A plant site soil and groundwater pollution risk management and control method
By constructing a risk prediction model that integrates physical mechanisms and machine learning, and combining HYDRUS software and spatiotemporal attention graph neural networks, the risk zones of the plant area are accurately delineated. By adopting appropriate remediation technologies, the problems of inaccurate assessment and resource waste in the risk management of soil and groundwater pollution in the plant area are solved, and efficient and economical pollution control is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING INST OF ENVIRONMENTAL SCI MINIST OF ECOLOGY & ENVIRONMENT OF THE PEOPLES REPUBLIC OF CHINA
- Filing Date
- 2025-06-04
- Publication Date
- 2026-06-09
AI Technical Summary
Existing methods for managing soil and groundwater pollution risks in factory areas suffer from problems such as inadequate implementation of source control measures, easy leakage in end-of-pipe treatment, incomplete pollution monitoring coverage, and inaccurate risk assessment, resulting in waste or insufficiency of resources for pollutant treatment.
A risk prediction model integrating physical mechanisms and machine learning was constructed. The migration of pollutants was simulated using HYDRUS software, and risk assessment was carried out by combining spatiotemporal attention graph neural networks. High, medium and low risk zones were divided, and corresponding remediation technologies, such as in-situ chemical oxidation, infiltration reactive walls, phytoremediation and microbial remediation, were adopted, and the remediation strategy was dynamically adjusted.
It improves the accuracy of risk assessment and the efficiency of remediation, optimizes resource allocation, achieves precise remediation, avoids excessive intervention, forms a full-chain pollution control system, and enhances the pertinence and economy of pollution control.
Smart Images
Figure CN120394541B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water treatment technology, specifically to a method for risk management of soil and groundwater pollution in a factory area. Background Technology
[0002] Currently, the main methods for managing soil and groundwater pollution risks include traditional physical, chemical, and biological remediation technologies: physical remediation separates or immobilizes pollutants through methods such as soil replacement, electrokinetic remediation, and thermal desorption, which is suitable for shallow pollution but is costly; chemical remediation uses chemical leaching, oxidation / reduction, and solidification / stabilization reactions to change the form of pollutants, which is effective but may cause secondary pollution; bioremediation uses natural processes such as plant extraction and microbial degradation to remediate pollution, which is low-cost but has a long cycle; the selection of technologies needs to comprehensively consider the type of pollutants, the remediation cycle, cost, and the risk of secondary pollution. The future trend is to use combined remediation, the application of nanomaterials, and in-situ remediation technologies to improve efficiency and reduce environmental impact.
[0003] In existing technologies, risk management steps include source control, end-of-pipe treatment, pollution monitoring, and emergency response. Source control reduces the risk of pollutant leakage through measures such as optimizing equipment layout, anti-corrosion and seepage prevention treatments, and reducing underground pipeline laying. End-of-pipe treatment prevents the spread of pollutants through zoned seepage prevention, leak collection systems, and strict management of hazardous waste storage sites. Pollution monitoring systems rely on groundwater pollution monitoring wells and regular monitoring procedures to achieve early detection and treatment of pollution. However, these methods have significant limitations: source control relies on enterprises to proactively fulfill their responsibilities, but some enterprises fail to implement measures effectively due to insufficient understanding of regulations or lack of motivation; the seepage prevention layers of end-of-pipe treatment may leak after long-term use due to aging or construction defects, and the density and frequency of pollution monitoring wells are insufficient to cover all hidden pollution sources; while emergency response can control the spread of pollution, it cannot completely eliminate pollutants that have already seeped into the soil and groundwater.
[0004] Furthermore, existing risk management methods have limitations in predicting and assessing pollution migration. Traditional risk assessments often rely on limited sampling data and empirical models, making it difficult to accurately predict the dynamic changes of pollutants in soil and groundwater. This results in high uncertainty and conservatism in risk assessment results, potentially leading to wasted resources or inadequate remediation measures. Summary of the Invention
[0005] To address the aforementioned problems, this invention provides a method for risk management of soil and groundwater pollution in factory areas.
[0006] A method for risk management of soil and groundwater pollution in a factory area includes the following steps:
[0007] Obtain the plant area dataset for the current and historical moments; the plant area dataset includes geological parameters, pollutant monitoring data, and environmental dynamic data.
[0008] Based on the aforementioned factory area dataset, a risk prediction model is established. The inputs to the risk prediction model are geological parameters, pollutant monitoring data, and environmental dynamic data. The output of the risk prediction model is the risk level distribution within the factory area, which includes high-risk areas, medium-risk areas, and low-risk areas. The risk prediction model includes a physical transfer module for generating training data and physical constraints from the input dataset, a learning prediction module for performing spatiotemporal feature calculations based on the training data and physical constraints, and a dynamic decision-making module for risk classification output.
[0009] Risk levels within the plant area are divided and addressed based on a risk prediction model. For high-risk areas, in-situ chemical oxidation and permeable reactive barrier methods are used for short-term remediation. For medium-risk areas, phytoremediation or microbial remediation methods are used for medium- to long-term remediation. Low-risk areas do not require remediation.
[0010] Explanation: The above method can accurately predict the risk level distribution within a plant area based on geological parameters, pollutant monitoring data, and environmental dynamic data. By constructing a risk prediction model that integrates the advantages of physical mechanisms and machine learning, it precisely quantifies the spatial distribution of risk levels within the plant area. Based on the physical migration module to enhance data reliability, the learning prediction module to capture spatiotemporal evolution patterns, and the dynamic decision-making module to achieve risk classification output, this method not only improves the accuracy of risk assessment and remediation efficiency but also dynamically adjusts remediation strategies. This provides a scientific basis for developing efficient remediation plans for in-situ chemical oxidation and permeable reaction walls in high-risk areas and for adopting eco-friendly phytoremediation or microbial remediation technologies in medium-risk areas. At the same time, it avoids excessive intervention in low-risk areas, ultimately forming a full-chain management system of "data-driven - risk classification - precise remediation," significantly improving the targeting, economy, and environmental benefits of pollution control.
[0011] Furthermore, the geological parameters include permeability coefficient and porosity, the pollutant monitoring data includes pollutant concentration, and the environmental dynamic data includes meteorological data and hydrological dynamic data.
[0012] Note: The above explanation clarifies the specific content of the factory area dataset, providing comprehensive basic information for establishing a risk prediction model.
[0013] Furthermore, the method for establishing a risk prediction model based on the dataset includes:
[0014] Constructing a physical migration module: Select HYDRUS software that can couple and simulate saturated-unsaturated water flow, solute transport and multiple processes. Based on the plant area dataset, set migration equations in HYDRUS software to obtain a physical migration module that outputs pollutant concentration distribution, pollution source intensity and diffusion rate in soil and groundwater.
[0015] Constructing a learning prediction module: A spatiotemporal attention graph neural network is designed, which employs a multi-head self-attention mechanism to capture long-range dependencies, uses a dynamic graph convolution module to adaptively adjust the weights between nodes, and adds a mass conservation regularization term to the loss function; the learning prediction module is trained based on the concentration distribution of pollutants in soil and groundwater and environmental data to obtain the learning prediction module, and the output of the learning prediction module is the future time pollutant concentration distribution.
[0016] Based on the future pollutant concentration distribution obtained from the learning prediction module, high-risk, medium-risk, and low-risk areas are divided.
[0017] Explanation: The above method utilizes HYDRUS software to construct a physical migration module, accurately coupling saturated-unsaturated water flow with solute transport processes. This ensures that the simulation results of pollutant concentration distribution, pollution source intensity, and diffusion rate conform to physical laws, providing a reliable data foundation for the model. A spatiotemporal attention graph neural network is used as the learning and prediction module, employing a multi-head self-attention mechanism to capture the long-range spatiotemporal dependence of pollutant migration. Based on the prediction results, dynamic risk level classification is achieved, ensuring both the sensitivity of high-risk area identification and optimizing resource allocation in medium- and low-risk areas. This provides scientific and practical technical support for pollution control in industrial plants.
[0018] Furthermore, based on the future pollutant concentration distribution obtained by the learning prediction module, the method for dividing high-risk, medium-risk, and low-risk areas includes: when the predicted concentration of pollutants in the future time distribution is less than or equal to 50% of the risk threshold, it is classified as a low-risk area; when the predicted concentration of pollutants in the future time distribution is greater than 50% of the risk threshold but less than or equal to 75%, it is classified as a medium-risk area; and when the predicted concentration of pollutants in the future time distribution is greater than 75%, it is classified as a high-risk area.
[0019] Explanation: The above method directly maps the predicted pollutant concentrations to high, medium, and low risk areas by setting clear and quantifiable grading standards. This avoids the arbitrariness of subjective judgment and achieves a gradient distinction of risk levels through grading thresholds. This ensures that high-risk areas can accurately pinpoint the core areas of pollution spread for priority strong intervention measures, medium-risk areas can implement dynamic monitoring and ecological restoration for potential risks, and low-risk areas are free from over-treatment.
[0020] Furthermore, the risk threshold is an environmental risk screening value.
[0021] Note: The above environmental risk screening values are taken from the "Soil Environmental Quality Standard for Construction Land Soil Pollution Risk Control (Trial)" (GB 36600-2018), which achieves seamless integration of risk assessment and current environmental management standards, and ensures the compliance of risk classification by utilizing the legal authority of the screening values.
[0022] Furthermore, the environmental data includes groundwater level data, surrounding vegetation distribution data, and land use change and irrigation data.
[0023] Note: The above method incorporates key environmental data such as groundwater level, surrounding vegetation distribution, land use change, and irrigation to construct a multi-dimensional risk prediction input system. It reflects the impact of hydrogeological conditions on pollutant migration, reveals the natural purification capacity of ecosystems and pollution exposure risks by using vegetation distribution data, and captures the dynamic driving role of human activities on pollution diffusion by combining land use change and irrigation data, thereby significantly improving the comprehensiveness and accuracy of risk assessment.
[0024] Furthermore, in the HYDRUS software, different rainfall boundary conditions are set according to rainfall data of different seasons; the relationship between air temperature and soil temperature is established through empirical formulas or experimental data, and then the corresponding soil temperature boundary is set in the software.
[0025] Note: By setting seasonal rainfall boundary conditions and soil temperature boundary conditions, the model's spatiotemporal simulation accuracy of pollution migration processes is significantly improved: by adjusting the boundary conditions based on rainfall data from different seasons, the impact of seasonal differences in rainfall infiltration intensity and frequency on pollutant leaching and diffusion can be accurately characterized.
[0026] Furthermore, the method also includes:
[0027] Determine the repair cycle. When the repair cycle is completed, obtain the plant area dataset after the repair cycle is completed, and use the plant area dataset after the repair cycle is completed to update the risk prediction model.
[0028] Based on the aforementioned risk prediction model, new high-risk areas, medium-risk areas, and low-risk areas are predicted until the risk level distribution within the factory area is entirely low-risk, at which point control is completed.
[0029] Explanation: The above method, by setting a repair cycle and periodically acquiring post-repair plant data, enables real-time feedback and iterative optimization of the risk prediction model on the actual repair effect, avoiding prediction deviations caused by environmental changes in the static model; based on the updated model, it continuously predicts and adjusts risk areas, ensuring that control measures always focus on the current high-risk points until the overall risk is reduced to a low-risk level, thus avoiding the waste of resources caused by over-repair.
[0030] Furthermore, the method for determining the remediation cycle involves selecting multiple remediation cycles. Among these cycles, the numerical range for short-term remediation is 0.1 to 1 year, and the numerical range for medium- and long-term remediation is 0.5 to 5 years. The risk prediction model is used to predict the risk level area of pollutants after each cycle, and a remediation cycle within the range of 0.1 to 5 years is selected that makes the risk level areas close to or all of them low-risk areas.
[0031] Explanation: By pre-setting multiple candidate periods within the range of 0.1 to 5 years, and using a risk prediction model to quantify the spatial distribution of pollution risk after each period, the subjectivity of setting a single period is avoided. Furthermore, by using the degree of convergence of risk level areas to low-risk areas as the screening criterion, the selected period can balance remediation efficiency and cost, and maximize the reduction of environmental risks.
[0032] The beneficial effects of this invention are:
[0033] This invention's method can accurately predict the risk level distribution within a factory area based on geological parameters, pollutant monitoring data, and environmental dynamic data. By constructing a risk prediction model that integrates the advantages of physical mechanisms and machine learning, it precisely quantifies the spatial distribution of risk levels within the factory area. Based on a physical migration module to enhance data reliability, a learning prediction module to capture spatiotemporal evolution patterns, and a dynamic decision-making module to achieve risk grading output, this method not only improves the accuracy of risk assessment and remediation efficiency but also dynamically adjusts remediation strategies. This provides a scientific basis for developing efficient remediation schemes for in-situ chemical oxidation and permeable reaction walls in high-risk areas and for adopting eco-friendly plant or microbial remediation technologies in medium-risk areas. Simultaneously, it avoids excessive intervention in low-risk areas, ultimately forming a full-chain management system of "data-driven - risk grading - precise remediation," significantly improving the targeting, economic efficiency, and environmental benefits of pollution control. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the risk prediction model structure according to an embodiment of the present invention;
[0035] Figure 2 These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0036] Figure 3These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0037] Figure 4 These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0038] Figure 5 These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0039] Figure 6 These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0040] Figure 7 These are the MIP test PID results of some monitoring points in the factory area in this embodiment of the invention;
[0041] Figure 8 This is a schematic diagram of the method flow for recent repairs in an embodiment of the present invention. Detailed Implementation
[0042] To further illustrate the methods and effects of this invention, the technical solution of this invention will be clearly and completely described below in conjunction with experiments.
[0043] Traditional control methods often employ a one-size-fits-all approach, which cannot meet the needs of different pollution scenarios. Furthermore, in the existing pollutant remediation process, some pollutants require soil remediation to significantly reduce their risk, while others diffuse over time and automatically reduce their risk to soil and groundwater without requiring remediation. Therefore, accurately delineating risk areas can greatly improve the effectiveness of remediation and control and reduce unnecessary remediation processes.
[0044] In response, this invention constructs a risk prediction model that includes factors such as pollution source characteristics, hydrogeological conditions, and receptor exposure pathways. This model can quantitatively assess the probability of pollution diffusion and potential hazards in each region, and then classify them into high, medium, and low risk levels.
[0045] For example, in-situ remediation technologies (such as enhanced bioremediation and nano-zero-valent iron injection) should be prioritized in high-risk areas to block the spread of pollution; exposure risks in medium-risk areas can be reduced through engineering controls (such as seepage barriers) combined with institutional management; and monitoring should be the primary method in low-risk areas. This tiered management strategy can optimize resource allocation, achieve precise control of pollution risks, avoid over-remediation or under-treatment, and provide a scientific basis for the sustainable operation of the plant.
[0046] Example 1: A method for risk management of soil and groundwater pollution in a factory area, comprising the following steps:
[0047] S1. Obtain the plant area dataset for the current and historical moments; the plant area dataset includes geological parameters, pollutant monitoring data, and environmental dynamic data.
[0048] The aforementioned geological parameters include permeability and porosity; pollutant monitoring data includes pollutant concentration; and the environmental dynamic data includes meteorological data and hydrological dynamic data.
[0049] For example, at the current time (December 2024; in this embodiment, "current time" refers to December 2024), the plant's permeability coefficient is 1.2 × 10⁻⁶. -4 The porosity of 32 was obtained experimentally; the permeability coefficient of the plant area at the historical time (March 2024, all historical times mentioned below in this embodiment refer to March 2024) was 1.0 × 10⁻⁶. -4 The porosity of 30 was obtained experimentally.
[0050] Meteorological dynamic data includes temperature, humidity, wind speed, and precipitation; the current temperature is 8℃, humidity is 65%, wind speed is 3.2m / s, and precipitation is 2.0mm; the historical temperature is 15℃, humidity is 50%, wind speed is 2.8m / s, and precipitation is 3.4mm; the current groundwater level is 5.2m, and the surface water flow velocity is 0.15m / s. 3 / s; historical moment 4.9m, surface water velocity 0.12m 3 / s;
[0051] like Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 As shown, the pollutant concentration data in the pollutant monitoring data are as follows: The target pollutant in the above-mentioned factory area is ethylbenzene, and the concentration monitoring points include, for example, ethylbenzene. Figure 2-7 The monitoring results for the MIP01-MIP20 monitoring points shown include: the PID has a significant peak value between approximately 1 and 3.2 meters, with a maximum value of 5 × 107 μV, indicating that a certain amount of pollutants has accumulated.
[0052] The MIP11 values were relatively high at 1.6 meters and 2.2 meters, which may indicate the presence of contaminants.
[0053] The PID value at MIP18 is relatively high between 1 and 2 meters, possibly indicating the presence of contaminants. Other MIP measurement points have relatively low PID values. MIP17 shows relatively high values at 6 and 8 meters, possibly indicating the presence of contaminants; MIP18 shows a relatively high value at 2.6 meters, possibly indicating the presence of contaminants. Other MIP measurement points have relatively low FID values. MIP11 shows a significant XSD peak between approximately 2 and 9 meters, with a detected value greater than 0.5 × 10⁶ μV, indicating the accumulation of a certain amount of contaminants. MIP18 shows a significant XSD peak between 1 and 3 meters, indicating the accumulation of a certain amount of contaminants.
[0054] The XSD values at other MIP measurement points are relatively low; the PID, FID, and XSD values at MIP20 points are relatively small.
[0055] S2. Based on the aforementioned factory area dataset, establish a risk prediction model (such as...). Figure 1 As shown in the figure, the input of the risk prediction model is geological parameters, pollutant monitoring data and environmental dynamic data, and the output of the risk prediction model is the risk level distribution within the plant area, which includes high-risk areas, medium-risk areas and low-risk areas. The risk prediction model includes a physical transfer module for generating training data and physical constraints from the input dataset, a learning prediction module for performing spatiotemporal feature calculations based on the training data and physical constraints, and a dynamic decision-making module for risk classification output.
[0056] The method for establishing a risk prediction model based on the dataset includes:
[0057] S2-1. Constructing a physical migration module: Select HYDRUS software (3D software) that can couple and simulate saturated-unsaturated water flow, solute transport and multiple processes. Based on the plant area dataset, set the migration equation in HYDRUS software to obtain a physical migration module that outputs pollutant concentration distribution, pollution source intensity and diffusion rate in soil and groundwater.
[0058] Specifically, the water flow equation in the migration equation is the Richards equation to simulate saturated-unsaturated water flow; the solute transport equation in the migration equation is the convection-dispersion equation (ADRE) to simulate the migration of ethylbenzene.
[0059] In the boundary conditions, the surface is rainwater infiltration (Neumann boundary), and the bottom is free drainage (Dirichlet boundary, pressure head = 0);
[0060] S2-2, Constructing a learning prediction module: Design a spatiotemporal attention graph neural network, which employs a multi-head self-attention mechanism to capture long-range dependencies, uses a dynamic graph convolution module to adaptively adjust the weights between nodes, and adds a mass conservation regularization term to the loss function; train the learning prediction module based on the concentration distribution of pollutants in soil and groundwater and environmental data to obtain the learning prediction module, whose output is the future time pollutant concentration distribution;
[0061] The 3D mesh output by HYDRUS is converted into a graph structure, where each mesh node is a node in the graph. Node features include: geological parameters (permeability coefficient, porosity); pollutant concentration (current time step); environmental dynamic data (rainfall, temperature encoded as time series); and hourly slices are used to generate time series graph data (e.g., 7 days × 24 hours = 168 time steps).
[0062] Using a learnable similarity matrix, edge weights are calculated based on node features (such as concentration gradients and distances); multi-head self-attention (time dimension) is used to capture dependencies at different time steps, such as the delayed impact of rainfall events on pollution spread; long-range associations between high-risk areas (such as nodes that are far apart but hydraulically connected) are identified; the loss function is the mean squared error (MSE) used to compare predicted and actual concentrations.
[0063] Step S2-2 above is implemented using three tools: PyTorch Geometric (Graph Neural Network), DGL (Dynamic Graph Convolution), and Transformer (Attention).
[0064] S2-3. Based on the future time pollutant concentration distribution obtained from the learning prediction module, high-risk, medium-risk, and low-risk areas are divided. The method includes: when the predicted concentration of pollutants in the future time distribution is less than or equal to 50% of the risk threshold, it is classified as a low-risk area; when the predicted concentration of pollutants in the future time distribution is greater than 50% of the risk threshold but less than or equal to 75%, it is classified as a medium-risk area; when the predicted concentration of pollutants in the future time distribution is greater than 75%, it is classified as a high-risk area. The risk threshold is an environmental risk screening value, which is 28 mg / kg in the "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard (Trial)" (GB 36600-2018); it is implemented through ArcGIS API (spatial analysis).
[0065] The environmental data includes groundwater level data, surrounding vegetation distribution data, and land use change and irrigation data.
[0066] In the HYDRUS software, different rainfall boundary conditions are set according to rainfall data of different seasons; the relationship between air temperature and soil temperature is established through empirical formulas or experimental data, and then the corresponding soil temperature boundary is set in the software.
[0067] S3. Based on a risk prediction model, classify and remediate the risk level distribution within the plant area; employ in-situ chemical oxidation and permeable reactive barrier methods for short-term remediation; for medium-risk areas, employ phytoremediation or microbial remediation methods for medium- to long-term remediation; low-risk areas require no remediation; the method further includes:
[0068] Determine the repair cycle. When the repair cycle is completed, obtain the plant area dataset after the repair cycle is completed, and use the plant area dataset after the repair cycle is completed to update the risk prediction model.
[0069] Based on the aforementioned risk prediction model, new high-risk areas, medium-risk areas, and low-risk areas are predicted until the risk level distribution within the factory area is entirely low-risk, at which point control is completed.
[0070] The above method for determining the remediation cycle involves selecting multiple remediation cycles. Among these cycles, the numerical range for short-term remediation is 0.1 to 1 year, and the numerical range for medium- and long-term remediation is 0.5 to 5 years. The risk prediction model is used to predict the risk level area of pollutants after each cycle. Within the range of 0.1 to 5 years, a remediation cycle is selected that makes the risk level areas close to or all of them low-risk areas.
[0071] For example, near-term remediation using in-situ chemical oxidation and permeable reactive barrier methods includes:
[0072] Recent remediation measures for in-situ chemical oxidation (ISCO):
[0073] Organic pollutants (such as benzene compounds, petroleum hydrocarbons, and chlorinated hydrocarbons) in soil and groundwater are directly oxidized and degraded by injecting strong oxidants (such as Fenton's reagent, persulfate, and permanganate) into non-toxic or low-toxic products (such as CO2, water, or inorganic salts). Fenton's reagent, activated persulfate, or ozone are selected based on the nature of the pollutants. A high-pressure injection drilling rig is used to inject the reagent into the contaminated layer through the drill rod. Near-term remediation measures for permeable reactive walls (PRBs) include: setting up walls of activated materials (such as zero-valent iron and modified zeolite) along the path of the contaminated water flow to remove pollutants (such as Cr(VI), petroleum hydrocarbons, and heavy metals) through adsorption, reduction, or precipitation. The wall design adopts a continuous or funnel-gate type, and the wall thickness is calculated as a safety factor of 1.5 times the hydraulic retention time (HRT). For example, other near-term remediation methods include... Figure 8 The method shown.
[0074] The specific methods for medium- to long-term remediation using phytoremediation or microbial remediation methods are as follows:
[0075] Phytoremediation technologies include: Planting hyperaccumulating plants (such as Sedum aizoon and Centipede Grass) to absorb heavy metals from the soil through their roots and transfer them to the above-ground parts, followed by centralized treatment after harvesting. Soil improvement involves applying chelating agents to activate heavy metals and enhance plant absorption efficiency. Plant immobilization utilizes root exudates (such as organic acids) or adds passivating agents (phosphate fertilizer, biochar) to immobilize heavy metals and reduce their bioavailability. Microbial synergy involves inoculating mycorrhizal fungi or rhizosphere growth-promoting bacteria (PGPR) to promote plant growth and heavy metal absorption. Biological communities or transgenic microorganisms can be used to target and degrade recalcitrant pollutants (such as polycyclic aromatic hydrocarbons). Through these methods, phytoremediation and microbial remediation technologies can achieve pollutant stabilization or removal in the medium to long term, balancing ecological restoration and sustainability.
Claims
1. A method for risk management of soil and groundwater pollution in a factory area, characterized in that, Includes the following steps: Obtain the plant area dataset for the current and historical moments; the plant area dataset includes geological parameters, pollutant monitoring data, and environmental dynamic data. Based on the aforementioned factory area dataset, a risk prediction model is established. The inputs to the risk prediction model are geological parameters, pollutant monitoring data, and environmental dynamic data. The output of the risk prediction model is the risk level distribution within the factory area, which includes high-risk areas, medium-risk areas, and low-risk areas. The risk prediction model includes a physical transfer module for generating training data and physical constraints from the input dataset, a learning prediction module for performing spatiotemporal feature calculations based on the training data and physical constraints, and a dynamic decision module for risk classification output. Methods for establishing risk prediction models include: Constructing a physical migration module: Select HYDRUS software that can couple and simulate saturated-unsaturated water flow, solute transport and multiple processes. Based on the plant area dataset, set migration equations in HYDRUS software to obtain a physical migration module that outputs pollutant concentration distribution, pollution source intensity and diffusion rate in soil and groundwater. Constructing a learning prediction module: A spatiotemporal attention graph neural network is designed, which employs a multi-head self-attention mechanism to capture long-range dependencies, uses a dynamic graph convolution module to adaptively adjust the weights between nodes, and adds a mass conservation regularization term to the loss function; the learning prediction module is trained based on the concentration distribution of pollutants in soil and groundwater and environmental data to obtain the learning prediction module, and the output of the learning prediction module is the future time pollutant concentration distribution. The 3D mesh output by HYDRUS is converted into a graph structure, where each mesh node is a node in the graph. Node features include: geological parameters; pollutant concentration; environmental dynamic data; and hourly slices to generate time series plot data. We use a learnable similarity matrix to calculate edge weights based on node features; through multi-head self-attention, we capture dependencies at different time steps; we identify long-range associations between high-risk areas; and the loss function is the mean squared error, which is used to compare predicted and true concentrations. Based on the future pollutant concentration distribution obtained from the learning prediction module, high-risk, medium-risk, and low-risk areas are divided. Risk levels within the plant area are divided and addressed based on a risk prediction model. For high-risk areas, in-situ chemical oxidation and permeable reactive barrier methods are used for short-term remediation. For medium-risk areas, phytoremediation or microbial remediation methods are used for medium- to long-term remediation. Low-risk areas do not require remediation.
2. The method as described in claim 1, characterized in that, The geological parameters include permeability coefficient and porosity; the pollutant monitoring data includes pollutant concentration; and the environmental dynamic data includes meteorological data and hydrological dynamic data.
3. The method as described in claim 1, characterized in that, Based on the future pollutant concentration distribution obtained from the learning prediction module, the method for classifying high-risk, medium-risk, and low-risk areas includes: when the predicted concentration of pollutants in the future time distribution is less than or equal to 50% of the risk threshold, it is classified as a low-risk area; when the predicted concentration of pollutants in the future time distribution is greater than 50% of the risk threshold but less than or equal to 75%, it is classified as a medium-risk area; and when the predicted concentration of pollutants in the future time distribution is greater than 75%, it is classified as a high-risk area.
4. The method as described in claim 3, characterized in that, The risk threshold is an environmental risk screening value.
5. The method as described in claim 1, characterized in that, The environmental data includes groundwater level data, surrounding vegetation distribution data, and land use change and irrigation data.
6. The method as described in claim 1, characterized in that, In the HYDRUS software, different rainfall boundary conditions are set according to rainfall data of different seasons; the relationship between air temperature and soil temperature is established through empirical formulas or experimental data, and then the corresponding soil temperature boundary is set in the software.
7. The method as described in claim 1, characterized in that, The method further includes: Determine the repair cycle. When the repair cycle is completed, obtain the plant area dataset after the repair cycle is completed, and use the plant area dataset after the repair cycle is completed to update the risk prediction model. Based on the aforementioned risk prediction model, new high-risk areas, medium-risk areas, and low-risk areas are predicted until the risk level distribution within the factory area is entirely low-risk, at which point control is completed.
8. The method as described in claim 7, characterized in that, The method for determining the remediation cycle is to select multiple remediation cycles, wherein the numerical range of the short-term remediation is 0.1 to 1 year, and the numerical range of the medium- and long-term remediation is 0.5 to 5 years. The risk prediction model is used to predict the risk level area of pollutants after each cycle, and the remediation cycle within the range of 0.1 to 5 years is selected so that the risk level areas are close to or are all low-risk areas.