Environment diagnosis method and device based on numerical simulation and svm efficient modeling

Abstract

The present application relates to an environmental diagnosis method and device based on numerical simulation and SVM efficient modeling, belonging to the technical field of soil-groundwater pollution prevention and control and risk management, the environmental diagnosis method comprises: selecting the related characteristic variable related to the groundwater environment state and the space-time evolution, the environmental diagnosis model based on numerical simulation and SVM efficient modeling is established for the complex groundwater environment system under the typical condition framework, the reliable diagnosis prediction of the soil and groundwater environment state, the evolution process and the related influence under the premise of meeting the given condition framework in different application scenes can be realized simply and conveniently, the accuracy, efficiency and practicability of the evaluation and prediction of the environmental evolution and risk level of the complex groundwater system are improved, which provides strong support for the protection of water and soil environment system and the safe use of resources, especially the precise prevention and repair of soil and groundwater pollution and the efficient management and control of risk, and has important practical value.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of soil-groundwater pollution prevention and control and risk management, and particularly relates to an environmental diagnosis method and device based on numerical simulation and SVM efficient modeling. BACKGROUND

[0002] At present, in the monitoring and risk assessment of groundwater pollution, applying a numerical simulation system to evaluate and quantify the state of groundwater pollution and environmental risk is a common research method. However, in order to accurately simulate, a large number of influencing factors and complex evolution processes need to be considered, which is time-consuming, professional, and has great uncertainty in modeling and prediction of complex medium groundwater systems. The Monte Carlo method and other methods for uncertainty assessment require hundreds or thousands of numerical simulation experiments, which is computationally expensive and even unacceptable, and low in efficiency. Therefore, there is an urgent need for a practical, efficient, simple and accurate groundwater environmental quantification and risk discrimination method. SUMMARY

[0003] To this end, the present application provides an environmental diagnosis method and device based on numerical simulation and SVM efficient modeling to solve the problems of large uncertainty, high cost and low efficiency in environmental risk diagnosis in the prior art.

[0004] To achieve the above purpose, the present application adopts the following technical solutions:

[0005] In a first aspect, the present application provides an environmental diagnosis method based on numerical simulation and SVM efficient modeling, comprising:

[0006] Selecting a related characteristic variable of a groundwater environment and constructing a corresponding groundwater environment system spatio-temporal evolution conceptual model under a typical condition framework; selecting and obtaining a numerical simulation system according to a mathematical control equation and a numerical model;

[0007] Based on the spatio-temporal evolution conceptual model, the numerical model and the numerical simulation system, n model input parameters affecting the spatio-temporal evolution of the related characteristic variable are obtained, and the model input parameters are specifically model input parameters; the variation interval of each model input parameter is selected and then a model input parameter interval set is obtained;

[0008] According to a given number of typical scenarios, different model input parameter combinations of the groundwater environment system form different typical scenarios; based on the numerical simulation system, the simulation values of the corresponding related characteristic variables of the different typical scenarios are obtained, and the model input parameter combinations of all the different typical scenarios and the simulation values of the corresponding related characteristic variables are collected to obtain a statistical modeling data set; a multiple regression statistical prediction model of the related characteristic variable and the model input parameter is established and optimized according to the statistical modeling data set;

[0009] obtaining a first set of data-enhanced scenario parameter combinations based on n model input variables, obtaining a first set of data-enhanced scenario simulation values of the associated characteristic variables corresponding to each scenario through the numerical simulation system, and collecting the first set of data-enhanced scenario parameter combinations and the first set of data-enhanced scenario simulation values as a model verification data set;

[0010] obtaining a second set of data-enhanced scenario parameter combinations based on n model input variables, obtaining a second set of data-enhanced scenario prediction values of the associated characteristic variables corresponding to each scenario through the multivariate regression statistical prediction model, determining the prediction confidence interval of the associated characteristic variables according to the second set of data-enhanced scenario prediction values and the prediction error random variable distribution interval corresponding to the given preset confidence level, and obtaining a statistical analysis enhanced modeling data set based on the n model input variables and the prediction confidence interval;

[0011] if a single environmental diagnosis threshold is given, sequentially obtaining a first set of output label values, a second set of output label values and a third set of output label values of the SVM classification machine learning based on the statistical modeling data set, the statistical analysis enhanced modeling data set and the model verification data set;

[0012] merging the statistical modeling data set, the statistical analysis enhanced modeling data set and the model verification data set corresponding to a preset number of scenario parameter combinations, and the first set of output label values, the second set of output label values and the third set of output label values corresponding thereto, respectively, to obtain a single threshold SVM classification machine learning data set;

[0013] if multiple environmental diagnosis thresholds are given, selecting different thresholds according to a preset threshold value variation interval and a preset threshold value number, increasing the environmental diagnosis threshold as a model input variable of the SVM classification machine learning modeling based on the statistical modeling data set and the statistical analysis enhanced modeling data set, and based on the model verification data set, and further obtaining a threshold variable-containing SVM classification machine learning data set;

[0014] obtaining a single threshold SVM environmental diagnosis model and a threshold variable-containing SVM environmental diagnosis model according to the single threshold SVM classification machine learning data set and the threshold variable-containing SVM classification machine learning data set according to and applying the SVM classification model modeling principle and optimization method, respectively;

[0015] For the practical application scenarios conforming to the typical condition framework, according to the actual conditions and collected data of the specific groundwater environment system, the specific values of the model variables are obtained, the output label values corresponding to the threshold values are obtained through the single threshold SVM environment diagnosis model, the output label values are the environment diagnosis prediction results, and / or according to the actual conditions and collected data of the specific groundwater environment system and the preset environment diagnosis threshold, the environment diagnosis prediction results are obtained through the threshold variable-containing SVM environment diagnosis model.

[0016] Further, the statistical modeling data set obtained by combining the model input parameter combinations of all the different typical scenarios and the corresponding simulation value sets of the associated characteristic variables thereof includes:

[0017] The variation level of each model input variable is selected; different typical scenarios composed of different model input parameter combinations of the groundwater environment system are obtained through orthogonal test method or input variable full combination method for deterministic scenario sampling;

[0018] The corresponding simulation values of the associated characteristic variables of the different typical scenarios are obtained based on the numerical simulation system; the statistical modeling data set of the deterministic scenario sampling is obtained according to the different model input parameter combinations of the different typical scenarios and the corresponding simulation values thereof;

[0019] And a preset number of random scenario samplings are obtained based on random values of n model input variables, that is, a certain number of different random scenario model input parameter combinations are obtained, the simulation values of the associated characteristic variables of each random scenario are obtained through the numerical simulation system, and the statistical modeling data set of the random scenario sampling is obtained by combining the random scenario model parameter combination and the simulation value set of the associated characteristic variables of the random scenario;

[0020] And the statistical modeling data sets of the deterministic scenario sampling and the random scenario sampling are integrated to obtain the statistical modeling data set of the integration of the deterministic scenario sampling and the random scenario sampling. Further, the SVM classification machine learning data set includes: SVM classification machine learning training set and SVM classification machine learning test set;

[0021] And the statistical modeling data set and the statistical analysis reinforcement modeling data set are respectively corresponding to a preset number of scenario parameter combinations, and after being combined with the first group of output label values and the second group of output label values respectively corresponding thereto, a single threshold SVM classification machine learning training set is obtained; the model verification data set corresponding to a preset number of scenario parameter combinations is combined with the third group of output label values to obtain a single threshold SVM classification machine learning test set;

[0022] and the obtained threshold variable containing SVM classification machine learning data set further comprises, if a plurality of environmental diagnosis threshold values are given, selecting different threshold values according to a preset threshold value variation interval and a preset threshold value number, taking the environmental diagnosis threshold value as a new model input variable based on the statistical modeling data set and the statistical analysis and reinforcement modeling data set, and obtaining a training set and a test set of the threshold variable containing SVM classification machine learning respectively based on the model verification data set.

[0023] Further, it further comprises: given the modeling accuracy preset value of the SVM environmental diagnosis model, including the training accuracy preset value and the test accuracy preset value, if the obtained modeling accuracy of the SVM environmental diagnosis model is lower than the preset value, increasing the different given typical scenario preset number and / or increasing the preset number of different scenario sampling for SVM environmental diagnosis model modeling optimization to obtain a corresponding accuracy SVM environmental diagnosis model; the corresponding accuracy SVM environmental diagnosis model includes a single threshold SVM environmental diagnosis model and a threshold variable containing SVM environmental diagnosis model.

[0024] Further, the correlation characteristic variable is at least 1; the correlation characteristic variable includes: groundwater pollutant concentration, saturated zone pollutant cross section or interface flux, pollution plume migration and diffusion front position, pollution plume stable time, stable pollution plume distribution area, stable pollution plume front distance from pollution source distance, stable pollution plume average concentration and pollution plume dissipation time;

[0025] and soil pollutant concentration, unsaturated zone gas phase pollutant concentration, unsaturated zone cross section pollutant flux, unsaturated zone-saturated zone pollutant interface flux, soil-gas interface pollutant flux, rock-gas interface pollutant flux and NAPL pollutant spatiotemporal distribution.

[0026] Further, the construction of the corresponding groundwater environmental system spatiotemporal evolution conceptual model under the typical condition framework comprises:

[0027] Determine the characteristics of the unsaturated zone, including medium type, lithology structure, medium heterogeneity and anisotropy characteristics;

[0028] Determine the characteristics of the saturated zone, including medium type, water-bearing system structure, seepage dimension, flow regime, medium heterogeneity and anisotropy characteristics;

[0029] Determine the characteristics of the pollution source, including the phase state of the pollutant of interest, the type of the pollutant of interest, the spatial distribution characteristics of the pollution source, the source intensity variation and dynamic characteristics;

[0030] Determine the characteristics of the boundary conditions, including the characteristics of the hydrodynamic field boundary conditions and the characteristics of the hydrochemical field boundary conditions;

[0031] determining initial condition characteristics, the initial condition characteristics including hydrodynamic field initial condition characteristics and hydrochemical field initial condition characteristics.

[0032] Further, the correlation characteristic variable and the model input variable include:

[0033] The correlation characteristic variable and the model input variable are processed by any one or any combination of preset data processing methods, the preset data processing methods including no conversion, equal scaling, logarithmic conversion, positive and negative consistency, and normalization.

[0034] Further, the scenario sampling of a preset number based on the n model input variables includes:

[0035] The random sampling number is selected, the random distribution type and the corresponding distribution parameters suitable for each model input variable are determined or selected, the random number generation of the model input variable is completed, and a scenario of a new parameter combination is formed.

[0036] Further, the correlation characteristic variable prediction error value and its distribution characteristics are obtained by the multiple regression statistical prediction model; according to the prediction error value and its distribution characteristics, a prediction error random variable distribution interval corresponding to a given preset confidence is determined, including:

[0037] Based on the model validation set data, the multiple regression statistical prediction model is used to obtain a prediction error value data set, the prediction error value data set including a plurality of prediction errors; the mean and variance of the prediction error random variable are estimated, the probability density function is fitted and identified, the sum of the two-side cumulative probabilities is determined according to the preset confidence, and the corresponding confidence prediction error random variable distribution interval is obtained.

[0038] Further, the SVM classification model modeling principle and the SVM classification model modeling optimization method in the optimization method include:

[0039] The SVM classification model modeling optimization method optimizes the modeling parameters;

[0040] The modeling parameter optimization includes

[0041] The kernel function type optimization and the associated parameter optimization;

[0042] The kernel function type optimization includes linear kernel, polynomial kernel, radial basis function kernel, and Sigmoid kernel selection;

[0043] The radial basis function kernel associated parameter optimization includes regularization parameter C and data mapping complexity parameter gamma optimization;

[0044] The SVM modeling optimization method preferably includes a grid search method, a particle swarm optimization method, and a genetic particle swarm hybrid algorithm.

[0045] Further, the obtaining of the SVM classification machine learning first group of output label values, second group of output label values, and third group of output label values includes:

[0046] If a single environment diagnosis threshold corresponding to each of the associated characteristic variables is given, then according to the statistical modeling data set and the model verification data set, the relative size of each of the associated characteristic variable simulation values and the corresponding single environment diagnosis threshold is compared, and the SVM classification machine learning first group of output label values and the third group of output label values are obtained in sequence; according to the statistical analysis reinforcement modeling data set, if the prediction confidence interval of each of the associated characteristic variables does not contain the corresponding single environment diagnosis threshold, then according to the relative size of the upper and lower limits of the prediction confidence interval and the corresponding single environment diagnosis threshold, the SVM classification machine learning second group of output label values is obtained;

[0047] In a second aspect, the present application provides an environment diagnosis device based on numerical simulation and SVM efficient modeling,

[0048] The device includes:

[0049] A conceptual model construction module is configured to select associated characteristic variables of a groundwater environment and construct a corresponding groundwater environment system spatio-temporal evolution conceptual model under a typical condition framework; and select and obtain a numerical simulation system according to mathematical control equations and a numerical model.

[0050] A numerical simulation module is configured to obtain n model input parameters affecting the spatio-temporal evolution of the associated characteristic variables based on the spatio-temporal evolution conceptual model, the numerical model, and the numerical simulation system, wherein the model input parameters are specifically model input parameters; select a variation interval of each model input parameter and then obtain a model input parameter interval set; and obtain corresponding numerical simulation values of the typical scenarios and scenario sampling based on the numerical simulation system.

[0051] A statistical model construction module is configured to obtain different typical scenarios formed by different combinations of model input parameters of a groundwater environment system according to a given number of typical scenarios; obtain corresponding numerical simulation values of the different typical scenarios based on the numerical simulation system; obtain a statistical modeling data set by collecting the combinations of model input parameters of all the different typical scenarios and the corresponding associated characteristic variable simulation values; and establish and optimize a multiple regression statistical prediction model of the associated characteristic variables and the model input parameters based on the statistical modeling data set.

[0052] The statistical model prediction error analysis module is configured to obtain a preset number of scenario samples based on the n model input variables, obtain a first set of data-intensive scenario parameter combinations, obtain a first set of data-intensive scenario simulation values of the associated characteristic variables corresponding to each scenario through the numerical simulation system, and set the first set of data-intensive scenario parameter combinations and the first set of data-intensive scenario simulation values as a model verification data set.

[0053] The machine learning intensive data set generation module is configured to obtain a preset number of scenario samples based on the n model input variables, obtain a second set of data-intensive scenario parameter combinations, and obtain a second set of data-intensive scenario prediction values of the associated characteristic variables corresponding to each scenario through the multivariate regression statistical prediction model.

[0054] If a single environmental diagnosis threshold is given, the SVM classification machine learning first set of output label values, the second set of output label values, and the third set of output label values are obtained in sequence based on the statistical modeling data set, the statistical analysis intensive modeling data set, and the model verification data set.

[0055] If multiple environmental diagnosis thresholds are given, different thresholds are selected according to a preset threshold value variation interval and a preset threshold value number, the statistical modeling data set and the statistical analysis intensive modeling data set are used as model input variables for SVM classification machine learning modeling based on the model verification data set, and a threshold variable-containing SVM classification machine learning data set is obtained.

[0056] The universal SVM environmental diagnosis model construction module is configured to obtain the single threshold SVM classification machine learning data set and the threshold variable-containing SVM classification machine learning data set, apply SVM classification model modeling principles and optimization methods, and obtain a corresponding single threshold SVM environmental diagnosis model and a threshold variable-containing SVM environmental diagnosis model in sequence.

[0057] The application scenario environment diagnosis module, for the actual application scenario meeting the typical condition framework, obtains the specific value of the model variable according to the actual condition and collected data of the specific groundwater environment system, obtains the output label value corresponding to the threshold value through the single threshold value SVM environment diagnosis model, the output label value is an environment diagnosis prediction result, and / or obtains the environment diagnosis prediction result through the threshold value variable containing SVM environment diagnosis model according to the actual condition and collected data of the specific groundwater environment system and the preset environment diagnosis threshold value.

[0058] The application has at least the following beneficial effects by adopting the above technical scheme:

[0059] The application provides an environment diagnosis method and device based on numerical simulation and SVM efficient modeling, establishes an environment diagnosis model based on SVM for a complex groundwater environment system under a typical condition framework based on the correlation characteristic variables of the groundwater environment, and can simply and conveniently realize reliable diagnosis and prediction of the groundwater environment state, evolution process and related influence of different application scenarios under the premise of meeting the given condition framework through the environment diagnosis model based on SVM, thereby improving the accuracy, efficiency and practicality of the environment diagnosis and risk assessment of the complex groundwater system, providing strong support for groundwater pollution risk diagnosis and prevention and repair, and having important practical value.

[0060] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only show some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0062] Figure 1 is a flowchart of an environment diagnosis method based on numerical simulation and SVM efficient modeling according to an exemplary embodiment of the application;

[0063] Figure 2 is a structural schematic diagram of the vadose zone- phreatic interface under a layered heterogeneous typical condition framework according to an exemplary embodiment of the application;

[0064] Figure 3 is a schematic block diagram of an environment diagnosis device based on numerical simulation and SVM efficient modeling according to an exemplary embodiment of the application.

[0065] The application will be further described below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0066] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present application.

[0067] It is a common research method to evaluate the state and environmental risk of groundwater pollution by using a numerical simulation system. However, in order to accurately simulate, a large number of influencing factors and complex evolution processes need to be considered, which is time-consuming, professional, and has great uncertainty in modeling and prediction of complex medium groundwater system. The Monte Carlo method and other methods for uncertainty evaluation need hundreds or thousands of numerical simulation experiments, but the calculation cost is huge and even unacceptable. Building a high-efficiency machine learning prediction model based on numerical simulation provides another choice for groundwater environmental process quantification and risk diagnosis. The premise of establishing a good machine learning model is to have a certain scale and quality of data. However, due to the complexity of the model and the influence of computing power, even if supercomputing is used, a large amount of data set cannot be obtained for training. Therefore, combining numerical simulation with machine learning method is more efficient and practical for precise modeling, which can reliably support groundwater environmental process quantification and risk diagnosis, and become the current urgent environmental problem to be solved.

[0068] The embodiment of the present application provides an environmental diagnosis method and device based on numerical simulation and SVM high-efficiency modeling, which can establish a SVM environmental problem diagnosis and prediction model with certain universality for a complex groundwater environmental system under a certain condition, effectively realize rapid and accurate prediction of groundwater environmental state, evolution process and related influence under different application scenarios in a given condition framework, so as to improve the accuracy, efficiency and practicability of complex groundwater system environmental diagnosis and risk assessment.

[0069] The method and device in the present application will be described below through specific embodiments.

[0070] Please refer to Figure 1 , Figure 1 It is a flowchart of an environmental diagnosis method based on numerical simulation and SVM high-efficiency modeling according to an exemplary embodiment of the present application, please refer to Figure 1 The method comprises the following steps.

[0071] Step S11, selecting the associated characteristic variable of the groundwater environment and constructing the corresponding groundwater environment system spatio-temporal evolution conceptual model under the typical condition framework; selecting and obtaining the numerical simulation system according to the mathematical control equation and the numerical model;

[0072] Step S12, obtaining n model input variables affecting the spatio-temporal evolution of the associated characteristic variable based on the spatio-temporal evolution conceptual model, the numerical model and the numerical simulation system, and the specific values of the model input variables are model input parameters; selecting the variation interval of each model input variable and then obtaining the model input variable interval set by collection;

[0073] Step S13, obtaining different typical scenarios composed of different combinations of model input parameters of the groundwater environment system according to the given number of typical scenarios; obtaining the simulation values of the corresponding associated characteristic variables of different typical scenarios based on the numerical simulation system, and obtaining the statistical modeling data set by collecting all the combinations of model input parameters and the corresponding simulation values of the associated characteristic variables of different typical scenarios; establishing and optimizing the multiple regression statistical prediction model of the associated characteristic variable and the model input variable according to the statistical modeling data set;

[0074] Step S14, obtaining a preset number of scenario samples based on n model input variables, obtaining a first group of data-intensive scenario parameter combinations, obtaining the first group of data-intensive scenario simulation values of the corresponding associated characteristic variables of each scenario through the numerical simulation system, and collecting the first group of data-intensive scenario parameter combinations and the first group of data-intensive scenario simulation values as the model validation data set; obtaining the prediction error values and their distribution characteristics of the associated characteristic variable through the multiple regression statistical prediction model based on the model validation data set; determining the prediction error random variable distribution interval corresponding to the given preset confidence according to the prediction error values and their distribution characteristics;

[0075] Step S15, obtaining a preset number of scenario samples based on n model input variables, obtaining a second group of data-intensive scenario parameter combinations, obtaining the second group of data-intensive scenario prediction values of the corresponding associated characteristic variables of each scenario through the numerical simulation system; determining the prediction confidence interval of the associated characteristic variable according to the second group of data-intensive scenario prediction values and the prediction error random variable distribution interval corresponding to the given preset confidence; obtaining the statistical analysis intensive modeling data set based on the preset number of scenario parameter combinations and the prediction confidence interval of n model input variables and then merging them;

[0076] Step S16, if a single environmental diagnosis threshold is given, then the first group of output label values, the second group of output label values and the third group of output label values are obtained in sequence according to the statistical modeling data set, the statistical analysis intensive modeling data set and the model validation data set;

[0077] Step S17, merge the statistical modeling dataset and the statistical analysis reinforcement modeling dataset and the model verification dataset with the first group of output label values and the second group of output label values and the third group of output label values respectively to obtain a single threshold SVM classification machine learning dataset corresponding to a preset number of scenario parameter combinations;

[0078] Step S18, if a plurality of environmental diagnosis thresholds are given, select different thresholds according to the preset threshold value variation interval and the preset threshold value number, and based on the statistical modeling dataset and the statistical analysis reinforcement modeling dataset and based on the model verification dataset, increase the environmental diagnosis threshold as a model input variable of SVM classification machine learning modeling, and then obtain a SVM classification machine learning dataset containing a threshold variable;

[0079] Step S19, according to the single threshold SVM classification machine learning dataset and the SVM classification machine learning dataset containing the threshold variable, according to and applying the SVM classification model modeling principle and optimization method, in turn obtain the corresponding single threshold SVM environmental diagnosis model and the SVM environmental diagnosis model containing the threshold variable;

[0080] Step S20, for the actual application scenario meeting the typical condition framework, according to the actual conditions and collected data of the specific groundwater environment system, obtain the specific value of the model variable, obtain the output label value corresponding to the threshold through the single threshold SVM environmental diagnosis model, the output label value is the environmental diagnosis prediction result, and / or according to the actual conditions and collected data of the specific groundwater environment system and the preset environmental diagnosis threshold, obtain the environmental diagnosis prediction result through the SVM environmental diagnosis model containing the threshold variable.

[0081] It should be noted that the technical solutions provided in the embodiments can be loaded in existing applications in the form of a small program or in the form of a plug-in for use, or in the form of a separate application, and the speed regulation function is realized through an external interface. The applicable scenarios include but are not limited to: diagnosis of groundwater environment system.

[0082] Specifically, the preset number or the preset quantity in each step is set according to the specific application scenario and the specific diagnosis requirement; the preset threshold value variation interval and the preset threshold value number are also set according to the specific application scenario and the specific diagnosis requirement.

[0083] It can be understood that the method provided in the embodiments is based on the correlation characteristic variables of the groundwater environment, and establishes an SVM-based environmental diagnosis model for the complex groundwater environment system under the typical condition framework. Through the SVM-based environmental diagnosis model, reliable diagnosis and prediction of the groundwater environment state, evolution process and related influence in different application scenarios under the premise of meeting the given condition framework can be realized simply and efficiently, and the accuracy, efficiency and practicability of the environmental diagnosis and risk assessment of the complex groundwater system are improved.

[0084] In specific practice, the step S13 of "obtaining a statistical modeling dataset from the model input parameter combinations of all different typical scenarios and the corresponding simulation values of the associated characteristic variables", includes: selecting the variation level of each model input parameter variable; obtaining different typical scenarios of different model input parameter combinations by orthogonal test method or input parameter combination method, for deterministic scenario sampling; obtaining the corresponding simulation values of the associated characteristic variables of different typical scenarios based on the numerical simulation system; obtaining the statistical modeling dataset of deterministic scenario sampling according to different model input parameter combinations of different typical scenarios and the corresponding simulation values; obtaining a preset number of random scenario samplings based on random values of n model input parameter variables, that is, obtaining a certain number of different random scenario model input parameter combinations, obtaining the simulation values of the associated characteristic variables of each random scenario by the numerical simulation system, and obtaining the statistical modeling dataset of random scenario sampling from the random scenario model parameter combination and the simulation value set of the associated characteristic variables of the random scenario; and integrating the statistical modeling dataset of deterministic scenario sampling and the statistical modeling dataset of random scenario sampling to obtain the statistical modeling dataset of deterministic scenario sampling and random scenario sampling.

[0085] In specific practice, the SVM classification machine learning dataset includes an SVM classification machine learning training set and an SVM classification machine learning test set; the method further includes: merging the statistical modeling dataset and the statistical analysis reinforcement modeling dataset corresponding to the preset number of scenario parameter combinations with the first group of output label values and the second group of output label values respectively to obtain a single threshold SVM classification machine learning training set; merging the model verification dataset corresponding to the preset number of scenario parameter combinations with the third group of output label values to obtain a single threshold SVM classification machine learning test set; if a plurality of environmental diagnosis thresholds are given, different thresholds are selected according to the preset threshold value variation interval and the preset threshold value number, and based on the statistical modeling dataset and the statistical analysis reinforcement modeling dataset, and based on the model verification dataset, the environmental diagnosis threshold is taken as a new model input parameter variable to obtain the training set and the test set of the SVM classification machine learning with the threshold parameter variable.

[0086] In specific practice, it further includes: given the modeling accuracy preset value of the SVM environmental diagnosis model, including the training accuracy preset value and the test accuracy preset value, if the modeling accuracy of the obtained SVM environmental diagnosis model is lower than the preset value, the SVM environmental diagnosis model is optimized by increasing the preset number of different given typical scenarios and / or increasing the preset number of different scenario samplings to obtain the corresponding accuracy SVM environmental diagnosis model; the corresponding accuracy SVM environmental diagnosis model includes a single threshold SVM environmental diagnosis model and a SVM environmental diagnosis model with threshold parameter variables.

[0087] In specific implementation, the "correlation characteristic variables" in step S11 include: groundwater pollutant concentration, saturated zone pollutant cross-section or interface flux, pollution plume migration and diffusion front position, pollution plume stable time, stable pollution plume distribution area, stable pollution plume front distance from pollution source, stable pollution plume average concentration, and pollution plume dissipation time; and soil pollutant concentration, aerated zone gas phase pollutant concentration, aerated zone cross-section pollutant flux, aerated zone-saturated zone pollutant interface flux, soil-gas interface pollutant flux, rock-gas interface pollutant flux, and NAPL pollutant spatiotemporal distribution.

[0088] In specific implementation, the correlation characteristic variables and the model input variables are processed by any one or any combination of preset data processing methods, including: no conversion, equal proportion scaling, logarithmic conversion, positive and negative consistency, and normalization.

[0089] In specific implementation, the preset number of scenarios are obtained based on the n model input variables, including selecting a random sampling number, determining or selecting a random distribution type and corresponding distribution parameters suitable for each model input variable, generating random numbers of the model input variables, and then constructing a new parameter combination scenario.

[0090] In specific implementation, the correlation characteristic variable prediction error value and its distribution characteristics are obtained by a multiple regression statistical prediction model; according to the prediction error value and its distribution characteristics, a prediction error random variable distribution interval corresponding to a given preset confidence level is determined, including: based on the model validation set data, a prediction error value dataset is obtained by using the multiple regression statistical prediction model, the prediction error value dataset includes multiple prediction errors; the mean and variance of the prediction error random variable are estimated, the probability density function is fitted and identified, the sum of the two-side cumulative probabilities is determined according to the preset confidence level, and the corresponding confidence prediction error random variable distribution interval is obtained.

[0091] In specific implementation, the SVM classification model modeling optimization method includes SVM classification model modeling optimization method selection and modeling parameter selection; the modeling parameter selection includes kernel function type selection and associated parameter selection; the kernel function type selection includes linear kernel, polynomial kernel, radial basis function kernel, and Sigmoid kernel selection; the radial basis function kernel associated parameter selection includes regularization parameter C and data mapping complexity parameter gamma selection; and the SVM modeling optimization method selection includes grid search method and particle swarm optimization method, and genetic particle swarm hybrid algorithm selection.

[0092] In specific practice, in order to obtain the first group of output label values, the second group of output label values and the third group of output label values of the SVM classification machine learning, if a single environmental diagnosis threshold corresponding to each associated characteristic variable is given, then according to the statistical modeling data set and the model verification data set, the relative size of each associated characteristic variable simulation value and the corresponding single environmental diagnosis threshold is compared, and the first group of output label values and the third group of output label values of the SVM classification machine learning are obtained in turn; according to the statistical analysis of the reinforced modeling data set, if each associated characteristic variable prediction confidence interval does not contain the corresponding single environmental diagnosis threshold, then according to the relative size of the upper and lower limits of the prediction confidence interval and the corresponding single environmental diagnosis threshold, the second group of output label values of the SVM classification machine learning is obtained;

[0093] Please refer to Figure 2 , Figure 2 is a schematic diagram of the structure of the vadose zone- phreatic interface under a layered heterogeneous typical condition framework, as shown in Figure 2 The vadose zone-phreatic interface under a layered heterogeneous typical condition framework includes: a pollution source 1 on the uppermost surface, an upper layer vadose zone 2, a lower layer vadose zone 3, a phreatic surface 4 and a phreatic zone 5 in turn below the pollution source 1; the pollution source 1 also contains accumulated water 6; it also includes a constant flow and concentration position 7 and a one-dimensional unsaturated flow direction 8.

[0094] In specific practice, the step S11 of "constructing a corresponding groundwater environment system space-time evolution conceptual model under a typical condition framework" includes: determining the characteristics of the vadose zone, including medium type, lithological structure, medium heterogeneity and anisotropy characteristics; determining the characteristics of the saturated zone, including medium type, water-bearing system structure, seepage dimension, flow regime, medium heterogeneity and anisotropy characteristics; determining the characteristics of the pollution source, including the phase state of the concerned pollutant, the type of the concerned pollutant, the spatial distribution characteristics of the pollution source, the source intensity variation and dynamic characteristics; determining the characteristics of the boundary conditions, including the characteristics of the hydrodynamic field boundary conditions and the characteristics of the hydrochemical field boundary conditions; determining the characteristics of the initial conditions, including the characteristics of the hydrodynamic field initial conditions and the characteristics of the hydrochemical field initial conditions.

[0095] Specifically, the groundwater environment associated characteristic variables are selected, the corresponding groundwater environment system space-time evolution conceptual model under a typical condition framework is constructed, the suitable mathematical control equation and numerical model for representing the space-time evolution of the corresponding groundwater environment system are further obtained, and the numerical simulation system for quantifying the space-time evolution of the groundwater environment associated characteristic variables is screened.

[0096] In specific practice, step S12 is to determine all model input parameters affecting the spatio-temporal evolution of the selected environmental associated characteristic variable based on the spatio-temporal evolution conceptual model and the numerical model and numerical simulation system used, to form a model input parameter set X = [X1, X2, …, Xn], n being the total number of model input parameters considered; considering the variable range of each model input parameter, the variation interval of each model input parameter to be considered is selected to form a model input parameter interval set V = [V1, V2, …, Vn].

[0097] In specific practice, in step S13, different typical scenarios formed by different combinations of model input parameters of the groundwater environment system are obtained. Specifically, the variation level of each model input parameter is selected, and the full combination or orthogonal test design method of model input parameters is applied to obtain the typical scenarios of different parameter combinations of the selected groundwater environment system, to obtain a set of simulation values Y of the groundwater environment associated characteristic variable, forming the statistical modeling data set of the deterministic scenario sampling.

[0098] In specific practice, in step S14, a preset number of scenario samplings are obtained based on n model input parameters, including: selecting a random sampling number, determining or selecting a random distribution type and corresponding distribution parameters suitable for each model input parameter, generating random numbers of model input parameters to form new parameter combination scenarios; for the orthogonal test design case, a random combination of each model input parameter level value can also be used to form a new parameter combination scenario.

[0099] Specifically, different scenario sampling sets are obtained by changing the combination of model input parameters, the numerical simulation system is applied to obtain a set of simulation values of the corresponding characteristic variable, and the statistical modeling data set of the deterministic scenario sampling is integrated to form a statistical modeling data set, an appropriate data processing method is selected, and a multiple regression statistical prediction model of the associated characteristic variable Y and the model input parameter X is established and optimized.

[0100] In specific practice, further including, further selecting an appropriate random sampling number, generating random numbers of model input parameters again and forming corresponding new more model input parameter combination scenarios, applying the numerical simulation system to obtain simulation values of the corresponding groundwater environment associated characteristic variable, integrating to form a model verification data set, applying the established regression statistical prediction model to obtain prediction error values and their distribution characteristics, and accordingly further determining the prediction error random variable distribution interval (a, b) of the groundwater environment associated characteristic variable statistical prediction model corresponding to a given confidence level.

[0101] In specific practice, the step S14 of "obtaining the prediction error value and its distribution characteristics of the associated characteristic variable by the multivariate regression statistical prediction model based on the model verification data set; and determining the prediction error random variable distribution interval corresponding to the given preset confidence level according to the prediction error value and its distribution characteristics", includes: obtaining the prediction error value data set by the multivariate regression statistical prediction model based on the model verification data set, the prediction error value data set including a plurality of prediction errors; estimating the mean and variance of the prediction error random variable, fitting and identifying its probability density function, and determining the sum of the cumulative probabilities on both sides as the prediction error random variable distribution interval.

[0102] Specifically, a suitable random sampling number is further selected, random number generation of the model input parameter variable is performed again, and a corresponding new more parameter combination scenario is realized, and the multivariate regression statistical prediction model is applied to obtain the prediction value Y of the associated characteristic variable of the underground water environment corresponding to each scenario realization S , based on the prediction error random variable distribution interval (a, b) corresponding to the given confidence level of the statistical prediction model of the associated characteristic variable of the underground water environment determined in the foregoing, the upper and lower limits of the true value of the associated characteristic variable of the underground water environment under the given confidence level, i.e., the prediction confidence interval (Y P1 , Y P2 ) is further determined, and the statistical analysis reinforced modeling data set is generated according to the random model input parameter combination scenario and the characteristic variable prediction confidence interval; similarly, the model verification data set can be supplemented and updated as needed.

[0103] In specific practice, in the steps S16 and S17, suitable data processing methods are adopted for data processing of the statistical modeling data set, the model verification data set, and the statistical analysis reinforced modeling data set, a given environment diagnosis threshold Vc is set, the output label value of the SVM machine learning can be directly obtained by comparing the threshold value with the size of the environmental associated characteristic value for the statistical modeling data set and the model verification data set; and for the statistical analysis reinforced modeling data set, the output label value of the SVM machine learning is identified by comparing the relative size of the upper and lower limits of the environmental associated characteristic variable prediction interval with the threshold value; the parameter combination scenario corresponding to the statistical modeling data set and the statistical analysis reinforced modeling data set and the corresponding output label value of the SVM machine learning form a single threshold SVM machine learning training set, and the parameter combination scenario corresponding to the model verification data set and the corresponding output label value of the SVM machine learning form a single threshold SVM machine learning test set.

[0104] In specific practice, the "SVM classification model modeling optimization method" in step 19 includes an SVM classification model modeling optimization method and modeling parameter optimization; the modeling parameter optimization includes kernel function type optimization and associated parameter optimization; the kernel function type optimization includes linear kernel, polynomial kernel, radial basis function kernel, and Sigmoid kernel selection; the radial basis function kernel associated parameter optimization includes regularization parameter C and data mapping complexity parameter gamma optimization; and the SVM modeling optimization method optimization includes grid search method and particle swarm optimization method and genetic particle swarm hybrid algorithm selection.

[0105] Specifically, suitable SVM machine learning training and test accuracy are selected, based on the SVM machine learning modeling optimization method, while avoiding overfitting, and thus obtaining a single specific environment diagnosis threshold Vc SVM environment diagnosis model.

[0106] In specific practice, in step S20, for a specific groundwater environment system, according to the investigation results of the specific groundwater environment system, it is confirmed that the specific application scene meets the requirements of the defined typical condition framework, and further the specific numerical value of the corresponding input parameter variable of the SVM environment diagnosis model is obtained, and for the specific environment diagnosis threshold considered, the corresponding accuracy prediction result can be obtained according to the SVM environment diagnosis model obtained by the single specific threshold SVM environment diagnosis model determination method, or for the specific environment diagnosis threshold considered, the corresponding accuracy prediction result can be quickly and simply obtained by applying the established universal SVM environment diagnosis model containing threshold parameter variables.

[0107] In one specific embodiment, polycyclic aromatic hydrocarbons (PAHs) are widely present in the environment as organic matter, and are highly valued by countries around the world due to their persistence, carcinogenicity and mutagenicity. The United States has published the carcinogenic concentration of PAHs in groundwater in the range of 0.2-6.9 ng / L, but the corresponding concentration in surface water is 0.1-800 ng / L, most of which is in the range of 2-50 ng / L, and the solubility of PAHs in water is mostly above 1000 ng / L. The contaminated surface organic contaminated water body continues to infiltrate, and will undergo seepage, adsorption retention, hydrodynamic dispersion and biodegradation in the aeration zone, and enter the shallow groundwater along the aeration zone soil with weak antifouling performance, thereby causing the concentration of PAHs in groundwater to continuously increase. According to the analysis of the solubility and carcinogenic concentration range of PAHs, the concentration of PAHs in groundwater is easy to exceed the carcinogenic standard. Therefore, it is necessary to pay attention to the migration and transformation of polycyclic aromatic hydrocarbon pollutants in the aeration zone, including the quantitative research on the flux change of polycyclic aromatic hydrocarbon pollutants at the aeration zone- phreatic interface under typical conditions. This will provide an important basis for the spatio-temporal distribution of related organic pollutants in the water-soil environment of the aeration zone and the risk assessment of groundwater pollution under typical conditions. The embodiment uses limited numerical simulation data to establish a high-efficiency and reliable aeration zone pollutant flux prediction model, and shows the important role of the application in environmental protection.

[0108] In view of the above problems, the environmental diagnosis method based on numerical simulation and SVM high-efficiency modeling is combined, and the specific steps are as follows:

[0109] Step S31, select the stable flux of pollutants at the aeration zone-saturated zone interface as the groundwater environment correlation characteristic variable, construct the corresponding groundwater environment system spatio-temporal evolution conceptual model under the typical condition framework, that is, the aeration zone-phreatic interface flux prediction conceptual model under the layered heterogeneous typical condition framework, further obtain the mathematical control equation and numerical model suitable for representing the spatio-temporal evolution of the corresponding groundwater environment system, and screen the numerical simulation system that can quantify the spatio-temporal evolution of the groundwater environment correlation characteristic variable.

[0110] Specifically, there are polycyclic aromatic hydrocarbon pollution problems in coking plant sites, petroleum chemical plant sites, power plant pollution sites, etc. in China. When precipitation forms surface water in these sites, the water flows stably and infiltrates, forming a stable seepage field. In this case, the concentration of organic matter in the infiltrating water is high, and even close to the solubility, which can be regarded as a constant flow and constant concentration infiltration process. Due to construction excavation or filling, ground ramming or hardening, horizontal drainage engineering measures and the natural sedimentary characteristics of the stratum, the stratum structure of the actual pollution site, landfill site, etc. usually has the characteristics of horizontal layered heterogeneity. In this study, the simulation area is set as a vertical layered heterogeneous profile, and the entire study area is divided into two layers with different lithology. The surface polycyclic aromatic hydrocarbon organic pollutant water enters the groundwater through the double-layered heterogeneous aeration zone soil, one-dimensional vertical flow, the upper boundary is given flow and given concentration condition, and the pollutant stable flux at the aeration zone- phreatic interface is used as the correlation characteristic variable.

[0111] Specifically, the selected organic pollutants are polycyclic aromatic hydrocarbon pollutants represented by benz[a]pyrene. This type of pollutant is a semi-volatile pollutant, and the volatilization effect is limited in a relatively closed underground environment, so the volatilization of the pollutant is not considered in this study.

[0112] Specifically, CHEMFLO-2000 software is used. CHEMFLO-2000 is a numerical simulation system for simulating water flow and solute transport in unsaturated zones. The water flow process is described by the Richards equation, as shown in equation (1); and the solute migration and transformation process is represented by the convection dispersion and adsorption degradation equation, as shown in equation (2). The two are coupled and solved by finite difference method.

[0113]

[0114] In the formula, h is the matrix potential; t is the time; x is the distance from the ground; and K(h) is the hydraulic conductivity corresponding to the matrix potential under the condition.

[0115] The partial differential control equation of solute transport considering linear adsorption and first-order degradation is as follows:

[0116]

[0117] In the formula, t is the time; θ is the soil moisture content; R is the solute lag coefficient in the soil, which is expressed as: ρ = ρ(x) is the soil density, k is the partition coefficient; C is the solute concentration in water; x is the distance from the ground; D is the hydrodynamic dispersion coefficient (including molecular diffusion and mechanical dispersion); q = q(x, t) is the water flow; and α = α(x) is the first-order degradation coefficient of microorganisms in water.

[0118] The Van Genuchten model, see equation (3), is used to describe the soil water characteristic curve, which represents the function between water content and matric potential and other hydraulic parameters.

[0119]

[0120] where: θ is the soil water content; h is the matric potential; θs is the saturated water content; θr is the residual water content; α, m, n are the empirical coefficients of the VG model, where m = 1-1 / n.

[0121] Step S32, based on the constructed conceptual model and the used numerical model and numerical simulation system, determine the influencing environmental associated characteristic variable space-time evolution of the selected parameters (see Table 1), constitute the parameter set X = [X1, X2, …, Xn], n is the total number of considered parameters, n = 16; considering the variable range of each parameter, select the variation interval of each parameter to be considered (see Table 1), constitute the parameter interval set V = [V1, V2, …, Vn], n is the total number of considered parameters, n = 16. 16

[0122] Step S33, select the variation level of each parameter (see Table 1 below Input parameter level and value table), apply orthogonal test design method, obtain the typical scene of different parameter combinations of groundwater environment system, wherein the number of orthogonal test design scene satisfies the non-overfitting condition of statistical modeling;

[0123] Table 1 Input parameter level and value table

[0124]

[0125] Note: F represents the pollutant flux; p represents the thickness ratio of the upper layer; T represents the thickness of the unsaturated zone; K upper and K lower represent the permeability coefficients of the upper and lower layers, respectively; and represent the saturated water contents of the upper and lower layers, respectively; and represent the saturated water contents of the upper and lower layers, respectively; α upper and α lower represent the distribution coefficients of the upper and lower layers, respectively; n upper and n lower represent the exponential parameters in the Van Genuchten model of the upper and lower layers, respectively; Q represents the pollutant flux; C represents the pollutant concentration; D represents the dispersion; k d represents the degradation coefficient.

[0126] ​Step S34, for the groundwater environment related characteristic variable, based on the numerical simulation, the orthogonal test design typical scenario groundwater environment related characteristic variable simulation value Y is obtained, and the statistical modeling data set of the deterministic scenario sampling is constituted.

[0127] Step S35, 20 different parameter combination scenario samplings are added, corresponding characteristic variable simulation values are obtained by using a numerical simulation system, and the statistical modeling data set is constituted by integrating the statistical modeling data set of the foregoing deterministic scenario sampling.

[0128]

[0129] Wherein, F represents the pollutant flux, p represents the upper layer thickness ratio, T represents the thickness of the aeration zone, K upper and K lower respectively represent the permeability coefficients of the upper layer and the lower layer; and respectively represent the upper layer and the lower layer saturated water content; and respectively represent the upper layer and the lower layer saturated water content; alpha upper and alpha lower respectively represent the distribution coefficients of the upper layer and the lower layer; n upper and n lower respectively represent the index parameters in the Van Genuchten model of the upper layer and the lower layer; Q represents the pollutant flow; C represents the pollutant concentration; D represents the dispersivity; k d represents the degradation coefficient.

[0130] Step S36, further select the random sampling number 35, generate random numbers of parameters again and constitute corresponding new more parameter combination scenarios, obtain corresponding groundwater environment related characteristic variable simulation values by using a numerical simulation system, integrate to constitute a model verification data set, obtain prediction error values and distribution characteristics by using the built regression statistical prediction model, and accordingly further determine the prediction error random variable distribution interval (-0.41, 0.40) of the groundwater environment related characteristic variable statistical prediction model corresponding to a given confidence.

[0131] Step S37, further select the random sampling number 1000, generate random numbers of parameters again and constitute corresponding new more parameter combination scenarios, obtain corresponding groundwater environment related characteristic variable prediction values Y S, the prediction error random variable distribution interval (-0.41, 0.40) corresponding to the given confidence level is determined based on the foregoing determined statistical prediction model of the groundwater environment related characteristic variable, and the upper and lower limits of the true value of the groundwater environment related characteristic variable, i.e., the prediction confidence interval (Y P1 ,Y P2 ) under the given confidence level is further determined.

[0132] Y P1 = Y S / (1+0.4)

[0133] Y P2 = Y S / (1-0.41)

[0134] The re-randomized parameter combination scenario and the characteristic variable prediction interval jointly constitute the statistical analysis reinforced modeling data set.

[0135] Step S38, a given environment diagnosis threshold Vc = -14.2 (equivalent to a stable pollutant flux threshold of 6.8 x 10 -5 ) is set.

[0136] Step S39, for the statistical modeling data set and the model verification data set, the output label value can be directly obtained by comparing the threshold value and the size of the environment related characteristic value; and for the statistical analysis reinforced modeling data set, the output label value is identified by comparing the relative size of the threshold value and the upper and lower limits of the environment related characteristic variable prediction interval; the parameter variable combination scenario corresponding to the statistical modeling data set and the statistical analysis reinforced modeling data set and the corresponding output label value jointly constitute the SVM machine learning training set, and the parameter variable combination scenario corresponding to the model verification data set and the corresponding output label value jointly constitute the SVM machine learning test set.

[0137] Step S40, the SVM machine learning modeling accuracy is set to 1.0, the preset training accuracy and test accuracy are both 1.0, the radial basis function (RBF) kernel is selected, the optimization related parameters including the regularization parameter C and the data mapping complexity parameter gamma are considered, the grid search method is used for modeling optimization, and overfitting is avoided.

[0138] Step S41, a preferred SVM environment diagnosis model for a single specific threshold Vc is further obtained, and the corresponding accuracy (including the training accuracy and the test accuracy) is less than 0.9.

[0139] Step S42, the number of random samples is further selected as 10000, the parameter variable random number generation is performed again to constitute a corresponding new more parameter combination scenario, the foregoing multiple regression statistical prediction model is applied to obtain the corresponding groundwater environment related characteristic variable prediction value Y S, the prediction error random variable distribution interval (-0.41, 0.40) corresponding to the given confidence level is determined based on the statistical prediction model of the groundwater environmental correlation characteristic variable determined as described above, and the upper and lower limits of the true value of the groundwater environmental correlation characteristic variable, i.e., the prediction confidence interval (Y P1 ,Y P2 ) under the given confidence level is further determined, wherein:

[0140] Y P1 = Y S / (1+0.4)

[0141] Y P2 = Y S / (1-0.41)

[0142] The recombination scenario of the random parameters and the prediction interval of the characteristic variable jointly constitute the statistical analysis and reinforced modeling data set.

[0143] Step S43, each step of steps S38 to S39 is repeated to achieve a preset training accuracy and a test accuracy of 1.0, and an optimal SVM environmental diagnosis model satisfying the modeling accuracy requirement for a single specific threshold Vc is obtained.

[0144] Step S44, according to the threshold variation interval, 12 typical threshold values V C-k (k = 1, 2, …, 12) = (-8, -9, …, -19) are selected, and for each V C-k , step S39 is repeated, and each V C-i corresponding statistical analysis and reinforced modeling data set is added with a new input parameter V C-k , which is further aggregated to become an SVM machine learning training set containing a threshold parameter variable, and a corresponding SVM machine learning test set is obtained.

[0145] Step 45, repeat step S40 to perform machine learning, and further obtain an optimal SVM environmental diagnosis model containing a threshold parameter variable.

[0146] Step S46, according to the investigation results of a specific groundwater environmental system in a study area, it is confirmed that the specific application scenario meets the requirements of the defined typical condition framework, and the specific parameter values of the corresponding input parameters of the corresponding SVM environmental diagnosis model are obtained, a given groundwater pollutant stable flux risk threshold is taken as an environmental diagnosis threshold (-14.2), and a precise prediction result is obtained by the SVM environmental diagnosis model obtained according to the single threshold universal SVM environmental diagnosis model determination method, or a precise prediction result is directly obtained by applying the established universal SVM environmental diagnosis model containing a threshold parameter variable, and the environmental diagnosis result is that the pollutant stable flux can cause groundwater pollution worse than the three-class water quality standard.

[0147] Please refer toFigure 3 , Figure 3 is a schematic block diagram of an environment diagnosis device based on numerical simulation and SVM efficient modeling according to an exemplary embodiment of the present application, referring to Figure 3 The environment diagnosis device 100 based on numerical simulation and SVM efficient modeling comprises:

[0148] A conceptual model construction module 101 is configured to select associated characteristic variables of a groundwater environment and construct a spatio-temporal evolution conceptual model of a corresponding groundwater environment system under a typical condition framework, and select and obtain a numerical simulation system according to a mathematical control equation and a numerical model.

[0149] A numerical simulation module 102 is configured to obtain n model input variables affecting the spatio-temporal evolution of the associated characteristic variables based on the spatio-temporal evolution conceptual model, the numerical model and the numerical simulation system, and the model input variables are specifically model input parameters; the variation interval of each model input variable is selected and then the model input variable interval set is obtained; and the corresponding numerical simulation values of the typical scenarios and the scenario sampling are obtained based on the numerical simulation system.

[0150] A statistical model construction module 103 is configured to obtain different typical scenarios composed of different model input parameter combinations of the groundwater environment system according to a given number of typical scenarios; obtain the corresponding numerical simulation values of the different typical scenarios based on the numerical simulation system, and obtain a statistical modeling data set by collecting the model input parameter combinations and the corresponding associated characteristic variable simulation values of all the different typical scenarios; and establish and optimize a multiple regression statistical prediction model of the associated characteristic variables and the model input variables according to the statistical modeling data set.

[0151] A statistical model prediction error analysis module 104 is configured to obtain a preset number of scenario samplings based on the n model input variables, obtain a first set of data-intensive scenario parameter combinations, obtain a first set of data-intensive scenario simulation values of the associated characteristic variables corresponding to each scenario through the numerical simulation system, and collect the first set of data-intensive scenario parameter combinations and the first set of data-intensive scenario simulation values as a model verification data set; obtain the prediction error values and the distribution characteristics of the associated characteristic variables based on the model verification data set through the multiple regression statistical prediction model; and determine the prediction error random variable distribution interval corresponding to a given prediction reliability according to the prediction error values and the distribution characteristics.

[0152] The machine learning reinforced dataset generation module 105 is configured to obtain a preset number of scenario samples based on the n model input variables, obtain a second group of data reinforced scenario parameter combinations, and obtain a second group of data reinforced scenario prediction values of the associated characteristic variables corresponding to each scenario through a multivariate regression statistical prediction model; determine a prediction confidence interval of the associated characteristic variables according to the second group of data reinforced scenario prediction values and a prediction error random variable distribution interval corresponding to a given preset confidence; and obtain a statistical analysis reinforced modeling dataset by merging the prediction confidence interval and the preset number of scenario parameter combinations based on the n model input variables.

[0153] If a single environment diagnosis threshold is given, the SVM classification machine learning first group of output label values, the second group of output label values and the third group of output label values are obtained in sequence according to the statistical modeling dataset, the statistical analysis reinforced modeling dataset and the model verification dataset; and the single threshold SVM classification machine learning dataset is obtained by merging the statistical modeling dataset, the statistical analysis reinforced modeling dataset and the model verification dataset, and the first group of output label values, the second group of output label values and the third group of output label values corresponding to the preset number of scenario parameter combinations, respectively.

[0154] If multiple environment diagnosis thresholds are given, different thresholds are selected according to the preset threshold value variation interval and the preset threshold value number, the environment diagnosis threshold is added as a model input variable of the SVM classification machine learning modeling based on the statistical modeling dataset and the statistical analysis reinforced modeling dataset, and based on the model verification dataset, and a threshold variable containing SVM classification machine learning dataset is obtained.

[0155] The universal SVM environment diagnosis model construction module 106 is configured to obtain the corresponding single threshold SVM environment diagnosis model and threshold variable containing SVM environment diagnosis model in sequence according to the single threshold SVM classification machine learning dataset and the threshold variable containing SVM classification machine learning dataset, according to and by applying the SVM classification model modeling principle and optimization method.

[0156] The application scenario environment diagnosis module 107 is configured to obtain the output label value corresponding to the threshold value by the single threshold SVM environment diagnosis model, and the output label value is the environment diagnosis prediction result, and / or obtain the environment diagnosis prediction result by the threshold variable containing SVM environment diagnosis model according to the actual conditions and collected data of the specific groundwater environment system and the preset environment diagnosis threshold, for the actual application scenario meeting the typical condition framework.

[0157] It should be noted that the device provided in the embodiment is applicable to scenarios including but not limited to the diagnosis of the groundwater environment system.

[0158] It can be understood that the device provided by the embodiment is based on the correlation characteristic variable of the groundwater environment, and establishes an environmental diagnosis model based on SVM for a complex groundwater environment system under a typical condition framework. The reliable diagnosis and prediction of the groundwater environment state, evolution process and related influence under different application scenarios can be realized under the premise of meeting the given condition framework, and the accuracy, efficiency and practicability of the environmental diagnosis and risk assessment of the complex groundwater system are improved.

[0159] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments can be included. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0160] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitation, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0161] It should be further noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0162] Each of the embodiments in the specification is described in a related manner, and the same and similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0163] The technical features of the above embodiments can be combined in any manner. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the specification.

[0164] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An environmental diagnosis method based on numerical simulation and SVM efficient modeling, characterized in that, include: Select relevant characteristic variables of the groundwater environment and construct a conceptual model of the spatiotemporal evolution of the corresponding groundwater environment system under typical condition framework; A numerical simulation system is selected and obtained based on the mathematical control equations and numerical model; Based on the spatiotemporal evolution conceptual model, the numerical model, and the numerical simulation system, n model input parameters that influence the spatiotemporal evolution of the associated characteristic variables are obtained, and the specific values ​​of the model input parameters are model input parameters; the variation range of each model input parameter is selected and then aggregated to obtain the model input parameter range set; Based on a given set of typical scenarios, different typical scenarios are obtained by combining different model input parameters of the groundwater environment system; based on the numerical simulation system, the simulated values ​​of the corresponding related characteristic variables of the different typical scenarios are obtained, and a statistical modeling dataset is obtained by collecting the model input parameter combinations of all the different typical scenarios and the simulated values ​​of their corresponding related characteristic variables; a multiple regression statistical prediction model is established and selected based on the statistical modeling dataset, and the related characteristic variables and the model input parameters are selected. Based on n model input parameters, a preset number of scenario samples are obtained to acquire a first set of data augmentation scenario parameter combinations. The numerical simulation system is used to obtain the first set of data augmentation scenario simulation values ​​for the associated feature variables corresponding to each scenario. The first set of data augmentation scenario parameter combinations and the first set of data augmentation scenario simulation values ​​are then used as the model validation dataset. Based on the model validation dataset, the prediction error values ​​and their distribution characteristics of the associated feature variables are obtained through the multivariate regression statistical prediction model. According to the prediction error values ​​and their distribution characteristics, the distribution interval of the prediction error random variable corresponding to a given preset confidence level is determined. Based on the n model input parameters, a preset number of scenario samples are obtained to obtain a second set of data-enhanced scenario parameter combinations. The second set of data-enhanced scenario prediction values ​​for each scenario-related feature variable are obtained through the multivariate regression statistical prediction model. The prediction confidence interval of the related feature variable is determined based on the second set of data-enhanced scenario prediction values ​​and the distribution interval of the prediction error random variable corresponding to the given preset confidence level. Based on the n model input parameters, a preset number of scenario parameter combinations and the prediction confidence intervals are obtained and then merged to obtain a statistical analysis enhanced modeling dataset. If a single environment diagnosis threshold is given, the first set of output label values, the second set of output label values, and the third set of output label values ​​for SVM classification machine learning are obtained sequentially based on the statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset. The statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset are combined with a preset number of scenario parameters, and then merged with the first group of output label values, the second group of output label values, and the third group of output label values ​​to obtain a single threshold SVM classification machine learning dataset. If multiple environmental diagnostic thresholds are given, different thresholds are selected according to the preset threshold value variation range and the preset threshold value number. Based on the statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset, environmental diagnostic thresholds are added as model input parameters for SVM classification machine learning modeling, thereby obtaining an SVM classification machine learning dataset with threshold parameters. Based on the single-threshold SVM classification machine learning dataset and the SVM classification machine learning dataset with threshold parameters, and by applying the SVM classification model modeling principle and optimization method, the corresponding single-threshold SVM environment diagnosis model and the SVM environment diagnosis model with threshold parameters are obtained sequentially. For practical application scenarios that conform to the typical condition framework, based on the actual conditions of the specific groundwater environment system and the collected data, the specific values ​​of the model parameters are obtained. Through the single-threshold SVM environmental diagnosis model, the output label value of the corresponding threshold is obtained. The output label value is the environmental diagnosis prediction result. And / or based on the actual conditions of the specific groundwater environment system, the collected data, and the preset environmental diagnosis threshold, the environmental diagnosis prediction result is obtained through the threshold parameter SVM environmental diagnosis model.

2. The diagnostic method according to claim 1, characterized in that, The statistical modeling dataset is obtained by combining the model input parameters of all the different typical scenarios and the set of simulated values ​​of their corresponding related feature variables, including: Select the variation level of each of the model input parameters; obtain different typical scenarios of groundwater environmental system composed of different combinations of model input parameters through orthogonal experimental methods or full combination methods of input parameters, and sample deterministic scenarios; Based on the numerical simulation system, the corresponding simulated values ​​of the associated feature variables for different typical scenarios are obtained; based on the different combinations of model input parameters for different typical scenarios and their corresponding simulated values, the statistical modeling dataset for the deterministic scenario is obtained; And based on the random values ​​of n model input parameters, a preset number of random scenario samples are obtained, that is, a certain number of different random scenario model input parameter combinations are obtained. The simulated values ​​of each random scenario associated feature variable are obtained through the numerical simulation system. The statistical modeling dataset of the random scenario samples is obtained by combining the random scenario model input parameter combinations and the set of simulated values ​​of random scenario associated feature variables. And by combining the statistical modeling dataset of the deterministic scenario sampling and the statistical modeling dataset of the random scenario sampling, a statistical modeling dataset integrating the deterministic scenario sampling and the random scenario sampling is obtained.

3. The diagnostic method according to claim 1, characterized in that, The SVM classification machine learning dataset includes an SVM classification machine learning training set and an SVM classification machine learning test set; the method further includes: The statistical modeling dataset and the statistical analysis reinforcement modeling dataset are combined with a preset number of scenario parameters, and then merged with the corresponding first set of output label values ​​and the second set of output label values ​​to obtain a single-threshold SVM classification machine learning training set; the model validation dataset is combined with a preset number of scenario parameters and the third set of output label values ​​to obtain a single-threshold SVM classification machine learning test set. If multiple environmental diagnostic thresholds are given, different thresholds are selected according to the preset threshold value variation range and the preset threshold value number. Based on the statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset, the environmental diagnostic thresholds are used as new model input parameters to obtain the training set and test set of SVM classification machine learning with threshold parameters.

4. The diagnostic method according to claim 1, characterized in that, Also includes: Given a preset accuracy value for the SVM environment diagnostic model, including a preset accuracy value for training and a preset accuracy value for testing, if the obtained SVM environment diagnostic model's modeling accuracy is lower than the preset value, then the preset number of different given typical scenarios and / or the preset number of different scenario samples are increased to optimize the SVM environment diagnostic model, thereby obtaining an SVM environment diagnostic model with corresponding accuracy; the corresponding accuracy SVM environment diagnostic model includes a single-threshold SVM environment diagnostic model and an SVM environment diagnostic model with threshold parameters.

5. The diagnostic method according to claim 1, characterized in that, The associated feature variable is at least one; the associated feature variable includes: groundwater pollutant concentration, pollutant flux at the cross section or interface of the saturated zone, location of the pollution plume migration and diffusion front, pollution plume stabilization time, stable pollution plume distribution area, distance of the stable pollution plume front from the pollution source, average concentration of the stable pollution plume, and pollution plume dissipation time. In addition, the concentrations of soil pollutants, the concentrations of gaseous pollutants in the vadose zone, the pollutant fluxes at the vadose zone cross section, the pollutant fluxes at the vadose zone-saturated zone interface, the pollutant fluxes at the soil-gas interface, the pollutant fluxes at the rock-gas interface, and the spatiotemporal distribution of NAPL pollutants.

6. The diagnostic method according to claim 1, characterized in that, The construction of a conceptual model for the spatiotemporal evolution of the corresponding groundwater environment system under typical conditions includes: The characteristics of the vadose zone are determined, including the medium type, lithological structure, and the heterogeneity and anisotropy of the medium. The characteristics of the saturated zone are determined, including the medium type, aquifer structure, seepage dimension, flow regime, and the heterogeneity and anisotropy of the medium. Determine the characteristics of pollution sources, including the phase state of pollutants of interest, the type of pollutants of interest, the spatial distribution characteristics of pollution sources, source intensity changes, and dynamic characteristics; Determine the boundary condition characteristics, which include hydrodynamic field boundary condition characteristics and hydrochemical field boundary condition characteristics; Determine the initial condition characteristics, which include the initial condition characteristics of the hydrodynamic field and the initial condition characteristics of the hydrochemical field.

7. The diagnostic method according to claim 1, characterized in that, Also includes: The associated feature variables and the model input parameters are processed using any or any combination of preset data processing methods, including: no transformation, proportional scaling, logarithmic transformation, positive and negative consistency, and normalization.

8. The diagnostic method according to claim 1, characterized in that, The process of obtaining a preset number of scenario samples based on n model input parameters includes: Select the number of random samples, determine or select the appropriate random distribution type and corresponding distribution parameters for each model input parameter, and complete the generation of random numbers for the model input parameters to form a new parameter combination scenario.

9. The diagnostic method according to claim 1, characterized in that, The process involves obtaining the prediction error values ​​and distribution characteristics of the associated feature variables through the multiple regression statistical prediction model; and determining the distribution interval of the prediction error random variable corresponding to a given preset confidence level based on the prediction error values ​​and distribution characteristics, including: Based on the model validation set data, a prediction error value dataset is obtained using the multivariate regression statistical prediction model. The prediction error value dataset includes multiple prediction errors. The mean and variance of the prediction error random variable are estimated, its probability density function is fitted and identified, and the sum of the cumulative probabilities on both sides is determined according to the preset confidence level, and the distribution interval of the prediction error random variable with the corresponding confidence level is obtained.

10. The diagnostic method according to claim 1, characterized in that, The SVM classification model modeling optimization method in the SVM classification model modeling principle and optimization method includes: SVM classification model optimization method and modeling parameter determination; The modeling parameters are determined, including Kernel function type selection and its associated parameter selection; Kernel function types include linear kernels, polynomial kernels, radial basis function kernels, and sigmoid kernels; The radial basis function kernel correlation parameters include the regularization parameter C and the data mapping complexity parameter gamma; The SVM classification modeling and optimization methods include grid search, particle swarm optimization, and a hybrid genetic particle swarm algorithm.

11. The diagnostic method according to claim 1, characterized in that, The process of obtaining the first set of output label values, the second set of output label values, and the third set of output label values ​​for SVM classification machine learning includes: If each of the aforementioned associated feature variables corresponds to a single environment diagnostic threshold, then based on the statistical modeling dataset and the model validation dataset, the simulated value of each of the aforementioned associated feature variables is compared with the corresponding single environment diagnostic threshold, and the first and third sets of output label values ​​for SVM classification machine learning are obtained sequentially. Based on the statistical analysis reinforcement modeling dataset, if the predicted confidence interval of each of the aforementioned associated feature variables does not include the corresponding single environment diagnostic threshold, then based on the relative size of the upper and lower limits of the predicted confidence interval and the corresponding single environment diagnostic threshold, the second set of output label values ​​for SVM classification machine learning is obtained.

12. An environmental diagnostic device based on numerical simulation and SVM efficient modeling, characterized in that, The device includes: The conceptual model construction module is used to select relevant characteristic variables of the groundwater environment and construct a conceptual model of the spatiotemporal evolution of the corresponding groundwater environment system under typical condition framework; and to select and obtain the numerical simulation system based on mathematical control equations and numerical models. The numerical simulation module is used to obtain n model input parameters that affect the spatiotemporal evolution of related characteristic variables based on the spatiotemporal evolution conceptual model, the numerical model, and the numerical simulation system. The specific values ​​of the model input parameters are model input parameters. The module selects the variation range of each model input parameter and then sets them to obtain the model input parameter range set. Based on the numerical simulation system, the module obtains the corresponding numerical simulation values ​​of typical scenarios and scenario sampling. The statistical model building module is used to obtain different typical scenarios composed of different combinations of model input parameters of the groundwater environment system based on a given number of typical scenarios; obtain the simulated values ​​of the corresponding related characteristic variables of the different typical scenarios based on the numerical simulation system; obtain a statistical modeling dataset by combining the model input parameters of all the different typical scenarios and the set of simulated values ​​of the corresponding related characteristic variables; and establish and select a multiple regression statistical prediction model of the related characteristic variables and the model input parameters based on the statistical modeling dataset. The statistical model prediction error analysis module is used to obtain a preset number of scenario samples based on n model input parameters, obtain a first set of data-enhanced scenario parameter combinations, obtain the first set of data-enhanced scenario simulation values ​​for the associated feature variables corresponding to each scenario through the numerical simulation system, and use the first set of data-enhanced scenario parameter combinations and the first set of data-enhanced scenario simulation values ​​as the model validation dataset; based on the model validation dataset, obtain the prediction error values ​​and their distribution characteristics of the associated feature variables through the multivariate regression statistical prediction model; and determine the distribution interval of the prediction error random variable corresponding to a given preset confidence level based on the prediction error values ​​and their distribution characteristics. The machine learning reinforcement dataset generation module is used to obtain a preset number of scenario samples based on n model input parameters, obtain a second set of data reinforcement scenario parameter combinations, and obtain the second set of data reinforcement scenario prediction values ​​for the associated feature variables corresponding to each scenario through the multivariate regression statistical prediction model; determine the prediction confidence interval of the associated feature variables based on the second set of data reinforcement scenario prediction values ​​and the random variable distribution interval of the prediction error corresponding to the given preset confidence level; and then merge the preset number of scenario parameter combinations obtained based on the n model input parameters and the prediction confidence interval to obtain a statistical analysis reinforcement modeling dataset; It is also used to, given a single environment diagnostic threshold, sequentially obtain the first set of output label values, the second set of output label values, and the third set of output label values ​​for SVM classification machine learning based on the statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset; combine the scenario parameters corresponding to the statistical modeling dataset, the statistical analysis reinforcement modeling dataset, and the model validation dataset with the corresponding first set of output label values, the second set of output label values, and the third set of output label values ​​to obtain a single threshold SVM classification machine learning dataset; It is also used to select different thresholds according to the preset threshold value variation range and the preset threshold value number if multiple environmental diagnostic thresholds are given, and to add environmental diagnostic thresholds as model input parameters for SVM classification machine learning modeling based on the statistical modeling dataset, the statistical analysis enhancement modeling dataset, and the model validation dataset, thereby obtaining an SVM classification machine learning dataset with threshold parameters. A general SVM environment diagnostic model construction module is used to obtain the corresponding single-threshold SVM environment diagnostic model and the SVM classification machine learning dataset with threshold parameters in sequence based on the single-threshold SVM classification machine learning dataset and the SVM classification machine learning dataset with threshold parameters, according to and applying the SVM classification model modeling principle and optimization method. The application scenario environmental diagnosis module, for actual application scenarios that conform to the typical condition framework, obtains the specific values ​​of the model parameters based on the actual conditions and collected data of the specific groundwater environment system, and obtains the corresponding threshold output label value through the single threshold SVM environmental diagnosis model. The output label value is the environmental diagnosis prediction result, and / or, based on the actual conditions and collected data of the specific groundwater environment system and the preset environmental diagnosis threshold, obtains the environmental diagnosis prediction result through the threshold parameter SVM environmental diagnosis model.

Images