A machine learning based groundwater vulnerability modeling method

Abstract

The application discloses a kind of groundwater vulnerability modeling methods based on machine learning, it is related to groundwater analysis field, including the following steps: obtaining groundwater multi-source data and carrying out hierarchical management;The classification management result is carried out, and the multi-source classification data and corresponding multi-source classification characteristic data are obtained;Groundwater multi-source data is stored in distribution, and interactive feature dataset is generated according to storage result;The initial DRASTIC model is analyzed by traversing the obtained interactive feature dataset, and the model parameter reference image is generated by the obtained traversal analysis result;According to model parameter reference image, the correlation analysis of model related parameters is carried out to obtain adjustment parameter;Real-time groundwater multi-source data is obtained, and the initial DRASTIC model is flexibly coupled and analyzed according to the position information in model parameter reference image of real-time groundwater multi-source data belonging to;The adaptability and accuracy of the model are improved to some extent.

Description

Technical Field

[0001] This invention relates to the field of groundwater analysis technology, and in particular to a groundwater vulnerability modeling method based on machine learning. Background Technology

[0002] Groundwater accounts for approximately 33% of global water extraction, and over 2 billion people rely on it as their primary source of water for survival. Over-extraction has led to a continuous decline in groundwater levels in many regions, triggering a global groundwater crisis. Against the backdrop of climate change, quantitative assessment of groundwater resources is crucial for sustainable water resource management. With advancements in data analysis methods and machine learning technologies, a vast amount of available in-situ and remote sensing data is rapidly increasing, providing opportunities for groundwater resource assessment from local to global levels. Research indicates that machine learning models can effectively monitor or predict various characteristics of groundwater resources. Advances in machine and deep learning algorithms, combined with the development of physics-based model computations, are conducive to improving the spatiotemporal resolution and accuracy of groundwater monitoring or prediction.

[0003] A search revealed Chinese invention patent CN117933720A, which discloses a method for assessing groundwater nitrate risk. The method includes: Step 1, creating a groundwater vulnerability map based on an improved DRASTIC model; Step 2, drawing a groundwater nitrate pollution distribution characteristic map based on limited nitrate concentration observation data of the target area; Step 3, constructing a data indicator attribute table; training multiple machine learning models based on the data indicator attribute table, combining the outputs of individual machine learning models into a comprehensive model, and using the weighted integral of the models for ensemble modeling to construct a groundwater nitrate pollution probability model; Step 4, constructing a groundwater nitrate pollution risk assessment model based on groundwater vulnerability, nitrate pollution distribution characteristics, and pollution probability to evaluate the groundwater nitrate pollution risk. This method provides theoretical guidance and reference value for regional groundwater nitrate pollution risk assessment and can offer guidance and understanding for groundwater risk management decisions in semi-arid regions.

[0004] Compared with existing technologies, the invention patent with Chinese patent number CN117933720A can improve the computational efficiency of traditional models by applying given existing datasets, and improve the universality and accuracy at the regional scale.

[0005] However, in practical applications, the above method combines the outputs of individual machine learning models into a comprehensive model and uses the weighted integral of the model for integrated modeling to construct a probability model of groundwater nitrate pollution. In this process, it relies on limited observation data, and the model integration uses simple weighted integrals, which to some extent affects the adaptability and accuracy of the model analysis process. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of insufficient adaptability in existing technologies by proposing a machine learning-based method for modeling groundwater vulnerability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A machine learning-based method for modeling groundwater vulnerability includes the following steps:

[0009] Step S1: Set up a groundwater modeling and management platform, obtain multi-source groundwater data through the platform, preprocess the obtained multi-source groundwater data, and manage the preprocessed multi-source groundwater data in layers.

[0010] Step S2: Classify the hierarchical management results to obtain multi-source classification data. Extract features from the obtained multi-source classification data to obtain multi-source classification feature data.

[0011] Step S3: Distribute the obtained multi-source classification feature data according to the hierarchical management results and classification management results, perform interactive mining based on the distributed storage results, generate interactive feature dataset, and divide the obtained interactive feature dataset into data analysis set and data validation set;

[0012] Step S4: Obtain the initial DRASTIC model, sequentially perform traversal analysis on the initial DRASTIC model using the obtained data analysis set, obtain the traversal analysis results, and map the obtained traversal analysis results according to the distributed storage results to generate a model parameter reference image.

[0013] Step S5: Set the corresponding flexible adjustment node according to the data mapping processing result in the model parameter reference image, perform correlation analysis on the model-related parameters at the mapping processing result through the flexible adjustment node, and set the adjustment parameters according to the correlation analysis result;

[0014] Step S6: Obtain real-time groundwater multi-source data, and flexibly couple the initial DRASTIC model with the corresponding flexible adjustment nodes in the model parameter reference image to generate a real-time DRASTIC model for analysis and processing.

[0015] The above technical solution further includes: the process of acquiring multi-source groundwater data includes:

[0016] A groundwater modeling and management platform is set up, which includes a data input terminal, a data acquisition terminal, and a data management terminal.

[0017] The data entry terminal is used to connect to the multi-source data entry interface set in the groundwater modeling and management platform to obtain the groundwater multi-source data corresponding to the multi-source data entry interface.

[0018] The data acquisition terminal is used to connect with the multi-source data acquisition equipment set in the groundwater modeling and management platform to acquire the groundwater multi-source data corresponding to the multi-source data acquisition equipment.

[0019] The data management terminal is used to clean and standardize the acquired groundwater multi-source data. The pre-processed groundwater multi-source data is managed hierarchically according to the type of medium, including three classification management levels: porous layer, fracture layer and karst layer. The groundwater multi-source data is then labeled according to the classification management level.

[0020] Furthermore, the process of acquiring multi-source classification data and multi-source classification feature data includes:

[0021] A multi-dimensional groundwater classification system is established based on hydrogeological type and auxiliary environment type. Groundwater multi-source data in different classification management levels are classified and managed according to the multi-dimensional groundwater classification system to obtain multi-source classification data.

[0022] Feature extraction schemes are set for different types of multi-source classification data. Based on the corresponding feature extraction schemes, features are extracted from the obtained multi-source classification data to obtain multi-source classification feature data of each unified dimension of different types of multi-source classification data.

[0023] Furthermore, the process of distributing the obtained multi-source classification feature data includes:

[0024] Based on the data labels of groundwater multi-source data, the corresponding classification and management levels are obtained. Different classification and management levels are set with corresponding hierarchical spaces. The obtained hierarchical spaces are standardized according to the multi-source classification feature data, and hierarchical unit spaces are set.

[0025] Based on the corresponding multi-dimensional groundwater classification system, corresponding classification spaces are set up in each level of unit space. The obtained classification spaces are standardized according to the multi-source classification feature data, and classification unit spaces are set up.

[0026] The obtained groundwater multi-source data are distributed and stored sequentially in their respective hierarchical space, hierarchical unit space, classification space, and classification unit space, based on the corresponding multi-source classification data and multi-source classification feature data.

[0027] Furthermore, the process of generating the interaction feature dataset includes:

[0028] The hierarchical space, hierarchical unit space, classification space, and classification unit space are uniformly marked as progressive storage space. The distributed storage results corresponding to each progressive storage space are used to construct a distributed three-dimensional storage space based on the data type and collection time.

[0029] The multi-source classification data stored in the distributed three-dimensional storage space are cross-combined based on the corresponding specified variable data according to different multi-source classification feature data in the same progressive storage space, the same multi-source classification feature data in the same progressive storage space, the same multi-source classification feature data between different progressive storage spaces, and different multi-source classification feature data between different progressive storage spaces. The progressive storage space is traversed in the order of classification unit space, classification space, hierarchical unit space, and hierarchical space. An interactive feature dataset is generated based on the traversal results.

[0030] The obtained interactive feature dataset is labeled according to the source of the distributed three-dimensional storage space to which the multi-source classification data belongs. The interactive feature datasets obtained from each labeling process are divided according to the proportion, and the corresponding data analysis set and data validation set are obtained respectively.

[0031] Furthermore, the process of generating the model parameter reference image includes:

[0032] The initial DRASTIC model and its original related parameters are obtained. The obtained data analysis sets are used to traverse and analyze the initial DRASTIC model based on machine learning algorithms to obtain the model-related parameters corresponding to each data analysis set. The model-related parameters of the initial DRASTIC model are validated through the data validation set to obtain the model-related parameters of the initial DRASTIC model corresponding to the interactive feature dataset.

[0033] The obtained model-related parameters are mapped according to the location information and parameter type of the distributed three-dimensional storage space to which the interactive feature dataset belongs, to generate model parameter reference images.

[0034] Furthermore, the process of performing correlation analysis on the model-related parameters at the mapping processing result by flexibly adjusting the nodes includes:

[0035] Based on the set interactive feature dataset, flexible adjustment nodes are set at the data mapping processing results in the model parameter reference image, and flexible adjustment sub-nodes are set in the flexible adjustment nodes according to the corresponding parameter types.

[0036] Each flexible adjustment sub-node within each flexible adjustment node is fused and extracted based on the spatial location of the reference image of its respective model parameter, multi-source classification data, and parameter type. The fusion and extraction results are then used to generate an association analysis dataset.

[0037] Based on the obtained correlation analysis dataset, the correlation strength matrix is ​​constructed according to the correlation degree, cooperative change law and response relationship between the relevant parameters of the corresponding model according to the time series correlation method. The corresponding adaptive adjustment coefficient is calculated according to the parameter correlation strength matrix. The obtained adaptive adjustment coefficient is marked as the adjustment parameter. The obtained adjustment parameter is mapped and stored according to the position of the flexible adjustment node in the model parameter reference image.

[0038] The mapping and storage results of the model parameter reference image are mapped one-to-one with the distributed three-dimensional storage space.

[0039] Furthermore, the process of generating the real-time DRASTIC model includes:

[0040] Acquire real-time multi-source groundwater data, classify the acquired real-time multi-source groundwater data, and obtain the location information of the corresponding distributed three-dimensional storage space;

[0041] Based on the one-to-one correspondence between the distributed three-dimensional storage space and the model parameter reference image, obtain the flexible adjustment node at the corresponding location information;

[0042] Based on the distribution of each flexible adjustment node, obtain the relevant flexible adjustment sub-nodes, obtain the corresponding adjustment parameters based on the flexible adjustment sub-nodes, obtain the corresponding model-related parameters based on the adjustment parameters, adjust the initial DRASTIC model based on the obtained model-related parameters, generate a real-time DRASTIC model, and perform vulnerability assessment on real-time groundwater multi-source data based on the real-time DRASTIC model.

[0043] The present invention has the following beneficial effects:

[0044] In this invention, the obtained multi-source groundwater data is managed hierarchically and categorized. Based on the hierarchical and categorized management results, the corresponding multi-source groundwater data is distributed and stored. Based on the distributed storage results, the multi-source groundwater data is interactively mined to generate different interactive feature datasets. The initial DRASTIC model is then analyzed using the interactive feature datasets to obtain the relevant model parameters. The obtained model parameters are then used to generate a model parameter reference image. This can improve the utilization rate and synergy of multi-source groundwater data in adjusting the initial DRASTIC model to a certain extent.

[0045] In this invention, the correlation between different types of model parameters at different locations of the initial DRASTIC model is stored through a set model parameter reference image. Based on the model-related parameters mapped and stored in the model parameter reference image, the corresponding model-related parameters are selected for the obtained real-time groundwater multi-source data. This can improve the applicability of the DRASTIC model to a certain extent and avoid the waste of resources or insufficient adaptability caused by the need to analyze and process groundwater multi-source data in all cases. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the structure of a machine learning-based groundwater vulnerability modeling method proposed in this invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0048] like Figure 1 As shown, the present invention proposes a groundwater vulnerability modeling method based on machine learning. Step S1: Set up a groundwater modeling management platform, obtain groundwater multi-source data through the groundwater modeling management platform, perform data preprocessing on the obtained groundwater multi-source data, and perform hierarchical management on the groundwater multi-source data after data preprocessing.

[0049] Step S2: Classify the hierarchical management results to obtain multi-source classification data. Extract features from the obtained multi-source classification data to obtain multi-source classification feature data.

[0050] Step S3: Distribute the obtained multi-source classification feature data according to the hierarchical management results and classification management results, perform interactive mining based on the distributed storage results, generate interactive feature dataset, and divide the obtained interactive feature dataset into data analysis set and data validation set;

[0051] Step S4: Obtain the initial DRASTIC model, sequentially perform traversal analysis on the initial DRASTIC model using the obtained data analysis set, obtain the traversal analysis results, and map the obtained traversal analysis results according to the distributed storage results to generate a model parameter reference image.

[0052] Step S5: Set the corresponding flexible adjustment node according to the data mapping processing result in the model parameter reference image, perform correlation analysis on the model-related parameters at the mapping processing result through the flexible adjustment node, and set the adjustment parameters according to the correlation analysis result;

[0053] Step S6: Obtain real-time groundwater multi-source data, and flexibly couple the initial DRASTIC model with the corresponding flexible adjustment nodes in the model parameter reference image to generate a real-time DRASTIC model for analysis and processing.

[0054] In the specific implementation process, the step S1 of setting up a groundwater modeling and management platform, acquiring multi-source groundwater data through the platform, preprocessing the acquired multi-source groundwater data, and performing hierarchical management on the preprocessed multi-source groundwater data includes:

[0055] S11: Set up a groundwater modeling and management platform, which includes a data entry terminal, a data acquisition terminal, and a data management terminal.

[0056] S111: The data entry terminal is used to connect with the multi-source data entry interface set in the groundwater modeling and management platform to obtain the groundwater multi-source data corresponding to the multi-source data entry interface.

[0057] S112: The data acquisition terminal is used to connect with the multi-source data acquisition device set in the groundwater modeling and management platform to acquire the groundwater multi-source data corresponding to the multi-source data acquisition device.

[0058] S113: Data management terminal, used to clean and standardize the acquired groundwater multi-source data, and to manage the pre-processed groundwater multi-source data in layers according to the type of medium, including three classification management levels: pore layer, fracture layer and karst layer, and to mark the groundwater multi-source data according to the classification management level.

[0059] It should be further explained that the data entry terminal and the data acquisition terminal in step S1 above are different. The data entry terminal obtains groundwater multi-source data from other different locations within the groundwater modeling and management platform through the groundwater modeling and management platform, while the data acquisition terminal obtains groundwater multi-source data from different multi-source data acquisition devices involved in the corresponding location within the groundwater modeling and management platform. This increases the amount of groundwater multi-source data to a certain extent, which facilitates subsequent modeling.

[0060] The obtained groundwater multi-source data includes, but is not limited to, hydrogeological data, in-situ monitoring data, water quality testing data, meteorological and hydrological data, and human activity data. Among them, hydrogeological data, in-situ monitoring data, and water quality testing data are marked as hydrogeological types, while meteorological and hydrological data and human activity data are marked as auxiliary environmental types.

[0061] In the process of hierarchical management of multi-source groundwater data, the corresponding medium type is obtained based on hydrogeological data, and the corresponding classification management level is set according to the medium type.

[0062] In the specific implementation process, the step S2 of classifying and managing the hierarchical management results, obtaining multi-source classification data, extracting features from the obtained multi-source classification data, and obtaining multi-source classification feature data includes:

[0063] S21: Establish a multi-dimensional groundwater classification system based on hydrogeological type and auxiliary environment type, classify and manage groundwater multi-source data in different classification management levels with the multi-dimensional groundwater classification system, obtain multi-source classification data, and label the obtained multi-source classification data according to data type;

[0064] It should be further explained that the multi-dimensional groundwater classification system sets corresponding classification label standards according to various types. The multi-source groundwater data is traversed and matched sequentially according to the corresponding classification label standards to obtain the multi-source classification data corresponding to the multi-source groundwater data.

[0065] S22: Set up feature extraction schemes for different types of multi-source classification data, extract features from the obtained multi-source classification data according to the corresponding feature extraction schemes, and obtain multi-source classification feature data of each unified dimension of different types of multi-source classification data respectively;

[0066] S221: Set different feature extraction schemes according to the corresponding classification identification standards in the multi-dimensional groundwater classification system. For example, extract structural features and spatial distribution features from hydrogeological data, and extract temporal trend features and dynamic change features from in-situ monitoring data.

[0067] S222: Extract features from multi-source classification data using different feature extraction schemes, and perform feature normalization processing based on the feature extraction results to convert different multi-source classification feature data into multi-source classification feature data of the same dimension, which is convenient for subsequent machine learning.

[0068] In the specific implementation process, the step S3, which involves distributing the obtained multi-source classification feature data according to the hierarchical management results and classification management results, performing interactive mining based on the distributed storage results to generate an interactive feature dataset, and dividing the obtained interactive feature dataset into a data analysis set and a data validation set, includes:

[0069] S31: Obtain the corresponding classification and management levels based on the data labels of groundwater multi-source data, set corresponding hierarchical spaces for different classification and management levels, standardize the obtained hierarchical spaces according to the multi-source classification feature data, and set hierarchical unit spaces.

[0070] S311: Set up classification management levels, which are used to perform distributed data storage of groundwater multi-source data involved in the three classification management levels of pore layer, fracture layer and karst layer respectively;

[0071] S312: Map the obtained groundwater multi-source data to the corresponding classification and management levels respectively, and map and store the obtained classification and management levels according to the medium type of the porous layer, fracture layer and karst layer.

[0072] S313: Standardize the groundwater multi-source data within each classification management level according to the multi-source classification feature data. The standardization process involves sorting the multi-source classification feature data according to the classification standards corresponding to the porous layer, fracture layer, and karst layer, and standardizing the sorting results. The standardization process involves calculating the difference between adjacent multi-source classification feature data according to the sorting results, calculating the mean of the obtained difference calculation results, setting equal intervals according to the mean statistics, and marking the obtained equal intervals sequentially. Each multi-source classification feature data is mapped to the equal interval of the corresponding sequential marking result, and each equal interval is connected sequentially according to the sequential marking. Standardization intervals are set according to the connection results.

[0073] S314: Set up hierarchical unit spaces according to the standardized intervals corresponding to the classification management levels of the pore layer, fracture layer and karst layer respectively, and store the groundwater multi-source data corresponding to the multi-source classification feature data in the pore layer, fracture layer and karst layer through the hierarchical unit spaces;

[0074] S32: Based on the corresponding multi-dimensional groundwater classification system, set up corresponding classification spaces in each level of unit space, and standardize the obtained classification spaces according to the multi-source classification feature data to set up classification unit spaces;

[0075] It should be further explained that, in the process of setting up the classification unit space, the process is the same as that of setting up the hierarchical unit space. It is obtained by standardizing the multi-source classification feature data of different types. The hierarchical space, hierarchical unit space, classification space and classification unit space set up improve the storage efficiency and data utilization efficiency of different groundwater multi-source data to a certain extent, and provide convenience for subsequent machine learning.

[0076] S33 will distribute the obtained groundwater multi-source data in the hierarchical space, hierarchical unit space, classification space and classification unit space according to the corresponding multi-source classification data and multi-source classification feature data.

[0077] S34: The hierarchical space, hierarchical unit space, classification space and classification unit space are uniformly marked as progressive storage space, and the corresponding distributed storage results in each progressive storage space are constructed into a distributed three-dimensional storage space according to the data type and collection time;

[0078] S35: The multi-source classification data stored in the distributed three-dimensional storage space are sequentially cross-combined based on corresponding specified variable data according to different multi-source classification feature data within the same progressive storage space, the same multi-source classification feature data within the same progressive storage space, the same multi-source classification feature data between different progressive storage spaces, and different multi-source classification feature data between different progressive storage spaces. The specified variable data is the difference constraint data between the corresponding types of multi-source classification feature data within the same progressive storage space or between different progressive storage spaces during the cross-combination process, thereby avoiding differences between multi-source classification feature data. If the value is too large, the resulting interactive feature dataset will be invalid. The system will take each classification unit space in the progressive storage space as a reference, and obtain the same or different types of multi-source classification feature data stored in the progressive storage space according to the specified variable data. The corresponding cross combination will be completed. The progressive storage space will be traversed in the order of classification unit space, classification space, hierarchical unit space and hierarchical space. During the traversal, the space range of the progressive storage space will be traversed and combined in ascending order. The interactive feature dataset will be generated based on the traversal results.

[0079] S36: The obtained interactive feature dataset is labeled according to the source of the distributed three-dimensional storage space to which the multi-source classification data belongs. The interactive feature datasets obtained from each labeling process are divided according to the proportion, and the corresponding data analysis set and data validation set are obtained respectively.

[0080] In the specific implementation process, the step S4 of obtaining the initial DRASTIC model, sequentially performing traversal analysis on the initial DRASTIC model using the obtained data analysis set to obtain the traversal analysis results, and mapping the obtained traversal analysis results according to the distributed storage results to generate a model parameter reference image includes:

[0081] S41: Obtain the initial DRASTIC model and the original related parameters involved. The original related parameters are the model related parameters originally set in the initial DRASTIC model. The obtained data analysis sets are used to traverse and analyze the initial DRASTIC model in turn based on the machine learning algorithm to obtain the model related parameters corresponding to each data analysis set.

[0082] S411: Obtain the initial DRASTIC model, which consists of seven factors i, including groundwater depth D, net recharge R, aquifer medium A, soil medium S, topographic slope T, vadose zone influence I, and hydraulic conductivity C. ,in, To score the indicators, To use fixed weights, they are labeled as the original relevant parameters;

[0083] S412: Divide the obtained data analysis set according to seven factors, input the index values ​​of the seven factors and label them as input features, label the pollution assessment data as target labels, set the training dataset, and set the objective function. ,in The value is the model prediction, and n is the number of samples.

[0084] S413: Iteratively train the initial DRASTIC model based on the random forest machine learning algorithm, and input feature vectors of the training set. Input-coupled model The model prediction value is calculated using the nonlinear mapping function f(⋅). ;

[0085] S414: Calculate the predicted value based on the set loss function. With real labels The loss value is used to iteratively update the model parameters. Set it as a model-related parameter;

[0086] S42: Validate the model-related parameters of the initial DRASTIC model using the data validation set, and obtain the model-related parameters of the initial DRASTIC model corresponding to the interactive feature dataset;

[0087] S43: Map the obtained model-related parameters according to the location information and parameter type of the distributed three-dimensional storage space to which the interactive feature dataset belongs, and generate model parameter reference images.

[0088] In the specific implementation process, the step S5, which involves setting corresponding flexible adjustment nodes based on the data mapping processing results within the model parameter reference image, performing correlation analysis on the model-related parameters at the mapping processing results through the flexible adjustment nodes, and setting adjustment parameters based on the correlation analysis results, includes:

[0089] S51: Based on the set interactive feature dataset, set flexible adjustment nodes at the data mapping processing results in the model parameter reference image respectively, and set flexible adjustment sub-nodes in the flexible adjustment nodes according to the corresponding parameter types;

[0090] S52: The flexible adjustment sub-nodes within each flexible adjustment node are fused and extracted based on the spatial location of the reference image of their respective model parameters, multi-source classification data, and parameter type. The fusion and extraction results are then used to generate an association analysis dataset.

[0091] S53: Based on the time-series correlation method, construct a parameter correlation strength matrix for the obtained correlation analysis dataset, corresponding to the correlation degree, cooperative change law and response relationship between the relevant parameters of the model. Calculate the corresponding adaptive adjustment coefficients based on the parameter correlation strength matrix, mark the obtained adaptive adjustment coefficients as adjustment parameters, and map and store the obtained adjustment parameters according to the position of the flexible adjustment node in the model parameter reference image.

[0092] S531: Obtain the correlation analysis dataset, set the correlation analysis scheme, and perform analysis and matching processing between the obtained correlation analysis dataset and the set correlation analysis scheme to obtain the corresponding correlation data. The correlation analysis scheme includes, but is not limited to: calculating the correlation data of the change magnitude of the same location at different times or different locations at the same time; analyzing whether different types of parameters change synchronously or influence each other based on Pearson linear correlation or mutual information (MI); and calculating the correlation of various types of parameters between different locations with the decay of the force based on the spatiotemporal correlation function.

[0093] S532: Analyze and process the corresponding association analysis datasets according to the association analysis scheme, obtain the correlation degree, cooperative change law and response relationship between the relevant parameters of the corresponding models, and standardize the obtained correlation degree, cooperative change law and response relationship between the relevant parameters of the models to generate the corresponding association numerical data;

[0094] S533: Construct a parameter correlation strength matrix from the pairwise correlation coefficients of all parameters. The rows and columns correspond to the relevant parameters, and r is the corresponding associated numerical data;

[0095] S534: Obtain the corresponding adjustment parameters based on the parameter correlation strength matrix;

[0096] S54: Map the model parameter reference image to the distributed three-dimensional storage space and establish a one-to-one correspondence.

[0097] In the specific implementation process, step S6, which involves acquiring real-time multi-source groundwater data and flexibly coupling the initial DRASTIC model with the corresponding flexible adjustment nodes within the model parameter reference image to generate a real-time DRASTIC model for analysis and processing, includes:

[0098] S61: Acquire real-time multi-source groundwater data, classify the acquired real-time multi-source groundwater data, and obtain the location information of the respective distributed three-dimensional storage space;

[0099] S62: Based on the one-to-one correspondence between the distributed three-dimensional storage space and the model parameter reference image, obtain the flexible adjustment node at the corresponding location information;

[0100] S63: Obtain the relevant flexible adjustment sub-nodes based on the distribution of each flexible adjustment node, obtain the corresponding adjustment parameters based on the flexible adjustment sub-nodes, and obtain the corresponding model-related parameters based on the adjustment parameters.

[0101] S64: Adjust the initial DRASTIC model based on the obtained model-related parameters to generate a real-time DRASTIC model, and conduct a vulnerability assessment of real-time groundwater multi-source data based on the real-time DRASTIC model.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based method for modeling groundwater vulnerability, characterized in that, Includes the following steps: Step S1: Set up a groundwater modeling and management platform, obtain multi-source groundwater data through the platform, preprocess the obtained multi-source groundwater data, and manage the preprocessed multi-source groundwater data in layers. Step S2: Classify the hierarchical management results to obtain multi-source classification data. Extract features from the obtained multi-source classification data to obtain multi-source classification feature data. Step S3: Distribute the obtained multi-source classification feature data according to the hierarchical management results and classification management results, perform interactive mining based on the distributed storage results, generate interactive feature dataset, and divide the obtained interactive feature dataset into data analysis set and data validation set; Step S4: Obtain the initial DRASTIC model, sequentially perform traversal analysis on the initial DRASTIC model using the obtained data analysis set, obtain the traversal analysis results, and map the obtained traversal analysis results according to the distributed storage results to generate a model parameter reference image. The process of generating the model parameter reference image includes: The initial DRASTIC model and its original related parameters are obtained. The original related parameters are the model-related parameters originally set in the initial DRASTIC model. The obtained data analysis sets are used to traverse and analyze the initial DRASTIC model based on machine learning algorithms to obtain the model-related parameters corresponding to each data analysis set. The model-related parameters involved in the initial DRASTIC model are validated through the data validation set to obtain the model-related parameters of the initial DRASTIC model corresponding to the interactive feature dataset. The obtained model-related parameters are mapped according to the location information and parameter type of the distributed three-dimensional storage space to which the interactive feature dataset belongs, to generate model parameter reference images; Step S5: Set the corresponding flexible adjustment node according to the data mapping processing result in the model parameter reference image, perform correlation analysis on the model-related parameters at the mapping processing result through the flexible adjustment node, and set the adjustment parameters according to the correlation analysis result; The process of performing correlation analysis on model-related parameters at the mapping processing result by flexibly adjusting nodes includes: Based on the set interactive feature dataset, flexible adjustment nodes are set at the data mapping processing results in the model parameter reference image, and flexible adjustment sub-nodes are set in the flexible adjustment nodes according to the corresponding parameter types. Each flexible adjustment sub-node within each flexible adjustment node is fused and extracted based on the spatial location of the reference image of its respective model parameter, multi-source classification data, and parameter type. The fusion and extraction results are then used to generate an association analysis dataset. Based on the obtained correlation analysis dataset, the correlation strength matrix is ​​constructed according to the correlation degree, cooperative change law and response relationship between the relevant parameters of the corresponding model according to the time series correlation method. The corresponding adaptive adjustment coefficient is calculated according to the parameter correlation strength matrix. The obtained adaptive adjustment coefficient is marked as the adjustment parameter. The obtained adjustment parameter is mapped and stored according to the position of the flexible adjustment node in the model parameter reference image. The mapping and storage results of the model parameter reference image are mapped one-to-one with the distributed three-dimensional storage space; Step S6: Obtain real-time groundwater multi-source data, and flexibly couple the initial DRASTIC model with the corresponding flexible adjustment nodes in the model parameter reference image to generate a real-time DRASTIC model for analysis and processing.

2. The groundwater vulnerability modeling method based on machine learning according to claim 1, characterized in that, The process of acquiring multi-source groundwater data includes: A groundwater modeling and management platform is set up, which includes a data input terminal, a data acquisition terminal, and a data management terminal. The data entry terminal is used to connect to the multi-source data entry interface set in the groundwater modeling and management platform to obtain the groundwater multi-source data corresponding to the multi-source data entry interface. The data acquisition terminal is used to connect with the multi-source data acquisition equipment set in the groundwater modeling and management platform to acquire the groundwater multi-source data corresponding to the multi-source data acquisition equipment. The data management terminal is used to clean and standardize the acquired groundwater multi-source data. The pre-processed groundwater multi-source data is managed hierarchically according to the type of medium, including three classification management levels: porous layer, fracture layer and karst layer. The groundwater multi-source data is then labeled according to the classification management level.

3. The groundwater vulnerability modeling method based on machine learning according to claim 2, characterized in that, The process of obtaining multi-source classification data and multi-source classification feature data includes: A multi-dimensional groundwater classification system is established based on hydrogeological type and auxiliary environment type. Groundwater multi-source data in different classification management levels are classified and managed according to the multi-dimensional groundwater classification system to obtain multi-source classification data. Feature extraction schemes are set for different types of multi-source classification data. Based on the corresponding feature extraction schemes, features are extracted from the obtained multi-source classification data to obtain multi-source classification feature data of each unified dimension of different types of multi-source classification data.

4. The groundwater vulnerability modeling method based on machine learning according to claim 3, characterized in that, The process of distributing and storing the obtained multi-source classification feature data includes: Based on the data labels of groundwater multi-source data, the corresponding classification and management levels are obtained. Different classification and management levels are set with corresponding hierarchical spaces. The obtained hierarchical spaces are standardized according to the multi-source classification feature data, and hierarchical unit spaces are set. Based on the corresponding multi-dimensional groundwater classification system, corresponding classification spaces are set up in each level of unit space. The obtained classification spaces are standardized according to the multi-source classification feature data to set up classification unit spaces. The obtained groundwater multi-source data are distributed and stored sequentially in their respective hierarchical space, hierarchical unit space, classification space, and classification unit space, based on the corresponding multi-source classification data and multi-source classification feature data.

5. The groundwater vulnerability modeling method based on machine learning according to claim 4, characterized in that, The process of generating the interaction feature dataset includes: The hierarchical space, hierarchical unit space, classification space, and classification unit space are uniformly marked as progressive storage space. The distributed storage results corresponding to each progressive storage space are used to construct a distributed three-dimensional storage space based on the data type and collection time. The multi-source classification data stored in the distributed three-dimensional storage space are cross-combined based on the corresponding specified variable data according to different multi-source classification feature data in the same progressive storage space, the same multi-source classification feature data in the same progressive storage space, the same multi-source classification feature data between different progressive storage spaces, and different multi-source classification feature data between different progressive storage spaces. The progressive storage space is traversed in the order of classification unit space, classification space, hierarchical unit space, and hierarchical space. An interactive feature dataset is generated based on the traversal results. The obtained interactive feature dataset is labeled according to the source of the distributed three-dimensional storage space to which the multi-source classification data belongs. The interactive feature datasets obtained from each labeling process are divided according to the proportion, and the corresponding data analysis set and data validation set are obtained respectively.

6. The groundwater vulnerability modeling method based on machine learning according to claim 1, characterized in that, The process of generating a real-time DRASTIC model includes: Acquire real-time multi-source groundwater data, classify the acquired real-time multi-source groundwater data, and obtain the location information of the corresponding distributed three-dimensional storage space; Based on the one-to-one correspondence between the distributed three-dimensional storage space and the model parameter reference image, obtain the flexible adjustment node at the corresponding location information; Based on the distribution of each flexible adjustment node, obtain the relevant flexible adjustment sub-nodes, obtain the corresponding adjustment parameters based on the flexible adjustment sub-nodes, obtain the corresponding model-related parameters based on the adjustment parameters, adjust the initial DRASTIC model based on the obtained model-related parameters, generate a real-time DRASTIC model, and perform vulnerability assessment on real-time groundwater multi-source data based on the real-time DRASTIC model.

Images