Intelligent decision-making method and system for groundwater pollution treatment

Abstract

The invention discloses an intelligent decision-making method and system for groundwater pollution treatment, and belongs to the technical field of environmental treatment. The method comprises the following steps: receiving feature data of a target pollution area, and intelligently retrieving similar historical governance cases from a case library; on the basis of a treatment scheme indicated by cases, a professional simulation process formed by sequentially coupling three models is automatically constructed and operated, and the treatment effect and the time cost are accurately simulated; geological data and a simulation result are combined, and the governance economic cost is automatically measured and calculated; and finally, inputting the case information, the simulation result and the economic cost into a large language model, carrying out multi-dimensional cross analysis, and automatically generating an intelligent evaluation report covering time, economy, efficiency and environmental influence. According to the method, case-based reasoning, professional numerical simulation, cost quantification and large language model intelligent analysis are deeply fused, the whole-process intelligence and quantification from pollution identification to scheme optimization are realized, and the efficiency and scientificity of treatment decision making are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of environmental governance technology, specifically relating to an intelligent decision-making method and system for groundwater pollution control. Background Technology

[0002] Groundwater is a vital resource supporting economic and social development and maintaining a healthy ecological environment; globally, 40% of agricultural irrigation water comes from groundwater. Groundwater pollution remediation has always been a crucial area of ​​environmental protection. Current groundwater pollution remediation methods primarily rely on manual analysis of pollution sources and remediation plans, and utilize traditional numerical simulation software (such as MODFLOW) to simulate groundwater flow and pollutant migration. However, these methods have the following shortcomings: Complex processes and low efficiency: Traditional groundwater pollution remediation processes involve significant manual intervention, especially in the data processing and report analysis stages. Groundwater pollution reports typically contain large amounts of unstructured data, such as text descriptions, charts, and images. Manually extracting this information is not only time-consuming but also prone to errors. The high complexity of manual operations leads to delays in the decision-making process, preventing remediation measures from responding promptly to changes in pollution levels.

[0003] Human evaluation lacks objective and accurate standards: Current assessments of groundwater pollution remediation plans typically rely on expert experience, employing a human approach to select and optimize plans. This method is highly dependent on individual experience, potentially leading to a lack of objectivity and standardization in the selection of remediation plans. For example, different experts may choose different remediation measures for the same pollution source based on different experiences and perspectives, resulting in inconsistent decision-making outcomes. Summary of the Invention

[0004] The main objective of this invention is to solve the technical problems of complex decision-making processes, low efficiency, and reliance on subjective judgment in existing groundwater pollution treatment technologies.

[0005] This invention provides an intelligent decision-making method and system for groundwater pollution control.

[0006] Specifically, the present invention provides an intelligent decision-making method for groundwater pollution control, comprising the following steps: S1. Receive pollution characteristic data of the target polluted area, and based on the pollution characteristic data, retrieve at least one similar historical remediation case from a pre-built remediation case database; S2. Based on the treatment schemes indicated by the similar historical treatment cases, construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, the MT3DMS solute migration model, and the PHT3D reaction transport model to simulate the implementation of the treatment scheme and obtain simulation results including the spatiotemporal evolution data of pollutant concentrations and the treatment time cost. S3. Based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan, calculate the economic cost of implementing the remediation plan. S4. Input the case information of the similar historical governance cases, the simulation results, and the governance economic costs into the big language model, and the big language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution. The intelligent evaluation report output by the large language model includes at least an analysis of the advantages and disadvantages of the governance plan in four dimensions: time cost, economic cost, governance efficiency, and environmental impact.

[0007] Furthermore, step S1 is detailed as follows: S11. Vectorize the pollution characteristic data of the target polluted area according to pollutant type, pollution source and pollution pathway to generate target feature vector; S12. Calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; S13. Sort the cases based on the comprehensive similarity score, and retrieve the top N cases with the highest comprehensive similarity as the similar historical governance cases.

[0008] Furthermore, step S2 is detailed as follows: S21. Based on the hydrogeological parameters of the target contaminated area, construct the MODFLOW groundwater flow model to simulate the groundwater head distribution and velocity field. S22. Based on the velocity field, the initial distribution of pollutants, and migration parameters, construct the MT3DMS solute migration model to simulate the convection and dispersion process of pollutants. S23. Based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, construct the PHT3D reaction transport model to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process; S24. Integrate the output of the above models to generate the spatiotemporal evolution data of the pollutant concentration and the treatment time cost.

[0009] Furthermore, step S3 specifically includes: S31. Based on the geological model data of the target contaminated area, determine the distribution area of ​​pollutants and the thickness of the contaminated strata; S32. Based on the difference between the initial pollutant concentration and the pollutant concentration after treatment in the simulation results, calculate the total amount of pollutants that need to be treated; S33. Multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

[0010] Furthermore, the method also includes: S5. Based on the spatiotemporal evolution data of pollutant concentration in the simulation results, drive the three-dimensional visualization engine to generate a three-dimensional dynamic visualization scene of the groundwater pollution and treatment process in the target polluted area. The three-dimensional dynamic visualization scene supports three-dimensional display of geological structures and pollutants, timeline animation playback, and interactive cross-sectional analysis.

[0011] A groundwater pollution remediation intelligent decision-making system includes: The case retrieval module is used to receive pollution characteristic data of the target polluted area and retrieve at least one similar historical remediation case from the pre-built remediation case database based on the pollution characteristic data. The simulation module, connected to the case retrieval module, is used to construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, MT3DMS solute migration model, and PHT3D reaction transport model based on the treatment scheme indicated by the similar historical treatment cases. The simulation module simulates the execution of the treatment scheme and outputs simulation results that include the spatiotemporal evolution data of pollutant concentration and the treatment time cost. The cost calculation module, connected to the case retrieval module and the simulation module, is used to calculate the economic cost of implementing the remediation plan based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan. The intelligent evaluation module, connected to the case retrieval module, the simulation module, and the cost calculation module, is used to input the case information of similar historical governance cases, the simulation results, and the governance economic costs into the large language model, and the large language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution.

[0012] Furthermore, the case retrieval module specifically includes: The data vectorization unit is used to vectorize the pollution characteristic data of the target polluted area according to the pollutant type, pollution source and pollution pathway to generate target feature vectors; A similarity calculation unit is used to calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; The case sorting and output unit is used to sort the cases based on the comprehensive similarity score and output the top N cases with the highest comprehensive similarity as the similar historical governance cases.

[0013] Furthermore, the simulation module specifically includes sequentially coupled components: The water flow simulation unit is used to construct and run the MODFLOW groundwater flow model based on the hydrogeological parameters of the target contaminated area to generate groundwater head distribution and velocity field. The solute migration simulation unit, connected to the water flow simulation unit, is used to construct and run the MT3DMS solute migration model based on the velocity field, the initial distribution of pollutants, and migration parameters, in order to simulate the convection and dispersion processes of pollutants. The chemical reaction simulation unit, connected to the solute migration simulation unit, is used to construct and run the PHT3D reaction transport model based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, in order to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process.

[0014] Furthermore, the cost calculation module specifically includes: The pollutant volume calculation unit is used to determine the pollutant distribution area and the thickness of the polluted stratum based on the geological model data of the target polluted area, and to calculate the total amount of pollutants that need to be treated by combining the concentration change data in the simulation results. The cost calculation unit is used to multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

[0015] The system also includes: A 3D visualization module, connected to the simulation module, is used to drive the 3D visualization engine based on the spatiotemporal evolution data of pollutant concentration in the simulation results, to generate and display a 3D dynamic visualization scene of groundwater pollution and treatment process in the target polluted area; wherein, the 3D visualization module supports 3D rendering of geological structures and pollutants, time-axis-based animation playback, and interactive cross-sectional analysis functions.

[0016] The beneficial effects of the technical solution provided by this invention are: 1. Significantly improve decision-making efficiency and automation: Through intelligent case retrieval, automatic construction and operation of simulation models, automatic cost calculation, and intelligent report generation, the entire chain of governance decision-making process is automated, shortening the traditional decision-making cycle of weeks or months to hours or even minutes, enabling rapid response to changes in the pollution situation.

[0017] 2. Enhancing the scientific and objective nature of decision-making: Centered on professional "MODFLOW-MT3DMS-PHT3D" coupled numerical simulation, it provides quantitative and high-precision predictions of governance effects, replacing empirical guesswork. Simultaneously, utilizing a large language model based on multi-source quantitative data (effects, time, cost) for standardized and multi-dimensional analysis, it outputs structured evaluation reports, overcoming the subjectivity and bias of human evaluation and making decision-making more objective and transparent.

[0018] 3. Achieve multi-objective comprehensive optimization of governance solutions: The system can simultaneously simulate, calculate costs, and comprehensively evaluate multiple alternative solutions, and quantitatively compare and weigh them across four key dimensions: time cost, economic cost, governance efficiency, and environmental impact. This helps decision-makers quickly identify the solution with the best technical, economic, and environmental benefits, thereby maximizing overall benefits.

[0019] 4. Enhance the understanding and communication of complex data: Through 3D dynamic visualization technology, abstract geological structures and obscure simulation data are transformed into intuitive and interactive animations of pollutant migration and treatment processes, which greatly reduces the threshold for understanding professional data and strongly supports cross-departmental collaborative consultations, scheme reviews and public communication. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a simplified flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention; Figure 3 This is a 3D model visualization of the geological strata in the treatment area; Figure 4 It is a 3D visualization diagram showing the distribution of pollutants in the geological strata of the remediation area; Figure 5 This is a schematic diagram of a case output of the intelligent evaluation and governance solution for large language models. Detailed Implementation

[0021] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] Example 1 Please refer to Figure 1 , Figure 1 This is a simplified flowchart of the method of the present invention, which specifically includes the following steps: S1. Receive pollution characteristic data of the target polluted area, and based on the pollution characteristic data, retrieve at least one similar historical remediation case from a pre-built remediation case database; It should be noted that step S1 is as follows: S11. Vectorize the pollution characteristic data of the target polluted area according to pollutant type, pollution source and pollution pathway to generate target feature vector; S12. Calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; S13. Sort the cases based on the comprehensive similarity score, and retrieve the top N cases with the highest comprehensive similarity as the similar historical governance cases.

[0023] As one embodiment, this invention retrieves case samples that meet the actual pollution control needs and objectives from a pollution control case database. The specific process is as follows: Feature engineering was used to design pollutant concentration, pollution source, and pollution pathway as feature vectors. The unit for pollutant concentration values ​​is mg / L. The feature vectors for pollution source and pollution pathway consist of 0s and 1s (see Table 1). 0 represents absence or non-existence, while 1 represents presence or existence.

[0024] Assume the pollution feature vector of the target contaminated area is The first case in the case database The feature vectors of each case are .

[0025] The nearest neighbor algorithm is used to calculate the similarity between cases in the database and the remediation areas. The cosine similarity between the polluted areas and the remediation cases is calculated in different dimensions (refer to Formula 1).

[0026] In this process, since different pollutants have different environmental risks and treatment difficulties, a weighting coefficient is introduced to weight and fuse the similarity of the two dimensions to strengthen the influence of key pollutant types in decision-making (refer to Formula 2). By ranking the similarity of cases, the five treatment cases most similar to the polluted area are finally retrieved as recommended cases (refer to Formula 3).

[0027] Table 1 Examples of Retrieval Vectors

[0028]

[0029] in This represents the feature vector of the target region in terms of pollutant type; Indicates the first Feature vectors of pollutant types from historical cases; The calculated cosine similarity value ranges from [0,1], with higher values ​​indicating greater similarity in pollutant composition. Similarly, similarity is calculated along the dimensions of pollution source and pathway. The formula is in the same form as above, but the input vector is replaced with... and . This is the pollutant dimension weighting coefficient, an adjustable parameter with a value range of [0, 1]. When the pollution scenario is dominated by a certain characteristic pollutant (such as heavy metals or persistent organic compounds), it can be used to enhance the pollutant's weighting coefficient. Set the value to a value greater than 0.5 (e.g., 0.7) to increase the weight of pollutant type similarity in the overall assessment. When the complexity of pollution sources and pathways becomes the main decision factor, the value can be adjusted downwards accordingly. The value. This coefficient can be optimized based on domain expert knowledge or historical decision data. Indicates the first The final overall similarity score between each case and the target region; the higher the score, the more similar the cases are overall; Sort() represents the descending sorting function; This indicates the preset number of recommended cases, which is usually set to a fixed value (N=5 in this embodiment of the invention) to provide sufficient but not overloaded reference information based on decision-making needs. This indicates that the output is a list containing... A list of the most similar case indexes or identifiers.

[0030] S2. Based on the treatment schemes indicated by the similar historical treatment cases, construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, the MT3DMS solute migration model, and the PHT3D reaction transport model to simulate the implementation of the treatment scheme and obtain simulation results including the spatiotemporal evolution data of pollutant concentrations and the treatment time cost. It should be noted that step S2 is as follows: S21. Based on the hydrogeological parameters of the target contaminated area, construct the MODFLOW groundwater flow model to simulate the groundwater head distribution and velocity field. S22. Based on the velocity field, the initial distribution of pollutants, and migration parameters, construct the MT3DMS solute migration model to simulate the convection and dispersion process of pollutants. S23. Based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, construct the PHT3D reaction transport model to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process; S24. Integrate the output of the above models to generate the spatiotemporal evolution data of the pollutant concentration and the treatment time cost.

[0031] The MODFLOW model is used to provide the groundwater head distribution and velocity field of the simulation area; the MT3DMS model simulates the convection and dispersion process of pollutants based on the velocity field; and the PHT3D model simulates the chemical reaction and concentration change between pollutants and remediation reagents based on the remediation reagent parameters indicated by the similar historical remediation cases.

[0032] The three professional models mentioned above are sequentially coupled. In subsequent governance schemes, they are automatically converted into structured input parameters that can be recognized by the MODFLOW, MT3DMS and PHT3D models through parameterization interfaces to drive the sequentially coupled simulation.

[0033] As one example, the pollution control simulation mainly undertakes the task of connecting and coordinating the information between the recommended cases in the control case database and the groundwater numerical simulation model in the prevention and control area.

[0034] Step S2 is responsible for converting the pollution control scheme generated by the large language model into standardized input information that can be directly called by the MODFLOW, MT3D and PHT3D numerical models, and relying on Flopy to realize the model construction, operation control and result summary.

[0035] By integrating with the MODFLOW model family, this step enables joint simulation of groundwater flow processes, contaminant migration and diffusion processes, and the interaction between remediation materials and contaminants.

[0036] It can receive structured governance scheme information output by the data extraction module and the large language model, automatically complete the configuration of model parameters and the organization and execution of multi-stage simulation tasks, and uniformly feed the simulation results back to the large language model for governance effect evaluation and comprehensive comparative analysis of economic and time costs, thereby supporting the continuous optimization and intelligent iteration of governance schemes. To achieve efficient conversion of natural language decision-making schemes to numerical models, the governance schemes output by the large language model adopt a unified structured description format. This structured scheme clearly distinguishes and standardizes the expression of hydrodynamic conditions, solute characteristics, and time scheduling information, ensuring that the governance schemes can be accurately mapped to the MODFLOW, MT3D, and PHT3D model systems supported by Flopy, thus guaranteeing the consistency and executability of information between the large language model decision and numerical simulation. Specifically, the details of this part are as follows: The system adopts a pipeline-style organizational structure, consisting of three coupled sub-modules: groundwater flow simulation, solute migration simulation, and pollutant control and reaction simulation. These are implemented based on MODFLOW, MT3D, and PHT3D, respectively, forming a complete simulation process covering "hydrodynamics - solute migration - chemical reaction".

[0037] In the groundwater flow simulation phase, a MODFLOW flow model was first constructed based on the spatial extent, resolution requirements, boundary conditions, hydrogeological structure, and exploration data of the study area. Spatial discretization determined the computational grid size and the number and type of aquifers, and constant head boundaries, general head boundaries, and recharge and discharge conditions were set. Simultaneously, combining actual engineering data and monitoring data from the treatment area, the locations of wells and injection / extraction rates were defined, and hydraulic parameters such as permeability coefficient and storage coefficient were assigned to each element. To ensure the stability and efficiency of the numerical calculation, the groundwater flow equations were solved iteratively using a PCG solver, with relevant convergence thresholds and maximum iteration counts set according to the model size and degree of heterogeneity. This phase outputs the groundwater head distribution and velocity field of the treatment area, providing basic hydrodynamic conditions for subsequent solute migration and reaction simulations.

[0038] Based on this, a solute migration model is constructed using MT3D to simulate the convection, dispersion, and source-sink processes of pollutants in the groundwater system. The spatial distribution of the contaminated zone is set through an initial concentration field, which can be defined as a uniform distribution, a zoned distribution, or a non-uniform distribution generated by interpolation based on monitoring data, inversion results, or historical survey data, thereby characterizing the initial range and intensity of the pollutants. The longitudinal and lateral dispersion, effective porosity, and adsorption or degradation parameters during the solute migration process are given by the structured remediation scheme output by the large language model. Simultaneously, the interactive module converts the stress cycle scheme of the reference case into time-discrete parameters in MODFLOW and MT3D, including the duration of each stress period, time step division, and steady-state or non-steady-state properties. It also dynamically adjusts the well injection / extraction flow rate, pollution source term intensity, and the start / stop status of remediation measures during different stress periods, thereby realizing the numerical expression of multi-stage remediation conditions.

[0039] After the solute migration simulation was completed, PHT3D was further introduced to construct a pollution prevention and reaction simulation model.

[0040] Based on the migration results from MT3D, this module combines the recommended remediation reagent selection, injection concentration, and chemical reaction system parameters from the searched cases to simulate the chemical reaction process between remediation materials and contaminants, including the transformation, decay, or immobilization mechanisms of contaminants. Through this coupling process, the changes in contaminant concentration over time after the implementation of remediation measures and their spatial distribution evolution in the groundwater system can be quantitatively characterized.

[0041] Ultimately, the simulation outputs a comprehensive result of groundwater flow and pollutant migration-response, including changes in groundwater head, spatial morphological evolution of the contaminated zone, and the temporal variation of key pollutant concentrations.

[0042] The pollutant concentration results can be used to analyze the degree of pollutant reduction, the time to reach the standard, and the remediation efficiency. They can also be fed back to the large language model as quantitative evaluation indicators for comprehensive comparison and intelligent optimization of different remediation schemes in terms of remediation effect, economic cost, and time cost.

[0043] As an example, it is assumed that the main pollutant in the target contaminated area is nitrate (NO3-N), the source of pollution is agricultural fertilizer, and the pollution pathway is intermittent infiltration. This invention has retrieved a similar historical case using a "permeable reactive barrier (PRB)" for remediation through step S1. The main process of step S2 is as follows: 1. Input reception and parameterization conversion Receive two types of input from upstream: Basic data for the target area include: the study area extent, grid discretization scheme, aquifer structure and parameters (permeability coefficient, storage coefficient), boundary conditions (constant head boundary, river boundary), initial head field, and initial spatial concentration distribution of nitrate pollutants (interpolated from monitoring data).

[0044] A structured remediation solution from a recommended case study: In this embodiment, the solution is "to install a permeable reactive barrier (PRB) filled with zero-valent iron (ZVI) downstream." This natural language solution is automatically converted into a set of instructions recognizable by the numerical model via a parameterized interface, including: PRB engineering parameters: wall location (specific grid row / column in the model), thickness (number of grids), permeability coefficient (set to 100 times that of the aquifer to simulate high permeability), and reactive material properties. Reaction parameters: the chemical reaction pathway between zero-valent iron (ZVI) and nitrate (NO3-N), and reaction kinetic constants (such as degradation rate constants). Time scheduling: the PRB construction and activation time (e.g., continuously effective from the 100th day after the start of the simulation).

[0045] 2. Construction and operation of sequential coupling model Based on the Flopy library, three specialized models are automatically built and coupled in the following order to form a complete simulation pipeline: (1) Construction and operation of MODFLOW groundwater flow model The purpose of MODFLOW is to simulate the groundwater seepage field under the influence of remediation projects, providing a "carrier" for contaminant migration. Based on the aforementioned basic data of the target area, a MODFLOW 6 "Groundwater Flow" (GWF) model was created. The hydraulic conductivity parameter of the grid cell containing the PRB wall was modified to a predetermined high value. A simulation stress period was set, with days 1-99 being the "background period" without PRB, and day 100 and onwards being the "remediation period." The PCG solver was used for solving the problem, with the convergence criterion set to a head change of less than 0.001 meters. Finally, the three-dimensional groundwater head distribution and velocity field at each time step throughout the entire simulation period were obtained. After the start of the remediation period, the velocity field will undergo a significant deflection near the PRB.

[0046] (2) Construction and operation of MT3DMS solute migration model The purpose of the MT3DMS solute migration model is to simulate the convection and dispersion processes of nitrate in groundwater. This invention creates the MT3DMS model and dynamically couples it with the GWF model from the previous step. Parameters such as the initial concentration field of nitrate, longitudinal and lateral dispersion, and effective porosity are loaded into the model. Crucially, the PRB is set as a "reactive transport boundary" with a specific concentration boundary during the remediation period (e.g., assuming the concentration at the wall inlet is consistent with the upstream), but its chemical reaction details are temporarily approximated by a simplified "first-order decay." Finally, preliminary spatiotemporal predictions of nitrate concentration are obtained, considering only physical migration and simple decay without considering complex chemical reactions.

[0047] (3) Construction and operation of the PHT3D reaction transport model The PHT3D reactive transport model aims to accurately simulate the biogeochemical reaction processes (such as reduction to nitrogen) between zero-valent iron and nitrate within the PRB. The PHT3D model, which tightly couples the migration engine of MT3DMS with the reaction engine of PHREEQC, is launched. Detailed chemical reaction networks from recommended cases (e.g., 4Fe₂O₃ + NO₃⁻ + 10H⁺ -> 4Fe²⁺ + NH₄⁺ + 3H₂O, and subsequent reactions) and their thermodynamic / kinetic parameters are loaded into the PHREEQC database. The PHT3D model inherits the water flow field and preliminary migration results from the MT3DMS solute transport model, but in the PRB region, the model is replaced with a "multi-component reactive transport simulation" incorporating the aforementioned complex reaction pathways. This ultimately yields high-precision, reaction-driven spatiotemporal evolution data of nitrate concentration. Furthermore, curves showing the decrease in nitrate concentration over time at downstream monitoring points in the PRB, and the process of pollutant range shrinkage, can be extracted.

[0048] Finally, the outputs of the three models are automatically integrated. Key outputs include: Spatiotemporal evolution data of pollutant concentrations: Stored in the form of a four-dimensional array (X, Y, Z, Time), recording the nitrate concentration of each grid cell at various time points. This data can be used to calculate the total pollutant reduction rate and the time to achieve compliance.

[0049] Treatment time cost: Directly read the total simulation time, or determine the "remediation cycle" by analyzing the time it takes for the concentration to reach the standard.

[0050] Supporting data: Changes in groundwater head and velocity fields after the implementation of the remediation project, used to assess the hydraulic impact of the project.

[0051] S3. Based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan, calculate the economic cost of implementing the remediation plan. It should be noted that step S3 specifically includes: S31. Based on the geological model data of the target contaminated area, determine the distribution area of ​​pollutants and the thickness of the contaminated strata; S32. Based on the difference between the initial pollutant concentration and the pollutant concentration after treatment in the simulation results, calculate the total amount of pollutants that need to be treated; S33. Multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

[0052] As an example, suppose the target contaminated area is a decommissioned industrial site, the main pollutant is petroleum hydrocarbons (TPH), and the plan is to use in-situ chemical oxidation (ISCO) technology for remediation.

[0053] First, receive the necessary data from the preceding steps, including: A 3D geological structure model of the target contaminated area, which has been digitized and can provide stratigraphic lithology, thickness and spatial boundaries at any location; Initial concentration distribution of pollutants from step S2 (c0): the initial time-stress concentration field from MT3DMS / PHT3D simulation, or an interpolated field constructed based on field monitoring data.

[0054] The target concentration distribution (c_t) after treatment in step S2 is based on the treatment plan (such as the concentration field at a future time point predicted after the injection of oxidant) or directly adopts the remediation target value specified by regulations (such as the TPH content in the soil being less than 100 mg / kg).

[0055] The unit pollutant treatment cost (price) retrieved from step S1 corresponding to the "In-situ Chemical Oxidation (ISCO)" treatment scheme. This cost is a dynamic value and may vary depending on the type of reagent (such as sodium persulfate, potassium permanganate), injection process, and local price levels, and is stored in the system's economic cost database.

[0056] Next, determine the spatial extent and volume of the pollutants.

[0057] First, based on the initial concentration c 0 and repair target c t The spatial extent of pollutants requiring remediation is determined. The specific method involves screening all pollutants in a three-dimensional geological model. c 0>c t The grid cells are labeled as "volumes of contaminants to be treated". For each grid cell to be treated... i The system reads its corresponding area. S i With effective contaminated stratum thickness h i .thickness h i It is not simply the total thickness of the formation, but the thickness that contributes to pollution, determined based on the vertical concentration distribution of pollutants.

[0058] pollutant volume V i = S i * h i Total pollutant volume V total = Σ ( S i * h i This sums over all grid cells to be governed. The summation covers all grid cells to be governed.

[0059] Secondly, calculate the total amount of pollutants that need to be treated.

[0060] It should be noted that this invention does not simply calculate volume, but rather the mass of pollutants, which is the basis for cost accounting.

[0061] For each grid cell i The mass of pollutants that need to be removed is: p i = V i * ρ bulk * (c 0,i} - c t,i ); ρ bulk This is the bulk density of the formation (mass per unit volume of soil / water-bearing medium, unit: kg / m³ or t / m³). This parameter is one of the fundamental properties of a geological model; c 0,i This is the initial pollutant concentration in the unit; c t,i This is the target concentration after treatment of this unit; Calculate the total amount of pollutant treatment p : p = Σ p i The summation covers all governance grid cells. Finally, the treatment cost is calculated. This invention can access an economic cost database to query the unit mass pollutant treatment cost (price) applicable to the "in-situ chemical oxidation treatment of petroleum hydrocarbons" scheme.

[0062] This price is typically a comprehensive unit price, including the cost of oxidizer, injection well construction, equipment rental, labor, monitoring, and all other related expenses (e.g., unit: yuan / kg pollutant); total economic cost of treatment. cost = Total amount of pollutants treated p Unit governance cost price .

[0063] Finally, the calculated total governance economic cost cost (Including time and economic costs), the environmental impact of historical governance solutions in the database is transmitted to the next step in a structured format (such as JSON).

[0064] S4. Input the case information of the similar historical governance cases, the simulation results, and the governance economic costs into the big language model, and the big language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution. The intelligent evaluation report output by the large language model includes at least an analysis of the advantages and disadvantages of the governance plan in four dimensions: time cost, economic cost, governance efficiency, and environmental impact.

[0065] Specifically, this invention uses a large language model to comprehensively retrieve case information, simulated governance time costs, and economic costs, and outputs intelligent evaluations of different implementation plans. It evaluates the advantages and disadvantages of different plans based on time cost, economic cost, governance efficiency, and environmental impact. Specifically: Suppose a combined pollution of ammonia nitrogen (NH4-N) and inorganic phosphorus in an agricultural area is identified. Two alternative remediation schemes are generated through retrieval and simulation: Scheme A (natural mineral precipitation) and Scheme B (sequencing batch filter - SUFR). This example demonstrates how a large language model can intelligently evaluate these two schemes.

[0066] Step S4 receives and integrates structured data packets from upstream steps S1-S3: Input from the case retrieval module: Case information for Option A: "Using a mixed mineral layer of natural zeolite and calcite, ammonia nitrogen and phosphorus are removed simultaneously through ion exchange and precipitation, which is suitable for intermittent pollution infiltration scenarios." Case information for Option B: "Using Sequencing Batch Filter (SUFR) technology, pollutants are removed by adsorption, filtration and biological action through filling with specific filler material, and it is highly adaptable to fluctuations in hydraulic load." Input from step S2: Simulation results for Scheme A: {"Treatment time cost": "Achieving Class III water standard in 18 months", "Key indicators": {"NH4-N removal rate": 92%, "Total phosphorus removal rate": 88%, "Plume volume reduction": 75%}}; Simulation results for Scheme B: {"Treatment time cost": "Achieving Class III water standard in 12 months", "Key indicators": {"NH4-N removal rate": 85%, "Total phosphorus removal rate": 95%, "Plume volume reduction": 70%}}; Inputs from step S3: Economic cost of scheme A: {"Total investment estimate": 1.5 million yuan", "Unit pollutant treatment cost": 12,000 yuan / ton}; Economic cost of scheme B: {"Total investment estimate": 2.2 million yuan", "Unit pollutant treatment cost": 18,000 yuan / ton}; Multi-dimensional cross-analysis and reasoning were performed using a large language model: The structured data mentioned above, serving as the core of the prompts, was input into a large language model fine-tuned based on reports from the groundwater remediation field. The model's built-in evaluation framework required cross-analysis from four dimensions: time cost, economic cost, remediation efficiency, and environmental impact.

[0067] The thought process of the large language model is illustrated below: Data Association and Contradiction Identification: The model identifies key contradictions—Option A has lower costs but takes longer; Option B has shorter time and higher phosphorus removal efficiency but significantly higher costs. Domain Knowledge Invocation: The model invokes its internal knowledge about "agricultural non-point source pollution" and "intermittent infiltration" characteristics to understand the challenge of unstable hydraulic load. Multi-dimensional Trade-off Analysis: Time Cost vs. Economic Cost: Option B saves 6 months of time but requires an additional investment of 700,000 yuan. The model calculates the "present value of time cost" to assess whether the reduced environmental risks from completing the treatment ahead of schedule outweigh the economic premium. Finally, a detailed comparison of treatment efficiency: The model indicates that Option A is better at removing ammonia nitrogen, while Option B is better at controlling phosphorus. Considering the current situation where phosphorus is the dominant pollutant, Option B is more suitable in terms of core objectives. Environmental Impact Considerations: The model analysis shows that the natural minerals used in Option A have extremely low environmental risks; Option B involves biological processes, and attention needs to be paid to potential bioclogging or secondary metabolite problems during long-term operation.

[0068] Based on analysis, the large language model generates a structured report that can be directly used for decision-making, as shown in Table 2 below: Table 2 Example of an analysis report

[0069] Finally, the generated intelligent assessment report is output in structured text and tabular formats, which can be integrated into the decision support platform for environmental engineers, managers, and review experts to access. The report is not a single conclusion, but rather provides clear trade-off analysis and contextualized recommendations, freeing decision-makers from tedious data comparisons and allowing them to focus on strategic decisions based on clear information.

[0070] It should be noted that the method also includes: S5. Based on the spatiotemporal evolution data of pollutant concentration in the simulation results, drive the three-dimensional visualization engine to generate a three-dimensional dynamic visualization scene of the groundwater pollution and treatment process in the target polluted area. The three-dimensional dynamic visualization scene supports three-dimensional display of geological structures and pollutants, timeline animation playback, and interactive cross-sectional analysis.

[0071] As one example, the core of visualization is to realize the three-dimensional spatialization, dynamism, and interactivity of groundwater pollution data. It mainly provides four functions: First, it provides an integrated display of all elements above and below ground, accurately overlaying two-dimensional geographic base maps, three-dimensional topography and underground geological structures, pollutants, monitoring wells, etc. in a unified view, intuitively revealing the spatial relationships of pollution.

[0072] Second, the pollution process is dynamically interpreted in time and space. Driven by timeline animation, the entire process of pollutant diffusion, migration and post-treatment degradation is dynamically simulated, transforming static simulation data into observable continuous evolution.

[0073] Thirdly, interactive query and 3D cross-sectional analysis are supported, allowing users to click to query facility attributes and use virtual cross-sections to see inside geological bodies and analyze the vertical distribution of pollutants, thus deepening the analysis from observation.

[0074] Fourth, it allows for comparison of the effects of multiple governance scenarios, supporting the side-by-side or overlay display of results from different simulation schemes, providing direct and intuitive visual evidence for scheme selection. These functions collectively transform professional data into intuitive insights, supporting scientific decision-making and collaborative consultation.

[0075] Specifically, firstly, this invention uses coordinate transformation and scene synchronization technology to accurately align the 3D local coordinates of Three.js with the geographic coordinates of MapLibre GL (such as Web Mercator) in real time, achieving seamless spatial integration of above-ground and underground objects. Secondly, for pollutant visualization, the grid concentration data output by MODFLOW / MT3DMS is used to generate dynamic isosurfaces using the MarchingCubes algorithm, and the concentration gradient is intuitively expressed through color mapping. Thirdly, to ensure smooth rendering of large-scale data, Level of Detail (LOD) technology is used to simplify the model according to the view distance, and clipping plane technology is used to achieve flexible geological cross-sections. Finally, through a data-driven architecture and state management, the invention dynamically binds and responds to changes in parameters such as time steps and scheme switching, driving real-time updates of the 3D scene, and connects with the backend simulation and intelligent evaluation modules through APIs, forming a complete closed loop from front-end interaction, back-end calculation to intelligent decision-making.

[0076] Example 2 Please refer to Figure 2 , Figure 2 This is a schematic diagram of the system structure of the present invention.

[0077] A groundwater pollution remediation intelligent decision-making system includes: The case retrieval module is used to receive pollution characteristic data of the target polluted area and retrieve at least one similar historical remediation case from the pre-built remediation case database based on the pollution characteristic data. It should be noted that the case retrieval module specifically includes: The data vectorization unit is used to vectorize the pollution characteristic data of the target polluted area according to the pollutant type, pollution source and pollution pathway to generate target feature vectors; A similarity calculation unit is used to calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; The case sorting and output unit is used to sort the cases based on the comprehensive similarity score and output the top N cases with the highest comprehensive similarity as the similar historical governance cases.

[0078] The simulation module, connected to the case retrieval module, is used to construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, MT3DMS solute migration model, and PHT3D reaction transport model based on the treatment scheme indicated by the similar historical treatment cases. The simulation module simulates the execution of the treatment scheme and outputs simulation results that include the spatiotemporal evolution data of pollutant concentration and the treatment time cost. It should be noted that the simulation module specifically includes sequentially coupled components: The water flow simulation unit is used to construct and run the MODFLOW groundwater flow model based on the hydrogeological parameters of the target contaminated area to generate groundwater head distribution and velocity field. The solute migration simulation unit, connected to the water flow simulation unit, is used to construct and run the MT3DMS solute migration model based on the velocity field, the initial distribution of pollutants, and migration parameters, in order to simulate the convection and dispersion processes of pollutants. The chemical reaction simulation unit, connected to the solute migration simulation unit, is used to construct and run the PHT3D reaction transport model based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, in order to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process.

[0079] The cost calculation module, connected to the case retrieval module and the simulation module, is used to calculate the economic cost of implementing the remediation plan based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan. It should be noted that the cost calculation module specifically includes: The pollutant volume calculation unit is used to determine the pollutant distribution area and the thickness of the polluted stratum based on the geological model data of the target polluted area, and to calculate the total amount of pollutants that need to be treated by combining the concentration change data in the simulation results. The cost calculation unit is used to multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

[0080] The intelligent evaluation module, connected to the case retrieval module, the simulation module, and the cost calculation module, is used to input the case information of similar historical governance cases, the simulation results, and the governance economic costs into the large language model, and the large language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution.

[0081] It should be noted that the system also includes: A 3D visualization module, connected to the simulation module, is used to drive the 3D visualization engine based on the spatiotemporal evolution data of pollutant concentration in the simulation results, to generate and display a 3D dynamic visualization scene of groundwater pollution and treatment process in the target polluted area. The 3D visualization module supports 3D rendering of geological structures and pollutants, time-axis-based animation playback, and interactive cross-sectional analysis.

[0082] Finally, please refer to Figures 3-5 . Figure 3 This is a 3D model visualization of the geological strata in the treatment area; Figure 4 It is a 3D visualization diagram showing the distribution of pollutants in the geological strata of the remediation area; Figure 5 This is a schematic diagram of a case output of the intelligent evaluation and governance solution for large language models.

[0083] In summary, the intelligent decision-making method and system for groundwater pollution control provided by this invention offers a more efficient and scientific technical solution for groundwater pollution control through pollutant distribution visualization, accurate numerical simulation, and intelligent decision evaluation. This system can effectively improve the accuracy and efficiency of pollution control and continuously optimize the control plan based on actual simulation results, thus possessing significant application value.

[0084] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart decision-making method for groundwater pollution control, characterized in that, Includes the following steps: S1. Receive pollution characteristic data of the target polluted area, and based on the pollution characteristic data, retrieve at least one similar historical remediation case from a pre-built remediation case database; S2. Based on the treatment schemes indicated by the similar historical treatment cases, construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, the MT3DMS solute migration model, and the PHT3D reaction transport model to simulate the implementation of the treatment scheme and obtain simulation results including the spatiotemporal evolution data of pollutant concentrations and the treatment time cost. S3. Based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan, calculate the economic cost of implementing the remediation plan. S4. Input the case information of the similar historical governance cases, the simulation results, and the governance economic costs into the big language model, and the big language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution. The intelligent evaluation report output by the large language model includes at least an analysis of the advantages and disadvantages of the governance plan in four dimensions: time cost, economic cost, governance efficiency, and environmental impact.

2. The intelligent decision-making method for groundwater pollution control as described in claim 1, characterized in that, Step S1 is as follows: S11. Vectorize the pollution characteristic data of the target polluted area according to pollutant type, pollution source and pollution pathway to generate target feature vector; S12. Calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; S13. Sort the cases based on the comprehensive similarity score, and retrieve the top N cases with the highest comprehensive similarity as the similar historical governance cases.

3. The intelligent decision-making method for groundwater pollution control as described in claim 1, characterized in that, Step S2 is as follows: S21. Based on the hydrogeological parameters of the target contaminated area, construct the MODFLOW groundwater flow model to simulate the groundwater head distribution and velocity field. S22. Based on the velocity field, the initial distribution of pollutants, and migration parameters, construct the MT3DMS solute migration model to simulate the convection and dispersion process of pollutants. S23. Based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, construct the PHT3D reaction transport model to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process; S24. Integrate the output of the above models to generate the spatiotemporal evolution data of the pollutant concentration and the treatment time cost.

4. The intelligent decision-making method for groundwater pollution control as described in claim 1, characterized in that, Step S3 specifically includes: S31. Based on the geological model data of the target contaminated area, determine the distribution area of ​​pollutants and the thickness of the contaminated strata; S32. Based on the difference between the initial pollutant concentration and the pollutant concentration after treatment in the simulation results, calculate the total amount of pollutants that need to be treated; S33. Multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

5. The intelligent decision-making method for groundwater pollution control as described in claim 1, characterized in that, The method also includes: S5. Based on the spatiotemporal evolution data of pollutant concentration in the simulation results, drive the three-dimensional visualization engine to generate a three-dimensional dynamic visualization scene of the groundwater pollution and treatment process in the target polluted area. The three-dimensional dynamic visualization scene supports three-dimensional display of geological structures and pollutants, timeline animation playback, and interactive cross-sectional analysis.

6. An intelligent decision-making system for groundwater pollution control, characterized in that, include: The case retrieval module is used to receive pollution characteristic data of the target polluted area and retrieve at least one similar historical remediation case from the pre-built remediation case database based on the pollution characteristic data. The simulation module, connected to the case retrieval module, is used to construct and run a groundwater pollutant prevention and control simulation model that couples the MODFLOW groundwater flow model, MT3DMS solute migration model, and PHT3D reaction transport model based on the treatment scheme indicated by the similar historical treatment cases. The simulation module simulates the execution of the treatment scheme and outputs simulation results that include the spatiotemporal evolution data of pollutant concentration and the treatment time cost. The cost calculation module, connected to the case retrieval module and the simulation module, is used to calculate the economic cost of implementing the remediation plan based on the geological data of the target polluted area, the total amount of pollutants, and the remediation plan. The intelligent evaluation module, connected to the case retrieval module, the simulation module, and the cost calculation module, is used to input the case information of similar historical governance cases, the simulation results, and the governance economic costs into the large language model, and the large language model performs multi-dimensional cross-analysis and outputs an intelligent evaluation report on the governance solution.

7. The intelligent decision-making system for groundwater pollution control as described in claim 6, characterized in that, The case retrieval module specifically includes: The data vectorization unit is used to vectorize the pollution characteristic data of the target polluted area according to the pollutant type, pollution source and pollution pathway to generate target feature vectors; A similarity calculation unit is used to calculate the comprehensive similarity score between the target feature vector and the feature vectors of each case in the governance case database; The case sorting and output unit is used to sort the cases based on the comprehensive similarity score and output the top N cases with the highest comprehensive similarity as the similar historical governance cases.

8. The intelligent decision-making system for groundwater pollution control as described in claim 6, characterized in that, The simulation module specifically includes sequentially coupled components: The water flow simulation unit is used to construct and run the MODFLOW groundwater flow model based on the hydrogeological parameters of the target contaminated area to generate groundwater head distribution and velocity field. The solute migration simulation unit, connected to the water flow simulation unit, is used to construct and run the MT3DMS solute migration model based on the velocity field, the initial distribution of pollutants, and migration parameters, in order to simulate the convection and dispersion processes of pollutants. The chemical reaction simulation unit, connected to the solute migration simulation unit, is used to construct and run the PHT3D reaction transport model based on the remediation reagents and chemical reaction parameters indicated by the similar historical remediation cases, in order to simulate the chemical reactions and concentration changes between pollutants and remediation reagents during the remediation process.

9. The intelligent decision-making system for groundwater pollution control as described in claim 6, characterized in that, The cost calculation module specifically includes: The pollutant volume calculation unit is used to determine the pollutant distribution area and the thickness of the polluted stratum based on the geological model data of the target polluted area, and to calculate the total amount of pollutants that need to be treated by combining the concentration change data in the simulation results. The cost calculation unit is used to multiply the total amount of pollutants by the pre-stored unit pollutant treatment cost corresponding to the treatment plan to obtain the treatment economic cost.

10. The intelligent decision-making system for groundwater pollution control as described in claim 6, characterized in that, The system also includes: A 3D visualization module, connected to the simulation module, is used to drive the 3D visualization engine based on the spatiotemporal evolution data of pollutant concentration in the simulation results, to generate and display a 3D dynamic visualization scene of groundwater pollution and treatment process in the target polluted area. The 3D visualization module supports 3D rendering of geological structures and pollutants, time-axis-based animation playback, and interactive cross-sectional analysis.

Images