Industrial site soil and groundwater pollution intelligent monitoring and risk assessment system

Abstract

The application discloses an industrial site soil and groundwater pollution intelligent monitoring and risk assessment system, which collects soil and groundwater monitoring data of an industrial site, combines with industrial site characteristic parameters to perform multi-source data space-time registration and characteristic level fusion, establishes a mobile-immobile dual-domain migration model, obtains mobile and immobile domain concentration field data, further generates soil-groundwater interface flux data, combines with historical production layout information of the industrial site to obtain pollution source position probability distribution and intensity confidence, inputs the mobile domain concentration field as a state vector, the interface flux as an observation vector, and the pollution source position probability distribution and intensity confidence as a control vector, performs Kalman filtering recursion, and outputs a pollution plume migration state estimation vector. The application can accurately depict the non-Fickian diffusion behavior of pollutants between preferential flow channels and matrix regions in a complex industrial site, and significantly improves the accuracy of pollutant migration prediction under complex heterogeneous site conditions.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring and pollution control technology, specifically relating to an intelligent monitoring and risk assessment system for soil and groundwater pollution in industrial heritage sites. Background Technology

[0002] Existing soil and groundwater pollution monitoring technologies suffer from the following shortcomings: First, most existing technologies employ the homogeneous medium assumption based on the convection-dispersion equation, treating soil as an ideal continuous porous medium for pollutant migration simulation. However, industrial sites typically contain heterogeneous structures such as underground pipe trenches, foundation remnants, and landfill materials, leading to significant non-Fick diffusion characteristics of pollutants between preferential flow channels and matrix regions. Traditional models cannot accurately describe this dual-domain migration behavior. Second, existing technologies often separate soil pollution monitoring from groundwater pollution monitoring, lacking a quantitative characterization of the mass flux exchange process at the vadose zone-saturation zone interface, making it difficult to reveal the key mechanisms of pollutant cross-media migration. Third, existing pollution source identification methods mainly rely on concentration threshold judgments or simple spatial interpolation, lacking probabilistic inversion methods based on physical mechanisms, resulting in insufficient accuracy in pollution source tracing and an inability to provide confidence assessments of source strength. Furthermore, existing technologies primarily employ static or quasi-static analysis methods for pollution state prediction, lacking the ability to integrate real-time observation data for dynamic state estimation, thus failing to achieve optimal estimation and prediction of pollution plume evolution.

[0003] The aforementioned technical deficiencies result in existing industrial site pollution monitoring systems exhibiting problems such as large deviations in migration prediction, inaccurate source tracing and location, and lagging risk assessment when facing complex site conditions. These shortcomings make it difficult to provide timely and reliable technical support for contaminated site remediation decisions. Therefore, there is an urgent need to develop an intelligent monitoring and risk assessment system that can comprehensively consider the unique characteristics of industrial site locations, achieve coupled soil-groundwater analysis, and possess dynamic state estimation capabilities. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent monitoring and risk assessment system for soil and groundwater pollution in industrial heritage sites. By introducing a mobile-stationary dual-domain migration model, it can accurately characterize the non-Fick diffusion behavior of pollutants in complex industrial heritage sites between preferential flow channels and matrix regions, significantly improving the accuracy of pollutant migration prediction under complex and heterogeneous site conditions.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] Firstly, a smart monitoring system for soil and groundwater pollution at industrial sites is provided, comprising: a multi-source heterogeneous sensing fusion module for collecting soil and groundwater monitoring data from industrial site sites, and performing spatiotemporal registration and feature-level fusion processing of multi-source data in conjunction with industrial site characteristic parameters to generate a fused feature vector; a dual-domain migration dynamics modeling module for establishing a moving-stationary dual-domain migration model based on the fused feature vector, calculating the concentration distribution of pollutants in the moving and stationary domains respectively, and coupling the mass transport process between the moving and stationary domains through a mass exchange term to output moving domain concentration field data and stationary domain concentration field data; and an interface mass flux coupling calculation module for calculating the moving domain... Concentration field data, stationary domain concentration field data, and the aforementioned soil monitoring data and groundwater monitoring data are used to generate soil-groundwater interface flux data. A Bayesian inversion pollution source tracing module is used to generate pollution source location probability distribution data and pollution source intensity confidence data based on the soil-groundwater interface flux data and historical production layout information of the industrial site. A Kalman filter state estimation module is used to take the moving domain concentration field data as the state vector, the soil-groundwater interface flux data as the observation vector, and the pollution source location probability distribution data and pollution source intensity confidence data as the control vector input, perform Kalman filter recursive calculation, and output a pollution plume migration state estimation vector.

[0007] Furthermore, the multi-source heterogeneous sensing fusion module includes: a spatiotemporal registration unit, used to perform time synchronization and spatial interpolation on the soil monitoring data and the groundwater monitoring data with different sampling frequencies, unifying all data onto the same spatiotemporal grid; and a feature fusion unit, used to perform principal component analysis on the multi-source data output by the spatiotemporal registration unit, and generate the fused feature vector by combining the industrial site feature parameters.

[0008] Furthermore, the moving-stationary dual-domain migration model includes: a parameter extraction and mapping unit, used to perform domain partitioning and parameter mapping processing on the fused feature vector, extracting soil physical feature components, hydraulic feature components, and initial pollutant distribution components from the fused feature vector, and constructing a model parameter set and initial condition data; a moving domain solving unit, used to solve the convection-dispersion equation in the moving domain based on the moving domain porosity, Darcy velocity, and hydrodynamic dispersion coefficient in the model parameter set, and in combination with the initial condition data; a stationary domain solving unit, used to solve the pure diffusion equation in the stationary domain based on the stationary domain porosity and hydrodynamic dispersion coefficient in the model parameter set, and in combination with the initial condition data; and a dual-domain coupling unit, used to calculate the mass exchange amount between the moving domain and the stationary domain based on the mass exchange coefficient in the model parameter set.

[0009] Furthermore, a model parameter set is constructed, including: converting the soil physical characteristic components into moving domain porosity and stationary domain porosity according to a preset empirical mapping function; converting the hydraulic characteristic components into Darcy velocity and hydrodynamic dispersion coefficient; and converting the initial distribution components of pollutants into initial concentration boundary conditions.

[0010] Furthermore, the mass exchange quantity is calculated using a first-order mass exchange model, whereby the mass exchange quantity is equal to the product of the mass exchange coefficient and the difference between the moving domain concentration field data and the stationary domain concentration field data.

[0011] Further, generating soil-groundwater interface flux data includes: calculating the vertical infiltration flux and capillary upwelling flux at the soil-groundwater interface based on the soil moisture content in the soil monitoring data and the groundwater level in the groundwater monitoring data, as contributions to convective flux; calculating the vertical concentration gradient at the soil-groundwater interface based on the values ​​of the moving domain concentration field data at the vadose zone-saturation zone interface, as contributions to diffuse flux; and superimposing the convective flux contributions and diffuse flux contributions to generate soil-groundwater interface flux data.

[0012] Further, generating pollution source location probability distribution data includes: establishing prior probability distributions for different areas of the industrial site based on historical production layout information; calculating the deviation between the predicted interface flux values ​​corresponding to each candidate pollution source location and the observed values ​​of soil-groundwater interface flux data based on the prior probability distributions for different areas of the industrial site, and calculating the likelihood function value based on the Gaussian distribution assumption; and generating the pollution source location probability distribution data by integrating the prior probability distributions for different areas of the industrial site and the likelihood function value based on Bayes' theorem.

[0013] Further, Kalman filtering recursive calculation is performed to output the pollution plume migration state estimation vector, including: calculating the current state prediction vector based on the state transition matrix and the state estimation vector at the previous time step, and calculating the prediction error covariance matrix, wherein the state transition matrix is ​​determined by the discretized equation of the dual-domain migration dynamics modeling module; calculating the Kalman gain matrix based on the prediction error covariance matrix, the observation matrix, and the observation noise covariance matrix; updating the state prediction vector to the pollution plume migration state estimation vector based on the Kalman gain matrix and the observation residual; and simultaneously calculating the estimation error covariance matrix based on the Kalman gain matrix, the observation matrix, and the prediction error covariance matrix.

[0014] Secondly, a risk assessment system for soil and groundwater pollution of industrial sites is provided, comprising: the intelligent monitoring system for soil and groundwater pollution of industrial sites as described in the first aspect, used to obtain an estimated vector of the migration state of the pollution plume; and a multi-temporal-scale risk early warning module, used to assess the pollution diffusion risk and delineate risk level zones at three time scales (short-term, medium-term, and long-term) based on the estimated vector of the migration state of the pollution plume and the probability distribution data of the pollution source location, and output short-term risk early warning results, medium-term risk early warning results, and long-term risk early warning results.

[0015] Furthermore, short-term risk warning results include: the risk of pollution spread within the next 7 days; medium-term risk warning results include: the risk of pollution spread within the next 30 days; long-term risk warning results include: the risk of pollution spread within the next 180 days; risk level zones are defined based on the risk index, the formula for which the risk index is calculated is:

[0016] ;

[0017] in, To predict the comprehensive risk index corresponding to time scale T1; This represents the total number of nodes in the spatial grid. For the i-th grid node, the concentration exceeds the pollutant concentration standard limit after the prediction time scale T1. The probability of; Let be the sensitive receptor exposure coefficient of the i-th grid node; The weight coefficient of the i-th grid node; a risk index less than 0.3 is defined as a low-risk zone, a risk index between 0.3 and 0.6 is defined as a medium-risk zone, and a risk index greater than 0.6 is defined as a high-risk zone.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0019] (1) This invention collects soil and groundwater monitoring data from industrial heritage sites, and performs multi-source data spatiotemporal registration and feature-level fusion processing using characteristic parameters of the industrial heritage sites to generate a fused feature vector; based on the fused feature vector, a moving-stationary dual-domain migration model is established to calculate the concentration distribution of pollutants in the moving and stationary domains respectively, and the mass transport process between the moving and stationary domains is coupled through a mass exchange term to output moving domain concentration field data and stationary domain concentration field data; based on the moving domain concentration field data, stationary domain concentration field data, soil monitoring data, and groundwater monitoring data, soil-groundwater interface flux data is generated; based on Soil-groundwater interface flux data, combined with historical production layout information of industrial sites, are used to generate probability distribution data of pollution source locations and confidence data of pollution source intensity. Using mobile domain concentration field data as the state vector, soil-groundwater interface flux data as the observation vector, and the probability distribution data of pollution source locations and the confidence data of pollution source intensity as the control vector input, Kalman filtering recursive calculation is performed to output a pollution plume migration state estimation vector. This method can accurately characterize the non-Fick diffusion behavior of pollutants in complex industrial sites between preferential flow channels and matrix regions, significantly improving the accuracy of pollutant migration prediction under complex and heterogeneous site conditions.

[0020] (2) This invention realizes cross-media joint analysis of soil and groundwater system through interface quality flux coupling calculation, quantitatively characterizes the material exchange process at the vadose zone-saturation zone interface, and effectively makes up for the defects of the two-media separation monitoring in the prior art.

[0021] (3) The present invention uses the Bayesian inversion method for pollution source tracing, which can not only output the spatial distribution of pollution sources, but also provide a confidence assessment of source strength, providing probabilistic risk cognition for pollution control decision-making;

[0022] (4) This invention introduces a Kalman filter state estimation architecture, which achieves dynamic optimal estimation of the pollution plume migration process by fusing model prediction and real-time observation data, significantly improving the real-time performance and prediction accuracy of the system. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall architecture of an intelligent monitoring and risk assessment system for soil and groundwater pollution at industrial sites, provided in an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0025] Example 1

[0026] like Figure 1 As shown, an intelligent monitoring system for soil and groundwater pollution in industrial heritage sites includes: a multi-source heterogeneous sensing fusion module, a dual-domain migration dynamics modeling module, an interface quality flux coupling calculation module, a Bayesian inversion pollution source tracing module, and a Kalman filter state estimation module.

[0027] The multi-source heterogeneous sensing fusion module collects soil and groundwater monitoring data from industrial heritage sites and performs spatiotemporal registration and feature-level fusion processing on the multi-source data, combined with the characteristic parameters of the industrial heritage sites, to generate a fused feature vector. The dual-domain migration dynamics modeling module establishes a moving-stationary dual-domain migration model based on the fused feature vector, calculates the concentration distribution of pollutants in the moving and stationary domains respectively, and couples the mass transport process between the moving and stationary domains through a mass exchange term, outputting moving-domain concentration field data and stationary-domain concentration field data. The interface mass flux coupling calculation module generates soil-groundwater interface flux data based on the moving-domain concentration field data, stationary-domain concentration field data, and the soil and groundwater monitoring data. The Bayesian inversion pollution source tracing module generates pollution source location probability distribution data and pollution source intensity confidence data based on the soil-groundwater interface flux data and historical production layout information of the industrial heritage sites. The Kalman filter state estimation module takes the moving domain concentration field data as the state vector, the soil-groundwater interface flux data as the observation vector, and the pollution source location probability distribution data and the pollution source intensity confidence data as the control vector input, performs Kalman filter recursive calculation, and outputs the pollution plume migration state estimation vector.

[0028] The core innovation of this invention lies in constructing a ternary coupled architecture of dual-domain migration, interface flux, and state estimation. This architecture differs from the existing model of independent analysis of a single medium. By establishing a bidirectional data flow between the moving-stationary dual-domain migration model and the Kalman filter state observer, it achieves a deep integration of model-driven and data-driven approaches. Specifically, the multi-source heterogeneous sensing fusion module, as the system's data entry point, receives raw monitoring data from the soil sensor network and groundwater monitoring well network, and completes spatiotemporal registration and feature fusion processing. The dual-domain migration dynamics modeling module receives the fused feature vectors and establishes a moving-stationary dual-domain migration model capable of describing the heterogeneous site characteristics of industrial heritage sites. The interface mass flux coupling calculation module simultaneously receives dual-domain concentration field data and raw monitoring data, calculating the mass exchange flux at the soil-groundwater interface. The Bayesian inversion pollution source tracing module performs probabilistic inversion of pollution sources based on interface flux data. The Kalman filter state estimation module aggregates dual-domain concentration field, interface flux, and pollution source information to perform optimal state estimation.

[0029] The multi-source heterogeneous sensing fusion module is the system's data acquisition and preprocessing layer. It is responsible for acquiring raw data from various monitoring devices and completing the fusion processing, providing input data in a unified format for downstream modules.

[0030] The inputs to the multi-source heterogeneous sensing fusion module include: soil monitoring data from a soil sensor network, which includes parameters such as soil moisture content, soil temperature, soil conductivity, and soil heavy metal concentration, sampled hourly, with a data format of a triplet structure of timestamp-location coordinates-monitoring value; groundwater monitoring data from a groundwater monitoring well network, including parameters such as groundwater level, water temperature, conductivity, dissolved oxygen, and concentration of characteristic pollutants, sampled every four hours, with a data format consistent with the soil monitoring data; and industrial site characteristic parameters from industrial site archives, including static information such as the layout of historical production workshops, the direction of underground pipe trenches, foundation depth, and the location of landfill areas. The output of the multi-source heterogeneous sensing fusion module 1 is a fused feature vector, which is... A 3D floating-point matrix, where This represents the number of nodes in the spatial grid. The number of feature dimensions.

[0031] The multi-source heterogeneous sensing fusion module includes a spatiotemporal registration unit and a feature fusion unit. The spatiotemporal registration unit performs time synchronization and spatial interpolation on monitoring data from different sensors and sampling frequencies, unifying all data onto the same spatiotemporal grid. Specifically, time synchronization uses linear interpolation to align data from different sampling frequencies to a unified time reference, and spatial interpolation uses an inverse distance weighting method to interpolate discrete monitoring point data onto a regular grid. The feature fusion unit performs feature-level fusion on the spatiotemporally registered multi-source data, using principal component analysis to extract key feature components and combining them with industrial heritage site feature parameters for feature enhancement, generating a fused feature vector.

[0032] The core processing logic of the multi-source heterogeneous sensing fusion module is as follows: First, the module receives raw monitoring data from the soil sensor network and the groundwater monitoring well network, and performs data quality checks on the soil and groundwater monitoring data, removing obvious outliers and missing values. Then, the module compares the sampling timestamps of the soil monitoring data and the groundwater monitoring data, aligning the two types of data to a unified time reference through linear interpolation. Next, based on the spatial coordinates of the monitoring points, the module uses an inverse distance weighting method to interpolate the discrete point data onto a predefined regular spatial grid, generating gridded data with a unified spatiotemporal resolution. Finally, the module performs principal component analysis on the gridded data, extracting principal components that explain more than 85% of the total variance as features, and concatenates the industrial site feature parameters as additional feature dimensions to generate the final fused feature vector.

[0033] Taking an abandoned electroplating plant site as an example: Assume the site is equipped with 20 soil sensors and 8 groundwater monitoring wells. The soil sensors collect five parameters—soil moisture content, electrical conductivity, and hexavalent chromium concentration—with a 1-hour cycle, while the groundwater monitoring wells collect four parameters—water level, electrical conductivity, and total chromium concentration—with a 4-hour cycle. In a typical processing cycle, the multi-source heterogeneous sensing fusion module 1 first receives 20 sets of soil monitoring data and 8 sets of groundwater monitoring data. Then, through spatiotemporal registration, it unifies all data to a 100×100 spatial grid and a 1-hour temporal resolution, ultimately generating a 10000×12-dimensional fusion feature vector. The 12-dimensional feature includes 5 principal components of soil parameters, 4 principal components of groundwater parameters, and 3 industrial site feature parameters.

[0034] The key parameters of the multi-source heterogeneous sensing fusion module include: time synchronization reference period, preferably set to 1 hour, which takes into account the highest sampling frequency of the soil sensor; spatial grid resolution, preferably set to 1 meter × 1 meter, which can achieve a balance between computational efficiency and accuracy; power exponent of inverse distance weighting, preferably set to 2, which makes the weight decay with the square of the distance, which is consistent with the spatial correlation characteristics of geological parameters; and the variance retention threshold of principal component analysis, preferably set to 85%, which can achieve a balance between dimensionality reduction efficiency and information retention.

[0035] The dual-domain migration dynamics modeling module is the core physical model layer of the system. It is responsible for establishing a moving-stationary dual-domain migration model that can accurately describe the characteristics of heterogeneous sites in industrial heritage sites, and provides concentration field predictions for downstream interface flux calculation and state estimation.

[0036] The input to the dual-domain migration dynamics modeling module is a fused feature vector from the multi-source heterogeneous sensing fusion module, which contains comprehensive monitoring information of each node on the spatial grid. The output of the dual-domain migration dynamics modeling module includes moving domain concentration field data and stationary domain concentration field data. The moving domain concentration field data is... The three-dimensional grid data represents the concentration distribution of pollutants in the preferential flow channel; the stationary domain concentration field data is three-dimensional grid data of the same dimension, representing the concentration distribution of pollutants in the matrix region.

[0037] The dual-domain migration dynamics modeling module includes a parameter extraction and mapping unit, a moving domain solution unit, a stationary domain solution unit, and a dual-domain coupling unit. The parameter extraction and mapping unit performs domain partitioning and parameter mapping on the fused feature vector, extracting soil physical characteristic components, hydraulic characteristic components, and initial pollutant distribution components from the fused feature vector to construct a model parameter set and initial condition data, converting the multidimensional information in the fused feature vector into the physical parameters required for the moving-stationary dual-domain migration model. The moving domain solution unit solves the convection-dispersion equation in the moving domain based on the moving domain porosity, Darcy velocity, and hydrodynamic dispersion coefficient from the model parameter set, combined with the initial condition data, to calculate the pollutant concentration distribution within the preferred flow channel. The stationary domain solution unit solves the pure diffusion equation in the stationary domain based on the stationary domain porosity and hydrodynamic dispersion coefficient from the model parameter set, combined with the initial condition data, to calculate the pollutant concentration distribution within the matrix region. The dual-domain coupling unit calculates the mass exchange term between the moving and stationary domains based on the mass exchange coefficient from the model parameter set, achieving coupled updating of the concentration fields in both domains.

[0038] The parameter extraction and mapping unit is a crucial bridge connecting the multi-source heterogeneous sensing fusion module and the solution of the dual-domain transfer equation. It is responsible for transforming the abstract feature information in the fused feature vector into the specific physical parameters of the dual-domain transfer model. The processing flow of the parameter extraction and mapping unit includes the following steps:

[0039] The first step is feature vector domain partitioning. The parameter extraction and mapping unit receives the fused feature vector from the multi-source heterogeneous sensing fusion module, and the fused feature vector is... A dimensional matrix, where This represents the number of nodes in the spatial grid. This refers to the feature dimension. The parameter extraction and mapping unit divides the fused feature vector into three feature components based on the feature source: soil physical feature component. Hydraulic characteristic components and initial distribution components of pollutants The soil physical characteristic components include information such as soil moisture content, soil electrical conductivity, and soil texture parameters, with dimensions of [missing information]. The hydraulic characteristic components include information such as groundwater level, hydraulic gradient, and estimated permeability coefficient, with dimensions of [missing information]. The initial distribution component of the pollutants includes information such as the principal component of pollutant concentration and the spatial gradient of concentration, with a dimension of [missing information]. ,in .

[0040] The second step is porosity parameter mapping. The parameter extraction and mapping unit calculates the moving domain porosity and the stationary domain porosity using an empirical mapping function based on the soil moisture content and soil texture parameters in the soil physical characteristic components. The moving domain porosity is calculated using the following mapping formula:

[0041] ;

[0042] The following mapping formula is used to calculate the porosity of the stationary region:

[0043] ;

[0044] in, The moving domain porosity is dimensionless and ranges from 0.1 to 0.4, representing the proportion of the preferred flow channels to the total volume. , is the fixed-domain porosity, dimensionless, ranging from 0.05 to 0.2, representing the proportion of the matrix region to the total volume; The value is the saturated water content, dimensionless, ranging from 0.35 to 0.55, obtained by referring to a table based on soil texture; The observed moisture content component in the fused feature vector is dimensionless and originates from the soil physical feature components. It is a soil texture index, dimensionless, with a value range of 0 to 1, and is obtained by normalizing the soil electrical conductivity component. and These are empirical coefficients, with values ​​of 0.6 and 0.4 respectively, obtained through calibration using historical site data; Residual moisture content, dimensionless, ranging from 0.02 to 0.08, obtained from a table based on soil texture.

[0045] The third step is hydraulic parameter mapping. The parameter extraction and mapping unit calculates the Darcy velocity and hydrodynamic dispersion coefficient based on the groundwater level and hydraulic gradient information in the hydraulic characteristic components. The Darcy velocity is calculated using the following mapping formula:

[0046] ;

[0047] The hydrodynamic dispersion coefficient is calculated using the following mapping formula:

[0048] ;

[0049] in, Darcy velocity, in m / d, ranges from 0 to 0.5, and represents the apparent velocity of groundwater in porous media. This is an estimated permeability coefficient, in m / d, derived from the principal component of the permeability coefficient in the hydraulic characteristic components. The hydraulic gradient is dimensionless and is calculated from the spatial difference of the groundwater level components. The hydrodynamic dispersion coefficient is expressed in units of 1000 ppm. The value ranges from 0.01 to 1.0; The longitudinal dispersion is expressed in meters (m) and ranges from 0.1 to 10. It is determined based on empirical formulas for site dimensions. Molecular dispersion coefficient, unit: For hexavalent chromium, the value is... ; The tortuosity factor is dimensionless and ranges from 0.1 to 0.7, calculated from the porosity.

[0050] The fourth step is mass exchange coefficient mapping. The parameter extraction and mapping unit calculates the mass exchange coefficient between the moving and stationary domains based on the soil texture parameters and porosity calculation results from the soil physical characteristic components. The mass exchange coefficient is calculated using the following mapping formula:

[0051] ;

[0052] in, This is the mass exchange factor, in units of... The value ranges from 0.001 to 0.1, representing the rate of mass exchange between the moving and stationary domains; is the shape factor, dimensionless, ranging from 3 to 15, determined according to the geometry of the immobilized domain; for spherical matrices, it is 15, and for plate-shaped matrices, it is 3. a is the characteristic size of the immobilized domain, in meters, ranging from 0.01 to 0.5, obtained by mapping from the soil texture index; coarse-grained soils correspond to larger characteristic sizes.

[0053] Step 5: Initial Condition Construction. The parameter extraction and mapping unit constructs the initial concentration boundary conditions for the dual-domain migration model based on the initial distribution components of the pollutants. The initial concentration of the moving domain is calculated using the following mapping formula:

[0054] ;

[0055] The initial concentration of the immobile region is calculated using the following mapping formula:

[0056] ;

[0057] in, The initial concentration in the mobile domain is expressed in mg / L, representing the initial pollutant content in the preferential flow channel. The initial concentration in the immobile region is expressed in mg / L, representing the initial contaminant content in the matrix region. The observed concentration component in the fused feature vector, in mg / L, originates from the initial distribution component of the pollutant. The mobile domain concentration distribution coefficient is dimensionless and ranges from 0.7 to 0.9, representing the initial distribution ratio of pollutants in the mobile domain. Let be the concentration distribution coefficient in the immobile region, dimensionless, and satisfy . .

[0058] After completing the above five steps, the parameter extraction and mapping unit outputs the dual-domain model parameter set and initial condition data. The dual-domain model parameter set includes the moving domain porosity. immobile porosity Darcy velocity Hydrodynamic dispersion coefficient and mass exchange coefficient The initial condition data includes the initial concentration of the mobile domain. and initial concentration of immobile region .

[0059] The moving domain solution unit receives the moving domain porosity, Darcy velocity, and hydrodynamic dispersion coefficient from the dual-domain model parameter set, and solves the moving domain convection-dispersion equation by combining it with the initial moving domain concentration from the initial condition data. The stationary domain solution unit receives the stationary domain porosity from the dual-domain model parameter set, and solves the stationary domain pure diffusion equation by combining it with the initial stationary domain concentration from the initial condition data. The dual-domain coupling unit receives the mass exchange coefficient from the dual-domain model parameter set, calculates the mass exchange between the moving and stationary domains, and couples the mass exchange as source and sink terms into the moving domain equation and the stationary domain equation, respectively. It then outputs the moving domain concentration field data and the stationary domain concentration field data through alternating iterative solutions.

[0060] Specifically, the core algorithm of the dual-domain migration dynamics modeling module is the solution of the moving-stationary model. This model divides the porous medium into two regions: a moving domain that can undergo convective transport and a stationary domain that only undergoes diffusion. The mass transport between the two domains is coupled through a first-order mass exchange term.

[0061] Specifically, the core algorithm of the dual-domain migration dynamics modeling module is the solution of the moving-stationary model. This model divides the porous medium into two regions: a moving domain that can undergo convective transport and a stationary domain that only undergoes diffusion. The mass transport between the two domains is coupled through a first-order mass exchange term.

[0062] Mobile domain control equations:

[0063] ;

[0064] Fixed domain control equations:

[0065] ;

[0066] in, The concentration of pollutants in the mobile domain is expressed in mg / L, representing the amount of pollutants in the priority flow channel. The concentration of pollutants in the immobile area is expressed in mg / L, representing the amount of pollutants in the matrix region. The moving domain porosity is dimensionless and ranges from 0.1 to 0.4, representing the proportion of the preferential flow channels to the total volume. , is the fixed-domain porosity, dimensionless, ranging from 0.05 to 0.2, representing the proportion of the matrix region to the total volume; The hydrodynamic dispersion coefficient is expressed in units of 1000 ppm. It is determined by the product of longitudinal dispersion and pore velocity; The Darcy velocity, expressed in m / d, is determined by the product of the permeability coefficient and the hydraulic gradient. This is the mass exchange factor, in units of... The value ranges from 0.001 to 0.1, representing the rate of mass exchange between the moving and stationary domains; t is a time variable, with units of days. This is the gradient operator.

[0067] Taking the monitoring of hexavalent chromium pollution at the site of an abandoned electroplating plant as an example: Assume the moving domain porosity is 0.25, the stationary domain porosity is 0.10, and the mass exchange coefficient is 0.01. The hydrodynamic dispersion coefficient is 0.1. The Darcy flow velocity is 0.1 m / d. At the initial moment, the mobile domain concentration at a pollution source is 100 mg / L, and the stationary domain concentration is 0 mg / L. Substituting into the mobile domain control equation to calculate the concentration change after day 1: Due to convection and dispersion, the pollutant migrates downstream by approximately 0.1 m, and simultaneously, due to mass exchange... With the effect of mg / (L·d), approximately 1 mg / L of pollutants are transferred from the moving domain to the stationary domain daily. After 10 days, the concentration in the moving domain decreases to approximately 82 mg / L, while the concentration in the stationary domain rises to approximately 8 mg / L. The concentration difference between the two domains gradually narrows, reflecting a dynamic equilibrium process of mass exchange between the two domains. This two-domain behavior cannot be described by traditional single-domain models and is the key to the accurate prediction of pollutant migration from industrial sites in this invention.

[0068] Specifically, the key parameters of the dual-domain migration dynamics modeling module are configured as follows: the preferred value range for the moving domain porosity is 0.15-0.30, which is determined based on the typical development degree of preferential flow channels in industrial heritage sites. A value that is too low indicates that preferential flow channels are not well-developed, while a value that is too high indicates that the site is close to homogeneous conditions; the preferred value range for the stationary domain porosity is 0.08-0.15, which is determined based on the typical pore characteristics of the matrix region of industrial heritage sites; and the preferred value range for the mass exchange coefficient is 0.005-0.05. This range is based on a combination of laboratory column experiments and field tracer experiments. A smaller mass exchange coefficient indicates that the mass exchange between the two domains is slower, while a larger mass exchange coefficient indicates that the two domains can reach equilibrium quickly.

[0069] The interface mass flux coupling calculation module is the cross-medium coupling layer of the system. It is responsible for calculating the vertical mass flux at the soil-groundwater interface, realizing the joint analysis of the vadose zone and the saturation zone. It is the key module for realizing integrated soil-groundwater monitoring in this invention.

[0070] The inputs to the interface quality flux coupling calculation module include: moving domain concentration field data and stationary domain concentration field data from the dual-domain migration dynamics modeling module, used to obtain concentration gradient information at the interface; and soil monitoring data and groundwater monitoring data from the multi-source heterogeneous sensing fusion module, used to obtain hydraulic condition information at the interface. The output of the interface quality flux coupling calculation module is soil-groundwater interface flux data. Two-dimensional field data, representing the vertical mass flux values ​​at various locations on the interface, in units of... .

[0071] Specifically, the interface mass flux coupling calculation module includes a hydraulic condition calculation unit, a concentration gradient calculation unit, and a flux synthesis unit. The hydraulic condition calculation unit is used to calculate the vertical infiltration flux and capillary upwelling flux at the interface based on soil moisture content and groundwater level data; the concentration gradient calculation unit is used to calculate the vertical concentration gradient at the interface based on moving domain concentration field data and stationary domain concentration field data; and the flux synthesis unit is used to combine convective flux and diffuse flux to calculate the total mass flux at the interface.

[0072] The core algorithm of the interface quality flux coupling calculation module is based on the principle of flux conservation, which decomposes the quality flux at the interface into two parts: the convective flux contribution and the diffuse flux contribution for calculation.

[0073] Formula for calculating soil-groundwater interface flux:

[0074] ;

[0075] in, Soil-groundwater interface flux, unit: Positive values ​​indicate that pollutants migrate from soil to groundwater, while negative values ​​indicate that pollutants migrate from groundwater to soil. The value represents the vertical infiltration flux, expressed in m / d, and is determined by rainfall infiltration and irrigation recharge, ranging from 0 to 0.05 m / d. Capillary uplift flux, expressed in m / d, is determined by the groundwater level and evaporation intensity, and ranges from 0 to 0.02 m / d. The concentration in the mobile domain at the interface is expressed in mg / L and is determined by the concentration output at the interface location from the dual-domain migration model. The effective dispersion coefficient at the interface, in units of The change in moisture content at the interface was taken into account. The vertical concentration gradient at the interface is expressed in mg / (L·m), and is determined by the vertical difference calculation of the concentration field in the moving domain.

[0076] Taking the interfacial flux calculation of an abandoned electroplating plant site during the rainy season as an example: Assume monitoring data shows the current vertical infiltration flux is 0.02 m / d, the capillary upflow flux is 0.005 m / d, the hexavalent chromium concentration in the mobile domain at the interface is 50 mg / L, and the effective dispersion coefficient at the interface is 0.05. The concentration difference within a 0.5 m range above and below the interface is 20 mg / L. Substituting these values ​​into the formula, the interfacial flux is calculated as follows: convective flux is (0.02-0.005)×50=0.75 mg / (m²·d), diffuse flux is 0.05×(20 / 0.5)=2.0 mg / (m²·d), and the total interfacial flux is 0.75+2.0=2.75 mg / (m²·d). This result indicates that under the current conditions, pollutants migrate from soil to groundwater at a rate of 2.75 mg / (m²·d), with convection contributing 27% and diffusion contributing 73%. This quantitative result provides a direct basis for assessing the extent to which groundwater is affected by soil pollution.

[0077] The key parameters of the interface quality flux coupling calculation module are configured as follows: the vertical infiltration flux is calculated using the rainfall infiltration coefficient method, with an optimal infiltration coefficient of 0.1-0.3, which is applicable to typical surface cover conditions of industrial heritage sites; the capillary rise flux is calculated using the empirical formula method, with an optimal capillary rise height of 0.5-2.0m, which is applicable to typical soil texture conditions of industrial heritage sites; the optimal effective dispersion coefficient of the interface is 0.3-0.7 times the moving domain dispersion coefficient, and this reduction factor takes into account the influence of moisture content changes at the interface on dispersion.

[0078] The Bayesian inversion pollution source tracing module is the pollution source identification layer of the system. It is responsible for inverting the spatial location and intensity of pollution sources based on interface flux observation data and historical information of industrial sites using Bayesian inference methods, and outputting probabilistic source tracing results.

[0079] The input to the Bayesian inversion pollution source tracing module consists of soil-groundwater interface flux data from the interface quality flux coupling calculation module, and historical production layout information from industrial site archives. The output of the Bayesian inversion pollution source tracing module includes pollution source location probability distribution data and pollution source intensity confidence data. The pollution source location probability distribution data consists of the posterior probability values ​​of each node on the spatial grid as a pollution source, and the pollution source intensity confidence data consists of the pollution source intensity estimate and its confidence interval.

[0080] The Bayesian inversion pollution source tracing module includes a prior construction unit, a likelihood calculation unit, and a posterior inference unit. The prior construction unit establishes a prior probability distribution of pollution source locations based on historical production layout information of industrial sites, assigning higher prior probabilities to locations such as historical production workshops, tank areas, and wastewater discharge outlets. The likelihood calculation unit calculates the likelihood function value of each candidate pollution source location based on the deviation between observed interface flux data and model prediction data. The posterior inference unit calculates the posterior probability distribution of pollution source locations based on Bayes' theorem, combining prior probabilities and likelihood functions.

[0081] The core algorithm of the Bayesian inversion pollution source tracing module is based on Bayes' theorem, treating the location of pollution sources as random variables and calculating their posterior distribution by fusing prior knowledge and observational data.

[0082] Bayesian posterior probability calculation formula:

[0083] ;

[0084] Likelihood function calculation formula:

[0085] ;

[0086] in, Let be the posterior probability of the location of the pollution source, which is dimensionless and represents the probability that each location is a pollution source given that interface flux data has been observed. is the likelihood function, dimensionless, representing the probability of observing the current interface flux data when the pollution source is assumed to be located at s; The likelihood function value is used to calculate the normalization of Bayes' formula when traversing all candidate pollution source locations s'. , which is a priori probability, dimensionless, representing the initial probability estimate of each location as a pollution source determined based on historical information of the industrial site; The prior probability value is used to iterate through all candidate pollution source locations s', and is used for the normalized integral of Bayes' formula; This is the location vector of the pollution source, containing the spatial coordinates of the pollution source; The integral traversal variable represents the set of all candidate pollution source locations, which is distinct from the estimated pollution source location s; This is the observed interface flux data vector; Let be the interface flux observation value at the i-th observation point, in units of . ; Assuming the pollution source is located at s, this is the interface flux value predicted by the two-domain migration model at the i-th observation point, in units of . ; Let be the standard deviation of the observation error for the i-th observation point, in units of . ; This represents the total number of observation points.

[0087] Taking the pollution source tracing of an abandoned electroplating plant site as an example: Assume there are 5 interface flux observation points at the site, with observed values ​​of 3.2, 2.8, 1.5, 0.8, and 0.3 mg / (m²·d), and a standard deviation of 0.5 mg / (m²·d) for each. According to historical records, there are 3 candidate pollution source locations: A is the original electroplating workshop with a prior probability of 0.5; B is the original wastewater treatment pond with a prior probability of 0.3; and C is the original chemical warehouse with a prior probability of 0.2. Using a two-domain migration model, the predicted interface flux values ​​at each observation point are calculated when the pollution source is assumed to be located at A, B, and C. The calculation results show that the predicted value at location A has the smallest deviation from the observed value. Substituting into the likelihood function formula, the likelihood value for location A is calculated to be 0.82, the likelihood value for location B is 0.45, and the likelihood value for location C is 0.12. Finally, the posterior probability is calculated using Bayes' theorem: the posterior probability for location A is... The posterior probability for location B is 0.19, and the posterior probability for location C is 0.05. This result indicates that the original electroplating workshop is the main source of pollution in as much as 76% of cases, providing a scientific basis for subsequent precise pollution control.

[0088] Specifically, the key parameters of the Bayesian inversion pollution source tracing module are configured as follows: the prior probability distribution is constructed using an expert assignment method based on historical information, with the prior probability of the historical production workshop area preferably set to 0.4-0.6, the prior probability of the wastewater treatment facility area preferably set to 0.2-0.4, and the prior probability of other areas preferably set to 0.05-0.2; the standard deviation of the observation error is estimated using the repeated measures method, with the preferred value being 10%-30% of the interface flux observation value; the posterior probability is calculated using the Markov chain Monte Carlo method, with the number of samplings preferably set to 10,000 to ensure convergence.

[0089] The Kalman filter state estimation module is the dynamic state estimation layer of the system. It is responsible for fusing model predictions and real-time observation data to achieve the optimal estimation of the migration state of the pollution plume. It is the core module for realizing dynamic intelligent monitoring in this invention.

[0090] The inputs to the Kalman filter state estimation module include: moving domain concentration field data from the dual-domain migration dynamics modeling module, serving as the model prediction value for the state vector; soil-groundwater interface flux data from the interface mass flux coupling calculation module, serving as the observation vector; and pollution source location probability distribution data and pollution source intensity confidence data from the Bayesian inversion pollution source tracing module, serving as the control vector. The output of the Kalman filter state estimation module is a pollution plume migration state estimation vector, which includes the optimal concentration estimate and its estimation error covariance for each node on the spatial grid.

[0091] Specifically, the Kalman filter state estimation module includes a state prediction unit, a gain calculation unit, and a state update unit. The state prediction unit calculates the predicted state value and prediction error covariance for the next time step based on the state transition equation of the dual-domain migration model. The gain calculation unit calculates the Kalman gain matrix based on the prediction error covariance and the observation noise covariance. The state update unit updates the predicted state value to the estimated state value based on the Kalman gain matrix and the observation residuals. Simultaneously, it calculates the estimated error covariance matrix based on the Kalman gain matrix, the observation matrix, and the prediction error covariance matrix, which characterizes the uncertainty of the state estimation.

[0092] The core algorithm of the Kalman filter state estimation module is the discrete Kalman filter recursive formula. This algorithm continuously integrates model predictions and observation data through a two-step prediction-update loop to achieve the optimal state estimation.

[0093] Among them, the state prediction equation is:

[0094] ;

[0095] Error covariance prediction equation:

[0096] ;

[0097] Kalman gain calculation:

[0098] ;

[0099] State update equation:

[0100] ;

[0101] Estimation error covariance update equation:

[0102] ;

[0103] in, Let be the state prediction vector at time k, representing the predicted state at time k based on the information at time k-1. Its elements are the predicted mobile domain concentration values ​​of each node in the spatial grid, in mg / L. Let be the state estimation vector at time k-1, representing the optimal state estimate after fusing the observation data at time k-1. Its elements are the optimal estimates of the mobile domain concentration of each node of the spatial grid at time k-1, in mg / L. Let F be the state estimation vector at time k, representing the optimal state estimate after fusing the observation data; F is the state transition matrix with dimension n×n, determined by the discretization equation of the two-domain transition model; B is the control input matrix with dimension n×p. The control vector is composed of the probability distribution of pollution source locations and the source strength confidence level. The prediction error covariance matrix has an n×n dimension and represents the uncertainty of the state prediction. Let be the estimation error covariance matrix at time k-1, with dimensions n×n, representing the uncertainty of the state estimation at time k-1; T is the superscript symbol for matrix transpose operation; The error covariance matrix, with dimensions n×n, is calculated from the Kalman gain matrix and the prediction error covariance matrix, representing the uncertainty of state estimation after fusing observation data; I is the identity matrix, with dimensions n×n, where all diagonal elements are 1 and all other elements are 0, used to ensure the dimensionality consistency of the covariance matrix during Kalman filter state update; Q is the process noise covariance matrix, with dimensions n×n, representing the uncertainty of the model itself. H is the Kalman gain matrix with dimensions n×m, which determines the correction weights of the observation data to the state estimate; H is the observation matrix with dimensions m×n, which maps the state vector to the observation space. R is the observation vector, which consists of interface flux observation data; R is the observation noise covariance matrix with dimensions m×n, representing the uncertainty of the observation data.

[0104] Taking the dynamic monitoring of a pollution plume from an abandoned electroplating plant site as an example: Assume that the state estimation vector of the monitoring system on day k-1 shows an estimated hexavalent chromium concentration of 45 mg / L in a core area, with an estimation error standard deviation of 8 mg / L. The two-domain migration model predicts that the concentration in this area will rise to 48 mg / L on day k. Simultaneously, interface flux observation data shows that the interface flux in this area on day k is 2.5 mg / (m²·d), and the corresponding concentration observation value calculated based on the observation matrix is ​​approximately 52 mg / L. Assume the process noise standard deviation is 5 mg / L and the observation noise standard deviation is 10 mg / L. Substituting into the Kalman filter formula, the prediction error variance is 8² + 5² = 89 (mg / L)², the Kalman gain is 89 / (89 + 100) = 0.47, and the final state estimate is 48 + 0.47 × (52 - 48) = 49.9 mg / L. The results indicate that the system integrates model predictions (48 mg / L) and observational data (52 mg / L), providing the optimal estimate (49.9 mg / L), and reducing the estimation error covariance from 89 mg / L² to 47.2 mg / L. 2 This effectively reduced uncertainty.

[0105] Specifically, the key parameters of the Kalman filter state estimation module are configured as follows: the diagonal elements of the process noise covariance matrix are preferably set to the square of 5%-15% of the model prediction value, which reflects the typical prediction error level of the two-domain migration model; the diagonal elements of the observation noise covariance matrix are preferably set to the square of the standard deviation of the interface flux observation error; the state transition matrix is ​​determined by the finite difference discretization of the two-domain migration model, and the time step is preferably set to 0.1-1 days; the initial state estimation vector is determined by the spatial interpolation of historical monitoring data, and the diagonal elements of the initial error covariance matrix are preferably set to the square of 20%-50% of the initial concentration estimate value.

[0106] Example 2

[0107] Based on the intelligent monitoring system for soil and groundwater pollution at industrial sites described in Embodiment 1, this embodiment provides a risk assessment system for soil and groundwater pollution at industrial sites, including:

[0108] The intelligent monitoring system for soil and groundwater pollution at industrial sites described in Example 1 is used to obtain the estimated vector of the migration state of the pollution plume.

[0109] The multi-temporal-scale risk early warning module is used to assess the pollution diffusion risk and delineate risk level zones at three time scales—short-term, medium-term, and long-term—based on the pollution plume migration state estimation vector and the pollution source location probability distribution data, and output short-term risk early warning results, medium-term risk early warning results, and long-term risk early warning results.

[0110] The multi-temporal-scale risk early warning module is the system's risk assessment and early warning layer. It is responsible for assessing the risk of pollution spread and outputting early warning information at three time scales: short-term, medium-term, and long-term, based on the pollution plume state estimation results and pollution source information.

[0111] Specifically, the inputs to the multi-temporal-scale risk early warning module include: a pollution plume migration state estimation vector from the Kalman filter state estimation module, providing current pollution distribution and prediction information; and pollution source location probability distribution data from the Bayesian inversion pollution source tracing module, providing spatial information of pollution sources. The outputs of the multi-temporal-scale risk early warning module include short-term risk early warning results, medium-term risk early warning results, and long-term risk early warning results, corresponding to risk assessment information at three prediction time scales: 7 days, 30 days, and 180 days, respectively.

[0112] The multi-temporal and spatial scale risk early warning module includes a short-term early warning unit, a medium-term early warning unit, a long-term early warning unit, and a risk zoning unit. The short-term early warning unit is used to assess the pollution spread risk within the next 7 days, focusing on rapid response to sudden pollution events; the medium-term early warning unit is used to assess the pollution spread risk within the next 30 days, focusing on the impact assessment of seasonal hydrological changes; the long-term early warning unit is used to assess the pollution spread risk within the next 180 days, focusing on predicting the overall evolution trend of the pollution plume; and the risk zoning unit is used to delineate risk zones of different levels based on the risk assessment results at each time scale.

[0113] The multi-temporal-scale risk early warning module adopts a risk index calculation method based on state estimation, which combines the probability of pollution concentration exceeding the standard with the degree of exposure to sensitive receptors to calculate a comprehensive risk index at different time scales.

[0114] Formula for calculating the multi-temporal and spatial scale risk index:

[0115] ;

[0116] in, The comprehensive risk index corresponding to the prediction time scale T1 is dimensionless and ranges from 0 to 1, with the closer to 1 indicating higher risk; T1 is the prediction time scale, with a value of 7 days for the short term, 30 days for the medium term, and 180 days for the long term. This represents the total number of nodes in the spatial grid. For the i-th grid node, the concentration exceeds the pollutant concentration standard limit after the prediction time scale T1. The probability is calculated from the state estimate and estimation error covariance of the Kalman filter using the normal cumulative function; The standard limit for pollutant concentration is set at 0.05 mg / L for hexavalent chromium (Class III standard for groundwater). Let be the sensitive receptor exposure coefficient of the i-th grid node. It is dimensionless and ranges from 0 to 1. It is determined based on whether there are sensitive receptors such as drinking water sources, farmland, and residential areas around the node. is the weight coefficient of the i-th grid node, which is dimensionless and is determined by combining the pollution source probability of the node and the downstream affected range.

[0117] Taking a multi-temporal and spatial scale risk assessment of an abandoned electroplating plant site as an example: Assume the site has 1000 spatial grid nodes, of which 50 nodes are located near drinking water source protection areas (exposure coefficient 1.0), 200 nodes are located near farmland irrigation areas (exposure coefficient 0.6), and the remaining nodes are industrial land (exposure coefficient 0.2). Kalman filter state estimation results show that in the core contaminated area, 10 nodes have current hexavalent chromium concentration estimates exceeding the 0.05 mg / L standard. In a short-term (7-day) risk assessment, the model predicts that the concentration at these 10 nodes will still exceed the standard, with a probability of 0.95; in a medium-term (30-day) risk assessment, due to the spread of the pollution plume, it is predicted that 15 nodes will exceed the standard, with a probability of 0.85; and in a long-term (180-day) risk assessment, it is predicted that 25 nodes will exceed the standard, with a probability of 0.70. Assuming the weighting coefficients are inversely proportional to the distance from the drinking water source, the short-term risk index is calculated to be 0.32, the medium-term risk index to be 0.48, and the long-term risk index to be 0.65. According to the risk classification standards (low risk <0.3, medium risk 0.3-0.6, high risk >0.6), this site is classified as medium risk in the short term, medium risk in the medium term, and high risk in the long term, requiring priority attention to the potential long-term pollution spread impact on the drinking water source.

[0118] The key parameters of the multi-temporal and spatial scale risk early warning module are configured as follows: the short-term prediction time scale is preferably set to 7 days, which is suitable for emergency response to sudden pollution events; the medium-term prediction time scale is preferably set to 30 days, which is suitable for the preparation of monthly monitoring reports; the long-term prediction time scale is preferably set to 180 days, which is suitable for semi-annual risk assessment and remediation plan formulation; the risk classification threshold is preferably set to low risk <0.3, medium risk 0.3-0.6, and high risk >0.6, which refers to the relevant provisions of the National Technical Guidelines for Risk Assessment of Contaminated Sites.

[0119] The complete data processing flow of the system of the present invention is as follows: First, the multi-source heterogeneous sensing fusion module collects raw monitoring data from the soil sensor network and the groundwater monitoring well network according to the preset sampling period, performs spatiotemporal registration and feature fusion processing on the soil monitoring data and the groundwater monitoring data, generates a fused feature vector and transmits it to the dual-domain migration dynamics modeling module, and at the same time transmits the soil monitoring data and the groundwater monitoring data to the interface quality flux coupling calculation module.

[0120] Then, the dual-domain migration dynamics modeling module receives the fused feature vector, establishes a moving-stationary dual-domain migration model, calculates the moving domain concentration field and the stationary domain concentration field respectively, and transmits the moving domain concentration field data and the stationary domain concentration field data to the interface quality flux coupling calculation module and the Kalman filter state estimation module.

[0121] Next, the interface mass flux coupling calculation module integrates the dual-domain concentration field data and the original monitoring data to calculate the vertical mass flux at the soil-groundwater interface, and transmits the soil-groundwater interface flux data to the Bayesian inversion pollution source tracing module and the Kalman filter state estimation module.

[0122] Furthermore, the Bayesian inversion pollution source tracing module performs Bayesian inversion calculations based on the soil-groundwater interface flux data and historical information of industrial sites, outputs the pollution source location probability distribution data and the pollution source intensity confidence data, and transmits them to the Kalman filter state estimation module and the multi-temporal-scale risk early warning module.

[0123] Subsequently, the Kalman filter state estimation module uses the moving domain concentration field data as the model prediction value of the state vector, the soil-groundwater interface flux data as the observation vector, and the pollution source information as the control vector, performs Kalman filter recursive calculation, outputs the pollution plume migration state estimation vector, and transmits it to the multi-temporal and spatial scale risk early warning module.

[0124] Finally, the multi-temporal-scale risk early warning module calculates the risk index on three time scales—short-term, medium-term, and long-term—based on the pollution plume migration state estimation vector and the pollution source location probability distribution data, delineates risk zones, and outputs the short-term risk early warning result, the medium-term risk early warning result, and the long-term risk early warning result.

[0125] The modular collaborative operation of the system of this invention produces a significant synergistic effect of 1+1>2: First, the collaboration between the dual-domain migration dynamics modeling module and the interface quality flux coupling calculation module enables cross-media joint analysis of the soil-groundwater system, which cannot be achieved by using either module alone; Second, the collaboration between the Bayesian inversion pollution source tracing module and the Kalman filter state estimation module significantly improves the accuracy of pollution plume prediction by introducing pollution source probability information as control input into state estimation; Third, the state estimation results of the Kalman filter state estimation module are fed back to the multi-temporal and spatial scale risk early warning module, realizing risk assessment based on optimal estimation and avoiding the bias of directly using model prediction or a single source of observation data.

[0126] To verify the technical effectiveness of the system of the present invention, a comparative test experiment was carried out at a typical industrial site. The site is the site of an electroplating plant that was shut down 10 years ago, covering an area of ​​about 2 hectares. Historically, there was a problem of hexavalent chromium pollution. The vadose zone of the site is about 5 meters thick, the groundwater is about 6 meters deep, and the soil type is an alternating layer of silty clay and fine sand, exhibiting typical dual-domain migration characteristics.

[0127] Test environment configuration: 20 soil sensors and 8 groundwater monitoring wells were deployed. The data collection period was 1 hour for soil and 4 hours for groundwater. The grid resolution was 1 meter × 1 meter × 0.5 meters, with a total of 40,000 grid nodes. The test period was 180 days.

[0128] Comparison with existing technologies: Two existing technologies were selected as comparison benchmarks. Scheme A is a single-medium monitoring system based on the traditional convection-dispersion equation, and Scheme B is a static spatial analysis system based on Kriging interpolation.

[0129] The performance metrics test results are as follows:

[0130] Regarding the accuracy of migration prediction, the root mean square error (RMSE) between the predicted and true values ​​was calculated using actual monitoring well concentration data as the true values. The 30-day prediction RMSE of the system of this invention is 3.2 mg / L, while the RMSE of scheme A is 8.7 mg / L and the RMSE of scheme B is 6.5 mg / L. The system of this invention represents a 63% improvement over scheme A and a 51% improvement over scheme B. This improvement is mainly attributed to the accurate characterization of non-Fick diffusion by the dual-domain migration model.

[0131] Regarding the accuracy of pollution source tracing, the actual pollution source location is used as the true value, and the deviation between the tracing result and the true location is calculated. The distance deviation between the highest posterior probability location identified by the system of this invention and the actual pollution source is 2.3 meters, while the deviation of scheme A is 15.6 meters and the deviation of scheme B is 8.9 meters. The system of this invention improves upon scheme A by 85% and scheme B by 74%, and this improvement is mainly attributed to the probabilistic inference capability of the Bayesian inversion method.

[0132] Regarding the real-time performance of state estimation, the convergence time of state estimation is calculated with daily updates as the cycle. The single Kalman filter update of the system of this invention takes about 12 seconds, which can achieve near real-time state estimation, while schemes A and B, due to the lack of a state estimation mechanism, can only perform offline static analysis.

[0133] Regarding the timeliness of risk warnings, a simulated sudden pollution event was used to test the system's response time. The system of this invention completed a short-term risk assessment and issued a warning within 2 hours of detecting abnormal interface throughput, while Solution A required more than 24 hours to complete the analysis, and Solution B could not provide dynamic warnings.

[0134] The test results above show that the system of the present invention is significantly superior to the existing technical solutions in terms of migration prediction accuracy, pollution source tracing accuracy, state estimation real-time performance and risk warning timeliness, and can effectively meet the actual needs of intelligent monitoring and risk assessment of contaminated sites of industrial heritage sites.

[0135] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent monitoring system for soil and groundwater pollution at industrial heritage sites, characterized in that, include: The multi-source heterogeneous sensing fusion module is used to collect soil monitoring data and groundwater monitoring data of industrial heritage sites, and to perform spatiotemporal registration and feature-level fusion processing of multi-source data in combination with the characteristic parameters of industrial heritage sites to generate fused feature vectors. The dual-domain migration dynamics modeling module is used to establish a moving-stationary dual-domain migration model based on the fused feature vector, calculate the concentration distribution of pollutants in the moving domain and the stationary domain respectively, and couple the mass transport process between the moving domain and the stationary domain through a mass exchange term, and output the moving domain concentration field data and the stationary domain concentration field data. The interface quality flux coupling calculation module is used to generate soil-groundwater interface flux data based on the moving domain concentration field data, the stationary domain concentration field data, the soil monitoring data, and the groundwater monitoring data. The Bayesian inversion pollution source tracing module is used to generate pollution source location probability distribution data and pollution source intensity confidence data based on the soil-groundwater interface flux data and combined with the historical production layout information of industrial sites. The Kalman filter state estimation module is used to take the mobile domain concentration field data as the state vector, the soil-groundwater interface flux data as the observation vector, the pollution source location probability distribution data and the pollution source intensity confidence data as the control vector input, perform Kalman filter recursive calculation, and output the pollution plume migration state estimation vector. The mobile-stationary dual-domain migration model includes: The parameter extraction and mapping unit is used to perform domain partitioning and parameter mapping processing on the fused feature vector, extract soil physical feature components, hydraulic feature components and initial distribution components of pollutants from the fused feature vector, and construct model parameter set and initial condition data. The moving domain solution unit is used to solve the convection-dispersion equations in the moving domain based on the moving domain porosity, Darcy velocity, and hydrodynamic dispersion coefficient in the model parameter set, and in combination with the initial condition data. The fixed domain solution unit is used to solve the pure diffusion equation in the fixed domain based on the fixed domain porosity and hydrodynamic dispersion coefficient in the model parameter set, and in combination with the initial condition data. The dual-domain coupling unit is used to calculate the amount of mass exchanged between the moving domain and the stationary domain based on the mass exchange coefficients in the model parameter set. Generate soil-groundwater interface flux data, including: Based on the soil moisture content in the soil monitoring data and the groundwater level in the groundwater monitoring data, the vertical infiltration flux and capillary upflow flux at the soil-groundwater interface are calculated as contributions to the flow rate. Based on the values ​​of the mobile domain concentration field data at the vadose zone-saturation zone interface, the vertical concentration gradient at the soil-groundwater interface is calculated as a contribution to the diffusion flux. The contributions of flux to flow and diffuse flux are superimposed to generate flux data at the soil-groundwater interface.

2. The intelligent monitoring system for soil and groundwater pollution in industrial heritage sites according to claim 1, characterized in that, The multi-source heterogeneous sensing fusion module includes: The spatiotemporal registration unit is used to perform time synchronization and spatial interpolation on the soil monitoring data and the groundwater monitoring data with different sampling frequencies, and to unify all data onto the same spatiotemporal grid. The feature fusion unit is used to perform principal component analysis on the multi-source data output by the spatiotemporal registration unit, and generate the fused feature vector by combining the feature parameters of the industrial site.

3. The intelligent monitoring system for soil and groundwater pollution in industrial heritage sites according to claim 2, characterized in that, Construct the model parameter set, including: The soil physical characteristic components are converted into moving domain porosity and stationary domain porosity according to a preset empirical mapping function. The hydraulic characteristic components are converted into Darcy velocity and hydrodynamic dispersion coefficient; The initial distribution components of the pollutants are converted into initial concentration boundary conditions.

4. The intelligent monitoring system for soil and groundwater pollution at industrial sites according to claim 3, characterized in that, The mass exchange quantity is calculated using a first-order mass exchange model, and the mass exchange quantity is equal to the product of the mass exchange coefficient and the difference between the moving domain concentration field data and the stationary domain concentration field data.

5. The intelligent monitoring system for soil and groundwater pollution in industrial heritage sites according to claim 4, characterized in that, Generate probability distribution data of pollution source locations, including: Establish prior probability distributions for different areas of industrial sites based on historical production layout information of industrial sites; Based on the prior probability distribution of different areas of the industrial site, the deviation between the predicted values ​​of interface flux corresponding to each candidate pollution source location and the observed values ​​of soil-groundwater interface flux data is calculated, and the likelihood function value is calculated based on the Gaussian distribution assumption. Based on Bayes' theorem, the prior probability distribution of different areas of the industrial site and the likelihood function value are combined to generate the probability distribution data of the pollution source location.

6. The intelligent monitoring system for soil and groundwater pollution in industrial heritage sites according to claim 5, characterized in that, Perform Kalman filtering recursive calculations to output a pollution plume migration state estimation vector, including: The current state prediction vector is calculated based on the state transition matrix and the state estimation vector of the previous time step, and the prediction error covariance matrix is ​​calculated. The state transition matrix is ​​determined by the discretization equation of the dual-domain migration dynamics modeling module. Calculate the Kalman gain matrix based on the prediction error covariance matrix, the observation matrix, and the observation noise covariance matrix; The state prediction vector is updated to the pollution plume migration state estimation vector based on the Kalman gain matrix and the observation residuals; at the same time, the estimation error covariance matrix is ​​calculated based on the Kalman gain matrix, the observation matrix, and the prediction error covariance matrix.

7. A risk assessment system for soil and groundwater pollution of industrial heritage sites, characterized in that, include: The intelligent monitoring system for soil and groundwater pollution at industrial sites according to any one of claims 1 to 6 is used to obtain an estimated vector of the migration state of a pollution plume. The multi-temporal-scale risk early warning module is used to assess the pollution diffusion risk and delineate risk level zones at three time scales—short-term, medium-term, and long-term—based on the pollution plume migration state estimation vector and the pollution source location probability distribution data, and output short-term risk early warning results, medium-term risk early warning results, and long-term risk early warning results.

8. The industrial site soil and groundwater pollution risk assessment system according to claim 7, characterized in that, Short-term risk warning results include: the risk of pollution spread within the next 7 days; The mid-term risk warning results include: the risk of pollution spread within the next 30 days; Long-term risk warning results include: the risk of pollution spread within the next 180 days; Risk levels are defined based on a risk index, which is calculated using the following formula: ； in, For prediction time scale The corresponding comprehensive risk index; This represents the total number of nodes in the spatial grid. For the i-th grid node, the concentration exceeds the pollutant concentration standard limit after the prediction time scale T1. The probability of; Let be the sensitive receptor exposure coefficient of the i-th grid node; Let be the weight coefficient of the i-th grid node; A risk index less than 0.3 is defined as a low-risk zone, a risk index between 0.3 and 0.6 is defined as a medium-risk zone, and a risk index greater than 0.6 is defined as a high-risk zone.

Images