Dynamic assessment and intelligent early warning system for health risk of heavy metal pollution in site soil

Abstract

The application discloses a site soil heavy metal pollution health risk dynamic assessment and intelligent early warning system and relates to the technical field of soil environment monitoring.The system comprises a multi-source data acquisition, data preprocessing, risk calculation engine and visual interactive terminal.The key technical point of the application lies in the introduction of a dynamic field evolution analysis module and an adaptive grid rendering control module:The former constructs a pollution potential field matrix representing the migration trend of pollutants based on soil heavy metal concentration and hydrogeological parameters, and calculates the spatio-temporal gradient change vector;The latter dynamically adjusts the local density of the grid according to the gradient vector module value, and automatically encrypts the calculation nodes in the area where the risk changes dramatically.Coupled with the time series prediction deduction and feedback correction mechanism, the application can simulate the dynamic evolution of the pollution plume in the porous medium in real time, solve the problems that the traditional static assessment is difficult to capture the migration law and the uniform grid calculation efficiency is low, and realize the high-precision, low-computing-power-consumption three-dimensional dynamic risk early warning.

Description

Technical Field

[0001] This invention relates to the field of soil environmental monitoring and information management technology, and in particular to a dynamic assessment and intelligent early warning system for health risks from heavy metal pollution in soil at a site. Background Technology

[0002] In current site environmental investigation and remediation engineering practices, with the increasing prominence of environmental problems left over from industrialization, heavy metal pollution in site soil exhibits complex characteristics such as a wide variety of pollutants, highly uneven spatial distribution, strong concealment, and long latency periods. Traditional soil environmental quality assessment methods often rely on low-frequency manual discrete sampling, laboratory chemical analysis, and static interpolation mapping based on fixed grids. While this approach can reveal the static pollution status at a specific moment to some extent, it is essentially an instantaneous discrete approximation of dynamic environmental processes. It is difficult to capture the complex migration and evolution patterns of pollutants over time, such as convection, dispersion, adsorption, and desorption, as the groundwater flow field, soil pore structure, and physicochemical properties change. Consequently, regulators often only see the lagging "past" state and cannot accurately predict the "future" diffusion risks.

[0003] Especially in large-scale, complexly contaminated sites with complex groundwater flow fields and highly heterogeneous soil media, the morphology, concentration gradient, and location of high-risk areas of heavy metal pollution plumes undergo nonlinear changes in real time. Existing assessment and visualization systems typically employ a globally uniform grid for simulation. This approach has significant limitations in practical applications: if a high-density grid is used across the entire field to capture detailed features of drastically changing gradients, such as pollution diffusion fronts or source-sink boundaries, computational resources and storage overhead will increase exponentially, severely hindering the system's real-time response and interactive early warning capabilities; conversely, if a sparse grid is used to reduce the load, numerical smoothing effects and information loss can easily occur in critical boundary areas, leading to misjudgments of the pollution extent and errors in remediation decisions, making it difficult to meet the actual needs of refined management. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic assessment and intelligent early warning system for health risks from heavy metal pollution in site soil, in order to solve the problems pointed out in the background art.

[0005] This invention provides a dynamic assessment and intelligent early warning system for health risks from heavy metal pollution in soil at a site, comprising a multi-source data acquisition module, a data preprocessing module, a risk calculation engine module, and a visual interactive terminal. The multi-source data acquisition module is used to acquire soil heavy metal concentration data and environmental parameters of the target site. The data preprocessing module is used to clean and standardize the soil heavy metal concentration data and the environmental parameters. The risk calculation engine module is used to calculate a health risk index based on the processed data. The visual interactive terminal is used to display a risk distribution map.

[0006] The system also includes a dynamic field evolution analysis module and an adaptive mesh rendering control module;

[0007] The dynamic field evolution analysis module is configured to construct a pollution potential energy field matrix characterizing the migration trend of heavy metal ions in the soil medium based on the soil heavy metal concentration data and the environmental parameters, and to calculate the gradient change vector of the pollution potential energy field matrix in the spatiotemporal dimension.

[0008] The adaptive mesh rendering control module is configured to dynamically adjust the local density of the visualization mapping mesh according to the magnitude of the gradient change vector, generate high-density computing nodes in areas where the gradient change vector exceeds a preset threshold, generate low-density computing nodes in areas where the gradient change vector is lower than the preset threshold, and generate a dynamic risk warning cloud map based on the adjusted mesh nodes.

[0009] Optionally, the environmental parameters acquired by the multi-source data acquisition module include soil permeability data, groundwater flow direction data, site elevation data, soil dry bulk density data, and soil effective porosity data.

[0010] The data preprocessing module is specifically configured to: use a spatial outlier detection algorithm to identify outliers in the soil heavy metal concentration data, and use neighborhood interpolation to complete the missing data to generate a standardized spatiotemporal data cube.

[0011] Optionally, the risk calculation engine module has a built-in library of toxicity response parameters for various heavy metals;

[0012] The risk calculation engine module is configured to: calculate the carcinogenic risk index and non-carcinogenic hazard quotient based on the soil heavy metal concentration data of each sampling point and the intake model of different exposure routes, and use the weighted sum of the carcinogenic risk index and the non-carcinogenic hazard quotient as the comprehensive health risk value of the sampling point.

[0013] Optionally, the risk calculation engine module also includes a Nemerow Integrated Pollution Index calculation unit;

[0014] The Nemerow Integrated Pollution Index calculation unit is configured to: extract the average and maximum values ​​of each individual heavy metal pollution index, calculate the integrated pollution level of the sampling point using the root mean square method, and attach the integrated pollution level as label data to the integrated health risk value.

[0015] Optionally, the specific steps for constructing the pollution potential field matrix by the dynamic field evolution analysis module include:

[0016] The target site is divided into an initial basic grid;

[0017] Map the comprehensive health risk value of each grid node to a potential energy scalar;

[0018] A correction coefficient based on the hydrodynamic dispersion mechanism is introduced to convert the groundwater flow direction data and the soil permeability data in the environmental parameters into corresponding migration probability weights.

[0019] The potential energy scalar difference between adjacent grid nodes is weighted using the migration probability weights to generate the pollution potential energy field matrix containing direction and intensity information.

[0020] Optionally, the adaptive mesh rendering control module dynamically adjusts the local density of the visual mapping mesh in the following way:

[0021] By traversing the pollution potential energy field matrix, the second derivative of each local region is calculated to obtain the risk curvature;

[0022] When the risk curvature is greater than the first set threshold, or when the absolute value of the comprehensive health risk of the grid node exceeds the preset control warning value, a grid splitting command is triggered to split the current grid cell into eight sub-grids using an octree algorithm, and to perform bilinear interpolation calculation on the risk value of the sub-grids.

[0023] When the risk curvature is less than the second set threshold, a grid merging command is triggered to merge eight adjacent grid cells into a parent grid, and the risk value of the parent grid is updated using mean averaging.

[0024] Optionally, the system further includes a time series prediction and extrapolation module;

[0025] The time-series prediction and extrapolation module is configured to: use a long short-term memory network model to learn the historical evolution sequence of the pollution potential energy field matrix and generate a predicted potential energy field matrix with a preset time step in the future.

[0026] The adaptive mesh rendering control module, based on the predicted potential energy field matrix, pre-defines the mesh in areas where high-risk diffusion may occur in the future, and displays the projected trajectory of the future risk boundary on the visualization interactive terminal in the form of a dynamic heat map.

[0027] Optionally, the system further includes a feedback correction and update module;

[0028] The feedback correction and update module is configured to: when a new batch of measured soil heavy metal concentration data is received, calculate the deviation between the measured data and the corresponding node of the predicted potential energy field matrix;

[0029] If the deviation value exceeds the tolerance range, a parameter inversion gradient signal is generated, the migration weight parameter in the dynamic field evolution analysis module is adjusted, and the reconstruction process of the entire field mesh is triggered.

[0030] Optionally, the visual interactive terminal is equipped with a hierarchical early warning response unit;

[0031] The hierarchical early warning response unit is configured as follows:

[0032] Based on the values ​​of each pixel in the dynamic risk warning cloud map, the risk is divided into a safe zone, a warning zone, and a control zone, and mapped to semi-transparent rendering layers in green, yellow, and red tones, respectively.

[0033] For the area where the high-density computing nodes are located, virtual buoys with coordinate labels and risk values ​​are automatically generated, and when an upward trend in risk values ​​is detected, the risk accumulation rate is represented by a high-brightness flashing frequency.

[0034] Optionally, the system is deployed on a cloud server cluster, and the multi-source data acquisition module is connected to the field sensor network through an IoT gateway.

[0035] The adaptive mesh rendering control module adopts a parallel computing architecture, which allocates the mesh adjustment tasks of different sub-regions to different computing threads, and finally synthesizes the complete dynamic risk warning cloud map in the frame buffer and pushes it to the visualization interactive terminal.

[0036] The present invention has achieved the following beneficial effects:

[0037] This invention enhances the dynamic spatiotemporal resolution and scientific rigor of site heavy metal pollution assessment by constructing a pollution potential energy field matrix driven by physical mechanisms. Unlike traditional static assessments that focus solely on concentration values, this system introduces correction coefficients based on Darcy's law and soil permeability, transforming static monitoring data into a potential energy field containing both directional and intensity information. This allows for the quantitative characterization of the migration trends and momentum of heavy metal ions in porous media. This mechanism enables the system to sensitively capture transient risk changes caused by abrupt changes in groundwater flow or surface runoff, accurately simulating the diffusion path of pollution plumes.

[0038] This invention employs gradient-aware adaptive mesh rendering control technology to resolve the contradiction between computational accuracy and resource consumption in large-scale 3D scenes. Based on the second derivative (curvature) of the risk field and the gradient change vector, the system intelligently triggers mesh splitting and densification commands in key areas such as pollution diffusion fronts and finger-like advances, while performing mesh merging and sparsification operations in background areas with gentler risk distribution. This on-demand allocation of computing power not only clearly restores the fine morphological features of high-risk areas with limited hardware resources, reducing the jagged edges and boundary blurring caused by traditional interpolation, but also improves the rendering frame rate of 3D dynamic cloud maps, achieving smooth real-time interactive early warning.

[0039] Furthermore, this invention establishes a closed-loop early warning system with adaptive parameter adjustment capabilities. By integrating a time-series prediction and feedback correction update module based on a long short-term memory network, the system can not only use historical evolution sequences to extrapolate future risk boundary trajectories and generate forward-looking early warning information, but also automatically calculate prediction bias and adjust the migration weight parameters of the evolution model in reverse upon receiving new batches of measured data. This dynamic calibration mechanism ensures that the system can continuously approximate the actual hydrogeological properties of the site over time, effectively overcoming the cumulative errors generated by long-term model operation, and providing site managers with intuitive, dynamic, and continuously optimized decision support.

[0040] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0041] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0043] Figure 1 This is a diagram illustrating the overall logical architecture of the dynamic assessment and intelligent early warning system for heavy metal pollution health risks in site soil, as described in this embodiment of the invention.

[0044] Figure 2 This is a schematic diagram of the logical flow of the adaptive mesh rendering control module based on risk curvature for splitting and merging the quadtree mesh in an embodiment of the present invention. Detailed Implementation

[0045] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0046] Example 1:

[0047] This invention provides a dynamic assessment and intelligent early warning system for health risks from heavy metal pollution in site soil. In current site environmental investigation and remediation engineering practices, with the increasing prominence of environmental problems left over from industrialization, heavy metal pollution in site soil exhibits complex characteristics such as a wide variety of pollutants, uneven spatial distribution, strong concealment, long latency periods, and severe harmful consequences. Traditional soil environmental quality assessment methods often rely on low-frequency manual sampling, laboratory analysis, and static mapping. While this model can reveal the static pollution status at a specific moment to some extent, it is essentially a discretized and static approximation of dynamic environmental processes. It is difficult to capture the migration and evolution patterns of pollutants over time as groundwater flow fields and soil physicochemical properties change, and it cannot provide effective early warnings with physical mechanisms to support future risk diffusion trends. Especially in large, complexly contaminated sites with complex groundwater flow fields and highly heterogeneous soil media, heavy metal pollutants (such as hexavalent chromium, lead, cadmium, arsenic, and mercury) undergo complex convection, dispersion, and adsorption-desorption processes due to soil pore water, groundwater runoff, and surface leaching. This causes the morphology, concentration gradient, and location of high-risk areas of the pollution plume to change in real time. If traditional fixed-grid interpolation and static heatmaps are still used, not only will computational resources be wasted in low-risk background areas, but information may also be lost due to smoothing in key areas with drastic gradient changes, such as pollution diffusion fronts, leading to misjudgments of the pollution extent and errors in remediation decisions.

[0048] To address the aforementioned technical challenges, this invention constructs a comprehensive system integrating hardware sensing, data cleaning, model solving, field evolution analysis, and intelligent visualization. This system not only assesses the current pollution state but also evaluates the dynamic mechanisms of pollution evolution, and achieves efficient and high-precision three-dimensional dynamic early warning through adaptive grid technology.

[0049] Specifically, the site's soil heavy metal pollution health risk dynamic assessment and visualization intelligent early warning system includes a multi-source data acquisition module, a data preprocessing module, a risk calculation engine module, a dynamic field evolution analysis module, an adaptive mesh rendering control module, and a visualization interactive terminal. In terms of physical deployment architecture, the system relies on a high-performance distributed computing cluster. The bottom sensing layer connects various in-situ sensors via an industrial-grade Internet of Things (IIoT) bus. The intermediate processing layer relies on high-performance servers equipped with GPU accelerator cards for matrix operations and model solving. The upper application layer uses a graphics engine based on WebGL or OpenGL standards for rendering and display on the client side.

[0050] The multi-source data acquisition module is configured to construct a three-dimensional monitoring network covering the target site, encompassing air, ground, and depth. This module integrates multiple heterogeneous data sources through an edge computing gateway to acquire soil heavy metal concentration data and environmental parameters for the target site.

[0051] In acquiring soil heavy metal concentration data, this module deploys array-style in-situ online monitoring probes at key soil layer interfaces at different depths (such as the topsoil layer 0-20cm, the middle of the vadose zone, and the water table interface). These probes are based on high-sensitivity X-ray fluorescence spectroscopy (XRF) analysis or laser-induced breakdown spectroscopy (LIBS). Specifically, the probes integrate miniature X-ray tubes (such as silver or rhodium targets) as excitation sources and high-resolution silicon drift detectors (SDDs). When the system issues a data acquisition command, the X-ray tubes emit primary X-rays to irradiate the soil medium, exciting the inner-shell electron transitions of heavy metal elements on the surface of soil particles and in pore water, producing characteristic fluorescence. The detectors receive and analyze the energy and intensity of these fluorescencees, and use built-in fundamental parameter (FP) or empirical coefficient method algorithms to invert and output the mass fractions of various heavy metal elements such as lead (Pb), cadmium (Cd), chromium (Cr), arsenic (As), mercury (Hg), copper (Cu), nickel (Ni), and zinc (Zn) in real time. The probe is encapsulated with a special IP68-rated corrosion-resistant coating, allowing it to be buried underground for extended periods and withstand high-salt and acid / alkali environments. Furthermore, the module includes a standard data interface to support the import of high-precision offline data from laboratory inductively coupled plasma mass spectrometry (ICP-MS). Time-stamp alignment technology is used to fuse offline and online data, enabling periodic calibration of the online probe's drift.

[0052] In addition to pollutant concentrations, the multi-source data acquisition module also simultaneously acquires key environmental parameters. These environmental parameters are the physical boundary conditions driving subsequent dynamic field evolution analysis. Specifically, these environmental parameters include, but are not limited to:

[0053] Soil permeability data: This data reflects the soil medium's ability to conduct fluids and is a key hydrogeological parameter determining the migration rate of pollutants. The system deploys time-domain reflectometers (TDRs) and tensiometers at different geological layers (such as clay, silt, and gravel) to monitor soil volumetric water content and matrix suction in real time. By combining pre-determined soil moisture characteristic curves (SWCC) and unsaturated hydraulic conductivity functions (such as the Van Genuchten model), real-time soil permeability (K-value) is calculated. For key preferred flow channel areas, historical data from in-situ pressure tests or injection tests are also incorporated for correction.

[0054] Groundwater flow direction data: This data reflects the driving force direction of horizontal migration of pollutants. The system deploys a multi-parameter groundwater monitoring well network around and within the site. Each monitoring well is equipped with a high-precision submersible level gauge (accuracy up to 1 mm) and an acoustic Doppler current meter (ADV). The multi-source data acquisition module synchronizes the water level and elevation data of each monitoring well in real time. Using triangulation algorithms or Kriging interpolation, it constructs groundwater head isosurfaces in real time, and then calculates the hydraulic gradient vector (including magnitude and direction) at any location, thereby accurately capturing seasonal or sudden changes in the groundwater flow field (such as flow direction deflection after heavy rain).

[0055] Site elevation data: This data is used to accurately simulate the scouring and transporting effects of surface runoff on surface heavy metals. The system is connected to a high-precision RTK-GPS (Real-time Dynamic Carrier Phase Differential Positioning) receiver or a LiDAR system mounted on a UAV. LiDAR acquires high-density point cloud data of the site by emitting high-frequency laser pulses and receiving the echoes. After filtering, noise reduction, and vegetation removal, a high-precision digital elevation model (DEM) with a resolution better than 5 cm is generated, providing fundamental data for analyzing micro-topographic runoff paths.

[0056] Soil physicochemical property data: This data is used to calculate the adsorption and retardation effects of pollutants in the medium. The system acquires the basic constant of soil dry bulk density in real time by using an array of dielectric constant sensors or a gamma-ray densitometer buried at different soil depths for in-situ monitoring. ) and soil effective porosity ( Distribution data of ).

[0057] Furthermore, in actual field monitoring environments, the collected data often contains outliers or missing values ​​due to factors such as electromagnetic interference, sensor aging, circuit noise, or communication packet loss. Therefore, the data preprocessing module is connected to the multi-source data acquisition module to clean and standardize the soil heavy metal concentration data and the environmental parameters.

[0058] Specifically, this module first employs a spatial outlier detection algorithm to identify outliers in the soil heavy metal concentration data. This embodiment preferably uses a density-based Local Outlier Factor (LOF) algorithm. The core idea of ​​this algorithm is to compare the local reachability density of a data point and its neighboring points. For any monitoring point p, the system first searches all points (Nk(p)) within its k-distance neighborhood. Then, it calculates the local reachability density (lrd) of point p relative to its neighbors. If the lrd of point p is significantly lower than the lrd of its neighbors, it indicates that the density in the area where point p is located is much lower than its surroundings, and this point is very likely a spatial outlier. The system calculates the LOF value for each point; if the LOF value is greater than a preset threshold (e.g., 1.5), the data is determined to be an anomaly. To prevent the accidental deletion of real pollution hotspots, the system also incorporates a time series analysis subroutine: if the anomaly point shows a continuous gradual trend on the time axis, it is retained and determined to be real pollution; if it is an isolated pulse, it is removed or corrected.

[0059] To address the issue of missing data, this module utilizes neighborhood interpolation to complete the missing data, generating a standardized spatiotemporal data cube. This embodiment employs spatiotemporal universal kriging interpolation. This method considers not only the spatial autocorrelation of data (stronger correlation with closer distances) but also temporal continuity and the drift trend of data with spatial location. The system first calculates the empirical variogram, fits the optimal variogram model (such as a spherical model, exponential model, or Gaussian model), and determines the nugget value, sill value, and range. Using these structural parameters, unbiased optimal estimations are performed on the missing spatiotemporal data, and the estimated variance is provided. Finally, all cleaned and completed data is resampled and mapped into a regular four-dimensional grid (x,y,z,t), forming a standardized spatiotemporal data cube, providing a unified data structure for subsequent matrix operations.

[0060] Specifically, to address the issue of low readings caused by moisture interference in moist soil environments, the in-situ XRF and LIBS probes are susceptible to moisture interference, the data preprocessing module incorporates a spectral-moisture coupling correction algorithm. The system reads the volumetric water content collected by the TDR sensor at the same location. Using a pre-set exponential compensation model (in To account for the characteristic attenuation factors of different heavy metal elements, for example, Pb is set to 0.25. (This refers to the soil heavy metal concentration after moisture correction), compared to the original concentration data. Inversion corrections were performed to eliminate the nonlinear absorption effect of soil pore water on X-ray fluorescence, ensuring that the basic data input into the model accurately reflects the dry basis heavy metal content of the soil.

[0061] The risk calculation engine module is the core computing unit of this system, used to transform environmental monitoring data into intuitive health risk indicators. This module has a built-in database of various heavy metal toxicity response parameters that comply with national environmental protection standards (such as the "Technical Guidelines for Risk Assessment of Soil Pollution in Construction Land") and internationally accepted guidelines (such as USEPA RAGS). This database stores the physicochemical properties (such as water solubility and partition coefficient) and toxicological parameters (such as reference dose RfD, reference concentration RfC, carcinogenicity slope factor SF, inhalation unit risk IUR, gastrointestinal absorption factor ABSgi, skin penetration coefficient Kp, etc.) of various heavy metals, and distinguishes between different valence states of heavy metals (such as the toxicity differences between Cr(III) and Cr(VI).

[0062] The risk calculation engine module is configured to calculate the carcinogenic risk index and non-carcinogenic hazard quotient based on the soil heavy metal concentration data at each sampling point (or grid node) and the intake models for different exposure pathways. In the specific calculation, the system first identifies the main sensitive receptors under the future planned land use of the site (e.g., children under residential land or adults under industrial land), and determines the exposure pathways (including oral ingestion of soil, skin contact with soil, and inhalation of soil particles). For each exposure pathway, the daily average exposure is calculated using the exposure assessment model. ).

[0063] For example, soil ingested orally The calculation formula is:

[0064]

[0065] in, Soil heavy metal concentration (mg / kg); Intake rate (mg / d); Exposure frequency (d / a); For years of exposure (a); For example, take the conversion factor. , used to convert mg to kg; Weight (kg); The average time is d.

[0066] based on Calculate the carcinogenic risk separately ( ) and non-carcinogenic hazard quotient ( For cases of multi-pollutant contamination, the system assumes that the effects of each pollutant are additive, and calculates the total carcinogenic risk separately. ) and total hazard index ( ).

[0067] Finally, the system uses the weighted sum of the carcinogenic risk index and the non-carcinogenic hazard quotient as the comprehensive health risk value for the sampling points. This is to incorporate the probabilistic CR value (typically...) The system combines the magnitude of logarithms with the ratioistic HQ value (usually on the order of 1), and incorporates logarithmic normalization and weight allocation mechanisms. For example, to avoid the anomaly of undefined logarithms in numerical calculations, the system introduces a small numerical smoothing term. (For example, take) The formula for calculating the comprehensive health risk value is:

[0068]

[0069] in, and For weighting coefficients, preferably, , This composite value intuitively quantifies the level of health threat.

[0070] The dynamic field evolution analysis module is configured to construct a pollution potential energy field matrix characterizing the migration trend of heavy metal ions in the soil medium based on the soil heavy metal concentration data and the environmental parameters, and to calculate the gradient change vector of the pollution potential energy field matrix in the spatiotemporal dimension. The detailed construction process will be described in subsequent embodiments.

[0071] The adaptive mesh rendering control module is configured to dynamically adjust the local density of the visualization mapping mesh according to the magnitude of the gradient change vector, generate high-density computing nodes in areas where the gradient change vector exceeds a preset threshold, generate low-density computing nodes in areas where the gradient change vector is lower than the preset threshold, and generate a dynamic risk warning cloud map based on the adjusted mesh nodes.

[0072] The visualized interactive terminal is used to display risk distribution maps. Developed based on WebGL technology, it supports B / S architecture access. Users can view real-time 3D risk cloud maps on the terminal; the color intensity of the cloud map represents the level of risk, and dynamic flow lines represent migration trends. The terminal also integrates an intelligent early warning function; when a predicted risk plume is about to hit a sensitive target, it automatically triggers an audible and visual alarm and generates an emergency response suggestion report.

[0073] Example 2:

[0074] As mentioned above, the environmental parameters acquired by the multi-source data acquisition module include soil permeability data, groundwater flow direction data, and site elevation data. These three parameters constitute the physical field boundary conditions describing the migration and transformation of heavy metal pollutants in porous media.

[0075] Soil permeability data characterizes the ease with which soil allows fluids to pass through. At the microscopic level, it depends on the geometry, connectivity, and porosity of soil pores; at the macroscopic level, it directly determines the rate at which pollutants migrate with groundwater convection. This system acquires data by burying sensors at different depths in the strata and combines this data with soil structure information (such as lenses and fracture zones) from geological survey data to construct a heterogeneous permeability tensor field.

[0076] Groundwater flow direction data are vector parameters describing the hydrodynamic field. Since heavy metal ions mainly migrate with groundwater in dissolved or colloidal adsorbed states, the flow direction of groundwater determines the dominant diffusion direction of the pollution plume. The system utilizes high-frequency water level monitoring data to calculate the hydraulic gradient distribution across the entire field in real time by solving the Darcy equation or using the finite difference method, thereby accurately capturing changes in flow direction caused by seasonal rainfall, artificial pumping, or river recharge.

[0077] Site elevation data is obtained through high-precision topographic mapping to construct the potential energy surface of surface runoff. During heavy rainfall events, surface runoff is a crucial carrier for the horizontal migration of surface heavy metals (especially elements such as arsenic and lead that easily accumulate in topsoil). Elevation data, combined with rainfall data, can be used to simulate the confluence path and scour intensity of surface runoff.

[0078] In terms of data preprocessing, the data preprocessing module is specifically configured as follows: using a spatial outlier detection algorithm to identify outliers in the soil heavy metal concentration data, and using neighborhood interpolation to complete the missing data, generating a standardized spatiotemporal data cube.

[0079] For spatial outlier detection, the LOF algorithm employed by the system exhibits significant advantages in handling non-uniformly sampled data. Unlike global detection methods based on statistical distributions (such as the 3σ principle), LOF focuses on local density. For example, in the core area of ​​a pollution source, high-concentration points cluster together, resulting in a high local density and an LOF value close to 1, which is considered normal. Conversely, in a clean background area, if an isolated high-concentration point suddenly appears, its local density is much lower than that of surrounding points, leading to a significantly higher LOF value and identification as an anomaly. This mechanism effectively avoids misclassifying genuine hotspots as noise.

[0080] For neighborhood interpolation completion, the system employs spatiotemporal kriging interpolation, introducing temporal correlation. This means that when inferring missing data at a specific location at a given time, the algorithm not only references contemporaneous data from surrounding locations but also historical data from the previous and next time points. This spatiotemporal collaborative interpolation significantly improves the ability to capture sudden pollution events (such as leaks) or transient processes. The resulting spatiotemporal data cube is a four-dimensional tensor that not only fills in monitoring blind spots but also performs spatiotemporal registration and resolution unification of multi-source heterogeneous data (e.g., resampling all data into a 1m×1m×1h grid), providing a standard mathematical object for subsequent field theory operations.

[0081] Example 3:

[0082] The risk calculation engine module has a built-in database of toxicity response parameters for various heavy metals. This database is the core knowledge base of the system, containing toxicological parameters for common heavy metals such as lead, cadmium, chromium, arsenic, mercury, nickel, copper, and zinc. These parameters are derived from the latest releases by authoritative domestic and international institutions (such as the USEPA IRIS database and the guidelines from the Ministry of Ecology and Environment of China) and support automatic updates in the cloud. The parameter database not only contains numerical values ​​but also information on the uncertainty distribution of the parameters (such as probability density functions), supporting Monte Carlo probabilistic risk assessments.

[0083] In terms of the calculation process, the risk calculation engine module is configured as follows: based on the soil heavy metal concentration data of each sampling point, combined with the intake model of different exposure routes, the carcinogenic risk index and the non-carcinogenic hazard quotient are calculated respectively, and the weighted sum of the carcinogenic risk index and the non-carcinogenic hazard quotient is used as the comprehensive health risk value of the sampling point.

[0084] Specifically, the system first determines the exposure pathways based on a site conceptual model. For carcinogenic risk, the calculation formula is as follows: ,in This is the lifetime average daily exposure. This is the carcinogenic slope factor. For non-carcinogenic risk, the calculation formula is... ,in This is a reference dose. For cases involving multiple routes and multiple contaminants, the overall carcinogenic risk is calculated systematically. Total Hazard Index .in, Representing the Heavy metal pollutants or the first Various exposure pathways and These are the corresponding individual carcinogenic risk index and individual non-carcinogenic hazard quotient, respectively.

[0085] To provide a unified risk quantification index for visualization and early warning, the system uses a weighted sum to calculate the comprehensive health risk value. The weighting coefficients can be set according to management objectives. For example, in a residential land scenario, the tolerance for carcinogenic risk is extremely low (typically...). Therefore, the carcinogenic risk item is given a higher weight; however, the weight allocation may be different in the context of industrial land use. Normalization ensures that risk indicators of different dimensions can be superimposed and compared in a scalar field.

[0086] Furthermore, to address the potential oversight of soil environmental quality degradation in health risk assessments, the risk calculation engine module also includes a Nemerow Integrated Pollution Index (IPI) calculation unit. This IPI calculation unit is configured to: extract the average and maximum values ​​of each individual heavy metal pollution index; calculate the integrated pollution level of the sampling point using the root mean square (RMS) method; and append the integrated pollution level as label data to the integrated health risk value. The specific calculation steps are as follows:

[0087] Calculate individual pollution indices ,in For actual measured concentration, This refers to the national standard screening or control value for this heavy metal under the current land use type.

[0088] Calculate the Nemerow Comprehensive Pollution Index .in, This is the average of all individual pollution indices. The value is the maximum. The mathematical construction of the Nemerow index determines its high sensitivity to the largest pollutant.

[0089] Determine the overall pollution level. The system is based on... The size, compared with the preset grading standard (such as... Safety, alert, Slight pollution (Moderate pollution, Pn>3 indicates severe pollution), determine the pollution level of this location.

[0090] Labeling. The system encodes this level result as text or an enumerated label, which is then appended as metadata to the comprehensive health risk value data structure for that location. During visualization, the system can simultaneously utilize color (representing risk values) and texture / legend (representing pollution level labels) to display the information, or a detailed pollution level report can be displayed when the user interacts with the query.

[0091] Example 4:

[0092] The specific steps for constructing the pollution potential energy field matrix by the dynamic field evolution analysis module include:

[0093] Step S1: Mesh initialization.

[0094] The target site is divided into an initial base grid. The system first generates a regular rectangular grid or an unstructured triangular grid covering the entire area based on the site's geographical boundaries (the base resolution can be set to 10 meters). Each grid node serves as a computational unit, storing the risk value, environmental parameters, and intermediate computational variables for that location.

[0095] Step S2: Potential mapping.

[0096] The comprehensive health risk value of each grid node is mapped to a potential energy scalar. The system introduces risk potential energy. The physical concept of potential energy (P) is that areas of high potential energy tend to diffuse towards areas of low potential energy. The system converts the dimensionless comprehensive health risk value into a potential energy scalar using a monotonically increasing mapping function (such as a linear or exponential mapping). At this point, the entire site is quantified as a undulating potential energy surface, with peaks representing high-risk sources and troughs representing clean areas. It is important to note that this system constructs an evolution model based on the principle of linear isomorphism between risk and concentration. According to the "Technical Guidelines for Risk Assessment of Soil Pollution in Construction Land," under a defined land use method and sensitive receptors, both exposure and toxicity parameters are spatiotemporal constants. Therefore, the health risk value of a single heavy metal... With soil concentration Satisfy linear equations ( (This is the comprehensive conversion coefficient). This system will... Defined as a virtual risk equivalent concentration, this linear relationship is substituted into the convection-diffusion equation (ADE), and the coefficients on both sides of the equation... It can be eliminated by reduction.

[0097] Specifically, to ensure that the calculation of risk potential energy migration conforms to physical laws, a linear isomorphic mapping strategy is adopted in this step. The pollution potential energy field constructed by this system is mathematically equivalent to a risk equivalent concentration field. This allows the system to directly utilize Darcy's law and convection-dispersion equations applicable to mass migration to drive the evolution of risk potential energy, thereby avoiding the high computational cost of simulating the reaction and transport of multiple pollutants separately while ensuring physical realism. It should be noted that the pollution potential energy field proposed in this invention is essentially an equivalent computational proxy model. Although the migration at the microscopic physical level is of pollutant entities, considering the definite mapping relationship between the health risk value and concentration of heavy metals in a specific site, this system can mathematically equivalently simulate the spatiotemporal distribution of pollution plume diffusion by directly mapping the risk value to virtual potential energy and using fluid dynamics equations to calculate the field evolution. This dimensionality reduction avoids the high computational cost of simulating the reaction and transport of multiple heavy metals separately, and is a key technical means to achieve real-time early warning.

[0098] Step S3: Construction of migration probability weights.

[0099] A correction coefficient based on the hydrodynamic dispersion mechanism is introduced to convert the groundwater flow direction data and soil permeability data in the environmental parameters into corresponding migration probability weights.

[0100] A potential energy difference alone is insufficient to drive migration; the conductivity of the medium must be considered. The system utilizes Darcy's law. .

[0101] in, For Darcy velocity vector, For soil permeability, This represents the hydraulic gradient.

[0102] For any node in the grid and its neighboring nodes The system calculates from Towards Probability weights of migrating pollutants This weight is an anisotropic parameter, and its calculation model takes into account the following factors:

[0103] Diffusion and Conduction Factors: Nodes Hydrodynamic dispersion coefficient at the location It comprehensively reflects the influence of soil permeability and flow velocity. The larger the dispersion coefficient, the stronger the tendency of pollutants to diffuse and migrate in the medium, and the greater the corresponding migration weight.

[0104] Flow direction factor: Calculating the groundwater velocity vector Vectors connected to the grid The included angle .if (Right now In the downstream region), the weight increases; if (Right now In the upstream region, the weight decreases or even approaches zero (considering the mechanical dispersion effect).

[0105] Distance factor: The closer the nodes are, the greater their mutual influence and the greater their weight.

[0106] Specifically, to achieve accurate quantification of the aforementioned physical processes, this embodiment constructs a discretized weighted model based on the convection-diffusion mechanism. Nodes are defined. Pointing to neighboring nodes migration probability weights The calculation formula is as follows:

[0107] Define nodes Pointing to neighboring nodes Unnormalized transition probability weights The calculation formula is as follows:

[0108]

[0109] Final normalized transfer probability weights The calculation formula is:

[0110]

[0111] In the formula: and They are nodes and nodes Hydrodynamic dispersion coefficient at (unit: This coefficient is calculated based on measured soil permeability and flow velocity, and the formula is as follows: ,in For longitudinal dispersion, For pore flow velocity, Molecular diffusion coefficient; calculation nodes When the hydrodynamic dispersion coefficient is at a certain point, then Similarly, compute nodes When the hydrodynamic dispersion coefficient is at a certain point, then ; The harmonic mean dispersion coefficient at the node interface characterizes the diffusion and conduction ability of pollutants in the medium.

[0112] The Euclidean distance between nodes is calculated by dividing the first term of the formula by... This makes the dimensions of the diffusion term uniform. ;

[0113] For nodes The actual average flow velocity at the location (unit: Its value is calculated by dividing Darcy velocity by the effective porosity of the soil. get;

[0114] For the flow direction projection factor, when the flow velocity vector Vector connecting nodes The included angle (Downstream) Otherwise, it is 0, thus achieving upwind simulation;

[0115] These are the weighting coefficients for the diffusion and convection terms;

[0116] The normalization factor is used to ensure local mass conservation, and its calculation formula is: ,in For nodes The set of effective neighboring nodes, Represents a set any neighbor node index in the table, Represents a node Pointing to neighbor nodes The unnormalized migration probability weights are calculated using the same method as in the aforementioned formula. same;

[0117] To ensure the convergence of the pollution potential energy field matrix during numerical solution, this step also sets hybrid boundary conditions: for the groundwater inflow boundary, the system applies a first type of boundary condition to lock the background risk value; for the outflow boundary, the system applies a second type of boundary condition, setting the gradient of the risk potential energy along the normal direction to zero. This allows the contamination plume to flow out freely without producing numerical reflections.

[0118] Furthermore, to accurately reflect the adsorption-desorption hysteresis effect of heavy metal ions on the surface of soil particles, a retardation factor was introduced into the migration probability weighting. The correction is made. The corrected convection term coefficients are... ,in ( For soil dry bulk density, (This refers to the solid-liquid partition coefficient for heavy metals). This parameter correction ensures that the simulated advance velocity of the risk front is consistent with the actual chemical migration velocity of heavy metal pollutants, preventing false expansion of the warning range due to neglecting adsorption.

[0119] To ensure the physical realism of the simulation, the weight allocation coefficients in this embodiment... and It is not a fixed constant, but rather depends on the grid Peclet number of the local flow field. Dynamic calculation. Definition ( The molecular diffusion coefficient of heavy metal ions in water can be found in relevant chemical physics handbooks. (System settings) Characterizing the contribution of convection, Characterize diffusion contribution.

[0120] Furthermore, it should be noted that the pollution potential energy field constructed in this invention is an engineering numerical equivalent surrogate model. Although health risk values ​​themselves fall under the category of statistics, their spatial distribution in a single medium is positively correlated with the pollutant concentration field and follows the same convection-dispersion differential equation constraints. Therefore, by constructing a virtual potential energy field for dimensionality reduction calculation, the high computational cost of solving complex multi-component reaction and transport equations can be avoided while ensuring the accuracy of trend prediction.

[0121] Step S4: Generation of the pollution potential energy field matrix.

[0122] The potential energy scalar difference between adjacent grid nodes is weighted using the migration probability weights to generate the pollution potential energy field matrix containing direction and intensity information.

[0123] The system traverses the entire grid and, for each node, calculates the scalar potential energy transfer rate between adjacent nodes. :

[0124]

[0125] in, and They are nodes and nodes The risk potential energy scalar; The migration probability weights are those calculated above.

[0126] At this time, if This indicates that potential energy originates from... Transmit to ;like This indicates that potential energy originates from... Transmit to .

[0127] Meanwhile, in order to construct a visualized flow field direction, the system calculates the local flux vector. :

[0128]

[0129] in, For grid nodes The set of valid neighboring nodes; From point to The unit vector; this vector It is only used to guide the generation of visual streamlines and does not directly participate in the numerical update of scalar fields.

[0130] The set of net flux vectors from all nodes constitutes the pollution potential energy field matrix. Each element in this matrix is ​​a vector containing magnitude and direction, representing the migration trend of the pollutant at that point in the next time step (i.e., the momentum of risk diffusion). Based on the calculated net flux vectors, the system uses the explicit finite volume method to perform discrete-time step updates to achieve the dynamic evolution of the field. The specific potential energy state update equation is as follows:

[0131]

[0132] In the formula: and They are respectively Time and Time grid nodes The potential energy scalar; The time step is adaptively adjusted according to CFL conditions; and Representing grid nodes respectively The upstream inflow neighbor node index and the downstream outflow neighbor node index; For grid nodes The effective neighbor set (i.e., all nodes with the same name) Spatially adjacent nodes); Indicates from neighboring nodes ( ) Inflow node The potential energy transfer rate is only if the potential energy is transferred from the potential energy source. Towards This item is non-zero when it is transmitted; Indicates from node Flow to neighboring nodes ( The potential energy transfer rate is only if the potential energy is transferred from the potential energy source. Towards This item is non-zero when it is transmitted; It is a first-order decay coefficient used to simulate the natural degradation or adsorption inhibition effect of pollutants.

[0133] Furthermore, the system calculates the gradient change vector of the pollution potential energy field matrix in the spatiotemporal dimension. Spatial gradient. It indicates the direction of the leading edge and the fastest path of the contamination plume; time gradient This indicates the rate of accumulation or decay of risk at that point. The magnitude of the gradient vector directly reflects the severity of the contamination evolution. This gradient information will be transmitted as a control signal to the subsequent adaptive mesh rendering control module to guide the mesh refinement or sparsification.

[0134] Example 5:

[0135] The adaptive mesh rendering control module is configured to dynamically adjust the local density of the visualized mapping mesh based on the magnitude of the gradient change vector.

[0136] In traditional computer graphics rendering, uniform meshes often lead to wasted computational resources (in background areas) or loss of detail (in frontal areas). This system introduces adaptive mesh subdivision technology, combined with an octree data structure for management. The specific control logic is as follows:

[0137] Threshold setting: The system sets a preset threshold for the gradient magnitude. (Segmentation threshold) and (Merge thresholds). These thresholds can be automatically adjusted based on the current viewport zoom level.

[0138] Mesh Traversal and Judgment: The system scans the gradient field data across the entire field in real time. For each basic mesh cell, the gradient magnitude within it is calculated.

[0139] Mesh encryption: When the gradient change vector exceeds a preset threshold In areas where the system identifies a pollution plume as being at the forefront of pollution diffusion, the edge of a source region, or a zone of abrupt changes in hydrogeological conditions, the risk distribution is considered complex. The system automatically splits the grid cell into eight sub-grids, creating high-density computational nodes. On these newly generated sub-nodes, potential energy and gradient values ​​are recalculated using trilinear interpolation or physics-based fluid interpolation algorithms. This process can be performed recursively until the gradient is less than a threshold or the minimum grid size (e.g., 0.1 meters) is reached. This results in extremely fine grids at the edges of the pollution plume, clearly revealing subtle features such as finger-like protrusions.

[0140] Mesh coarsening: When the gradient change vector is below the preset threshold In areas where the system identifies a potential risk, it determines that the area is either a background value zone or a central area with a uniform risk distribution. The system automatically merges eight adjacent sub-mesh areas into a single parent mesh, generating low-density computing nodes, thereby freeing up memory and video memory resources and improving the rendering frame rate.

[0141] A dynamic risk warning cloud map is generated based on the adjusted mesh nodes. The rendering engine (such as a shader developed based on WebGL) reads the vertex data of the adaptive mesh.

[0142] Color mapping: Based on the comprehensive health risk value, nodes are mapped to different colors.

[0143] Transparency mapping: Dynamically adjusts transparency based on gradient modulus or risk value. Low-risk areas are set to high transparency, allowing users to see high-risk pollution plumes underground through the surface soil.

[0144] Dynamic streamlines: Based on the vector direction in the pollution potential field matrix, virtual tracer particles are emitted by a particle system to draw lines that flow over time, intuitively indicating the migration path of pollutants.

[0145] The visual interactive terminal is used to display risk distribution maps. This terminal provides an immersive three-dimensional interactive environment.

[0146] Intelligent Early Warning: When the dynamic field evolution analysis module predicts that a high-risk potential energy field (red cloud map) will touch a sensitive receptor (such as groundwater wells, school boundaries, or river shorelines) at a certain future moment, the terminal will automatically trigger an early warning mechanism. At this time, the adaptive grid control module will force the grid of the contact area to be subdivided to the extreme to display the precise contact location and concentration flux, and pop up an alarm window on the screen displaying "Warning: The risk in area XX is expected to exceed the standard in 24 hours," and generate emergency response suggestions (such as suggested barrier wall locations and depths).

[0147] Interactive Analysis: Users can perform virtual drilling operations, click on any location on the map to view the geological structure and vertical pollution distribution at that point; they can also use the time machine slider to replay historical pollution processes or predict future trends.

[0148] Example 6:

[0149] In the dynamic evolution simulation of heavy metal pollution in soil, the spatial morphology of pollution plumes is often extremely complex, exhibiting high heterogeneity and anisotropy. Especially in areas where groundwater flow fields are abruptly redirected by geological lenses, fissures, or artificial structures, or at reaction fronts where multiple heavy metal pollutants exhibit geochemical antagonistic / synergistic effects, the spatial distribution of risk values ​​is no longer a smooth, gentle surface, but rather contains numerous abrupt change points, ridges, and saddle points. Traditional global uniform grid partitioning methods have certain shortcomings in such scenarios: if a high-resolution grid is used across the entire field to capture details, the computational load will explode exponentially, causing the system to fail to meet the timeliness requirements of real-time early warning; if a sparse grid is used to reduce the load, numerical dispersion and truncation errors will occur at key pollution fronts, leading to misjudgments of the pollution range. Therefore, this invention introduces a second-order differential control strategy based on risk curvature to achieve intelligent on-demand allocation of grid computing power.

[0150] The specific implementation steps of the adaptive mesh rendering control module to dynamically adjust the local density of the visualized mapping mesh are as follows:

[0151] Step 1: Constructing the full-field risk curvature field.

[0152] The system first initiates a full-field scan, reading the pollution potential energy field matrix generated in real time by the dynamic field evolution analysis module. This matrix is ​​a discretized numerical scalar field. To accurately identify the structural features within the field, the gradient change vector reflects the spatial rate of change of the risk gradient, and its magnitude is mathematically equivalent to the second derivative (i.e., curvature) of the risk surface. When the gradient vector undergoes a drastic change in space, it is inevitably accompanied by a large curvature. Therefore, this system specifically utilizes the Frobenius norm of the Hessian matrix to quantify this feature:

[0153] For two-dimensional grid nodes Its risk curvature The calculation formula is:

[0154]

[0155] in These are the second-order partial derivatives and mixed partial derivatives calculated using a five-point difference scheme, respectively. This indicator can simultaneously capture the curvature and distortion of the risk potential surface, thereby accurately locating the irregular abrupt boundary of the pollution front.

[0156] At the numerical computation level, considering the discreteness and noise characteristics of engineering data, the system preferably adopts the Laplacian operator of the higher-order central difference scheme or calculates the modulus of the Hessian matrix.

[0157] Specifically, for any basic node in the mesh The system utilizes the potential energy values ​​of its eight neighboring nodes to construct a local quadratic surface fitting model and analytically calculates the second-order partial derivative of that point in the X-axis direction. Second-order partial derivatives in the Y-axis direction and mixed partial derivatives The system combines these second-order differential components (e.g., the Frobenius norm taking the eigenvalues ​​of the Hessian matrix) and defines it as the risk curvature scalar at that point. Physically, this curvature directly reflects the degree of bending of the risk potential energy surface: a curvature close to zero means that the risk distribution in the area is flat or linearly gradual (such as a large background area or a homogeneous high-concentration core area); a curvature that is extremely large means that the risk distribution in the area is steep and distorted, often corresponding to the leading edge of pollution diffusion, the tip of finger-like protrusions, or the confluence of sources and sinks.

[0158] Step 2: Grid state determination based on dual threshold hysteresis comparison.

[0159] Based on the constructed risk curvature field, the system enters a dynamic adjustment process for the grid topology. To prevent the grid from falling into an oscillating state of frequent splitting and merging due to small numerical fluctuations near the threshold critical point, this system introduces dual-threshold control logic with hysteresis characteristics.

[0160] When the system detects that the risk curvature within a certain grid cell is greater than a first preset threshold (i.e., the splitting threshold), When the value is set to 0.5 in this embodiment, the region is determined to be in a nonlinear disturbance zone with extremely complex physical processes. At this time, the system triggers a mesh splitting command. This system uses the Octree algorithm to accurately divide the current parent mesh cell into eight child voxels in three-dimensional geometric space, and performs trilinear interpolation calculation on the risk value of the child voxels.

[0161] Conversely, when the system detects that the risk curvature of a certain grid cell and its sibling nodes (other child nodes belonging to the same parent node) is less than the second set threshold, the system triggers a grid merging command to merge eight adjacent grid cells into one parent grid.

[0162] Furthermore, in the calculation of risk curvature, the system preferably uses a higher-order central difference scheme to calculate the Frobenius norm of the Hessian matrix, thereby quantifying the local second-order rate of change of the scalar field.

[0163] Step 3: Data inheritance and interpolation based on physical conservation.

[0164] Changes in mesh topology are accompanied by the reconstruction of node data. During mesh splitting, newly generated child nodes (especially the center points located inside the parent mesh) lack original monitoring data. To ensure the energy conservation and continuity of the physics field before and after splitting, the system is configured to perform bilinear interpolation calculations on the risk values ​​of the child meshes. Specifically, the system uses the known potential energy values ​​of the parent mesh vertices, combined with the normalized position weights of the new child nodes in the local coordinate system, to construct a bilinear approximation equation and solve for the potential energy values ​​of the new nodes. For scenarios with higher accuracy requirements, the system can use bicubic spline interpolation to ensure the continuity of the first derivative (flow velocity).

[0165] During the grid merging process, to avoid information loss, the system uses averaging to update the risk values ​​of the parent grid. That is, the new value of the parent node is not simply taken from the value of a single child node, but rather the arithmetic mean or volume-weighted average of the risk values ​​of all merged child nodes is calculated. This process acts as a low-pass filter, preserving the macroscopic risk characteristics of the region.

[0166] Example 7:

[0167] In traditional site environmental investigation and remediation projects, monitoring data is often outdated. Managers can usually only see the past and present pollution distribution, but lack intuitive scientific judgment on how the pollution plume will migrate with groundwater in the next few days or weeks, and whether it will reach sensitive downstream targets (such as residential areas and drinking water wells).

[0168] Therefore, the system of this invention also includes a time-series prediction and extrapolation module. Considering that the migration and transformation process of heavy metals in soil is a typical complex spatiotemporal dynamic system controlled by the coupling of physical (convection, dispersion), chemical (adsorption, desorption, precipitation) and biological processes, and driven nonlinearly by external environmental factors such as rainfall and groundwater level fluctuations, traditional analytical solution models are difficult to apply, while numerical model parameter calibration is difficult. Therefore, this embodiment adopts a data-driven strategy.

[0169] The time-series prediction and extrapolation module is configured to perform deep learning on the historical evolution sequence of the pollution potential energy field matrix using a Long Short-Term Memory (LSTM) network model, particularly a Convolutional Long Short-Term Memory (ConvLSTM) variant incorporating convolution operations. The ConvLSTM network structure, by introducing convolution operations in the input-to-state and state-to-state transitions, can simultaneously extract time-dependent features (such as the cumulative effect of pollutants and seasonal trends) and spatial structural features (such as the morphological anisotropy and diffusion direction of plumes) from the data.

[0170] The specific workflow is as follows: The system constructs a sliding time window, which displays past data... ( The pollution potential energy field matrix is ​​serialized into a high-dimensional tensor for (positive integer) time steps (e.g., one snapshot per day over the past 30 days) and input into the network. It should be noted that...

[0171] Given that ConvLSTM networks require input data to have a fixed tensor dimension, while the front-end rendering uses a dynamic topology octree structure, this module has a built-in two-layer mesh bidirectional mapping engine.

[0172] Forward voxelization mapping: Before inputting the AI ​​model, the system iterates through the current time step. For unstructured octagonal leaf nodes, a rasterization filling algorithm is used to map discrete leaf node data to a preset resolution (e.g., ...). In the global rule voxel container, a standardized five-dimensional input tensor is constructed.

[0173] Inverse Feature-Driven Reconstruction: Model Output After predicting the tensor according to the rules at time step, the system does not directly render the tensor. Instead, it calculates the local gradient field of the predicted tensor. Only in regions where the gradient magnitude exceeds a preset threshold, the system sends a pre-splitting command to the adaptive mesh controller, pre-encrypting the mesh at the rendering end. This mechanism solves the data structure mismatch problem between the deep learning model and the adaptive rendering mesh.

[0174] When constructing the model input, this system encapsulates multi-source spatiotemporal data into a standardized five-dimensional tensor. .in, Batch size; The length of the historical time window (e.g., the past 30 days); The spatial resolution of the grid; The feature channel count is 5, specifically including five physical channels: 1) the comprehensive health risk field at the current moment; 2) the soil permeability distribution field; 3) the X-axis velocity component of groundwater; 4) the Y-axis velocity component of groundwater; and 5) the surface elevation gradient field. By introducing physical environmental parameters as independent channels, the neural network can implicitly learn the driving mechanism of environmental factors on risk evolution.

[0175] The network filters random noise through an internal forget gate, accepts new environmental driving variables through an input gate, and derives a sequence of predicted potential energy field matrices for a predetermined time step (e.g., 7 to 15 days) based on memory cells through an output gate. During model training, to prevent the deep learning model from making predictions that violate the law of conservation of mass, this system constructs a residual constraint loss function based on a Physical Information Neural Network (PINN) architecture. The specific formula is as follows:

[0176]

[0177] in, This represents the total loss function value of the physical information neural network. To predict the mean square error between the potential energy field and the historical measured potential energy field; The weighting coefficients for the physical residual term; This represents the square of the L2 norm of the physical residual term (i.e., the square of the Euclidean norm). The physical residual term of the convection-dispersion-reaction partial differential equation is defined as follows: In this definition: For risk potential energy scalar field; It is a time variable; For the Hamiltonian operator, here Represents divergence operation. Indicates gradient operation; This is the pore velocity vector; The hydrodynamic dispersion coefficient; It is the first-order decay coefficient (used to characterize the natural degradation or adsorption of pollutants).

[0178] By minimizing this physical residual term, the neural network's predictions are forced to satisfy the data fitting while following the unsteady migration patterns of pollutants in porous media, rather than incorrectly forcing the flux divergence to be zero (which would lead to incorrect steady-state assumptions). The weighting coefficients for the physical residual term.

[0179] This prediction result is applied to the feedforward stage of rendering control, forming a forward-looking mesh scheduling mechanism. The adaptive mesh rendering control module no longer adjusts the mesh solely based on the current measured data, but simultaneously reads the predicted potential energy field matrix. The system analyzes the predicted field and identifies areas that are not yet contaminated (low curvature) at the current moment, but will face high-risk diffusion at the predicted moment (i.e., the future contamination front arrival area).

[0180] Based on this, the system pre-defines the mesh in areas where high-risk spread is likely to occur in the future. This means that in the current rendering frame, the system forces the mesh of these potential risk areas to be subdivided to a high level of precision. When the actual pollution front reaches the area over time, the fine mesh carrier is already ready, and the system no longer needs to spend time on complex topology reconstruction; it only needs to update the node values, thus ensuring a significant smoothness and continuity of the visualized boundaries.

[0181] On the aforementioned visual interactive terminal, the system displays the projected trajectory of future risk boundaries in the form of a dynamic heatmap. Typically, the system overlays a semi-transparent layer with dashed outlines or flowing textures on top of the real-time measured cloud map. This layer displays the future pollution range calculated by the predictive model. Users can drag the timeline slider to visually present the dynamic diffusion and evolution trajectory of the pollution plume in the underground medium, and intuitively determine when it will cross the control threshold.

[0182] Example 8:

[0183] Any model-based prediction system will inevitably generate cumulative errors in the long run due to the non-stationarity of environmental parameters (such as changes in permeability caused by soil pore blockage).

[0184] To this end, the system also includes a feedback correction and update module, which is configured to immediately start a verification procedure when the multi-source data acquisition module receives a new batch of measured soil heavy metal concentration data in a new sampling period (e.g., daily timed sensor feedback or weekly laboratory test data entry).

[0185] The system first maps discrete measured data points to the current grid system through spatial interpolation and extracts the node data of the predicted potential field matrix under the same spatiotemporal coordinates. Then, the system calculates the deviation between the measured and predicted data. This deviation includes not only the root mean square error (RMSE) in numerical terms but also the structural similarity (SSIM) or overlap index (IoU) in spatial distribution.

[0186] If the deviation exceeds the system's preset tolerance range (e.g., the relative error of concentration at key control points exceeds 10%, or the deviation of the plume front position exceeds a set distance), the system determines that the current evolution analysis model parameters have drifted and cannot accurately reflect the actual physical state of the site. At this time, the system generates a strong parameter inversion gradient signal.

[0187] This signal is used to drive the parameter inversion subroutine. The system utilizes techniques such as gradient descent, genetic algorithms, or ensemble Kalman filtering (EnKF) to minimize bias and inversely adjust the basic hydrogeological parameters (such as permeability) in the dynamic field evolution analysis module using a data assimilation algorithm. ).

[0188] Specifically, the target functional is constructed as follows. :

[0189]

[0190] In the formula: Let be the scalar value of the objective functional to be minimized; It is a set of spatiotemporal observation points containing measured data. Represents a set The summation is performed on all observation points. Based on current penetration rate parameters The model's predicted values, This represents the corresponding measured risk value; the last term is the Tikhonov regularization term, used to constrain the smoothness of the parameter field; The L2 norm squared represents the spatial gradient of the permeability field; This is the regularization coefficient.

[0191] To efficiently solve the objective functional with respect to the full-field permeability parameter Based on the gradient, this system constructs a discrete adjoint equation corresponding to the forward evolution equation. The system obtains the adjoint variables by solving the adjoint equation through inverse time integration. After that, the gradient analytical expression of the objective functional with respect to permeability can be directly derived:

[0192]

[0193] In the formula: For the objective functional relative to the permeability parameter The gradient vector; The accompanying state variable vector (Lagrange multipliers); superscript This represents the matrix transpose operation; The coefficient matrix of the discretized system includes convection, dispersion, and reaction terms. This indicates that the matrix is ​​related to the parameters. The partial derivatives; This represents the overall predicted state vector.

[0194] Using this gradient information, the system employs the L-BFGS optimization algorithm to iteratively update the permeability field. This continues until the simulation bias converges.

[0195] For example, if the measured rate of pollutant diffusion is found to be significantly faster than predicted, the system will automatically increase the weight of soil permeability along the flow direction in that area or decrease the hindering factor. This process is essentially an automated data assimilation process, which allows the model parameters to continuously approximate the actual hydrogeological properties of the site.

[0196] After parameter correction, to ensure consistency across the entire field of data, the system triggers a full-field grid reconstruction process. Using the corrected parameter set and the latest measured data, the pollution potential energy distribution across the entire field is recalculated, and the topology of the adaptive grid is refreshed. This ensures that the system always simulates based on the latest environmental understanding. As runtime and data accumulation increase, the system's understanding of site heterogeneity deepens, and prediction accuracy continuously improves.

[0197] Example 9:

[0198] For frontline environmental regulators and engineering implementers, complex matrix data and model parameters are invisible; what they need are intuitive, clear, and action-oriented early warning signals.

[0199] To this end, the visual interactive terminal is equipped with a hierarchical early warning response unit, the core task of which is to perform dimensionality reduction and semantic processing on massive amounts of risk data. The hierarchical early warning response unit is configured to: execute strict risk zoning logic based on the values ​​of each pixel point (corresponding to the spatial location of the actual site) in the dynamic risk early warning cloud map, in accordance with the national "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard" or the project-specific control objectives.

[0200] The system divides site risks into three logical levels:

[0201] Safe Zone (Risk Value < Screening Value): The system maps this area to a semi-transparent green rendering layer. This area represents soil quality that meets planning requirements and requires no intervention. The layer is designed with high transparency (e.g., 90%), allowing users to see the underlying geological structure.

[0202] Warning Zone (screening value ≤ risk value ≤ control value): The system maps this zone to a semi-transparent yellow rendering layer. This area represents a potential risk, where pollutants are beginning to accumulate, requiring intensive monitoring.

[0203] Controlled Area (Risk Value > Control Value): The system maps it to a low-opacity red rendering layer and overlays a Fresnel edge glow effect to make it appear bright and solid in the 3D scene, ensuring that users can easily locate the core contamination cluster.

[0204] For the areas where the high-density computing nodes are located (these areas typically correspond to the core of pollution sources, high-risk fronts, or geological abrupt change zones), the system uses color blocks for display and automatically generates virtual buoys with coordinate labels and risk values. These buoys float above their corresponding coordinates in the 3D scene and are connected to specific grid points underground via guide lines. The buoy panel displays the depth of the point, the name of the main pollutant, the current concentration multiple, and the comprehensive risk index in real time. To avoid label stacking, the system introduces a force-guided layout algorithm to dynamically adjust the buoy positions.

[0205] Furthermore, to express the dynamic evolution of risk, the system introduces a visual variable with a time dimension. When the risk value of a certain area is detected to be continuously increasing within the most recent time window (i.e., the time gradient is positive), the system calculates its risk accumulation rate. The graded early warning response unit is configured to characterize the risk accumulation rate with a high-brightness flashing frequency: the faster the risk increases, the higher the frequency of the self-illuminating pulse of the virtual buoy or red warning zone (e.g., linearly mapping from a 0.5Hz breathing light effect to a rapid 5Hz flash).

[0206] Example 10:

[0207] The system is deployed on a cloud server cluster, preferably in a private cloud or industry-specific cloud environment, and is equipped with high-performance computing (HPC) nodes and a distributed storage system. This centralized deployment ensures data security and the elastic scaling of computing power.

[0208] At the perception layer, the multi-source data acquisition module is not directly connected to the server, but rather connects to the field sensor network through an industrial-grade IoT gateway interface. The IoT gateway possesses edge computing capabilities, supports lightweight protocols such as MQTT and CoAP, and is responsible for performing analog-to-digital conversion, noise reduction, compression, and encryption on the raw analog signals, transmitting them stably to the cloud via a 4G / 5G private network or fiber optic cable.

[0209] For the most computationally intensive stage of dynamic mesh adjustment and rendering, single-threaded computation is insufficient to meet the real-time interactive requirements of large-scale sites (millions of meshes). Therefore, the adaptive mesh rendering control module adopts a parallel computing architecture. The system utilizes OpenMP or CUDA parallel programming models to implement a spatial domain decomposition strategy.

[0210] The system divides the entire target site into several non-overlapping computational tiles in the horizontal direction, with each tile corresponding to an independent sub-region. The main control program dynamically allocates the mesh adjustment tasks (including curvature calculation, split determination, merge determination, and interpolation update) of different sub-regions to different computational threads or GPU cores for parallel execution.

[0211] To address the common boundary data synchronization problem in parallel computing, the system introduces a "Ghost Nodes" mechanism. This means that each sub-region maintains an additional set of boundary node copies of its adjacent regions in memory. When calculating derivatives or interpolations, threads directly read data from the Ghost Nodes, eliminating the need for frequent lock contention between threads.

[0212] After each sub-thread completes its calculations, it writes the generated local mesh vertex data and color indices into video memory. Finally, an image compositing operation is performed in the frame buffer to stitch together the complete dynamic risk warning cloud map.

[0213] The synthesized 3D visualization is streamed in real-time to the interactive visualization terminal using pixel streaming technology. The cloud server handles all graphics rendering, encoding the image into a high-definition video stream (such as H.265 format) and pushing it to the user's browser or large-screen terminal. The terminal only needs to handle video decoding and relaying user interaction commands (such as rotation, zoom, and click).

[0214] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A dynamic assessment and visualization intelligent early warning system for health risks from heavy metal pollution in soil at a site, comprising a multi-source data acquisition module, a data preprocessing module, a risk calculation engine module, and a visualization interactive terminal. The multi-source data acquisition module is used to acquire soil heavy metal concentration data and environmental parameters of the target site. The data preprocessing module is used to clean and standardize the soil heavy metal concentration data and the environmental parameters. The risk calculation engine module is used to calculate a health risk index based on the processed data. The visualization interactive terminal is used to display a risk distribution map. Its features are, The system also includes a dynamic field evolution analysis module and an adaptive mesh rendering control module; The dynamic field evolution analysis module is configured to construct a pollution potential energy field matrix characterizing the migration trend of heavy metal ions in the soil medium based on the soil heavy metal concentration data and the environmental parameters, and to calculate the gradient change vector of the pollution potential energy field matrix in the spatiotemporal dimension. The adaptive mesh rendering control module is configured to dynamically adjust the local density of the visualization mapping mesh according to the magnitude of the gradient change vector, generate high-density computing nodes in areas where the gradient change vector exceeds a preset threshold, generate low-density computing nodes in areas where the gradient change vector is lower than the preset threshold, and generate a dynamic risk warning cloud map based on the adjusted mesh nodes. The specific steps for constructing the pollution potential energy field matrix by the dynamic field evolution analysis module include: The target site is divided into an initial basic grid; Map the comprehensive health risk value of each grid node to a potential energy scalar; A correction coefficient based on Darcy's law is introduced to convert the groundwater flow direction data and soil permeability data in the environmental parameters into corresponding migration probability weights; the specific calculation method of the migration probability weights is as follows: Define nodes Pointing to neighboring nodes Unnormalized transition probability weights The calculation formula is: in, and They are nodes and nodes The hydrodynamic dispersion coefficient at that location; The distance between nodes is the Euclidean distance. For nodes The actual average flow velocity at that location; For the flow direction projection factor, when the flow velocity vector Vector connecting nodes The included angle hour, Otherwise, it is 0; and These are the weighting coefficients for the diffusion and convection terms, respectively, which are based on the grid Peklayt number of the local flow field. Dynamic calculation, definition ,in Let the molecular diffusion coefficient be set. , ; The migration probability weights are obtained by processing the unnormalized migration probability weights using a normalization factor. The potential energy scalar difference between adjacent grid nodes is weighted using the migration probability weights to generate the pollution potential energy field matrix containing direction and intensity information; The adaptive mesh rendering control module dynamically adjusts the local density of the visualized mapping mesh in the following way: By traversing the pollution potential energy field matrix, the second derivative of each local region is calculated to obtain the risk curvature; When the risk curvature is greater than the first set threshold, or when the absolute value of the comprehensive health risk of the grid node exceeds the preset control warning value, a grid splitting command is triggered to split the current grid cell into eight sub-grids using an octree algorithm, and to perform bilinear interpolation calculation on the risk value of the sub-grids. When the risk curvature is less than the second set threshold, a grid merging command is triggered to merge eight adjacent grid cells into a parent grid, and the risk value of the parent grid is updated using mean averaging.

2. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 1, characterized in that, The environmental parameters acquired by the multi-source data acquisition module include soil permeability data, groundwater flow direction data, site elevation data, soil dry bulk density data, and soil effective porosity data. The data preprocessing module is specifically configured to: use a spatial outlier detection algorithm to identify outliers in the soil heavy metal concentration data, and use neighborhood interpolation to complete the missing data to generate a standardized spatiotemporal data cube.

3. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 1, characterized in that, The risk calculation engine module has a built-in library of toxicity response parameters for various heavy metals. The risk calculation engine module is configured to: calculate the carcinogenic risk index and non-carcinogenic hazard quotient based on the soil heavy metal concentration data of each sampling point and the intake model of different exposure routes, and use the weighted sum of the carcinogenic risk index and the non-carcinogenic hazard quotient as the comprehensive health risk value of the sampling point.

4. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 3, characterized in that, The risk calculation engine module also includes a Nemerow Integrated Pollution Index calculation unit; The Nemerow Integrated Pollution Index calculation unit is configured to: extract the average and maximum values ​​of each individual heavy metal pollution index, calculate the integrated pollution level of the sampling point using the root mean square method, and attach the integrated pollution level as label data to the integrated health risk value.

5. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 1, characterized in that, The system also includes a time series prediction and deduction module; The time-series prediction and extrapolation module is configured to: use a long short-term memory network model to learn the historical evolution sequence of the pollution potential energy field matrix and generate a predicted potential energy field matrix with a preset time step in the future. The adaptive mesh rendering control module, based on the predicted potential energy field matrix, pre-defines the mesh in areas where high-risk diffusion may occur in the future, and displays the projected trajectory of the future risk boundary on the visualization interactive terminal in the form of a dynamic heat map.

6. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 5, characterized in that, The system also includes a feedback correction and update module; The feedback correction and update module is configured to: when a new batch of measured soil heavy metal concentration data is received, calculate the deviation between the measured data and the corresponding node of the predicted potential energy field matrix; If the deviation value exceeds the tolerance range, a parameter inversion gradient signal is generated, the migration weight parameter in the dynamic field evolution analysis module is adjusted, and the reconstruction process of the entire field mesh is triggered.

7. The intelligent early warning system for dynamic assessment and visualization of health risks from heavy metal pollution in site soil according to claim 1, characterized in that, The visual interactive terminal is equipped with a hierarchical early warning response unit; The hierarchical early warning response unit is configured as follows: Based on the values ​​of each pixel in the dynamic risk warning cloud map, the risk is divided into a safe zone, a warning zone, and a control zone, and mapped to semi-transparent rendering layers in green, yellow, and red tones, respectively. For the area where the high-density computing nodes are located, virtual buoys with coordinate labels and risk values ​​are automatically generated, and when an upward trend in risk values ​​is detected, the risk accumulation rate is represented by a high-brightness flashing frequency.

8. A dynamic assessment and visualized intelligent early warning system for health risks from heavy metal pollution in site soil according to any one of claims 1 to 7, characterized in that, The system is deployed on a cloud server cluster, and the multi-source data acquisition module is connected to the field sensor network through an IoT gateway. The adaptive mesh rendering control module adopts a parallel computing architecture, which allocates the mesh adjustment tasks of different sub-regions to different computing threads, and finally synthesizes the complete dynamic risk warning cloud map in the frame buffer and pushes it to the visualization interactive terminal.

Images