Intelligent responsive design system for permeable reactive barrier for groundwater contamination

Abstract

The present application relates to the technical field of environmental remediation, and in particular to an intelligent responsive design system for a permeable reactive barrier (PRB) for groundwater contamination. A local intelligent responsive design system for a PRB for groundwater contamination constructs a plurality of information bases to comprehensively collect and organize relevant information about a target site, including hydrogeology, contaminants, and reactive media. By using the information to construct a machine learning model and a three-dimensional finite difference model, a PRB design scheme can be predicted on the basis of contaminant characteristics, and the actual site effectiveness of the scheme can be evaluated. By defining target conditions and constraint conditions, and using a genetic algorithm for optimization, a PRB design scheme that both meets remediation effectiveness requirements and conforms to actual situations is ultimately obtained, thereby providing a scientific, efficient and comprehensive solution for PRB design in groundwater contamination remediation.

Description

A Smart Responsive Permeable Reactive Wall Design System for Groundwater Pollution

[0001] Cross-reference to related applications

[0002] This application claims priority to Chinese Patent Application No. 202411887792.9, filed on December 20, 2024, entitled “A Smart Responsive Permeable Reactive Wall Design System for Groundwater Pollution,” the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of environmental remediation technology, and in particular to a smart responsive permeable reactive barrier design system for groundwater pollution. Background Technology

[0004] With increasing industrialization and urbanization, the concentration of pollutants in soil and groundwater is constantly rising, posing a significant threat to the environment and human health. In particular, the pollution of drinking water by nitrate nitrogen and heavy metals has severely impacted the groundwater environment and human health, necessitating effective remediation technologies. Groundwater remediation technologies are mainly divided into two categories: ex-situ remediation and in-situ remediation. In-situ remediation is favored due to its ease of operation and lower cost. Permeable reactive walls (PRBs), as an effective method in in-situ remediation, are widely used to intercept and purify pollutants in groundwater.

[0005] Key factors in PRB design include the width of the reactive barrier and the selection of the infill material, which directly affect the PRB's reaction efficiency and service life. Accurately determining the PRB thickness is crucial to ensuring its long-term effective treatment of contaminants. Traditional PRB width design methods, such as continuous column experiments based on breakthrough curves and empirical formula calculations, have limitations such as high time costs and insufficient explanatory power for mathematical regression models. These methods often fail to accurately reflect the complexity of the actual groundwater environment and contaminant migration, leading to discrepancies between design results and actual needs.

[0006] Machine learning algorithms, capable of handling complex data relationships and pattern recognition, have been widely applied in environmental assessment and prediction in recent years. Three-dimensional numerical simulation is also an important means of evaluating the effectiveness of permeable reactive barrier (PRB). This application applies a machine learning model to predict the width of the PRB to improve the efficiency and accuracy of the design, and forms a visualization module through a three-dimensional finite difference model. In summary, the research background of the intelligent responsive permeable reactive barrier design system for groundwater pollution mainly revolves around the severity of groundwater pollution problems, the effectiveness of in-situ remediation technologies, the importance of PRB design, the limitations of traditional design methods, and the introduction and potential applications of machine learning technology. Summary of the Invention

[0007] This application overcomes the shortcomings of the prior art and provides a smart responsive permeable reactive barrier design system for groundwater pollution.

[0008] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0009] This application discloses a smart responsive permeable reactive barrier design system for groundwater pollution, the system comprising:

[0010] A site information database is constructed, and three-dimensional hydrogeological parameters and borehole data of the target site are collected. Using the borehole data, a three-dimensional geological spatial distribution map and flow field distribution of the target site are generated based on spatial interpolation information. The three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site are then imported into the site information database.

[0011] A pollutant information database is constructed to obtain characteristic information of various pollutants, and the characteristic information of various pollutants is imported into the pollutant information database; a packing material information database is constructed to obtain characteristic information of various reaction packing materials, and the characteristic information of various reaction packing materials is imported into the packing material information database.

[0012] A matching information database is constructed, and historical monitoring data, pollution source information and pollutant migration patterns of the target site are collected. Based on the historical monitoring data, pollution source information and pollutant migration patterns of the target site, the available reaction packings for various pollutants are determined, and the available reaction packings for various pollutants are imported into the matching information database.

[0013] A machine learning model is constructed, and the characteristic information of various pollutants and available reactive packing materials are imported into the machine learning model for training to obtain a trained machine learning model.

[0014] The three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site are obtained from the site information database. Based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site, and combined with the finite difference method, a three-dimensional finite difference model of the target site is constructed.

[0015] The pollutant information of the target site is obtained, and the pollutant information of the target site is imported into a trained machine learning model to obtain the PRB wall design scheme of the target site; the PRB wall design scheme of the target site is imported into a three-dimensional finite difference model to obtain the final PRB wall design scheme of the target site.

[0016] More specifically, using borehole data, a three-dimensional geological spatial distribution map of the target site is generated based on spatial interpolation information, specifically as follows:

[0017] Obtain borehole data for each borehole in the target site. The borehole data includes the coordinate information of the borehole location, geological information, hydraulic conductivity, water level data, porosity data, rock layer thickness data, and stratum lithology.

[0018] A three-dimensional coordinate system is determined based on the scope of the target site, and the coordinates of each borehole are located in the coordinate system.

[0019] Using the stratigraphic lithology and thickness data at the borehole location as the basic data points, the stratigraphic lithology and thickness at each three-dimensional coordinate point within the target site are calculated using the Kriging interpolation method, starting from the bottom layer, for each stratigraphic layer.

[0020] In the horizontal direction, based on the planar position relationship and thickness variation of the boreholes, the interpolation range is gradually expanded to determine the distribution boundary of the same stratum on the plane;

[0021] In the vertical direction, based on the vertical relationship and thickness variation of the strata in different boreholes, the continuous distribution of the strata in the vertical direction is constructed by interpolation;

[0022] For geological structural data, including faults and folds, these are incorporated as constraints into the interpolation process. If there is displacement of the strata on both sides of a fault, the displacement is calibrated in three-dimensional space based on the fault location coordinates. The bending morphology of folds is constructed in three-dimensional space based on the corresponding morphology data in the borehole data.

[0023] By continuously integrating the interpolation results of various strata and geological structures, a three-dimensional geological spatial distribution map of the target site is finally formed.

[0024] More specifically, using borehole data, the flow field distribution of the target site is generated based on spatial interpolation information, specifically as follows:

[0025] Using the three-dimensional coordinate system of the target site as a framework, each borehole is located according to its coordinates. Using the water level data in the borehole as a feature, the water level values ​​of each three-dimensional coordinate point in the target site are obtained based on the inverse distance weighted interpolation method, thereby obtaining the spatial distribution of the water level.

[0026] Based on hydraulic conductivity data and Darcy's law, the hydraulic gradient at each three-dimensional coordinate point is calculated using the water level distribution. During the calculation process, the influence of the anisotropy of the strata on the hydraulic conductivity is taken into account, and the hydraulic gradient is calculated according to the characteristics of different strata.

[0027] Using the hydraulic conductivity and the calculated hydraulic gradient, the groundwater velocity vector is calculated again according to Darcy's law to determine the direction and speed of groundwater flow in the corresponding stratum.

[0028] The flow field distribution of the target site is constructed based on the direction and speed of groundwater flow in each stratum and the spatial distribution of water level.

[0029] More specifically, based on historical monitoring data of the target site, pollution source information, and pollutant migration patterns, the available reaction packing materials for various pollutants are determined, specifically:

[0030] Obtain historical monitoring data and pollution source information for the target site;

[0031] Based on the historical monitoring data, identify all pollutants that have appeared in the target site, and determine the emission intensity and emission mode of various pollutants that have appeared in the target site from the pollution source information;

[0032] The migration patterns of various pollutants that have appeared in the target site are obtained, and the dispersion characteristics of various pollutants that have appeared in the target site are determined based on the migration patterns; wherein, the dispersion characteristics include dispersion coefficient and dispersion direction.

[0033] And obtain characteristic information of various pollutants that have appeared in the target site from the pollutant information database;

[0034] Based on the characteristics, emission intensity, emission mode, and dispersion characteristics of various pollutants that have appeared in the target site, generate characteristic text information of various pollutants that have appeared in the target site.

[0035] The characteristic information of each reaction packing is obtained from the packing information database. A local sensitive attention mechanism is introduced, and the attention score between the characteristic information of each reaction packing and the characteristic text information of various pollutants that have appeared in the target site is analyzed according to the local sensitive attention mechanism.

[0036] The attention scores between the characteristic information of each reaction packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with preset score thresholds.

[0037] If the attention score between the characteristic information of a certain reactive packing material and the characteristic text information of a certain pollutant that has appeared in the target site is greater than a preset score threshold, then the reactive packing material is identified as a usable reactive packing material for the corresponding pollutant.

[0038] The available reactive packing materials for various pollutants that have appeared in the target site are obtained after the attention scores between the characteristic information of each reactive packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with the preset score threshold.

[0039] More specifically, based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map, and flow field distribution of the target site, and combined with the finite difference method, a three-dimensional finite difference model of the target site is constructed, specifically as follows:

[0040] Based on the three-dimensional geological spatial distribution map of the target site, the spatial layout, thickness variation, interlayer contact relationship and geological structural characteristics of different strata are obtained, thereby determining the spatial structure and stratigraphic boundary conditions of the model;

[0041] And based on the flow field distribution of the target site, obtain the spatial distribution of groundwater level, the magnitude and direction of hydraulic gradient, and the dynamic characteristics of water flow;

[0042] Based on the target site's geometry, size, spatial structure, geological boundary conditions, spatial distribution of groundwater level, magnitude and direction of hydraulic gradient, and dynamic characteristics of water flow, the target site is divided into three-dimensional spatial grids, resulting in several grid nodes.

[0043] For each grid node, the partial differential equation describing groundwater movement is discretized using the finite difference method to obtain the equivalent source and sink intensity values ​​for each grid node. At the same time, the zero flux boundary conditions for each grid node are determined based on the three-dimensional geological spatial distribution map and flow field distribution.

[0044] Substitute the equivalent values ​​of the source and sink intensity of each grid node into the corresponding positions in the discrete equation, and set the boundary constraints of the discrete equation according to the zero flux boundary conditions determined by the three-dimensional geological spatial distribution map and the flow field distribution.

[0045] Then, based on the source and sink information in the flow field distribution, the source and sink terms in the discrete equations are adjusted, and the discrete equations of all grid nodes are combined according to the three-dimensional spatial coordinate relationship to construct a three-dimensional finite difference model of the target site.

[0046] More specifically, a machine learning model is constructed, and the characteristic information of various pollutants and available reactive packing materials are imported into the machine learning model for training to obtain a trained machine learning model, specifically as follows:

[0047] The characteristic information of various pollutants is obtained from the pollutant information database, and the available reaction packings for various pollutants are obtained from the pairing information database.

[0048] Based on the characteristic information of various pollutants and the available reactive fillers, the data network is searched to obtain PRB wall design schemes for the treatment of various pollutants.

[0049] The characteristic information of various pollutants and available reactive fillers are used as input feature nodes, and the PRB wall design scheme for treating various pollutants is used as output feature nodes; a mapping relationship topology is constructed based on the input feature nodes and output feature nodes.

[0050] A machine learning model is constructed based on a graph neural network. The connection relationship between the input feature nodes and the output feature nodes in the mapping relationship topology graph is transformed into a data structure that the machine learning model can recognize. The information of the nodes and edges in the mapping relationship topology graph is encoded according to the graph general relationship and then embedded into the parameter space of the machine learning model for learning and training.

[0051] After each training cycle, the feature contribution of the machine learning model is obtained. If the feature contribution of the machine learning model gradually increases and exceeds a preset threshold, the training is completed, and the trained machine learning model is output.

[0052] More specifically, the process involves acquiring pollutant information from the target site, importing this information into a trained machine learning model to obtain a PRB wall design scheme for the target site, and then importing this PRB wall design scheme into a three-dimensional finite difference model to obtain the final PRB wall design scheme for the target site.

[0053] The pollutant information of the target site is obtained and imported into the trained machine learning model to obtain the PRB wall design scheme for the target site; the PRB wall design scheme includes the PRB wall design thickness, location and filler.

[0054] The PRB wall design scheme for the target site is imported into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall.

[0055] The target conditions are defined based on minimizing pollutant concentration and maximizing PRB lifespan, and the constraints are set based on geological conditions, cost budget, and construction feasibility.

[0056] Determine whether the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints; if the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints, then the PRB wall design scheme is directly output as the final PRB wall design scheme.

[0057] Otherwise, a genetic algorithm is introduced, and an iterative optimization cycle is set. The PRB wall design scheme is adjusted and optimized according to the genetic algorithm. After each iterative optimization cycle is completed, the optimized PRB wall design scheme is obtained.

[0058] The optimized PRB wall design scheme is input into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall;

[0059] The iteration stops when the numerical simulation results of the permeable reactive wall of the optimized PRB wall design scheme meet the target conditions and constraints, and the optimized PRB wall is output as the final PRB wall design scheme.

[0060] More specifically, the three-dimensional hydrogeological parameters include strata, water level, horizontal / vertical permeability coefficient, groundwater flow velocity, porosity, and water storage rate.

[0061] More specifically, the characteristic information of the reaction packing includes material, ratio, size, specific surface area, and reaction rate constant; the characteristic information of the pollutants includes the negative logarithm of the acid dissociation constant and molecular weight.

[0062] This application addresses the technical deficiencies in the prior art and offers the following beneficial effects: The local intelligent responsive permeable reactive barrier (PRB) design system for groundwater pollution comprehensively collects and organizes relevant information on the target site's hydrogeology, pollutants, and reactive filler materials by constructing multiple databases. Utilizing this information, machine learning models and three-dimensional finite difference models are built, enabling the prediction of PRB wall design schemes based on pollutant characteristics and the evaluation of the schemes' actual site effects. By defining target and constraint conditions and employing genetic algorithms for optimization, a PRB wall design scheme that meets both remediation requirements and practical realities is ultimately obtained, thus providing a scientific, efficient, and comprehensive solution for permeable reactive barrier design in groundwater pollution remediation. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0064] Figure 1 is a framework diagram of a local wastewater pollution intelligent responsive permeable reactive barrier design system.

[0065] Figure 2 is a flowchart of the local groundwater pollution intelligent responsive permeable reactive barrier design system. Detailed Implementation

[0066] To better understand the above-mentioned objectives, features, and advantages of this application, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0067] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below.

[0068] As shown in Figures 1 and 2, this application discloses a smart responsive permeable reactive barrier design system for groundwater pollution. The design system includes:

[0069] S102: Construct a site information database, collect three-dimensional hydrogeological parameters and borehole data of the target site, and use the borehole data to generate a three-dimensional geological spatial distribution map and flow field distribution of the target site based on spatial interpolation information, and import the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site into the site information database.

[0070] S104: Construct a pollutant information database, obtain characteristic information of various pollutants, and import the characteristic information of various pollutants into the pollutant information database; construct a packing information database, obtain characteristic information of various reaction packings, and import the characteristic information of various reaction packings into the packing information database.

[0071] S106: Construct a pairing information database, collect historical monitoring data, pollution source information and pollutant migration patterns of the target site, determine the available reactive packing materials for various pollutants based on the historical monitoring data, pollution source information and pollutant migration patterns of the target site, and import the available reactive packing materials for various pollutants into the pairing information database.

[0072] S108: Construct a machine learning model, and train the machine learning model by importing the characteristic information of various pollutants and the available reactive packing materials into the machine learning model to obtain a trained machine learning model.

[0073] S110: Obtain the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site from the site information database, and construct a three-dimensional finite difference model of the target site based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site and in combination with the finite difference method.

[0074] S112: Obtain pollutant information of the target site, import the pollutant information of the target site into the trained machine learning model, obtain the PRB wall design scheme of the target site; import the PRB wall design scheme of the target site into the three-dimensional finite difference model, obtain the final PRB wall design scheme of the target site.

[0075] The three-dimensional hydrogeological parameters include strata, water level, horizontal / vertical permeability coefficient, groundwater flow velocity, porosity, and storage capacity; the characteristic information of the reaction packing includes material, proportion, size, specific surface area, and reaction rate constant; and the characteristic information of the pollutants includes the negative logarithm of the acid dissociation constant and molecular weight of the pollutants.

[0076] It should be noted that the local intelligent responsive permeable reactive barrier (PRB) design system for groundwater pollution comprehensively collects and organizes relevant information on the target site's hydrogeology, pollutants, and reactive barrier materials by constructing multiple databases. Using this information, machine learning models and three-dimensional finite difference models are built. The former can predict PRB design schemes based on pollutant characteristics, while the latter can evaluate the actual site effects of the schemes. By defining target conditions and constraints and employing genetic algorithms for optimization, a PRB design scheme is ultimately obtained that satisfies both remediation requirements (such as minimizing pollutant concentration and maximizing PRB lifetime) and practical considerations (geological conditions, cost budget, and construction feasibility). This provides a scientific, efficient, and comprehensive solution for permeable reactive barrier design in groundwater pollution remediation.

[0077] More specifically, using borehole data, a three-dimensional geological spatial distribution map of the target site is generated based on spatial interpolation information, specifically as follows:

[0078] Obtain borehole data for each borehole in the target site. The borehole data includes the coordinate information of the borehole location, geological information, hydraulic conductivity, water level data, porosity data, rock layer thickness data, and stratum lithology.

[0079] A three-dimensional coordinate system is determined based on the scope of the target site, and the coordinates of each borehole are located in the coordinate system.

[0080] Using the stratigraphic lithology and thickness data at the borehole location as the basic data points, the stratigraphic lithology and thickness at each three-dimensional coordinate point within the target site are calculated using the Kriging interpolation method, starting from the bottom layer, for each stratigraphic layer.

[0081] In the horizontal direction, based on the planar position relationship and thickness variation of the boreholes, the interpolation range is gradually expanded to determine the distribution boundary of the same stratum on the plane;

[0082] In the vertical direction, based on the vertical relationship and thickness variation of the strata in different boreholes, the continuous distribution of the strata in the vertical direction is constructed by interpolation;

[0083] For geological structural data, including faults and folds, these are incorporated as constraints into the interpolation process. If there is displacement of the strata on both sides of a fault, the displacement is calibrated in three-dimensional space based on the fault location coordinates. The bending morphology of folds is constructed in three-dimensional space based on the corresponding morphology data in the borehole data.

[0084] By continuously integrating the interpolation results of various strata and geological structures, a three-dimensional geological spatial distribution map of the target site is finally formed.

[0085] Specifically, the first step is to acquire abundant borehole data from each borehole in the target site. This data encompasses coordinate information of the borehole location (the foundation for determining the borehole's spatial position), geological information (such as stratigraphic lithology, a key element in constructing the geological spatial distribution), hydraulic conductivity (related to groundwater flow), water level data (reflecting groundwater conditions), porosity data (affecting formation water storage capacity), and stratum thickness data (an important parameter for determining stratigraphic structure). A three-dimensional coordinate system is determined based on the scope of the target site, providing a framework for the spatial positioning of all subsequent data. The coordinates of each borehole are located within this coordinate system, ensuring that the borehole data accurately corresponds to the three-dimensional space of the target site. Using the stratigraphic lithology and thickness data at the borehole locations as the base data points, Kriging interpolation is employed. Calculations are performed for each stratum, starting from the bottom layer. In the horizontal direction, the interpolation range is expanded by considering the planar positional relationships and thickness variations of the boreholes, thereby determining the distribution boundary of the same stratum on the plane. In the vertical direction, interpolation is performed based on the vertical relationships and thickness variations of the strata in different boreholes to construct a continuous vertical distribution of the strata. This interpolation method can reasonably infer the stratigraphic distribution of the entire target site based on limited borehole data. Geological structural data (faults and folds) are incorporated as constraints into the interpolation process. For faults, if there is displacement of the strata on both sides of the fault, this displacement relationship is accurately marked in three-dimensional space based on the fault's location coordinates. For folds, their bending morphology is constructed in three-dimensional space based on the corresponding morphological data from the borehole data. This allows geological structures to be accurately represented in the three-dimensional geological spatial distribution map, more realistically reflecting the geological conditions of the target site. By continuously integrating the interpolation results of various strata and geological structures, the distribution of all strata and the information on geological structures are combined to ultimately form a complete three-dimensional geological spatial distribution map of the target site.

[0086] Through the above steps, a comprehensive and accurate three-dimensional geological spatial distribution map of the target site can be constructed using limited borehole data. This method fully considers various factors such as lithology, thickness, and geological structure of the strata, and reasonably infers the geological conditions of the entire site through Kriging interpolation, so that the constructed three-dimensional geological spatial distribution map can accurately reflect the stratigraphic distribution, geological structural relationships, and other information of the target site.

[0087] More specifically, using borehole data, the flow field distribution of the target site is generated based on spatial interpolation information, specifically as follows:

[0088] Using the three-dimensional coordinate system of the target site as a framework, each borehole is located according to its coordinates. Using the water level data in the borehole as a feature, the water level values ​​of each three-dimensional coordinate point in the target site are obtained based on the inverse distance weighted interpolation method, thereby obtaining the spatial distribution of the water level.

[0089] Based on hydraulic conductivity data and Darcy's law, the hydraulic gradient at each three-dimensional coordinate point is calculated using the water level distribution. During the calculation process, the influence of the anisotropy of the strata on the hydraulic conductivity is taken into account, and the hydraulic gradient is calculated according to the characteristics of different strata.

[0090] Using the hydraulic conductivity and the calculated hydraulic gradient, the groundwater velocity vector is calculated again according to Darcy's law to determine the direction and speed of groundwater flow in the corresponding stratum.

[0091] The flow field distribution of the target site is constructed based on the direction and speed of groundwater flow in each stratum and the spatial distribution of water level.

[0092] Specifically, using the established three-dimensional coordinate system of the target site as a framework, each borehole is positioned within this system according to its coordinates. Then, based on the water level data in the boreholes, the inverse distance weighted interpolation method is used to calculate the water level values ​​at each three-dimensional coordinate point within the target site. The principle of the inverse distance weighted interpolation method is that unknown points closer to known points are more influenced by those known points. This method allows for a reasonable estimation of the spatial distribution of water level across the entire site based on the borehole water level data. The hydraulic gradient at each three-dimensional coordinate point is calculated based on hydraulic conductivity data and Darcy's law (which describes the relationship between fluid flow rate, hydraulic gradient, and hydraulic conductivity in porous media). In this process, the anisotropy of the formation on hydraulic conductivity is fully considered. Due to the different characteristics of different formations (such as rock structure and porosity), hydraulic conductivity may vary in different directions. Therefore, it is necessary to accurately calculate the hydraulic gradient based on the characteristics of different formations to reflect the actual driving force of groundwater flow.

[0093] Using the known hydraulic conductivity and the calculated hydraulic gradient, the groundwater velocity vector is calculated again according to Darcy's law. The velocity vector contains information about the direction and magnitude of groundwater flow, determining the direction and magnitude of groundwater flow in the corresponding strata. Finally, the flow field distribution of the target site is constructed by combining the direction and magnitude of groundwater flow in each stratum with the spatial distribution of water levels. The flow field distribution can visually demonstrate the flow state of groundwater within the target site, including information such as water flow velocity, direction, and water level at different locations.

[0094] Through the above steps, the flow field distribution of the target site can be accurately constructed based on borehole data. This method fully considers factors such as water level data, hydraulic conductivity, and formation characteristics. Utilizing scientific principles such as inverse distance weighted interpolation and Darcy's law, it accurately calculates key elements such as water level distribution, hydraulic gradient, and velocity vector. The constructed flow field distribution can realistically and comprehensively reflect the flow state of groundwater within the target site, providing important basic data for studying groundwater movement patterns, pollutant migration and diffusion, and the design of permeable reactive barriers.

[0095] More specifically, based on historical monitoring data of the target site, pollution source information, and pollutant migration patterns, the available reaction packing materials for various pollutants are determined, specifically:

[0096] Obtain historical monitoring data and pollution source information for the target site;

[0097] Based on the historical monitoring data, identify all pollutants that have appeared in the target site, and determine the emission intensity and emission mode of various pollutants that have appeared in the target site from the pollution source information;

[0098] The migration patterns of various pollutants that have appeared in the target site are obtained, and the dispersion characteristics of various pollutants that have appeared in the target site are determined based on the migration patterns; wherein, the dispersion characteristics include dispersion coefficient and dispersion direction.

[0099] And obtain characteristic information of various pollutants that have appeared in the target site from the pollutant information database;

[0100] Based on the characteristics, emission intensity, emission mode, and dispersion characteristics of various pollutants that have appeared in the target site, generate characteristic text information of various pollutants that have appeared in the target site.

[0101] The characteristic information of each reaction packing is obtained from the packing information database. A local sensitive attention mechanism is introduced, and the attention score between the characteristic information of each reaction packing and the characteristic text information of various pollutants that have appeared in the target site is analyzed according to the local sensitive attention mechanism.

[0102] The attention scores between the characteristic information of each reaction packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with preset score thresholds.

[0103] If the attention score between the characteristic information of a certain reactive packing material and the characteristic text information of a certain pollutant that has appeared in the target site is greater than a preset score threshold, then the reactive packing material is identified as a usable reactive packing material for the corresponding pollutant.

[0104] The available reactive packing materials for various pollutants that have appeared in the target site are obtained after the attention scores between the characteristic information of each reactive packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with the preset score threshold.

[0105] Specifically, the process begins by acquiring historical monitoring data and pollution source information for the target site. The historical monitoring data identifies all types of pollutants that have appeared at the site, and the emission intensity and emission methods of each pollutant are determined from the pollution source information. Simultaneously, the migration patterns of the pollutants are acquired to determine their dispersion characteristics (dispersion coefficient and dispersion direction), and characteristic information of the pollutants is obtained from a pollutant database. This comprehensive information covers the presence, sources, and migration patterns of pollutants at the target site.

[0106] Based on the characteristics, emission intensity, emission mode, and dispersion features of pollutants, characteristic textual information about various pollutants within the target site is generated. This process integrates scattered pollutant-related information into a textual form that facilitates subsequent analysis and comparison, comprehensively representing the unique properties and behavioral patterns of each pollutant.

[0107] After obtaining the characteristic information of each reactive packing material from the packing material information database, a local sensitive attention mechanism is introduced. This mechanism can analyze the attention score between the reactive packing material characteristic information and the pollutant characteristic text information. The attention score reflects the degree of correlation between the reactive packing material and the pollutant. The local sensitive attention mechanism can more accurately capture the key parts of this correlation, thus more accurately measuring the degree of matching between the two. The attention score is compared and analyzed with a preset score threshold. If the attention score between a certain reactive packing material and a certain pollutant is greater than the preset score threshold, the reactive packing material is marked as a usable reactive packing material for the corresponding pollutant. By performing such comparative analysis on the attention scores between all reactive packing materials and pollutants, the usable reactive packing materials for various pollutants in the target site are finally determined.

[0108] Through the above steps, usable reactive media for various pollutants can be accurately determined based on historical monitoring data, pollution source information, and pollutant migration patterns at the target site. This method comprehensively considers multiple characteristics of pollutants and integrates them into feature text information. Simultaneously, it utilizes a localized sensitive attention mechanism to precisely analyze the correlation between reactive media and pollutants. By comparing with threshold values, usable reactive media are selected, providing a scientific basis for the design of permeable reactive walls in groundwater pollution remediation and ensuring the selection of appropriate reactive media to effectively treat pollutants at the target site.

[0109] More specifically, based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map, and flow field distribution of the target site, and combined with the finite difference method, a three-dimensional finite difference model of the target site is constructed, specifically as follows:

[0110] Based on the three-dimensional geological spatial distribution map of the target site, the spatial layout, thickness variation, interlayer contact relationship and geological structural characteristics of different strata are obtained, thereby determining the spatial structure and stratigraphic boundary conditions of the model;

[0111] And based on the flow field distribution of the target site, obtain the spatial distribution of groundwater level, the magnitude and direction of hydraulic gradient, and the dynamic characteristics of water flow;

[0112] Based on the target site's geometry, size, spatial structure, geological boundary conditions, spatial distribution of groundwater level, magnitude and direction of hydraulic gradient, and dynamic characteristics of water flow, the target site is divided into three-dimensional spatial grids, resulting in several grid nodes.

[0113] For each grid node, the partial differential equation describing groundwater movement is discretized using the finite difference method to obtain the equivalent source and sink intensity values ​​for each grid node. At the same time, the zero flux boundary conditions for each grid node are determined based on the three-dimensional geological spatial distribution map and flow field distribution.

[0114] Substitute the equivalent values ​​of the source and sink intensity of each grid node into the corresponding positions in the discrete equation, and set the boundary constraints of the discrete equation according to the zero flux boundary conditions determined by the three-dimensional geological spatial distribution map and the flow field distribution.

[0115] Then, based on the source and sink information in the flow field distribution, the source and sink terms in the discrete equations are adjusted, and the discrete equations of all grid nodes are combined according to the three-dimensional spatial coordinate relationship to construct a three-dimensional finite difference model of the target site.

[0116] Specifically, the spatial layout, thickness variations, interlayer contact relationships, and geological structural features of different strata are obtained from the 3D geological spatial distribution map. This provides the spatial framework and stratigraphic boundary conditions for constructing a 3D finite difference model. For example, the spatial layout of the strata determines the basic framework of the model, while interlayer contact relationships and geological structural features (such as faults and folds) affect the flow path and boundary conditions of groundwater. Simultaneously, the spatial distribution of groundwater level, the magnitude and direction of hydraulic gradient, and the dynamic characteristics of water flow are obtained based on the flow field distribution. These are key information describing the state of groundwater movement and provide a basis for subsequent flow-related calculations in model construction. A 3D spatial mesh is generated based on the geometry and size of the target site, as well as the previously determined spatial framework, stratigraphic boundary conditions, spatial distribution of groundwater level, magnitude and direction of hydraulic gradient, and dynamic characteristics of water flow, resulting in several mesh nodes. Mesh generation needs to comprehensively consider various factors. For example, in areas with complex geological structures or drastic water flow changes, a finer mesh may be needed to accurately capture the characteristics of these areas; while in relatively uniform areas, the mesh size can be appropriately increased to reduce computational load.

[0117] For each grid node, the partial differential equations describing groundwater movement are discretized using the finite difference method to obtain the equivalent source-sink intensity values ​​for each node. These equivalent values ​​reflect the inflow or outflow of groundwater at each node and are crucial parameters in the model describing the material and energy exchange within the groundwater system. Simultaneously, zero-flux boundary conditions for each grid node are determined based on the 3D geological spatial distribution map and flow field distribution. For example, impermeable boundaries are determined based on geological structures, and zero-flux conditions are set on these boundaries to ensure the model conforms to actual geological and flow conditions. The equivalent source-sink intensity values ​​for each grid node are substituted into the corresponding positions in the discrete equations, and boundary constraints are set according to the determined zero-flux boundary conditions. This ensures that the discrete equations accurately reflect the actual boundary conditions. Finally, the source-sink terms in the discrete equations are adjusted based on the source-sink information in the flow field distribution to ensure that the settings of the source-sink terms match the actual groundwater recharge and discharge. Finally, the discrete equations of all grid nodes are combined according to the three-dimensional spatial coordinate relationship to construct a three-dimensional finite difference model of the target site. This model can comprehensively reflect the three-dimensional hydrogeological characteristics and groundwater movement patterns of the target site.

[0118] The three-dimensional finite difference model constructed through the above steps can comprehensively and accurately reflect the three-dimensional hydrogeological conditions and groundwater movement patterns of the target site. It comprehensively considers various factors such as stratigraphic structure, geological formation, groundwater level, and hydraulic gradient. Through reasonable mesh generation, discretization operations, and boundary condition settings, it transforms the complex actual situation into a computable mathematical model. This model provides a reliable foundation for subsequent research on groundwater flow, pollutant migration and diffusion, and can be used to simulate and predict various groundwater-related phenomena at the target site, providing effective decision support for groundwater pollution control and water resource management.

[0119] More specifically, a machine learning model is constructed, and the characteristic information of various pollutants and available reactive packing materials are imported into the machine learning model for training to obtain a trained machine learning model, specifically as follows:

[0120] The characteristic information of various pollutants is obtained from the pollutant information database, and the available reaction packings for various pollutants are obtained from the pairing information database.

[0121] Based on the characteristic information of various pollutants and the available reactive fillers, the data network is searched to obtain PRB wall design schemes for the treatment of various pollutants.

[0122] The characteristic information of various pollutants and available reactive fillers are used as input feature nodes, and the PRB wall design scheme for treating various pollutants is used as output feature nodes; a mapping relationship topology is constructed based on the input feature nodes and output feature nodes.

[0123] A machine learning model is constructed based on a graph neural network. The connection relationship between the input feature nodes and the output feature nodes in the mapping relationship topology graph is transformed into a data structure that the machine learning model can recognize. The information of the nodes and edges in the mapping relationship topology graph is encoded according to the graph general relationship and then embedded into the parameter space of the machine learning model for learning and training.

[0124] After each training cycle, the feature contribution of the machine learning model is obtained. If the feature contribution of the machine learning model gradually increases and exceeds a preset threshold, the training is completed, and the trained machine learning model is output.

[0125] Specifically, the process begins by retrieving characteristic information of various pollutants from a pollutant information database and obtaining available reactive fillers for each pollutant from a matching database. Then, based on this information, the data network is searched to obtain PRB (permeable reactive barrier) wall design schemes for the treatment of various pollutants. This step establishes a preliminary connection between pollutant characteristics, available reactive fillers, and PRB wall design schemes. A mapping topology graph is constructed, using pollutant characteristics and available reactive fillers as input feature nodes and PRB wall design schemes as output feature nodes. This topology graph visually illustrates the relationship structure between input and output, providing a framework for the subsequent construction of machine learning models and helping to understand the influence of different factors on PRB wall design. A machine learning model is then built based on a graph neural network. In this process, the connection relationships between input and output feature nodes in the mapping topology graph are transformed into a data structure that the machine learning model can recognize. The information of nodes and edges in the topology graph is encoded according to the graph's general relationships and embedded into the parameter space of the machine learning model for training. This approach enables the model to learn the complex relationships between input features (pollutant characteristics and reactive fillers) and output features (PRB wall design schemes). After each training cycle, the feature contribution of the machine learning model is obtained. Feature contribution reflects the importance of each feature (input feature node) in the model to the output result (PRB wall design scheme). When the feature contribution gradually increases and exceeds a preset threshold, it indicates that the model has fully learned the relationship between the input and output. At this point, training is complete, and the trained machine learning model is output. This judgment mechanism ensures that the model has sufficient accuracy and reliability.

[0126] The machine learning model constructed and trained through the above steps can effectively learn the complex mapping relationship between pollutant characteristics, available reactive fillers, and PRB wall design schemes. Using an innovative graph neural network-based approach, these relationships are integrated into the model structure, and model training is optimized by continuously evaluating feature contribution. The resulting well-trained machine learning model can accurately predict suitable PRB wall design schemes based on pollutant characteristics and available reactive fillers, providing an efficient and accurate decision support tool for permeable reactive wall design in groundwater pollution remediation.

[0127] More specifically, the process involves acquiring pollutant information from the target site, importing this information into a trained machine learning model to obtain a PRB wall design scheme for the target site, and then importing this PRB wall design scheme into a three-dimensional finite difference model to obtain the final PRB wall design scheme for the target site.

[0128] The pollutant information of the target site is obtained and imported into the trained machine learning model to obtain the PRB wall design scheme for the target site; the PRB wall design scheme includes the PRB wall design thickness, location and filler.

[0129] The PRB wall design scheme for the target site is imported into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall.

[0130] The target conditions are defined based on minimizing pollutant concentration and maximizing PRB lifespan, and the constraints are set based on geological conditions, cost budget, and construction feasibility.

[0131] Determine whether the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints; if the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints, then the PRB wall design scheme is directly output as the final PRB wall design scheme.

[0132] Otherwise, a genetic algorithm is introduced, and an iterative optimization cycle is set. The PRB wall design scheme is adjusted and optimized according to the genetic algorithm. After each iterative optimization cycle is completed, the optimized PRB wall design scheme is obtained.

[0133] The optimized PRB wall design scheme is input into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall;

[0134] The iteration stops when the numerical simulation results of the permeable reactive wall of the optimized PRB wall design scheme meet the target conditions and constraints, and the optimized PRB wall is output as the final PRB wall design scheme.

[0135] Specifically, the process begins by acquiring pollutant information from the target site and inputting it into a trained machine learning model to generate a PRB (Polymer Reactive Barrier) wall design scheme. This scheme includes key information such as the PRB wall's design thickness, location, and filler material. The machine learning model is trained based on previously established mapping relationships between pollutant characteristics, available reactive fillers, and the PRB wall design scheme, enabling it to generate a preliminary PRB wall design scheme based on the input pollutant information. The PRB wall design scheme obtained from the machine learning model is then imported into a three-dimensional finite difference model for evaluation, yielding numerical simulation results of the permeable reactive barrier. The three-dimensional finite difference model can simulate the three-dimensional hydrogeological conditions and groundwater movement patterns of the target site. This model allows for analysis of the PRB wall design scheme's effectiveness in the actual site, such as pollutant migration in the presence of a PRB wall. Target conditions are defined based on minimizing pollutant concentration and maximizing PRB lifetime, considering both remediation effectiveness and the long-term effectiveness of the PRB. Simultaneously, constraints are set based on geological conditions (such as stratigraphic structure and groundwater level), cost budget (including material costs and construction costs), and construction feasibility (such as construction difficulty and site accessibility). These target conditions and constraints provide a standard for evaluating the feasibility of PRB wall design schemes. The numerical simulation results of the permeable reactive barrier are then assessed to determine if they simultaneously meet the target conditions and constraints. If not, a genetic algorithm is introduced for optimization. The genetic algorithm, based on the principle of biological evolution, adjusts and optimizes the PRB wall design scheme by setting iterative optimization cycles. After each iteration cycle, the optimized PRB wall design scheme is obtained and then input into a three-dimensional finite difference model for re-evaluation until the numerical simulation results of the optimized scheme meet the target conditions and constraints. At this point, the optimized scheme is output as the final PRB wall design scheme. This process realizes the transformation from target site pollutant information to a scientifically sound, economically feasible, and efficient PRB wall design scheme, providing reliable decision support for permeable reactive barrier design in groundwater pollution control.

[0136] In addition, the user interface of this system is also designed:

[0137] Input interface design: Design a user-friendly input interface that allows users to input 3D geological information, pollutant information and other relevant parameters.

[0138] Output interface design: Design an intuitive output interface to display the optimized PRB design results, including location, thickness, and predicted lifetime.

[0139] Visual design: Provides 3D visualization capabilities to help users intuitively understand simulation results and design schemes;

[0140] Module integration design: Integrate various modules (data processing, machine learning, numerical simulation, optimization algorithms, user interface) into a unified platform.

[0141] Interface design: Design the interfaces between modules to ensure data flow and functional collaboration.

[0142] In addition, in practical applications, this design system may also include the following steps:

[0143] After the PRB wall is laid out on the target site based on the final PRB wall design scheme, the permeability coefficient of the PRB wall at several preset locations is obtained at several preset time points during the actual treatment process.

[0144] A Lorentz curve fitting algorithm is introduced, which treats the permeability coefficient data collected by each preset location node at each preset time node as discrete data points. Using time and location as variable dimensions, the variable correspondence in the Lorentz function form is determined. Based on the variable dimensions and the variable correspondence in the Lorentz function form, each discrete data point is fitted into a Lorentz curve of permeability coefficient distribution, thus obtaining the Lorentz curve of permeability coefficient distribution for each preset location node in the PRB wall.

[0145] Obtain the treatment precision requirements of the target site, and determine the preset permeability coefficient range of the PRB wall based on the treatment precision requirements of the target site;

[0146] The permeability range is determined from the Lorentz curve of the permeability distribution at each preset location node based on the preset permeability range of the PRB wall.

[0147] Analyze the lengths of curve segments within the permeability range and outside the permeability range of the Lorentz curve for each preset location node.

[0148] Divide the length of the curve segment within the permeability range of the Lorentz curve for each preset location node by the length of the curve segment outside the permeability range of the Lorentz curve for each preset location node to obtain the compliance line segment length ratio coefficient of the PRB wall permeability Lorentz curve for each preset location node.

[0149] Compare the proportional coefficient of the Lorentz curve compliance line segment length of the PRB wall permeability coefficient at each preset location node with the preset coefficient value.

[0150] If the proportional coefficient of the length of the compliance line segment of the PRB wall permeability coefficient Lorenz curve at a certain preset location node is not greater than the preset coefficient value, then the corresponding preset location node of the PRB wall is defined as an abnormal node of the PRB wall permeability coefficient.

[0151] Based on the abnormal nodes of the PRB wall permeability coefficient, a PRB wall anomaly assessment report and early warning information are generated. The parameters of the corresponding machine learning model are adjusted and optimized based on the PRB wall anomaly assessment report.

[0152] The length of the curve segment within the permeability range is divided by the length of the curve segment outside the range to obtain the proportionality coefficient of the Lorenz curve compliance segment length for the PRB wall permeability coefficient. This coefficient reflects the proportion of permeability coefficients at each preset location node that conform to the preset range. This proportionality coefficient is compared with a preset coefficient value. If the proportionality coefficient of a preset location node is not greater than the preset coefficient value, this node is defined as an abnormal PRB wall permeability coefficient node. This determination can accurately identify locations where permeability coefficient anomalies may exist, potentially indicating problems such as blockage, structural damage, or reactive filler failure in the PRB wall at that location. An anomaly assessment report and early warning information are generated based on the identified abnormal PRB wall permeability coefficient nodes. This report and early warning information can promptly inform relevant personnel of potential problems with the PRB wall. Simultaneously, the parameters of the corresponding machine learning model are adjusted and optimized based on the anomaly assessment report. This means improving the machine learning model through feedback data from the actual remediation process, enabling the model to more accurately predict the design parameters of the PRB wall, improving the model's accuracy and reliability, and thus providing better support for subsequent PRB wall design and remediation.

[0153] Overall, the Lorenz curve fitting algorithm was used to visualize and quantify permeability coefficient data, accurately identifying potential problem locations. The generated anomaly assessment reports and early warning information facilitated the timely detection and handling of potential problems in PRB walls. Simultaneously, the machine learning model was optimized using actual treatment data, improving its predictive capabilities and achieving a virtuous cycle from actual treatment effects to model optimization. This ultimately improved the overall effectiveness of PRB wall design and treatment, ensuring that the treatment accuracy requirements of the target site were met.

[0154] In addition, after collecting the three-dimensional hydrogeological parameters, borehole data, and historical monitoring data of the target site, the three-dimensional hydrogeological parameters, borehole data, and historical monitoring data of the target site are denoised based on the decision tree algorithm:

[0155] A comprehensive dataset containing 3D hydrogeological parameters, borehole data, and historical monitoring data is constructed. Each data item in the dataset is standardized to clarify its attribute type and value range. Based on prior knowledge of the hydrogeological system of the target site and preliminary statistical analysis of the data, an initial attribute set for constructing the decision tree is determined. This attribute set should cover parameters that have a key impact on data quality and characteristics, such as formation permeability and water level variation in 3D hydrogeological parameters, borehole depth and core integrity in borehole data, and pollutant concentration variation trends in historical monitoring data.

[0156] Next, based on evaluation metrics such as information gain and Gini coefficient, the most discriminative attribute is selected from the initial attribute set as the root node of the decision tree. The dataset is then divided into several subsets based on the different values ​​of this attribute. For each subset, the attribute selection and dataset partitioning operations are recursively repeated to construct the decision tree structure. During the tree construction process, a preset stopping criterion is set, such as stopping the split when the number of samples in a subset is lower than a preset threshold, the purity of a subset reaches a predetermined standard (e.g., the proportion of data in a certain category exceeds a set ratio), or the depth of the decision tree reaches a preset maximum depth.

[0157] After the decision tree is constructed, each leaf node of the decision tree is traversed. For the data sample in the leaf node, its similarity with other samples in the same leaf node is analyzed. If the difference between a sample and most samples in the leaf node in key attributes exceeds a preset similarity threshold, the sample is determined to be noise data. The similarity assessment here can be based on distance metrics such as Euclidean distance and Mahalanobis distance, combined with the physical meaning of the data for a comprehensive judgment.

[0158] Finally, samples identified as noise were removed from the original dataset to obtain the denoised three-dimensional hydrogeological parameters, borehole data, and historical monitoring data.

[0159] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A smart responsive permeable reactive barrier design system for groundwater pollution, characterized in that, The design system includes: A site information database is constructed, and three-dimensional hydrogeological parameters and borehole data of the target site are collected. Using the borehole data, a three-dimensional geological spatial distribution map and flow field distribution of the target site are generated based on spatial interpolation information. The three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site are then imported into the site information database. A pollutant information database is constructed to obtain characteristic information of various pollutants, and the characteristic information of various pollutants is imported into the pollutant information database; a packing material information database is constructed to obtain characteristic information of various reaction packing materials, and the characteristic information of various reaction packing materials is imported into the packing material information database. A matching information database is constructed, and historical monitoring data, pollution source information and pollutant migration patterns of the target site are collected. Based on the historical monitoring data, pollution source information and pollutant migration patterns of the target site, the available reaction packings for various pollutants are determined, and the available reaction packings for various pollutants are imported into the matching information database. A machine learning model is constructed, and the characteristic information of various pollutants and available reactive packing materials are imported into the machine learning model for training to obtain a trained machine learning model. The three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site are obtained from the site information database. Based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map and flow field distribution of the target site, and combined with the finite difference method, a three-dimensional finite difference model of the target site is constructed. The pollutant information of the target site is obtained, and the pollutant information of the target site is imported into a trained machine learning model to obtain the PRB wall design scheme of the target site; the PRB wall design scheme of the target site is imported into a three-dimensional finite difference model to obtain the final PRB wall design scheme of the target site.

2. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that, Using borehole data, a three-dimensional geological spatial distribution map of the target site is generated based on spatial interpolation information, specifically as follows: Obtain borehole data for each borehole in the target site. The borehole data includes the coordinate information of the borehole location, geological information, hydraulic conductivity, water level data, porosity data, rock layer thickness data, and stratum lithology. A three-dimensional coordinate system is determined based on the scope of the target site, and the coordinates of each borehole are located in the coordinate system. Using the stratigraphic lithology and thickness data at the borehole location as the basic data points, the stratigraphic lithology and thickness at each three-dimensional coordinate point within the target site are calculated using the Kriging interpolation method, starting from the bottom layer, for each stratigraphic layer. In the horizontal direction, based on the planar position relationship and thickness variation of the boreholes, the interpolation range is gradually expanded to determine the distribution boundary of the same stratum on the plane; In the vertical direction, based on the vertical relationship and thickness variation of the strata in different boreholes, the continuous distribution of the strata in the vertical direction is constructed by interpolation; For geological structural data, including faults and folds, these are incorporated as constraints into the interpolation process. If there is displacement of the strata on both sides of a fault, the displacement is calibrated in three-dimensional space based on the fault location coordinates. The bending morphology of folds is constructed in three-dimensional space based on the corresponding morphology data in the borehole data. By continuously integrating the interpolation results of various strata and geological structures, a three-dimensional geological spatial distribution map of the target site is finally formed.

3. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 2, characterized in that, Using borehole data, the flow field distribution of the target site is generated based on spatial interpolation information, specifically as follows: Using the three-dimensional coordinate system of the target site as a framework, each borehole is located according to its coordinates. Using the water level data in the borehole as a feature, the water level values ​​of each three-dimensional coordinate point in the target site are obtained based on the inverse distance weighted interpolation method, thereby obtaining the spatial distribution of the water level. Based on hydraulic conductivity data and Darcy's law, the hydraulic gradient at each three-dimensional coordinate point is calculated using the water level distribution. During the calculation process, the influence of the anisotropy of the strata on the hydraulic conductivity is taken into account, and the hydraulic gradient is calculated according to the characteristics of different strata. Using the hydraulic conductivity and the calculated hydraulic gradient, the groundwater velocity vector is calculated again according to Darcy's law to determine the direction and speed of groundwater flow in the corresponding stratum. The flow field distribution of the target site is constructed based on the direction and speed of groundwater flow in each stratum and the spatial distribution of water level.

4. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that, Based on historical monitoring data, pollution source information, and pollutant migration patterns at the target site, the available reaction packings for various pollutants were determined, specifically: Obtain historical monitoring data and pollution source information for the target site; Based on the historical monitoring data, identify all pollutants that have appeared in the target site, and determine the emission intensity and emission mode of various pollutants that have appeared in the target site from the pollution source information; The migration patterns of various pollutants that have appeared in the target site are obtained, and the dispersion characteristics of various pollutants that have appeared in the target site are determined based on the migration patterns; wherein, the dispersion characteristics include dispersion coefficient and dispersion direction. And obtain characteristic information of various pollutants that have appeared in the target site from the pollutant information database; Based on the characteristics, emission intensity, emission mode, and dispersion characteristics of various pollutants that have appeared in the target site, generate characteristic text information of various pollutants that have appeared in the target site. The characteristic information of each reaction packing is obtained from the packing information database. A local sensitive attention mechanism is introduced, and the attention score between the characteristic information of each reaction packing and the characteristic text information of various pollutants that have appeared in the target site is analyzed according to the local sensitive attention mechanism. The attention scores between the characteristic information of each reaction packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with preset score thresholds. If the attention score between the characteristic information of a certain reactive packing material and the characteristic text information of a certain pollutant that has appeared in the target site is greater than a preset score threshold, then the reactive packing material is identified as a usable reactive packing material for the corresponding pollutant. The available reactive packing materials for various pollutants that have appeared in the target site are obtained after the attention scores between the characteristic information of each reactive packing material and the characteristic text information of various pollutants that have appeared in the target site are compared and analyzed with the preset score threshold.

5. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that, Based on the three-dimensional hydrogeological parameters, three-dimensional geological spatial distribution map, and flow field distribution of the target site, and combined with the finite difference method, a three-dimensional finite difference model of the target site is constructed, specifically as follows: Based on the three-dimensional geological spatial distribution map of the target site, the spatial layout, thickness variation, interlayer contact relationship and geological structural characteristics of different strata are obtained, thereby determining the spatial structure and stratigraphic boundary conditions of the model; And based on the flow field distribution of the target site, obtain the spatial distribution of groundwater level, the magnitude and direction of hydraulic gradient, and the dynamic characteristics of water flow; Based on the target site's geometry, size, spatial structure, geological boundary conditions, spatial distribution of groundwater level, magnitude and direction of hydraulic gradient, and dynamic characteristics of water flow, the target site is divided into three-dimensional spatial grids, resulting in several grid nodes. For each grid node, the partial differential equation describing groundwater movement is discretized using the finite difference method to obtain the equivalent source and sink intensity values ​​for each grid node. At the same time, the zero flux boundary conditions for each grid node are determined based on the three-dimensional geological spatial distribution map and flow field distribution. Substitute the equivalent values ​​of the source and sink intensity of each grid node into the corresponding positions in the discrete equation, and set the boundary constraints of the discrete equation according to the zero flux boundary conditions determined by the three-dimensional geological spatial distribution map and the flow field distribution. Then, based on the source and sink information in the flow field distribution, the source and sink terms in the discrete equations are adjusted, and the discrete equations of all grid nodes are combined according to the three-dimensional spatial coordinate relationship to construct a three-dimensional finite difference model of the target site.

6. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that, A machine learning model is constructed, and the characteristic information of various pollutants and available reactive packing materials are imported into the machine learning model for training to obtain a trained machine learning model. Specifically: The characteristic information of various pollutants is obtained from the pollutant information database, and the available reaction packings for various pollutants are obtained from the pairing information database. Based on the characteristic information of various pollutants and the available reactive fillers, the data network is searched to obtain PRB wall design schemes for the treatment of various pollutants. The characteristic information of various pollutants and available reactive fillers are used as input feature nodes, and the PRB wall design scheme for treating various pollutants is used as output feature nodes; a mapping relationship topology is constructed based on the input feature nodes and output feature nodes. A machine learning model is constructed based on a graph neural network. The connection relationship between the input feature nodes and the output feature nodes in the mapping relationship topology graph is transformed into a data structure that the machine learning model can recognize. The information of the nodes and edges in the mapping relationship topology graph is encoded according to the graph general relationship and then embedded into the parameter space of the machine learning model for learning and training. After each training cycle, the feature contribution of the machine learning model is obtained. If the feature contribution of the machine learning model gradually increases and exceeds a preset threshold, the training is completed, and the trained machine learning model is output.

7. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that, The process involves acquiring pollutant information from the target site, importing this information into a trained machine learning model to obtain a PRB wall design scheme for the target site, and then importing this PRB wall design scheme into a three-dimensional finite difference model to obtain the final PRB wall design scheme for the target site. Specifically: The pollutant information of the target site is obtained and imported into the trained machine learning model to obtain the PRB wall design scheme for the target site; the PRB wall design scheme includes the PRB wall design thickness, location and filler. The PRB wall design scheme for the target site is imported into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall. The target conditions are defined based on minimizing pollutant concentration and maximizing PRB lifespan, and the constraints are set based on geological conditions, cost budget, and construction feasibility. Determine whether the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints; if the numerical simulation results of the permeable reactive wall all meet the target conditions and constraints, then the PRB wall design scheme is directly output as the final PRB wall design scheme. Otherwise, a genetic algorithm is introduced, and an iterative optimization cycle is set. The PRB wall design scheme is adjusted and optimized according to the genetic algorithm. After each iterative optimization cycle is completed, the optimized PRB wall design scheme is obtained. The optimized PRB wall design scheme is input into the three-dimensional finite difference model for evaluation to obtain the numerical simulation results of the permeable reactive wall; The iteration stops when the numerical simulation results of the permeable reactive wall of the optimized PRB wall design scheme meet the target conditions and constraints, and the optimized PRB wall is output as the final PRB wall design scheme.

8. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that: The three-dimensional hydrogeological parameters include strata, water level, horizontal / vertical permeability coefficient, groundwater flow velocity, porosity, and water storage rate.

9. The intelligent responsive permeable reactive barrier design system for groundwater pollution according to claim 1, characterized in that: The characteristics of the reaction packing include material, ratio, size, specific surface area, and reaction rate constant; the characteristics of the pollutants include the negative logarithm of the acid dissociation constant and molecular weight.

Images