix. The 3x3 neighborhood window method is used to calculate the terrain slope parameters of each grid cell, the water head gradient value and the aquifer permeability coefficient in the groundwater flow field monitoring data are analyzed, and the seepage velocity vector of each grid cell is calculated using Darcy's law. A three-dimensional spatial parameter matrix is established that includes the slope-seepage velocity correlation; Minute-level rainfall intensity data are extracted from the meteorological parameter set, and the surface runoff scouring coefficient is calculated in combination with the terrain slope parameters. The anemometer monitoring data are analyzed to generate a wind speed vector rose diagram, and the aerosol horizontal diffusion rate is calculated in combination with the covariance intensity coefficient. The atmospheric dry and wet deposition flux values are calculated based on the product relationship between rainfall intensity and wind speed vector; The soil pore water mass conservation equation is established based on the three-dimensional parameter matrix, and the collected soil adsorption coefficient is introduced. The water body interphase distribution equation is constructed, and the initial boundary conditions of the flux matrix are set. The turbulent diffusion equation is established by coupling the calculated atmospheric diffusion rate; the three medium equations are combined into a nonlinear coupled equation set through the Jacobian matrix; The three-dimensional spatial parameter matrix is divided into finite volume units according to the source contribution rate grid size. The Godunov format is used to discretize the convection term, and the Darcy flow boundary conditions are set using the seepage velocity vector. The implicit difference format is used to discretize the adsorption term, and the dimension consistency of the verification threshold is maintained; Set the initial iteration tolerance value as 1 / 10 of the Monte Carlo simulation accuracy; load the slope-seepage correlation matrix as the preconditioning matrix; perform conjugate gradient iteration calculation until the residual reaches the tolerance requirement; output the soil vertical flux, water body horizontal flux and atmospheric deposition flux dataset containing the time stamp.

[0013] The operation logic of the grade map output module is: Construct a PFAS toxicity characteristic knowledge graph containing the semi-lethal concentration of ecologically sensitive species, biological enrichment factor and degradation half-life; configure an attention mechanism weighted LSTM network, set the sampling period, and load the optimized correlation matrix as the initial weight parameter; Input the multi-medium dynamic migration flux matrix into the LSTM network for spatio-temporal sequence learning, output the dynamic exposure concentration distribution prediction data for the next three hydrological years; set the risk judgment threshold based on the toxicity characteristic knowledge graph, simulate the exposure concentration exceedance probability using the Monte Carlo method, and generate a risk probability distribution heat map; perform spatial kriging interpolation optimization on the risk probability distribution heat map, superimpose the source contribution rate distribution layer to generate a gridded risk grade map.

[0014] The operation logic of the early warning heat map generation module is: Obtain the spatial gridded risk grade map data, extract the risk grade code and time stamp sequence of each grid element; perform sub-grid optimization on the spatial grid using the improved inverse distance weighted interpolation algorithm, and perform spatio-temporal three-dimensional interpolation calculation combined with the time decay coefficient of the migration flux; based on the environmental supervision standard, establish a red-orange-yellow-blue four-level gradient color scale mapping table, render the risk data after interpolation optimization into a heat map in the RGB color space; superimpose the source contribution rate distribution layer and the geographic coordinate reference to generate a graded risk early warning heat map.

[0015] As described above, the data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substances provided by the application has at least the following beneficial effects: The data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substances provided by the application, the feature database construction module collects exposure data of PFAS in different media through a multi-source environmental sensor array, and unifies data from different sensors, different times and spatial scales through a heterogeneous spatio-temporal alignment technology, thereby constructing a high-quality PFAS multi-modal exposure feature database. This database breaks through the traditional single data source and low spatio-temporal resolution monitoring method, provides a solid data foundation for subsequent pollution tracing and dynamic modeling, and improves the adaptability of the system to complex environmental exposure scenarios.

[0016] Secondly, the distribution map acquisition module can effectively distinguish various PFAS pollution sources such as industrial emissions, surface runoff and atmospheric deposition by combining isotope fingerprint analysis of pollution sources and nonlinear source analysis algorithm, and quantifies the contribution rate of various source items in different regions. This not only enhances the scientificity and accuracy of pollution tracing, but also provides accurate decision basis for pollution control. Traditional pollution source analysis methods are mostly based on linear assumption, which is difficult to handle the characteristics of complex nonlinear migration and mixing of PFAS in the environment, resulting in large deviation of source identification results and affecting the treatment effect.

[0017] In the flux matrix generation module, a three-dimensional geographic information dynamic migration model is introduced, combined with multi-dimensional environmental factors such as ground elevation, groundwater flow field and meteorological parameters, to simulate the migration flux evolution process of PFAS between soil, water and air three-phase media. By constructing a multi-medium dynamic migration flux matrix, the system can dynamically reflect the migration path and concentration evolution trend of PFAS in the environment, and realize the fine modeling of the spatial diffusion behavior of pollutants. This technical means effectively makes up for the shortcomings of existing models in multi-medium coupling migration modeling, and overcomes the technical defects of traditional static models that cannot accurately reflect the dynamic migration process of pollutants.

[0018] In summary, the present application not only fills the gap in multi-medium dynamic modeling and nonlinear source analysis in the prior art, but also provides strong technical support for environmental pollution control and ecological risk prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0020] Figure 1 The system of the present application is connected to the schematic diagram. DETAILED DESCRIPTION

[0021] The following will combine the above content of the present application, which is only an example and explanation of the concept of the present application. Those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace, as long as they do not deviate from the concept of the present application or exceed the scope defined by the present application, which shall belong to the protection scope of the present application. EMBODIMENT

[0022] Please refer to Figure 1 The data integration risk assessment system for multi-source exposure of perfluoro / polyfluoroalkyl substances is shown in the figure, which comprises: Feature database construction module: Collect environmental exposure data of PFAS through multi-source environmental sensor array, process the collected data for heterogeneous space-time alignment, and establish a PFAS multi-modal exposure feature database; The operation logic of the feature database construction module is as follows: Deploy water-phase ion-selective electrode sensors, soil solid-phase adsorption sensors, and atmospheric gas chromatography sensors to form a three-dimensional sensor network, and synchronously collect original data of water dissolved PFAS concentration, soil adsorbed PFAS content, and atmospheric aerosol PFAS flux; exclude abnormal values from the original data of each sensor, fill in the missing data segments using the sliding window mean method, and obtain standardized time-series concentration sequences; Align the standardized time-series concentration sequences in time and space, mark them with unified timestamps through a satellite positioning and timing system, and unify the spatial coordinate reference through a geographic information system, to generate a space-time aligned exposure data set; based on the space-time aligned exposure data set, establish a multi-modal data correlation matrix, calculate the covariance intensity coefficients between water-soil-air three-phase data, and identify cross-medium correlation characteristics; verify the internal consistency of the correlation matrix using Cronbach's coefficient, and remove the data of discrete monitoring nodes with a coupling degree lower than a set threshold; perform wavelet packet transform filtering processing on the verified correlation matrix to eliminate device background noise interference, and output the PFAS multi-modal exposure feature database.

[0023] The embodiment of the application detects the ion mobility of dissolved perfluoro / polyfluoroalkyl substance in water through the water-phase ion selective electrode sensor deployed in the three-dimensional sensing network, and inverses the microgram-level concentration of the dissolved PFAS by combining the preset electrode response curve; the soil solid-phase adsorption sensor measures the dielectric loss factor of PFAS in soil pore media through the high-frequency dielectric spectrum scanning technology, and calculates the mass adsorption amount of soil adsorbed PFAS based on the dielectric constant offset of the characteristic frequency band; the atmospheric gas chromatography sensor captures aerosol particles through the negative pressure adsorption sampling device, locks the peak position of a specific PFAS homolog by using the retention time of the chromatographic column, and calculates the atmospheric deposition flux by integrating the peak area. For the raw data collected by each sensor, a sliding window monitoring algorithm based on a dynamic standard deviation threshold is used to filter and remove abnormal data points based on three times the moving standard deviation, and the weighted moving average interpolation method is used to complete the missing period for continuous missing, wherein the length of the sliding window is dynamically adjusted according to the sampling frequency of the sensor, so that the standardized time series concentration sequence after processing retains the original fluctuation characteristics to the maximum extent while maintaining data continuity. In the time and space alignment process, the monitoring nodes are synchronized to the millisecond level by the Beidou satellite timing system, and the UTM coordinate conversion technology is used to unify the geographic coordinates of each sensor to the WGS84 ellipsoid reference surface, so as to realize accurate matching of multi-source data in time and space dimensions. When constructing the multi-modal data correlation matrix, first, the water-soil-air three-phase data are resampled according to the hourly average, the covariance intensity coefficient between each pair of monitoring points is calculated, and the sliding correlation coefficient test method is used to identify the statistically significant cross-medium correlation characteristics; in order to verify the internal consistency between the data, the Cronbach's reliability analysis model is applied, the standard deviation ratio of each monitoring node to the overall data set is calculated, and when the node coupling degree is lower than the preset threshold of 0.7, it is determined as a discrete node and is removed. Finally, the decomposition layer number in the wavelet packet transform filtering process is adaptively selected, the optimal decomposition scale is automatically determined according to the power spectrum characteristics of the device background noise, the signal-to-noise separation is realized by reconstructing the high-frequency noise component, and the PFAS multi-modal exposure feature database after denoising is output.

[0024] The multi-modal data correlation matrix is established based on the spatiotemporal alignment exposure data set, the covariance intensity coefficient between the water-soil-air three-phase data is calculated, and the cross-medium correlation characteristics are identified; the internal consistency of the correlation matrix is verified by using the Cronbach's coefficient, and the discrete monitoring node data with a coupling degree lower than the set threshold is removed, and the specific operation logic is as follows: The water phase concentration, soil adsorption amount and atmospheric flux data in the spatiotemporal alignment exposure dataset are split into three-dimensional data matrices according to the monitoring time window, and the matrix dimensions correspond to geographical coordinates, time stamps and medium types respectively; the three-dimensional data matrices are standardized and dimensionless, and the units of water phase concentration, soil adsorption amount and atmospheric flux are uniformly converted into standard Gini coefficients; based on the standardized Gini coefficients, the water-soil double medium covariance matrix is calculated, and the soil-air double medium covariance matrix is calculated synchronously, and two groups of medium inter-covariance intensity distribution maps are obtained; the positive correlation characteristic section in the inter-covariance intensity distribution map is extracted, and the water-soil-air three-phase inter-covariance intensity coefficient is selected according to the preset intensity threshold, wherein the intensity threshold is dynamically adjusted based on the quartile distance of historical monitoring data; the selected inter-covariance intensity coefficient is subjected to spatial topology analysis, and a multi-modal data correlation matrix containing medium interaction weight is constructed; The three-phase data coupling relationship of each monitoring node is extracted from the multi-modal data correlation matrix, and the internal consistency coefficient of the node is calculated; the internal consistency coefficient is compared with a preset reliability threshold, and the reliability threshold is generated by equi-proportion scaling based on the intensity threshold; the monitoring nodes with consistency coefficients lower than the reliability threshold in the continuous three time windows are marked as discrete nodes, and a discrete node spatial distribution heat map is generated; based on the discrete node distribution heat map, the multi-modal data correlation matrix is topologically reconstructed, and an optimized correlation matrix is generated after removing abnormal nodes.

[0025] The embodiment of the application reorganizes the water phase dissolved PFAS concentration, soil adsorption capacity and atmospheric deposition flux in the spatiotemporal alignment exposed data set into a three-dimensional data cube according to geographical grid numbering, wherein the row dimension is the spatial latitude and longitude grid index, the column dimension is the integer timestamp sequence, and the layer dimension corresponds to the water-soil-air medium type identifier, forming an original data matrix with geographical spatiotemporal correlation. The Gini coefficient standardization algorithm is used to perform dimensionless conversion on the three-phase data, the maximum and minimum values of the historical monitoring values of each medium are used to form a normalization reference interval, the normalized Gini coefficient is calculated by mapping the actual monitoring value to the relative position of the interval, and the comparability of different dimension data is ensured. For water-soil double medium covariance analysis, the sliding window covariance calculation method is used in the time dimension, the window length is set to 24 hours to cover the complete day-night cycle, the Gini coefficient sequence of the water phase concentration and the soil adsorption capacity is subjected to sliding correlation operation, the covariance intensity value of each grid point is calculated, and the spatial outlier interference is removed by using the Mahalanobis distance-based anomaly detection mechanism; when the soil-air medium covariance is calculated, the time lag compensation algorithm is introduced, the flux data is subjected to time axis translation alignment processing according to the atmospheric diffusion velocity, and the spatiotemporal consistency of the correlation analysis is ensured. In the generation process of the covariance intensity distribution map, the density peak clustering algorithm is used to identify the covariance hot area, the upper quartile of the covariance value is used as a dynamic threshold to select a significant correlation section, and the transfer path weight between the water-soil-air three phases is analyzed through a spatial topological network model, and a multi-modal correlation matrix with medium interaction intensity is constructed. In the data consistency verification stage, for the Gini coefficient sequence of each monitoring node in the three media, first, the cross-medium standard deviation ratio is calculated as an internal consistency indicator, and then a confidence threshold band is generated according to the covariance intensity threshold ratio - when the internal consistency indicator of the node is lower than the threshold for more than three consecutive monitoring windows, it is determined that the node is a spatial discrete node. The spatial distribution of the discrete node is generated into a heat map through the kernel density estimation algorithm, the abnormal aggregation area is identified through the Delaunay triangle subdivision technology, and the topological reconstruction of the correlation matrix is carried out through the maximum flow minimum cut algorithm in graph theory, so that the abnormal node and its associated edges are removed on the premise of maintaining the connectivity of the spatial network, and finally an optimized correlation matrix verified for spatial self-consistency is formed.

[0026] The distribution map acquisition module: based on the PFAS multi-modal exposure feature database, the isotope fingerprint of the pollution source is analyzed, and the source contribution rate distribution map of the industrial emission, surface runoff and atmospheric deposition of three types of pollution sources is acquired through a nonlinear source analysis algorithm; The operation logic of the distribution map acquisition module is: S21: extracting δ¹³C, δ¹ 8 O isotope ratio and fluorocarbon chain length distribution data from the PFAS multi-modal exposure feature database to generate a multi-dimensional isotope fingerprint feature vector; S22: Constructing an isotopic reference fingerprint library of industrial emission sources, surface runoff sources and atmospheric deposition sources, wherein the industrial source fingerprint contains characteristic peaks of hexafluoropropylene oligomers, and the surface runoff source fingerprint contains characteristic of perfluorocarboxylic acid homolog abundance ratio; S23: Using a hybrid weight algorithm based on Mahalanobis distance improvement, calculating the similarity matching value of the multi-dimensional isotopic fingerprint characteristic vector and the reference fingerprint library; establishing a source contribution rate optimization model by a nonlinear least squares method, and inverting the initial contribution rate of the three types of pollution sources with the monitoring point concentration data as the constraint condition; verifying the initial contribution rate by residual concentration deconvolution, and returning to adjust the hybrid weight parameter when the residual error exceeds the set threshold; S24: Generating a spatial interpolation map according to the verified contribution rate data, and outputting the source contribution rate distribution map of the three types of pollution sources, namely, industrial emission, surface runoff and atmospheric deposition.

[0027] In the embodiment of the present application, δ¹³C, δ¹ 8After eliminating the instrument analysis error by standardization processing, the isotope abundance ratio and fluorocarbon chain length distribution spectrum data were fused with the isotope characteristics and carbon number distribution histogram of fluorocarbon chain according to the spatial encoding of monitoring points to form a multi-dimensional fingerprint vector containing isotope composition and molecular structure characteristics. When constructing the reference fingerprint library for industrial emission sources, the characteristic ion peaks of oligomers of hexafluoropropylene oxide were captured by gas chromatography-mass spectrometry technology, and the industrial source characteristic spectrum was determined by retention time locking and fragment ion abundance ratio, and the characteristic was strengthened by combining the proportion of long-chain PFAS (C8-C12) in historical monitoring data around the factory. The construction of the fingerprint of the surface runoff source was carried out by calculating the carbon number distribution probability of perfluorocarboxylic acid homologues under different hydrological scenarios, taking the C6 / C8 carboxylic acid concentration ratio as the core characteristic quantity, and fusing the abnormal peak value of the odd carbon homologues generated in the PFAS aging process in the runoff sediments to form a characteristic template. In the mixed weight similarity calculation stage, first, the similarity of the isotope characteristic vector was matched by improving the Mahalanobis distance algorithm, in which the oligomer characteristic peak accounted for 60% of the matching weight of the industrial source, the isotope composition accounted for 30%, and the carbon chain length accounted for 10%; the matching weight of the surface runoff source was adjusted dynamically, and the weight coefficient of the C6 / C8 carboxylic acid ratio was automatically increased according to the seasonal change of hydrology. In the construction of the nonlinear source analysis model, the total concentration measured at the monitoring point was taken as the constraint condition to establish a ternary contribution equation group of industry-surface runoff-atmospheric deposition, and the Levenberg-Marquardt optimization algorithm was used to inverse the contribution proportion of each pollution source, and the uncertainty of the contribution rate was calculated by Monte Carlo sampling. For the model inversion results, the Fourier deconvolution technology was used to compare the frequency spectrum of the predicted concentration with the measured residual concentration, and when the residual error spectrum exceeded the threshold in the characteristic frequency band, the weight parameter adjustment mechanism was automatically triggered-the matching weight of the oligomer characteristic peak of the industrial source was optimized by the gradient descent method, and the calculation was iterated again until the accuracy requirement was met. After the source contribution rate data verified by the final results were obtained, the Kriging spatial interpolation algorithm considering the terrain resistance effect was used to fuse the slope resistance factor of the digital elevation model to correct the interpolation results, and the source contribution rate distribution map was generated.

[0028] The operation logic of S23 is: The characteristic variance matrix of industrial emission sources, the covariance matrix of surface runoff sources, and the mean vector of atmospheric deposition sources were extracted from the reference fingerprint library as the initial weight parameters; the generated multi-dimensional isotope fingerprint characteristic vector was calculated with the improved Mahalanobis distance with the characteristic matrix of the three types of pollution sources, wherein the improved weight was a linear combination of 0.6 times the source characteristic variance and 0.4 times the monitoring concentration gradient; the distance calculation results of the three types of pollution sources were normalized in inverse proportion to generate three matching values of industrial emission similarity, surface runoff similarity, and atmospheric deposition similarity; a similarity matching matrix containing latitude and longitude coordinates was established to store the matching value data of each monitoring point and the type of pollution source; The contribution degree correlation equation of the pollution source is constructed based on the similarity matching matrix, and the contribution rate weight coefficient initial values of industrial emission, surface runoff and atmospheric deposition are all set as 1 / 3; the obtained real-time monitoring concentration data is taken as a boundary condition to establish a concentration constraint equation group; the Levenberg-Marquardt algorithm is applied to iteratively solve the nonlinear equation group to obtain the initial contribution rate numerical solution of the three types of pollution sources; the initial contribution rate is subjected to forced normalization processing, so that the sum of the contribution rates of the three types of pollution sources is constant at 100%; According to the initial contribution rate, the theoretical concentration distribution is reconstructed, the standard deviation residual of the measured concentration and the theoretical concentration of the monitoring point is calculated, the residual is compared with the coupling degree verification threshold, and when the residual exceeds 1.5 times of the threshold, a parameter adjustment instruction is triggered; the mixed weight proportion is dynamically adjusted according to the residual distribution characteristics; the adjusted weight parameters are used for re-verification and execution until the residual error reaches the threshold requirement.

[0029] The embodiment of the application extracts the intensity variance matrix of hexafluoropropylene oxide oligomer characteristic peaks from the industrial emission source reference fingerprint, simultaneously obtains the covariance matrix of perfluorocarboxylic acid homolog abundance ratio in the surface runoff source reference fingerprint, and the mean vector of fluorocarbon chain length distribution in the atmospheric deposition source, which together constitute the initial parameter set for the calculation of the mixing weight. When using the improved Mahalanobis distance algorithm, for the multi-dimensional isotope fingerprint characteristic vector of each monitoring point, first calculate the standardized deviation degree of the characteristic vector from the industrial source variance matrix, and at the same time, combine the concentration gradient change trend within 0.5 km range of the upstream and downstream of the monitoring point, assign 60% weight to the inherent property of the source characteristics, and 40% weight to the local concentration spatial gradient characteristics, to form a distance measurement that takes into account stability and dynamics. When performing inverse proportional normalization processing on the distance values of the three types of pollution sources calculated, the dynamic baseline correction technology is used, taking the natural background value of the hydrological unit where each monitoring point is located as the normalization base, to ensure the comparability of the similarity of industrial emissions, surface runoff and atmospheric deposition. The similarity matching matrix constructed uses a four-dimensional data structure for storage, including longitude, latitude, elevation and time stamp index, to realize the organic integration of spatial stereoscopic grid and time series. When establishing the correlation equation of the contribution rate of the pollution source, the initial equal weight coefficient is embedded into the diagonal elements of the Jacobian matrix by constructing the three-party weight balance constraint condition, and the boundary contraction method is used to control the solution space range of the iteration process. In the implementation of the Levenberg-Marquardt optimization algorithm, an adaptive adjustment mechanism is set for the iteration step size, and when the residual error reduction rate is lower than the threshold value, the damping factor is automatically enlarged to prevent oscillation divergence. The contribution rate forced normalization processing is realized by introducing the Lagrange multiplier method, which maximizes the preservation of the original source analysis results while maintaining the summation constraint. In the residual verification stage, the spectral decomposition technology is used to separate the multi-scale difference characteristics of the theoretical concentration and the measured concentration, and special attention is paid to the energy difference of the frequency band corresponding to the oligomer characteristic peaks of the industrial source. When the residual energy of a specific frequency band exceeds the threshold value, the proportion coefficient of the concentration gradient in the mixing weight is dynamically adjusted by the gradient reverse propagation algorithm, and the optimization is increased by 0.1 as a step until the error converges.

[0030] The flux matrix generation module: according to the source contribution rate distribution map, a three-dimensional geographical information dynamic migration model is constructed, combined with the ground elevation, groundwater flow field and meteorological parameter data set, the migration flux evolution calculation of PFAS in the soil-water-air three-phase medium is performed, and a multi-medium dynamic migration flux matrix is generated; The steps of the flux matrix generation module are: Based on the source contribution rate distribution map, the three-dimensional modeling space range was delineated, and the Kriging interpolation algorithm was used to generate the spatial distribution grid of the initial load of pollutants. The digital elevation model and groundwater flow field monitoring data were overlaid and analyzed, and the terrain slope parameters and groundwater seepage velocity vector of each grid unit were calculated. The rainfall intensity and wind speed data in the meteorological parameters were analyzed, and the atmospheric deposition flux and aerosol diffusion rate were calculated combined with the covariance intensity coefficient. The three-phase coupling migration equation set was established by the soil pore water convection-adsorption equation, the water body interphase distribution equation and the atmospheric diffusion equation. The finite volume method was used for spatial discretization of the coupling equation set, and the groundwater flow field boundary conditions were set based on the seepage velocity. The preconditioned conjugate gradient method was used to iteratively solve the discretized equation set, and the time-varying distribution data of the flux of each medium interface were output. The soil vertical filtration flux, water body horizontal migration flux and atmospheric dry and wet deposition flux data were fused to construct a multi-medium dynamic migration flux matrix containing time dimension.

[0031] The embodiment of the present application generates the initial load grid of pollutants based on the spatial hot area boundary of the source contribution rate distribution map by using the adaptive anisotropic Kriging interpolation algorithm. The interpolation process dynamically adjusts the variation function parameters according to the spatial heterogeneity of the contribution rate distribution, automatically enhances the smoothing effect of the radial basis function when the density of the monitoring points is lower than the threshold value, and ensures the natural transition of the concentration gradient of the industrial hot spot area and the non-point source diffusion area. The digital elevation model data is subjected to pyramid hierarchical processing, the terrain slope vector field is extracted at the kilometer resolution layer, and the grid slope value is calculated at the hundred-meter resolution layer by using the bidirectional slope gradient algorithm, so as to realize the multi-scale fusion of the terrain features. The groundwater flow field monitoring data is subjected to three-dimensional seepage velocity reconstruction by using the Darcy law, wherein the permeability coefficient matrix is fused by using the Kalman filtering technology to combine the borehole permeability and the geological radar inversion data, and an anisotropic seepage velocity vector field is generated at a 10-meter grid scale. When processing the meteorological parameters, the minute-level rainfall intensity data is subjected to three times of spline interpolation to realize the spatial downscaling processing, the spatial redistribution of the wind speed vector is calculated based on the conservation principle of turbulent kinetic energy, and the spatio-temporal joint probability model of the atmospheric deposition flux is constructed by combining the verified covariance intensity coefficient. In the process of establishing the three-phase coupled equation set, the soil pore water convection term is discretized by using the explicit upwind format to maintain the calculation stability, and the nonlinear term of the Freundlich isotherm adsorption model is introduced into the adsorption term; the construction of the water body phase distribution equation is based on the liquid-gas interface concentration gradient, and the mass transfer coefficient is calculated by using the double membrane theory; the eddy diffusion term of the atmospheric diffusion equation is dynamically corrected by combining the turbulent fluctuation parameters of the mesoscale meteorological model. When the finite volume method is discretized, the three-dimensional calculation domain is subjected to unstructured grid partitioning according to the terrain slope, the grid is densified in the steep slope area, and the seepage velocity vector field is extended to the model boundary condition setting by using the extrapolation method. In the iterative solving process of the preconditioned conjugate gradient method, the weighted Jacobian preconditioning matrix based on the slope parameter is constructed to accelerate the convergence, and the calculation is automatically terminated when the cosine similarity of the adjacent two iteration solution vectors is lower than 0.001. The finally generated multi-medium dynamic migration flux matrix adopts the space-time cubic data structure, the geographical dimension adopts the UTM coordinate index, the time dimension is stored in layers at an interval of 15 minutes, and the medium dimension is divided into three dimension systems of soil vertical flux (mg / m2·h), water body horizontal flux (μg / L·km) and atmospheric flux (ng / m3·s), and the migration flux database is formed by lossless compression encoding.

[0032] The detailed sub-steps of the flux matrix generation module include: Read the unified geographical coordinate reference data, perform coordinate conversion on the digital elevation model, and generate a standardized terrain elevation matrix; calculate the terrain slope parameters of each grid cell by using the 3*3 neighborhood window method, analyze the water head gradient value and the aquifer permeability coefficient in the groundwater flow field monitoring data, and calculate the seepage velocity vector of each grid cell by using the Darcy law; establish a three-dimensional spatial parameter matrix containing the slope-seepage velocity correlation; Minute-level rainfall intensity data was extracted from the meteorological parameter set, and the surface runoff scouring coefficient was calculated by combining the terrain slope parameter. The wind speed vector rose diagram was generated by analyzing the anemometer monitoring data, and the aerosol horizontal diffusion rate was calculated by matching the covariance intensity coefficient. The atmospheric dry and wet deposition flux value was calculated based on the product relationship of rainfall intensity and wind speed vector; The soil pore water mass conservation equation was established based on the three-dimensional parameter matrix, and the collected soil adsorption coefficient was introduced. The water body interphase distribution equation was constructed, and the initial boundary conditions of the flux matrix were set. The turbulent diffusion equation was established by coupling the calculated atmospheric diffusion rate. The three medium equations were combined into a nonlinear coupled equation set through the Jacobian matrix; The three-dimensional space parameter matrix was divided into finite volume units according to the source contribution rate grid size. The Godunov format was applied to discretize the convection term, and the seepage velocity vector was used to set the Darcy flow boundary condition. The implicit difference format was used to discretize the adsorption term, and the dimensional consistency of the verification threshold was maintained; The initial iteration tolerance value was set to 1 / 10 of the Monte Carlo simulation accuracy. The slope-seepage correlation matrix was loaded as a preconditioning submatrix. The conjugate gradient iteration calculation was performed until the residual reached the tolerance requirement. The soil vertical flux, water body horizontal flux, and atmospheric deposition flux data sets containing time stamps were output.

[0033] The embodiment of the application performs spatial reference system conversion on geographic coordinate reference data, unifies the original coordinate system of the digital elevation model to the national geodetic coordinate system through seven-parameter Bursa transformation, and generates a standardized terrain elevation matrix eliminating projection distortion. Based on the elevation matrix, the terrain slope parameter is calculated by using an improved 3*3 neighborhood gradient algorithm: the x and y direction terrain gradient components are extracted by the Sobel operator, the vector sum of the two components is divided by the grid resolution to obtain the slope value, and at the same time, the Gauss smoothing filter is implemented on the region with a slope exceeding 45 degrees to eliminate local mutation noise. In the analysis of the groundwater flow field parameter, the effective water head gradient value is separated from the monitoring data by using the adaptive threshold segmentation technology, the seepage coefficient field of the aquifer is inversed by using the ground penetrating radar, and the seepage velocity vector is calculated by using the Darcy's law considering anisotropy: the full-field average seepage coefficient is used for the component along the main axis direction of the seepage, and the transverse seepage attenuation factor is introduced for the component along the secondary axis direction, so as to form a three-dimensional velocity field accurately representing the flow state of the groundwater. In the meteorological parameter processing stage, the time series piecewise linear fitting is implemented on the minute-level rainfall data, the standardized terrain slope parameter is combined, the scouring coefficient of each grid unit is calculated by using the slope runoff scouring model, and the mud resuspension effect correction factor is automatically triggered in the region with a slope greater than 5 degrees. The construction of the wind speed vector rose diagram adopts the 16-azimuth angle analysis method, after the matching and verification of the covariance intensity coefficient, the three-dimensional wind speed field is reconstructed by using the kinetic energy conservation formula of fluid mechanics, and finally the aerosol diffusion rate containing the particle size distribution effect is calculated by using the vertical flux integration method. In the construction of the soil pore water mass conservation equation, the terrain slope parameter is used as the gravity driving term coefficient, the soil adsorption coefficient is spatially interpolated according to the organic matter content, and the retardation factor is introduced to correct the dynamic adsorption process. The water body interphase distribution equation adopts the double membrane theory to construct the mass transfer model, and the initial boundary conditions are set and dynamically optimized in combination with the statistical eigenvalue of the previous flux matrix. In the construction of the atmospheric turbulent diffusion equation, the atmospheric stability grade is judged according to the Richardson number, and the turbulent diffusion coefficient parameterization scheme is dynamically adjusted. When the finite volume is discretized, the slope limiter is introduced into the numerical flux calculation of the Godunov format, the second-order reconstruction is used in the steep slope area to ensure the calculation stability, and the seepage velocity input of the Darcy flow boundary condition is realized by using the bilinear interpolation to realize the spatial matching. When the adsorption term is processed by using the implicit difference, the nonlinear equation is solved by using the Newton iteration method, and the discrete process strictly maintains the same microgram per square meter per hour dimension system as the verification threshold. In the precondition conjugate gradient solving stage, the slope-seepage correlation matrix is decomposed into a block diagonal matrix as a preconditioning sub, the iteration tolerance value is dynamically adjusted according to the concentration uncertainty output by the Monte Carlo simulation, and the convergence is determined when the cosine similarity of the adjacent iteration step solution vectors exceeds 0.999.

[0034] Grade map output module: based on the multi-medium dynamic migration flux matrix, a composite risk assessment model is established, the long short-term memory network weighted by attention mechanism is used for dynamic exposure concentration prediction, the ecological toxicity threshold database is used for probability risk assessment, and the spatial gridded risk grade map is output; The operation logic of the grade map output module is: A PFAS toxicity characteristic knowledge graph containing the semi-lethal concentration of ecologically sensitive species, biological enrichment factor and degradation half-life is constructed; an attention mechanism weighted LSTM network is configured, a sampling period is set, and the optimized correlation matrix is loaded as the initial weight parameter; The multi-medium dynamic migration flux matrix is input into the LSTM network for spatio-temporal sequence learning, and the dynamic exposure concentration distribution prediction data of the next three hydrological years is output; based on the toxicity characteristic knowledge graph, the risk judgment threshold is set, the Monte Carlo method is used to simulate the probability of exceeding the exposure concentration, and the risk probability distribution heat map is generated; the risk probability distribution heat map is optimized by spatial kriging interpolation, and the gridded risk grade map is generated by superimposing the source contribution rate distribution layer.

[0035] Early warning heat map generation module: the spatial gridded risk grade map is optimized by spatio-temporal interpolation, a graded risk early warning heat map is generated, and the geographic information system service interface is pushed to the environmental supervision terminal.

[0036] The operation logic of the early warning heat map generation module is: The spatial gridded risk grade map data is obtained, the risk grade code and timestamp sequence of each grid element are extracted; the improved inverse distance weighted interpolation algorithm is used for sub-grid optimization of spatial gridding, and spatio-temporal three-dimensional interpolation calculation is performed combined with the time decay coefficient of migration flux; based on the environmental supervision standard, a red-orange-yellow-blue four-level gradient color mapping table is established, and the risk data after interpolation optimization is rendered into a heat map in the RGB color space; the source contribution rate distribution layer and geographic coordinate reference are superimposed to generate a graded risk early warning heat map.

[0037] It should be noted that the interval and threshold size are set for ease of comparison, and the size of the threshold depends on the number of sample data and the base number set by the person skilled in the art for each group of sample data, as long as it does not affect the proportional relationship between the parameters and the quantized values. And the above formula is a dimensionless calculation of the numerical value, and the formula is obtained by software simulation of a large amount of data to obtain a formula closest to the actual situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0038] It should be understood that the size of the sequence number of the above processes does not mean the order of execution in various embodiments of the present application, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0039] It should be understood that determining B according to A does not mean that B is determined only according to A, but B can also be determined according to A and / or other information.

[0040] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0041] Finally: the above is only a preferred embodiment of the present application and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances, characterized in that, include: Feature database construction module: Collects environmental exposure data of perfluorinated / polyfluoroalkyl substances through a multi-source environmental sensor array, performs heterogeneous spatiotemporal alignment processing on the collected data, and establishes a PFAS multimodal exposure feature database; Distribution map acquisition module: Based on the PFAS multimodal exposure feature database, the isotopic fingerprint of pollution sources is analyzed, and the source contribution rate distribution map of three types of pollution sources, namely industrial emissions, surface runoff and atmospheric deposition, is obtained through nonlinear source apportionment algorithm. Flux matrix generation module: Construct a three-dimensional geographic information dynamic migration model based on the source contribution rate distribution map, and combine it with ground elevation, groundwater flow field and meteorological parameter datasets to perform PFAS migration flux evolution calculation in soil-water-atmosphere three-phase media to generate a multi-media dynamic migration flux matrix. The risk level map output module establishes a composite risk assessment model based on a multi-media dynamic migration flux matrix, uses an attention-weighted long short-term memory network to predict dynamic exposure concentrations, combines an ecotoxicity threshold database to perform probabilistic risk assessment, and outputs a spatial gridded risk level map. Early warning heat map generation module: Performs spatiotemporal interpolation optimization on the spatial gridded risk level map to generate a graded risk early warning heat map, and pushes it to the environmental supervision terminal through the geographic information system service interface.

2. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 1, characterized in that, The operation logic of the feature database construction module is as follows: A three-dimensional sensing network consisting of an aqueous ion-selective electrode sensor, a soil solid-phase adsorption sensor, and an atmospheric gas chromatography sensor was deployed to simultaneously collect raw data on dissolved PFAS concentration in water, adsorbed PFAS content in soil, and atmospheric aerosol PFAS flux. Outliers were screened out of the raw data from each sensor, and missing data segments were filled using the sliding window mean method to obtain a standardized time-series concentration sequence. The standardized time-series concentration series is aligned with a spatiotemporal reference by using a unified timestamp marking system through a satellite positioning and timing system and a unified spatial coordinate reference through a geographic information system to generate a spatiotemporally aligned exposure dataset. A multimodal data association matrix was established based on the spatiotemporal aligned exposure dataset, and the covariance intensity coefficient among the water-soil-air three-phase data was calculated to identify cross-media correlation characteristics. The internal consistency of the correlation matrix is ​​verified using Cronbach's coefficients, and discrete monitoring node data with coupling degrees below a set threshold are removed. Wavelet packet transform filtering is applied to the verified correlation matrix to eliminate background noise interference from the equipment, and the PFAS multimodal exposure feature database is output.

3. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 2, characterized in that, A multimodal data association matrix was established based on the spatiotemporal aligned exposure dataset, and the covariance intensity coefficient among the water-soil-air three-phase data was calculated to identify cross-media correlation characteristics. The internal consistency of the correlation matrix is ​​verified using Cronbach's coefficients, and discrete monitoring node data with coupling degrees below a set threshold are removed. The specific operation logic is as follows: The water phase concentration, soil adsorption, and atmospheric flux data in the spatiotemporally aligned exposure dataset were split into three-dimensional data matrices according to the monitoring time window. The matrix dimensions correspond to geographic coordinates, timestamps, and media types, respectively. The three-dimensional data matrices were standardized and dimensionless, and the units of water phase concentration, soil adsorption, and atmospheric flux were uniformly converted to the standard Gini coefficient. The water-soil dual-media covariance matrix was calculated based on the standardized Gini coefficient, and the soil-air dual-media covariance matrix was calculated simultaneously to obtain the covariance intensity distribution map between the two media. Positively correlated feature segments in the covariance intensity distribution map were extracted, and the water-soil-air three-phase covariance intensity coefficients were selected according to the preset intensity threshold, where the intensity threshold was dynamically adjusted based on the interquartile range of historical monitoring data. Spatial topology analysis was performed on the selected covariance intensity coefficients to construct a multimodal data association matrix including media interaction weights. The three-phase data coupling relationship of each monitoring node is extracted from the multimodal data association matrix, and the internal consistency coefficient of the node is calculated. The internal consistency coefficient is compared with the preset confidence threshold, which is generated by scaling the intensity threshold proportionally. Monitoring nodes with consistency coefficients lower than the confidence threshold within three consecutive time windows are marked as discrete nodes, and a spatial distribution heat map of discrete nodes is generated. The multimodal data association matrix is ​​reconstructed based on the discrete node distribution heat map, and an optimized association matrix is ​​generated after removing abnormal nodes.

4. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 1, characterized in that, The operation logic of the distribution map acquisition module is as follows: S21: Extract δ¹³C and δ¹ from the PFAS multimodal exposure feature database. 8 O isotope ratio and fluorocarbon chain length distribution data are used to generate a multidimensional isotope fingerprint feature vector. S22: Construct an isotopic reference fingerprint library for industrial emission sources, surface runoff sources, and atmospheric deposition sources. The industrial source fingerprint includes characteristic peaks of hexafluoropropylene oxide oligomers, and the surface runoff source fingerprint includes abundance ratio characteristics of perfluorocarboxylic acid homologues. S23: A hybrid weighting algorithm based on Mahalanobis distance is used to calculate the similarity matching value between the multidimensional isotopic fingerprint feature vector and the reference fingerprint database; a source contribution rate optimization model is established by nonlinear least squares method, and the initial contribution rate of the three types of pollution sources is inverted with the concentration data of monitoring points as constraints; the initial contribution rate is verified by residual concentration deconvolution, and the hybrid weighting parameters are adjusted when the residual error exceeds the set threshold. S24: Generate a spatial interpolation map based on the verified contribution rate data, and output the source contribution rate distribution maps of three types of pollution sources: industrial emissions, surface runoff, and atmospheric deposition.

5. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 4, characterized in that, The operation logic of S23 is as follows: The feature variance matrix of industrial emission sources, the covariance matrix of surface runoff sources, and the mean vector of atmospheric deposition sources are extracted from the reference fingerprint database as initial weight parameters. The generated multidimensional isotopic fingerprint feature vector is then used to calculate the improved Mahalanobis distance with the feature matrices of the three types of pollution sources. The improved weight is a linear combination of 0.6 times the source feature variance and 0.4 times the monitoring concentration gradient. The distance calculation results of the three types of pollution sources are then inversely normalized to generate three matching values: industrial emission similarity, surface runoff similarity, and atmospheric deposition similarity. A similarity matching matrix containing latitude and longitude coordinates is established to store the matching value data between each monitoring point and the pollution source type. A pollution source contribution correlation equation was constructed based on a similarity matching matrix, with the initial weighting coefficients for industrial emissions, surface runoff, and atmospheric deposition all set to 1 / 3. Real-time monitoring concentration data was used as boundary conditions to establish a set of concentration constraint equations. The Levenberg-Marquardt algorithm was applied to iteratively solve the nonlinear equations, obtaining numerical solutions for the initial contribution rates of the three pollution sources. The initial contribution rates were then forcibly normalized to ensure the sum of the contribution rates of the three pollution sources was constant at 100%. The theoretical concentration distribution is reconstructed based on the initial contribution rate, and the standard deviation residual between the measured concentration and the theoretical concentration at the monitoring point is calculated. The residual is compared with the coupling degree verification threshold. When the residual exceeds 1.5 times the threshold, a parameter adjustment command is triggered. The mixing weight ratio is dynamically adjusted according to the residual distribution characteristics. The adjusted weight parameters are used to re-verify and execute until the residual error reaches the threshold requirement.

6. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 1, characterized in that, The steps of the flux matrix generation module are as follows: The spatial range of the 3D modeling is delineated based on the source contribution rate distribution map. The spatial distribution grid of the initial pollutant load is generated using the Kriging interpolation algorithm. The digital elevation model and groundwater flow field monitoring data are spatially overlaid and analyzed to calculate the topographic slope parameters and groundwater seepage velocity vectors of each grid cell. The rainfall intensity and wind speed data in the meteorological parameters are analyzed, and the atmospheric deposition flux and aerosol diffusion rate are calculated by combining the covariance intensity coefficient. A three-phase coupled migration equation set consisting of soil pore water convection-adsorption equation, water phase distribution equation, and atmospheric diffusion equation is established. The coupled equation set is spatially discretized using the finite volume method, and the groundwater flow field boundary conditions are set based on the seepage velocity. The discretized equation set is iteratively solved using the precondition conjugate gradient method, and the time-varying flux distribution data of each medium interface are output. The soil vertical infiltration flux, water horizontal migration flux, and atmospheric dry and wet deposition flux data are integrated to construct a multi-media dynamic migration flux matrix with time dimension.

7. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 6, characterized in that, The refinement sub-steps of the flux matrix generation module include: Read unified geographic coordinate benchmark data, perform coordinate transformation on the digital elevation model, and generate a standardized terrain elevation matrix; use the 3×3 neighborhood window method to calculate the terrain slope parameters of each grid cell, analyze the hydraulic gradient value and aquifer permeability coefficient in the groundwater flow field monitoring data, apply Darcy's law to calculate the seepage velocity vector of each grid cell; and establish a three-dimensional spatial parameter matrix that includes the slope-seepage velocity correlation. Minute-level rainfall intensity data are extracted from meteorological parameters and combined with topographic slope parameters to calculate the surface runoff erosion coefficient; wind speed vector rose diagrams are generated by analyzing anemometer monitoring data and the covariance intensity coefficient is matched to calculate the horizontal diffusion rate of aerosols; atmospheric dry and wet deposition flux values ​​are calculated based on the product relationship between rainfall intensity and wind speed vector. Based on the three-dimensional parameter matrix, a soil pore water mass conservation equation is established, and the collected soil adsorption coefficient is introduced; a water phase distribution equation is constructed, and the initial boundary conditions of the flux matrix are set; the atmospheric diffusion rate is coupled to establish a turbulent diffusion equation; the three medium equations are combined into a nonlinear coupled equation system through the Jacobian matrix. The three-dimensional spatial parameter matrix is ​​divided into finite volume units according to the source contribution rate grid size; the convection term is discretized using the Godunov scheme, and the Darcy flow boundary conditions are set using the seepage velocity vector; the adsorption term is discretized using an implicit difference scheme to maintain the dimensional consistency of the verification threshold. The initial iteration tolerance is set to 1 / 10 of the Monte Carlo simulation accuracy; the slope-seepage correlation matrix is ​​loaded as a precondition submatrix; conjugate gradient iteration calculation is performed until the residual reaches the tolerance requirement; the output datasets of soil vertical flux, water horizontal flux and atmospheric deposition flux containing timestamps are output.

8. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 1, characterized in that, The operation logic of the gradation map output module is as follows: A knowledge graph of PFAS toxicity characteristics, including the half-lethal concentration, bioaccumulation factors, and degradation half-life of ecologically sensitive species, was constructed. An attention-weighted LSTM network was configured, the sampling period was set, and the optimized association matrix was loaded as the initial weight parameters. The multi-media dynamic migration flux matrix is ​​input into an LSTM network for spatiotemporal sequence learning, outputting dynamic exposure concentration distribution prediction data for the next three hydrological years. Risk judgment thresholds are set based on toxicity feature knowledge graphs, and the Monte Carlo method is used to simulate the probability of exceeding exposure concentration standards, generating a risk probability distribution heatmap. Spatial kriging interpolation optimization is performed on the risk probability distribution heatmap, and a source contribution rate distribution layer is superimposed to generate a gridded risk level map.

9. The data integration and risk assessment system for multi-source exposure to perfluorinated / polyfluoroalkyl substances according to claim 1, characterized in that, The operation logic of the early warning heatmap generation module is as follows: Acquire spatial gridded risk level map data, extract the risk level code and timestamp sequence of each grid cell; use an improved inverse distance weighted interpolation algorithm to perform sub-grid optimization of the spatial grid, and perform spatiotemporal three-dimensional interpolation calculation in combination with the time decay coefficient of migration flux; establish a red-orange-yellow-blue four-level gradient color scale mapping table based on environmental regulatory standards, and render the interpolated and optimized risk data into a heat map in RGB color space; By overlaying the source contribution rate distribution layer and geographic coordinate benchmark, a graded risk warning heat map is generated.

Images