A method for diagnosing and controlling groundwater and soil pollution risks on-site based on machine learning
Through machine learning-based methods, a risk classification and grading model for polluted sites is constructed, which solves the problem of inaccurate classification and grading in the existing technology, and achieves rapid and accurate pollution risk assessment, providing medium and long-term risk control for polluted sites. Reliable reference.
Patent Information
- Application Number
- CN202311322998.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-13
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2043-10-13
AI Technical Summary
The existing risk control technology for polluted sites cannot accurately classify and classify medium- and long-term site pollution risks, resulting in ineffective risk control.
Using a machine learning-based method, we build a site pollution risk classification and grading index system, combine BP neural network and SVM algorithm to establish a classification and grading model to achieve fast and accurate pollution risk classification and grading, providing a reliable reference for risk management and control.
It greatly reduces the number of risk assessment indicators, shortens the traditional risk assessment workflow, saves work costs and time, and realizes fast, intelligent and accurate pollution risk classification and grading, suitable for medium and long-term organic, heavy metal and compound pollution sites. Risk control.
Smart Images

Figure CN117556984B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of soil heavy metal pollution analysis, and in particular to a method for diagnosing and controlling groundwater and soil pollution risks on a site based on machine learning. Background Art
[0002] With the adjustment of urban layout and the upgrading of industrial structure, a large number of heavily polluting enterprises such as chemical, petroleum, plastic, coking, thermal power, steel, electroplating, nonferrous metal smelting have been relocated or closed, leaving behind a large number of polluted sites. Heavy metals or organic pollutants in polluted sites cause long-term pollution to soil and groundwater, posing a great threat to human health. The remediation of polluted sites has become a hot issue in the environmental protection industry. However, due to the deep pollution of the site and the special geographical environment, some polluted sites cannot be remediated to meet the standards in the short term. Risk control methods need to be used to block the migration or volatilization of pollutants in the site to avoid the spread of pollutants. The existing risk control technology for polluted sites cannot accurately classify and grade the medium- and long-term site pollution risks, which is not conducive to risk control. Summary of the invention
[0003] The purpose of the present invention is to provide a method for diagnosing and controlling the groundwater and soil pollution risks on a site based on machine learning. By establishing a site pollution risk classification, grading index system and index rules, clarifying the classification and grading standards, and constructing a classification and grading model through BP neural network and SVM algorithm, it can achieve fast and accurate site pollution risk classification and grading, and provide a reliable reference for risk control.
[0004] To achieve the above object, the present invention provides the following solutions:
[0005] A method for diagnosing and controlling groundwater and soil pollution risks on a site based on machine learning, the method comprising the following steps:
[0006] Construct a case library for site pollution risk classification and grading, including risky and non-risky cases;
[0007] According to the needs of site pollution risk assessment and site pollution risk classification and grading cases, a site pollution risk classification and grading indicator system is established;
[0008] Determine the site pollution risk classification and grading indicator rules respectively;
[0009] According to the site pollution risk classification and grading index rules, the site pollution risk classification and grading indicators are scored, and the 1-9 scale method in the hierarchical analysis method is used to test the consistency of the scores of each indicator;
[0010] By adjusting the various parameters of BP neural network and SVM, the best classification and grading model is optimized, risk classification is performed based on BP neural network, risk grading is performed based on SVM, and risk classification and grading prediction of target cases is completed in MATLAB, and risk management and control is performed based on the prediction results of risk classification and grading.
[0011] Furthermore, the risk management and control according to the prediction results of risk classification and grading specifically includes:
[0012] For in-production + high-risk sites, risk control is carried out from three aspects: reducing pollutant emissions at the source, upgrading and transforming production technology, and improving pollutant control capabilities; for in-production + low-risk sites, long-term monitoring is carried out, long-term monitoring points are set up in different functional areas of the factory, and samples are collected and analyzed for testing; for closed and relocated + high-risk sites with restoration conditions, risk control is carried out from two aspects: governance and restoration, and passive restoration-attenuation; for closed and relocated + high-risk sites that do not have restoration conditions, a combination of engineering control, long-term monitoring and institutional control is adopted; for closed and relocated + low-risk sites, a combination of long-term monitoring and access management is adopted.
[0013] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the method for diagnosing and controlling the risk of groundwater and soil pollution on a site based on machine learning provided by the present invention, in view of the technical requirements for risk control of polluted sites, combines risk cases and risk-free cases, constructs a site pollution risk classification, grading index system and index rules, clarifies the classification and grading standards, and integrates the BP neural network and SVM algorithm to construct a classification and grading model, which greatly reduces the number of risk assessment indicators, shortens the traditional risk assessment workflow (such as eliminating the preliminary investigation and detailed investigation links), reduces the workload of traditional risk assessment, saves work cost investment, shortens decision-making time, and realizes fast, intelligent and accurate classification and grading of site pollution risks, providing a reliable reference for risk control. The present invention is simple to operate, simple and practical, technically reliable, and easy to implement in engineering. It can be used for medium- and long-term risk control of organic polluted sites, heavy metal polluted sites and complex polluted sites, and has important environmental and social significance. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0015] Figure 1 This is a flow chart of the method for diagnosing and controlling groundwater and soil pollution risks on-site based on machine learning of the present invention;
[0016] Figure 2 Schematic diagram of the effect of training function on the prediction accuracy of BP neural network;
[0017] Figure 3 Schematic diagram of the effect of learning rate on the prediction accuracy of BP neural network;
[0018] Figure 4 Schematic diagram of the effect of the number of neurons on the prediction accuracy of the BP neural network;
[0019] Figure 5 This is a schematic diagram of risk classification results supported by the BP neural network according to an embodiment of the present invention;
[0020] Figure 6 This is a schematic diagram of risk classification results supported by a support vector machine according to an embodiment of the present invention;
[0021] Figure 7 This is a schematic diagram of risk grading results supported by the BP neural network according to an embodiment of the present invention;
[0022] Figure 8 This is a schematic diagram of risk grading results supported by a support vector machine according to an embodiment of the present invention. DETAILED DESCRIPTION
[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] The purpose of the present invention is to provide a method for diagnosing and controlling the groundwater and soil pollution risks on a site based on machine learning. By establishing a site pollution risk classification, grading index system and index rules, clarifying the classification and grading standards, and constructing a classification and grading model through BP neural network and SVM algorithm, it can achieve fast and accurate site pollution risk classification and grading, and provide a reliable reference for risk control.
[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0026] like Figure 1 As shown, the method for diagnosing and controlling groundwater and soil pollution risks based on machine learning provided by the present invention comprises the following steps:
[0027] Collect and organize site pollution risk classification and grading cases, and build a site pollution risk classification and grading case library, including risky cases and non-risky cases;
[0028] According to the needs of site pollution risk assessment and site pollution risk classification and grading cases, a site pollution risk classification and grading indicator system is established;
[0029] Determine the site pollution risk classification and grading indicator rules respectively;
[0030] According to the site pollution risk classification and grading index rules, the site pollution risk classification and grading indicators are scored, and the 1-9 scale method in the hierarchical analysis method is used to test the consistency of the scores of each indicator;
[0031] By adjusting the various parameters of BP neural network and SVM, the best classification and grading model is optimized, risk classification is performed based on BP neural network, risk grading is performed based on SVM, and risk classification and grading prediction of target cases is completed in MATLAB, and risk management and control is performed based on the prediction results of risk classification and grading.
[0032] For example, the collection and organization of site pollution risk classification and grading cases specifically include:
[0033] With the help of the national soil environment information platform, 200 pilot demonstration projects, literature and other materials, we collected and sorted out site soil pollution risk classification and grading cases (including preliminary investigation reports, detailed investigation reports, and risk assessment reports), and obtained 274 risky cases and 55 risk-free cases, involving chemical raw materials and chemical products manufacturing industry, non-ferrous metal smelting and rolling processing industry, ferrous metal smelting and rolling processing industry, metal products industry, pharmaceutical manufacturing industry, petroleum, coal and other fuel processing industries, etc.
[0034] Refer to the information survey form for closed and relocated enterprise plots in Appendix 3 of the "Technical Regulations on Information Collection on Land Use by Key Industry Enterprises (Trial)" to determine the case framework for classification and grading of site pollution risks; organize the collected cases according to the case framework; finally, summarize all cases to form a case library.
[0035] The basic principles of the big data algorithm used in the present invention are as follows:
[0036] 1) BP Neural Network
[0037] BP is the abbreviation of Back Propagation. The principle is to use the error after the output to estimate the direct predecessor layer of the output layer, and then use this error to estimate the error of the previous layer. By backpropagating layer by layer, the error estimates of all other layers can be obtained.
[0038] 2) Support Vector Machine
[0039] SVM is a generalized linear classifier that performs binary classification of data in a supervised learning manner. Its decision boundary is the maximum margin hyperplane solved for the learning sample. SVM uses different inner product functions (kernel functions) to lead to different algorithms. When using SVM to train data, the generalization ability of the model should be as good as possible under the condition that the accuracy meets certain requirements to prevent overfitting during the training process. Therefore, during training, it is necessary to continuously adjust the gamma and c values and continuously cross-validate the data to find the appropriate gamma and c values. The larger the gamma value, the higher the dimension of the mapping, the better the training result, but the easier it is to cause overfitting, that is, low generalization ability; the smaller the gamma value, the lower the accuracy of the training result and the poor fitting degree. On the other hand, the larger the c value, the stricter the classification and the higher the accuracy, but the corresponding generalization ability is reduced; the smaller the c value, the greater the error tolerance, but the lower the accuracy.
[0040] Based on the case library, the method for diagnosing and controlling the risk of groundwater and soil pollution on site based on machine learning specifically includes:
[0041] 1) Identify the source of site pollution risk control cases
[0042] 65 risky cases and 55 non-risk cases were selected from the case library as site pollution risk classification and grading source cases, involving 53 chemical raw materials and chemical products manufacturing industries, 4 non-metallic mineral products industries, 8 non-ferrous metal smelting and rolling processing industries, 12 ferrous metal smelting and rolling processing industries, 20 metal products industries, 3 petroleum, coal and other fuel processing industries and 20 other industries.
[0043] 2) Establish a site pollution risk classification and grading indicator system
[0044] Refer to the information survey form for closed and relocated enterprise plots in Appendix 3 of the "Technical Regulations on Land Use Information Collection for Key Industry Enterprises (Trial)" (Environmental Protection Office Soil
[2017] No. 67), and establish a site pollution risk classification and grading indicator system.
[0045] 3) Determine site pollution risk classification and grading index rules
[0046] Based on the "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard (Trial)" (GB 36600-2018), and referring to the Class III water standards in the "Land Use Status Classification" (GB / T 21010-2017) and the "Groundwater Quality Standard" (DZ / T 0290-2015), the site pollution risk classification and grading indicator rules are determined respectively.
[0047] 4) Index scoring consistency test
[0048] The 1-9 scale method in the hierarchical analysis method (as shown in Table 1) was used to test the consistency of the scores of each indicator. If the test was passed, the score was considered acceptable.
[0049] Table 1 Factor relative importance scale and its corresponding meaning
[0050]
[0051]
[0052] In MATLAB software, the consistency index is calculated according to the maximum value of the eigenvalue of each characteristic index information and the number of characteristic index information, as shown in formula (1).
[0053]
[0054] In the formula, I C is the consistency index, λ max is the maximum eigenvalue of the judgment matrix, and n is the number of characteristic indicator information.
[0055] According to the consistency index and the preset random consistency index, the consistency ratio is calculated as shown in formula (2):
[0056]
[0057] In the formula, R C is the consistency ratio, I R is the preset random consistency index (as shown in Table 2).
[0058] Table 2 Random consistency index scores
[0059] n 1 2 3 4 5 6 ... 22 23 24 25 <![CDATA[I R ]]> 0 0 0.5303 0.8845 1.1143 1.2525 ... 1.6401 1.6482 1.6511 1.6550
[0060] 5) Big data algorithm optimization
[0061] By adjusting the parameters of BP neural network and SVM, the best classification and grading model is optimized, and the classification and grading prediction of the target case is completed in MATLAB. 2 ), mean relative error (MRE) and root mean square error (RMSE) are used as evaluation criteria to evaluate the performance of classification and grading models, as shown in formulas (3) and (4).
[0062]
[0063]
[0064] Where n is the number of test samples, A(i) is the measured value of the sample, and P(i) is the predicted value of the sample.
[0065] Among them, the specific process of BP neural network and SVM parameter optimization is as follows:
[0066] (1) BP neural network optimization
[0067] The selection of appropriate BP neural network parameters has an important impact on the accuracy of risk assessment results. The parameters that affect the BP neural network include training function, learning rate, and the number of neurons in the hidden layer. At present, the commonly used training functions include gradient descent method (traingd), gradient descent method with momentum (traindm), adaptive learning rate gradient descent method (traingda), adaptive learning rate momentum gradient descent method (traindx), conjugate gradient method (traincg), quasi-Newton algorithm (trainbfg), and Levenberg-Marquardt algorithm (trainlm). In the present invention, the influence of the training function on the prediction accuracy of the BP neural network is shown in Figure 2 trainlm has the smallest MSE (0.00043) and no effect of the maximum number of errors (such as Figure 2 ). Therefore, trainlm is selected as the training function of the BP neural network.
[0068] The effect of learning rate on the prediction accuracy of BP neural network can be seen in Figure 3 When the number of training times is close to the same, the MSE of the model is 0.000289 ( Figure 3 ). Therefore, the learning rate of the model is selected as 0.1.
[0069] The effect of the number of neurons on the prediction accuracy of BP neural network is shown in Figure 4 When the number of neurons in the hidden layer is 6, the model has the smallest mean square error (0.000314) ( Figure 4 ).
[0070] In summary, by adjusting the model hyperparameters, when trainlm is the training function, the number of hidden layer neurons is 6, and the learning rate is 0.1, the model has the best performance, good fitting effect, and fast convergence speed.
[0071] (2) SVM parameter optimization
[0072] As shown in Table 3, model 2 has the highest accuracy rate of 73.3% compared with the other two models. It can be seen that the model with linear kernel function, 1 cost and 0.08 gamma has the best effect.
[0073] Table 3 Comparison of different parameters of support vector machine
[0074] Model Kernel Function Optimal parameters Correct rate (%) Model 1 radial The cost is 5 and the gamma is 0.1 66.7 Model 2 linear The cost is 1 and the gamma is 0.08 73.3 Model 3 polynomial cost is 0 and gamma is 1 66.7
[0075] The present invention relates to site pollution risk classification and site pollution risk grading, using different indicator systems and models, specifically including:
[0076] 1. Classification of site contamination risks
[0077] (1) Construction of indicator system
[0078] According to the needs of site pollution risk assessment and the collected information on risky and non-risky cases, a site pollution risk classification index system is constructed to determine whether there is a risk of site pollution (Table 4). s Indicates the total number of times soil pollution exceeds the standard.
[0079] Table 4 Risk classification indicator system
[0080]
[0081]
[0082]
[0083] (2) Establishment of calculation rules
[0084] 1) Wetness and dryness index rules
[0085] The dryness and wetness index rule is shown in formula (5):
[0086]
[0087] Where DWI represents the wetness / dryness index (dimensionless); P represents the multi-year average precipitation (mm); E0 represents the multi-year average evaporation (mm). The number of years taken when calculating precipitation and evaporation can be adjusted according to actual needs.
[0088] 2) Characteristic pollutant rules
[0089] SVOCs, VOCs, heavy metals, inorganic substances, and petroleum hydrocarbons in soil characteristic pollutants and groundwater characteristic pollutants mainly focus on 85 pollutants in the "Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard (Trial)" (GB 36600-2018). The soil pollutant multiple calculation is shown in formula (6).
[0090]
[0091] Where n is the number of pollutant types in the soil whose concentration exceeds the screening value; C si RS is the concentration of the i-th pollutant that exceeds the screening value, taking the highest value (mg / kg) among all soil sample test data of the site; siis the screening value of the ith pollutant (mg / kg), referring to the guidance value for soil pollution risk screening for construction land.
[0092] The calculation of groundwater pollutant multiples is shown in formula (7):
[0093]
[0094] Where n is the number of pollutant types in groundwater whose concentration exceeds the screening value; C gwi RS is the concentration of the i-th pollutant whose concentration exceeds the screening value, taking the highest value (mg / L) among all groundwater sample test data of the site; gwi is the limit value of the ith pollutant (mg / L), referring to the Class III water quality standard in the Groundwater Quality Standard (DZ / T 0290-2015).
[0095] 3) Land use planning
[0096] Based on the “Land Use Status Classification” (GB / T 21010-2017) and the “Soil Environmental Quality Construction Land Soil Pollution Risk Control Standard (Trial)” (GB 36600-2018), land use planning rules were established according to the site land use planning level determined by the future planning of the site (Table 5).
[0097] Table 5 Land use planning rules
[0098]
[0099]
[0100] 4) City level
[0101] The city grades are determined according to the Notice on Adjusting the Standards for Classification of Urban Sizes (Guofa
[2014] No. 51), where Class A cities include Beijing, Shanghai, Guangzhou and Shenzhen; Class B cities include Chongqing, Tianjin, Hangzhou, Nanjing and other 21 cities; Class C cities include Taiyuan, Hohhot, Guiyang and Lanzhou and other 41 cities; Class D cities include Yinchuan, Xining, Lhasa and Sanya and other 62 cities; Class E cities include Zhaoqing, Qingyuan, Shanwei and Meizhou and other 190 cities; and Class F cities are other cities (Table 6).
[0102] Table 6 City level scoring rules
[0103] City level Points A 6 B.C 4 D, E, F 1
[0104] 5) The influence of the vadose zone and the influence of the phreatic aquifer
[0105] Combined with the actual situation of the vadose zone and the unconfined aquifer in the source case and the target case, the lithology classification rule with the largest permeability coefficient of the vadose zone and the most important lithology and classification rule of the unconfined aquifer are determined. The corresponding relationship between the lithology with the largest permeability coefficient of the vadose zone and its assignment is shown in Tables 7 and 8.
[0106] Table 7 Rules for assigning permeability coefficient of vadose zone
[0107]
[0108]
[0109] Table 8 Aquifer medium rules
[0110] Lithology type Permeability coefficient Points Clay or Uncertain <![CDATA[1×10 -11 -4.7×10 -9 ]]> 2 Silt <![CDATA[6×10 -7 -6×10 -6 ]]> 2 Pink sand <![CDATA[6×10 -6 -1.2×10 -5 ]]> 2 Fine sand <![CDATA[1.2×10 -5 -6×10 -5 ]]> 4 medium sand <![CDATA[6×10 -3 -2.4×10 -2 ]]> 4 coarse sand <![CDATA[2.4×10 -2 -6×10 -2 ]]> 4 gravel <![CDATA[6×10 -2 -1.8×10 -1 ]]> 6
[0111] (3) Indicator scoring
[0112] According to the indicator rules, the site pollution risk classification indicators are assigned points (Table 9).
[0113] Table 9 Site pollution risk classification index scores
[0114]
[0115]
[0116]
[0117] (4) Score consistency test
[0118] According to the risk classification index, the consistency test matrix is assigned, as shown in formula (8), and the consistency test of the judgment matrix is performed to determine the consistency index (I C ) is 11.4517, random consistency index (I R ) is 1.5079, the consistency ratio (R C ) is 0.0299<0.1, and the risk classification index scoring is considered acceptable.
[0119]
[0120] (5) Classification model screening
[0121] Both BP neural network and support vector machine can perform classification prediction well, as shown in R 2 are 0.9974 and 0.9103 respectively; in addition, the MRE (0.12%) and RMSE (0.0376) of the BP neural network are both smaller than the MRE (49.1%) and RMSE (13.7232) of the support vector machine, as shown in Table 10. Obviously, the BP neural network has the best prediction performance, so the BP neural network is selected for prediction in risk classification ( Figure 5 and Figure 6 ).
[0122] Table 10 Comparison of risk judgment error results
[0123] Evaluation indicators BP Neural Network Support Vector Machine <![CDATA[R 2 ]]> 0.9974 0.9103 MRE 0.12% 49.1% RMSE 0.0376 13.7232
[0124] II. Site Pollution Risk Classification
[0125] (1) Construction of indicator system
[0126] With reference to the Technical Regulations on Risk Screening and Risk Grading of Enterprise Land, a soil indicator system and a groundwater indicator system related to risk grading were established (Tables 11 and 12).
[0127] Table 11 Soil indicator system related to risk classification
[0128]
[0129]
[0130]
[0131] Table 12 Risk classification related groundwater indicator system
[0132]
[0133]
[0134] (2) Indicator scoring
[0135] With reference to the Technical Regulations on Risk Screening and Risk Grading of Enterprise Land, the relevant soil and groundwater indicators for site pollution risk grading are assigned points (Tables 13 and 14).
[0136] Table 13 Risk classification soil index scoring
[0137]
[0138]
[0139] Table 14 Risk classification of groundwater indicators
[0140]
[0141]
[0142]
[0143] (3) Score consistency test
[0144] According to the soil index scoring test matrix for site pollution risk classification, as shown in formula (9), when performing the consistency test of the risk classification matrix, I C is 11.5279, I R 1.5079, R C It is 0.035<0.1, and the indicator score in the risk grading stage is considered acceptable.
[0145]
[0146] (4) Calculation and classification of site risk scores
[0147] According to the Technical Regulations on Risk Screening and Risk Grading of Enterprise Sites and the collected basic information on contaminated sites, the site risk score is calculated (Formula 10), and the sites are divided into low risk, medium risk and high risk according to the score (Table 15).
[0148]
[0149] Where: S is the total score of the plot risk screening, S s Score the soil of the plot, S gw Score groundwater for your plot.
[0150] Table 15 Risk level classification rules
[0151] Total score of land risk screening Land risk classification S≥70 points High-risk plots 50≤S<70 points Medium risk plots S<50 points Low risk plots
[0152] (5) Hierarchical model screening
[0153] like Figure 7 and Figure 8 As shown in Figure 2, both BP neural network and support vector machine can perform classification prediction well, as shown in R 2 are 0.9700 and 0.9977 respectively; in addition, although the MRE (0.50%) and MSE (0.4341) of the support vector machine are greater than the MRE (0.05%) and RMSE (0.0358) of the BP neural network, the error values are also within the allowable range (Table 16). 2 , MRE and RMSE, the prediction performance of support vector machine is better, so support vector machine is selected for prediction in risk stratification.
[0154] Table 16 Comparison of risk classification error results
[0155] Evaluation indicators BP Neural Network Support Vector Machine <![CDATA[R 2 ]]> 0.970 0.997 MRE 0.05% 0.50% RMSE 0.0358 0.4341
[0156] Furthermore, the risk management and control according to the prediction results of risk classification and grading specifically includes:
[0157] (1) Scenario 1: In production + high-risk sites. Risk control is mainly carried out from three aspects: source pollutant emission reduction, production technology upgrade and transformation, and pollutant control capacity improvement. Among them, source pollutant emission reduction mainly adopts environmentally friendly production raw materials and auxiliary materials replacement to reduce pollutant generation from the source of the production line; production technology transformation mainly includes production technology upgrades, outdated production capacity elimination, production equipment transformation and other measures to improve production efficiency and reduce the amount of source raw materials, thereby reducing the generation and emission of pollutants at the back end; pollutant control capacity improvement includes upgrading and transformation of enterprise pollution control facilities, remediation of solid waste storage sites, etc., to reduce pollutant emission concentrations and reduce environmental disturbances.
[0158] (2) Scenario 2: In production + low-risk sites. Long-term monitoring measures are mainly taken to prevent risks from occurring. Long-term monitoring points are set up in different functional areas of the factory and samples are collected and analyzed and tested to achieve the purpose of controlling potential exposure risks. When setting up long-term monitoring points, it is necessary to increase the number of points in production workshops, solid waste and raw material storage sites, and sewage pipeline layout areas, so as to timely evaluate and discover environmental risks.
[0159] (3) Scenario 3: Closure and relocation + high-risk sites with remediation conditions. Risk management is mainly carried out from two aspects: governance and remediation and passive remediation-attenuation. Among them, governance and remediation is to use a series of engineering technologies to carry out site pollution control. Specific engineering technologies include soil import, soil replacement, soil turning, thermal analysis, solidification stabilization, harmless landfill, soil leaching, redox, plant remediation, etc. Passive remediation-attenuation uses engineering technology and natural conditions on the site to rely on naturally occurring biological, chemical or physical processes to fix, reduce or remove pollutants. Specific technologies include infiltration reaction grids, enhanced monitoring of natural attenuation, etc., which are mainly suitable for sites with slow migration of organic pollutants and relatively low governance and remediation budgets.
[0160] (4) Scenario 4: Closure and relocation of high-risk sites that do not meet the conditions for restoration. The main method is to combine engineering control, long-term monitoring and institutional control to prevent the spread of risks. Among them, engineering control mainly uses various barrier projects to limit the migration of pollutants and cut off the exposure path between the pollution source and the receptor, so as to reduce the risk of pollution exposure and protect the safety of the receptor. Specific engineering control technologies include mud walls, grouting walls, sheet pile walls, in-situ soil mixing, geomembranes, linings, etc. The main links of long-term monitoring are consistent with the long-term monitoring of low-risk sites in production. The main difference is that when setting up monitoring points for closure and relocation of high-risk sites, it is necessary to combine the site survey results and increase the number of points in areas with higher pollutant concentrations. When taking long-term monitoring measures, the institutional control carried out simultaneously includes government control, ownership control, mandatory and licensing systems, information disclosure (such as setting up signboards), etc., to ensure the safety of the soil environment and human living environment around the site.
[0161] (5) Scenario 5: Closure and relocation + low-risk sites. The main method is to combine long-term monitoring and access management to prevent environmental risks. Among them, when setting up long-term monitoring points, monitoring points need to be set up in the former production area, raw material storage area and sensitive target areas around the site; when the use is changed, the local ecological environment management department needs to issue a risk-free certificate for the site before entering the land use approval procedure.
[0162] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for diagnosing and controlling groundwater and soil pollution risks on a site based on machine learning, characterized in that: The following steps are involved: Construct a case library for site pollution risk classification and grading, including risky and non-risky cases; According to the needs of site pollution risk assessment and site pollution risk classification and grading cases, a site pollution risk classification and grading indicator system is established respectively, wherein the site pollution risk grading indicator system includes a soil indicator system related to risk grading and a groundwater indicator system related to risk grading; Determine the site pollution risk classification and grading index rules respectively; According to the site pollution risk classification and grading index rules, the site pollution risk classification and grading indicators are scored, and the 1-9 scale method in the hierarchical analysis method is used to test the consistency of the scores of each indicator; By adjusting the parameters of BP neural network and SVM, the best classification and grading model is optimized, risk classification is performed based on BP neural network, risk grading is performed based on SVM, and risk classification and grading prediction of target cases is completed in MATLAB, and risk management and control is performed based on the prediction results of risk classification and grading; The site pollution risk classification index rules include: dryness and wetness index rules, characteristic pollutant rules, land use planning rules, urban grade rules, influence of vadose zone and influence of phreatic aquifer rules; Based on the first formula, the dryness and wetness index rule is established, where the first formula is DWI stands for dryness-wetness index. represents the average precipitation over many years, It represents the average evaporation over many years; Based on the second formula, characteristic pollutant rules are established, where the second formula is n is the number of pollutant types in the soil whose concentration exceeds the screening value, C si is the concentration of the i-th pollutant whose concentration exceeds the screening value, RS si is the screening value of the i-th pollutant; Based on the third formula, the groundwater pollutant multiple rule is established, where the third formula is m is the number of pollutant types in groundwater whose concentration exceeds the screening value, C gwi is the concentration of the i-th pollutant whose concentration exceeds the screening value, RS gwi is the limit value of the ith pollutant; The risk management and control according to the prediction results of risk classification and grading specifically includes: According to the scores of each indicator of site pollution risk classification and grading, the site's plot soil score and plot groundwater score are calculated separately; The total risk screening score of the plot is calculated based on the soil score and groundwater score of the plot; Based on the determined total risk screening score of the plot, the risk level of the site plot is determined.
2. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1 is characterized in that: The site pollution risk classification and grading cases in the site pollution risk classification and grading case library involve chemical raw materials and chemical products manufacturing industry, non-ferrous metal smelting and rolling processing industry, ferrous metal smelting and rolling processing industry, metal products industry, pharmaceutical manufacturing industry, petroleum, coal and other fuel processing industry.
3. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1, characterized in that: The site pollution risk classification index system includes first-, second- and third-level indicators, among which the first-level indicators include site overview, pollution sources, migration pathways and sensitive receptors; The secondary indicators under the site overview include the industry, land use planning, whether the site is located in an industrial park or agglomeration area, and the city level. Among them, the tertiary indicators under the industry include metal products industry, chemical raw materials and chemical products manufacturing industry, non-ferrous metal smelting and rolling industry, ferrous metal smelting and rolling industry, and oil and natural gas extraction industry; the tertiary indicators under the land use planning include Class I land and Class II land; the tertiary indicators under whether the site is located in an industrial park or agglomeration area include yes and no; the tertiary indicators under the city level include ≤0.5, 0.5-1, and ≥1; The secondary indicators under the pollution sources include soil pollutants and groundwater pollutants. The tertiary indicators under soil pollutants include Es≥100, 50≤Es<100, 10≤Es<50, 0≤Es<10; the tertiary indicators under groundwater pollutants include Es≥100, 50≤Es<100, 10≤Es<50, where Es represents the total multiples of soil pollution exceeding the standard. The secondary indicators under the migration pathway include the dryness and wetness index, and the tertiary indicators under the dryness and wetness index include ≥1, 0.5-1, ≤0.5; The secondary indicators under sensitive receptors include the possibility of population entry, population within 500m around the plot, and the distance of the plot from the nearest drinking water source. The tertiary indicators under the possibility of population entry include possible and impossible. The third-level indicators under the population within 500m around the plot include >500 and <500. The third-level indicators under the distance of the plot from the nearest drinking water source include >500m and <500m.
4. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1, characterized in that: The site pollution risk grading index system includes a soil index system related to risk grading and a groundwater index system related to risk grading. The soil index system related to risk grading and the groundwater index system related to risk grading both include first-, second- and third-level indicators, among which the first-level indicators include pollution sources, pollutant migration pathways and sensitive receptors.
5. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1, characterized in that: The optimal classification and grading model is optimized by adjusting various parameters of the BP neural network and the SVM, including: Optimize the parameters of the BP neural network to obtain the best classification model, wherein the parameters of the BP neural network are optimized as follows: Select the Levenberg-Marquardt algorithm as the training function of the BP neural network; The learning rate of the selected model is 0.1; The number of neurons in the hidden layer is selected as 6.
6. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1, characterized in that: The optimal classification and grading model is optimized by adjusting various parameters of the BP neural network and the SVM, including: The SVM parameters were optimized to obtain the best classification model, wherein the optimized SVM parameters included: the kernel function was linear, the cost was 1, and the gamma was 0.
08.
7. The method for diagnosing and controlling groundwater and soil pollution risks based on machine learning according to claim 1, characterized in that: The risk management and control according to the prediction results of risk classification and grading specifically includes: For production and high-risk sites, reduce pollutant emissions at the source, upgrade production technology, and improve pollutant control capabilities Carry out risk management in three aspects: Carry out long-term monitoring for in-production and low-risk sites, set up long-term monitoring points in different functional areas of the plant, and collect samples and conduct analysis and testing; For closed and relocated high-risk sites that meet restoration conditions, risk management and control will be carried out from two aspects: governance restoration and passive restoration-attenuation; For closed and relocated high-risk sites that are not suitable for restoration, a combination of engineering control, long-term monitoring and institutional control will be adopted; For closure and relocation + low-risk sites, a combination of long-term monitoring and access management is adopted.
